net/packet/af_packet.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              PACKET - implements raw packet sockets.
   7  *
   8  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  11  *
  12  * Fixes:
  13  *              Alan Cox        :       verify_area() now used correctly
  14  *              Alan Cox        :       new skbuff lists, look ma no backlogs!
  15  *              Alan Cox        :       tidied skbuff lists.
  16  *              Alan Cox        :       Now uses generic datagram routines I
  17  *                                      added. Also fixed the peek/read crash
  18  *                                      from all old Linux datagram code.
  19  *              Alan Cox        :       Uses the improved datagram code.
  20  *              Alan Cox        :       Added NULL's for socket options.
  21  *              Alan Cox        :       Re-commented the code.
  22  *              Alan Cox        :       Use new kernel side addressing
  23  *              Rob Janssen     :       Correct MTU usage.
  24  *              Dave Platt      :       Counter leaks caused by incorrect
  25  *                                      interrupt locking and some slightly
  26  *                                      dubious gcc output. Can you read
  27  *                                      compiler: it said _VOLATILE_
  28  *      Richard Kooijman        :       Timestamp fixes.
  29  *              Alan Cox        :       New buffers. Use sk->mac.raw.
  30  *              Alan Cox        :       sendmsg/recvmsg support.
  31  *              Alan Cox        :       Protocol setting support
  32  *      Alexey Kuznetsov        :       Untied from IPv4 stack.
  33  *      Cyrus Durgin            :       Fixed kerneld for kmod.
  34  *      Michal Ostrowski        :       Module initialization cleanup.
  35  *         Ulises Alonso        :       Frame number limit removal and
  36  *                                      packet_set_ring memory leak.
  37  *              Eric Biederman  :       Allow for > 8 byte hardware addresses.
  38  *                                      The convention is that longer addresses
  39  *                                      will simply extend the hardware address
  40  *                                      byte arrays at the end of sockaddr_ll
  41  *                                      and packet_mreq.
  42  *              Johann Baudy    :       Added TX RING.
  43  *              Chetan Loke     :       Implemented TPACKET_V3 block abstraction
  44  *                                      layer.
  45  *                                      Copyright (C) 2011, <lokec@ccs.neu.edu>
  46  *
  47  *
  48  *              This program is free software; you can redistribute it and/or
  49  *              modify it under the terms of the GNU General Public License
  50  *              as published by the Free Software Foundation; either version
  51  *              2 of the License, or (at your option) any later version.
  52  *
  53  */
  54
  55 #include <linux/types.h>
  56 #include <linux/mm.h>
  57 #include <linux/capability.h>
  58 #include <linux/fcntl.h>
  59 #include <linux/socket.h>
  60 #include <linux/in.h>
  61 #include <linux/inet.h>
  62 #include <linux/netdevice.h>
  63 #include <linux/if_packet.h>
  64 #include <linux/wireless.h>
  65 #include <linux/kernel.h>
  66 #include <linux/kmod.h>
  67 #include <linux/slab.h>
  68 #include <linux/vmalloc.h>
  69 #include <net/net_namespace.h>
  70 #include <net/ip.h>
  71 #include <net/protocol.h>
  72 #include <linux/skbuff.h>
  73 #include <net/sock.h>
  74 #include <linux/errno.h>
  75 #include <linux/timer.h>
  76 #include <asm/uaccess.h>
  77 #include <asm/ioctls.h>
  78 #include <asm/page.h>
  79 #include <asm/cacheflush.h>
  80 #include <asm/io.h>
  81 #include <linux/proc_fs.h>
  82 #include <linux/seq_file.h>
  83 #include <linux/poll.h>
  84 #include <linux/module.h>
  85 #include <linux/init.h>
  86 #include <linux/mutex.h>
  87 #include <linux/if_vlan.h>
  88 #include <linux/virtio_net.h>
  89 #include <linux/errqueue.h>
  90 #include <linux/net_tstamp.h>
  91 #include <linux/reciprocal_div.h>
  92 #include <linux/percpu.h>
  93 #ifdef CONFIG_INET
  94 #include <net/inet_common.h>
  95 #endif
  96
  97 #include "internal.h"
  98
  99 /*
 100    Assumptions:
 101    - if device has no dev->hard_header routine, it adds and removes ll header
 102      inside itself. In this case ll header is invisible outside of device,
 103      but higher levels still should reserve dev->hard_header_len.
 104      Some devices are enough clever to reallocate skb, when header
 105      will not fit to reserved space (tunnel), another ones are silly
 106      (PPP).
 107    - packet socket receives packets with pulled ll header,
 108      so that SOCK_RAW should push it back.
 109
 110 On receive:
 111 -----------
 112
 113 Incoming, dev->hard_header!=NULL
 114    mac_header -> ll header
 115    data       -> data
 116
 117 Outgoing, dev->hard_header!=NULL
 118    mac_header -> ll header
 119    data       -> ll header
 120
 121 Incoming, dev->hard_header==NULL
 122    mac_header -> UNKNOWN position. It is very likely, that it points to ll
 123                  header.  PPP makes it, that is wrong, because introduce
 124                  assymetry between rx and tx paths.
 125    data       -> data
 126
 127 Outgoing, dev->hard_header==NULL
 128    mac_header -> data. ll header is still not built!
 129    data       -> data
 130
 131 Resume
 132   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 133
 134
 135 On transmit:
 136 ------------
 137
 138 dev->hard_header != NULL
 139    mac_header -> ll header
 140    data       -> ll header
 141
 142 dev->hard_header == NULL (ll header is added by device, we cannot control it)
 143    mac_header -> data
 144    data       -> data
 145
 146    We should set nh.raw on output to correct posistion,
 147    packet classifier depends on it.
 148  */
 149
 150 /* Private packet socket structures. */
 151
 152 /* identical to struct packet_mreq except it has
 153  * a longer address field.
 154  */
 155 struct packet_mreq_max {
 156         int             mr_ifindex;
 157         unsigned short  mr_type;
 158         unsigned short  mr_alen;
 159         unsigned char   mr_address[MAX_ADDR_LEN];
 160 };
 161
 162 union tpacket_uhdr {
 163         struct tpacket_hdr  *h1;
 164         struct tpacket2_hdr *h2;
 165         struct tpacket3_hdr *h3;
 166         void *raw;
 167 };
 168
 169 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 170                 int closing, int tx_ring);
 171
 172 #define V3_ALIGNMENT    (8)
 173
 174 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 175
 176 #define BLK_PLUS_PRIV(sz_of_priv) \
 177         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 178
 179 #define PGV_FROM_VMALLOC 1
 180
 181 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
 182 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
 183 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
 184 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
 185 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
 186 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
 187 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
 188
 189 struct packet_sock;
 190 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
 191 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 192                        struct packet_type *pt, struct net_device *orig_dev);
 193
 194 static void *packet_previous_frame(struct packet_sock *po,
 195                 struct packet_ring_buffer *rb,
 196                 int status);
 197 static void packet_increment_head(struct packet_ring_buffer *buff);
 198 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
 199                         struct tpacket_block_desc *);
 200 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 201                         struct packet_sock *);
 202 static void prb_retire_current_block(struct tpacket_kbdq_core *,
 203                 struct packet_sock *, unsigned int status);
 204 static int prb_queue_frozen(struct tpacket_kbdq_core *);
 205 static void prb_open_block(struct tpacket_kbdq_core *,
 206                 struct tpacket_block_desc *);
 207 static void prb_retire_rx_blk_timer_expired(unsigned long);
 208 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 209 static void prb_init_blk_timer(struct packet_sock *,
 210                 struct tpacket_kbdq_core *,
 211                 void (*func) (unsigned long));
 212 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 213 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 214                 struct tpacket3_hdr *);
 215 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 216                 struct tpacket3_hdr *);
 217 static void packet_flush_mclist(struct sock *sk);
 218
 219 struct packet_skb_cb {
 220         unsigned int origlen;
 221         union {
 222                 struct sockaddr_pkt pkt;
 223                 struct sockaddr_ll ll;
 224         } sa;
 225 };
 226
 227 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 228
 229 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 230 #define GET_PBLOCK_DESC(x, bid) \
 231         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 232 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 233         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 234 #define GET_NEXT_PRB_BLK_NUM(x) \
 235         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 236         ((x)->kactive_blk_num+1) : 0)
 237
 238 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 239 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 240
 241 static int packet_direct_xmit(struct sk_buff *skb)
 242 {
 243         struct net_device *dev = skb->dev;
 244         const struct net_device_ops *ops = dev->netdev_ops;
 245         netdev_features_t features;
 246         struct netdev_queue *txq;
 247         u16 queue_map;
 248         int ret;
 249
 250         if (unlikely(!netif_running(dev) ||
 251                      !netif_carrier_ok(dev))) {
 252                 kfree_skb(skb);
 253                 return NET_XMIT_DROP;
 254         }
 255
 256         features = netif_skb_features(skb);
 257         if (skb_needs_linearize(skb, features) &&
 258             __skb_linearize(skb)) {
 259                 kfree_skb(skb);
 260                 return NET_XMIT_DROP;
 261         }
 262
 263         queue_map = skb_get_queue_mapping(skb);
 264         txq = netdev_get_tx_queue(dev, queue_map);
 265
 266         __netif_tx_lock_bh(txq);
 267         if (unlikely(netif_xmit_frozen_or_stopped(txq))) {
 268                 ret = NETDEV_TX_BUSY;
 269                 kfree_skb(skb);
 270                 goto out;
 271         }
 272
 273         ret = ops->ndo_start_xmit(skb, dev);
 274         if (likely(dev_xmit_complete(ret)))
 275                 txq_trans_update(txq);
 276         else
 277                 kfree_skb(skb);
 278 out:
 279         __netif_tx_unlock_bh(txq);
 280         return ret;
 281 }
 282
 283 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
 284 {
 285         struct net_device *dev;
 286
 287         rcu_read_lock();
 288         dev = rcu_dereference(po->cached_dev);
 289         if (likely(dev))
 290                 dev_hold(dev);
 291         rcu_read_unlock();
 292
 293         return dev;
 294 }
 295
 296 static void packet_cached_dev_assign(struct packet_sock *po,
 297                                      struct net_device *dev)
 298 {
 299         rcu_assign_pointer(po->cached_dev, dev);
 300 }
 301
 302 static void packet_cached_dev_reset(struct packet_sock *po)
 303 {
 304         RCU_INIT_POINTER(po->cached_dev, NULL);
 305 }
 306
 307 static bool packet_use_direct_xmit(const struct packet_sock *po)
 308 {
 309         return po->xmit == packet_direct_xmit;
 310 }
 311
 312 static u16 packet_pick_tx_queue(struct net_device *dev)
 313 {
 314         return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
 315 }
 316
 317 /* register_prot_hook must be invoked with the po->bind_lock held,
 318  * or from a context in which asynchronous accesses to the packet
 319  * socket is not possible (packet_create()).
 320  */
 321 static void register_prot_hook(struct sock *sk)
 322 {
 323         struct packet_sock *po = pkt_sk(sk);
 324
 325         if (!po->running) {
 326                 if (po->fanout)
 327                         __fanout_link(sk, po);
 328                 else
 329                         dev_add_pack(&po->prot_hook);
 330
 331                 sock_hold(sk);
 332                 po->running = 1;
 333         }
 334 }
 335
 336 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
 337  * held.   If the sync parameter is true, we will temporarily drop
 338  * the po->bind_lock and do a synchronize_net to make sure no
 339  * asynchronous packet processing paths still refer to the elements
 340  * of po->prot_hook.  If the sync parameter is false, it is the
 341  * callers responsibility to take care of this.
 342  */
 343 static void __unregister_prot_hook(struct sock *sk, bool sync)
 344 {
 345         struct packet_sock *po = pkt_sk(sk);
 346
 347         po->running = 0;
 348
 349         if (po->fanout)
 350                 __fanout_unlink(sk, po);
 351         else
 352                 __dev_remove_pack(&po->prot_hook);
 353
 354         __sock_put(sk);
 355
 356         if (sync) {
 357                 spin_unlock(&po->bind_lock);
 358                 synchronize_net();
 359                 spin_lock(&po->bind_lock);
 360         }
 361 }
 362
 363 static void unregister_prot_hook(struct sock *sk, bool sync)
 364 {
 365         struct packet_sock *po = pkt_sk(sk);
 366
 367         if (po->running)
 368                 __unregister_prot_hook(sk, sync);
 369 }
 370
 371 static inline __pure struct page *pgv_to_page(void *addr)
 372 {
 373         if (is_vmalloc_addr(addr))
 374                 return vmalloc_to_page(addr);
 375         return virt_to_page(addr);
 376 }
 377
 378 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 379 {
 380         union tpacket_uhdr h;
 381
 382         h.raw = frame;
 383         switch (po->tp_version) {
 384         case TPACKET_V1:
 385                 h.h1->tp_status = status;
 386                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 387                 break;
 388         case TPACKET_V2:
 389                 h.h2->tp_status = status;
 390                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 391                 break;
 392         case TPACKET_V3:
 393         default:
 394                 WARN(1, "TPACKET version not supported.\n");
 395                 BUG();
 396         }
 397
 398         smp_wmb();
 399 }
 400
 401 static int __packet_get_status(struct packet_sock *po, void *frame)
 402 {
 403         union tpacket_uhdr h;
 404
 405         smp_rmb();
 406
 407         h.raw = frame;
 408         switch (po->tp_version) {
 409         case TPACKET_V1:
 410                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 411                 return h.h1->tp_status;
 412         case TPACKET_V2:
 413                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 414                 return h.h2->tp_status;
 415         case TPACKET_V3:
 416         default:
 417                 WARN(1, "TPACKET version not supported.\n");
 418                 BUG();
 419                 return 0;
 420         }
 421 }
 422
 423 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
 424                                    unsigned int flags)
 425 {
 426         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 427
 428         if (shhwtstamps) {
 429                 if ((flags & SOF_TIMESTAMPING_SYS_HARDWARE) &&
 430                     ktime_to_timespec_cond(shhwtstamps->syststamp, ts))
 431                         return TP_STATUS_TS_SYS_HARDWARE;
 432                 if ((flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
 433                     ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
 434                         return TP_STATUS_TS_RAW_HARDWARE;
 435         }
 436
 437         if (ktime_to_timespec_cond(skb->tstamp, ts))
 438                 return TP_STATUS_TS_SOFTWARE;
 439
 440         return 0;
 441 }
 442
 443 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
 444                                     struct sk_buff *skb)
 445 {
 446         union tpacket_uhdr h;
 447         struct timespec ts;
 448         __u32 ts_status;
 449
 450         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
 451                 return 0;
 452
 453         h.raw = frame;
 454         switch (po->tp_version) {
 455         case TPACKET_V1:
 456                 h.h1->tp_sec = ts.tv_sec;
 457                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
 458                 break;
 459         case TPACKET_V2:
 460                 h.h2->tp_sec = ts.tv_sec;
 461                 h.h2->tp_nsec = ts.tv_nsec;
 462                 break;
 463         case TPACKET_V3:
 464         default:
 465                 WARN(1, "TPACKET version not supported.\n");
 466                 BUG();
 467         }
 468
 469         /* one flush is safe, as both fields always lie on the same cacheline */
 470         flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
 471         smp_wmb();
 472
 473         return ts_status;
 474 }
 475
 476 static void *packet_lookup_frame(struct packet_sock *po,
 477                 struct packet_ring_buffer *rb,
 478                 unsigned int position,
 479                 int status)
 480 {
 481         unsigned int pg_vec_pos, frame_offset;
 482         union tpacket_uhdr h;
 483
 484         pg_vec_pos = position / rb->frames_per_block;
 485         frame_offset = position % rb->frames_per_block;
 486
 487         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 488                 (frame_offset * rb->frame_size);
 489
 490         if (status != __packet_get_status(po, h.raw))
 491                 return NULL;
 492
 493         return h.raw;
 494 }
 495
 496 static void *packet_current_frame(struct packet_sock *po,
 497                 struct packet_ring_buffer *rb,
 498                 int status)
 499 {
 500         return packet_lookup_frame(po, rb, rb->head, status);
 501 }
 502
 503 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 504 {
 505         del_timer_sync(&pkc->retire_blk_timer);
 506 }
 507
 508 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 509                 int tx_ring,
 510                 struct sk_buff_head *rb_queue)
 511 {
 512         struct tpacket_kbdq_core *pkc;
 513
 514         pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
 515                         GET_PBDQC_FROM_RB(&po->rx_ring);
 516
 517         spin_lock_bh(&rb_queue->lock);
 518         pkc->delete_blk_timer = 1;
 519         spin_unlock_bh(&rb_queue->lock);
 520
 521         prb_del_retire_blk_timer(pkc);
 522 }
 523
 524 static void prb_init_blk_timer(struct packet_sock *po,
 525                 struct tpacket_kbdq_core *pkc,
 526                 void (*func) (unsigned long))
 527 {
 528         init_timer(&pkc->retire_blk_timer);
 529         pkc->retire_blk_timer.data = (long)po;
 530         pkc->retire_blk_timer.function = func;
 531         pkc->retire_blk_timer.expires = jiffies;
 532 }
 533
 534 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
 535 {
 536         struct tpacket_kbdq_core *pkc;
 537
 538         if (tx_ring)
 539                 BUG();
 540
 541         pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
 542                         GET_PBDQC_FROM_RB(&po->rx_ring);
 543         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
 544 }
 545
 546 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 547                                 int blk_size_in_bytes)
 548 {
 549         struct net_device *dev;
 550         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 551         struct ethtool_cmd ecmd;
 552         int err;
 553         u32 speed;
 554
 555         rtnl_lock();
 556         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 557         if (unlikely(!dev)) {
 558                 rtnl_unlock();
 559                 return DEFAULT_PRB_RETIRE_TOV;
 560         }
 561         err = __ethtool_get_settings(dev, &ecmd);
 562         speed = ethtool_cmd_speed(&ecmd);
 563         rtnl_unlock();
 564         if (!err) {
 565                 /*
 566                  * If the link speed is so slow you don't really
 567                  * need to worry about perf anyways
 568                  */
 569                 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
 570                         return DEFAULT_PRB_RETIRE_TOV;
 571                 } else {
 572                         msec = 1;
 573                         div = speed / 1000;
 574                 }
 575         }
 576
 577         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 578
 579         if (div)
 580                 mbits /= div;
 581
 582         tmo = mbits * msec;
 583
 584         if (div)
 585                 return tmo+1;
 586         return tmo;
 587 }
 588
 589 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 590                         union tpacket_req_u *req_u)
 591 {
 592         p1->feature_req_word = req_u->req3.tp_feature_req_word;
 593 }
 594
 595 static void init_prb_bdqc(struct packet_sock *po,
 596                         struct packet_ring_buffer *rb,
 597                         struct pgv *pg_vec,
 598                         union tpacket_req_u *req_u, int tx_ring)
 599 {
 600         struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
 601         struct tpacket_block_desc *pbd;
 602
 603         memset(p1, 0x0, sizeof(*p1));
 604
 605         p1->knxt_seq_num = 1;
 606         p1->pkbdq = pg_vec;
 607         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 608         p1->pkblk_start = pg_vec[0].buffer;
 609         p1->kblk_size = req_u->req3.tp_block_size;
 610         p1->knum_blocks = req_u->req3.tp_block_nr;
 611         p1->hdrlen = po->tp_hdrlen;
 612         p1->version = po->tp_version;
 613         p1->last_kactive_blk_num = 0;
 614         po->stats.stats3.tp_freeze_q_cnt = 0;
 615         if (req_u->req3.tp_retire_blk_tov)
 616                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 617         else
 618                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 619                                                 req_u->req3.tp_block_size);
 620         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 621         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 622
 623         prb_init_ft_ops(p1, req_u);
 624         prb_setup_retire_blk_timer(po, tx_ring);
 625         prb_open_block(p1, pbd);
 626 }
 627
 628 /*  Do NOT update the last_blk_num first.
 629  *  Assumes sk_buff_head lock is held.
 630  */
 631 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 632 {
 633         mod_timer(&pkc->retire_blk_timer,
 634                         jiffies + pkc->tov_in_jiffies);
 635         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 636 }
 637
 638 /*
 639  * Timer logic:
 640  * 1) We refresh the timer only when we open a block.
 641  *    By doing this we don't waste cycles refreshing the timer
 642  *        on packet-by-packet basis.
 643  *
 644  * With a 1MB block-size, on a 1Gbps line, it will take
 645  * i) ~8 ms to fill a block + ii) memcpy etc.
 646  * In this cut we are not accounting for the memcpy time.
 647  *
 648  * So, if the user sets the 'tmo' to 10ms then the timer
 649  * will never fire while the block is still getting filled
 650  * (which is what we want). However, the user could choose
 651  * to close a block early and that's fine.
 652  *
 653  * But when the timer does fire, we check whether or not to refresh it.
 654  * Since the tmo granularity is in msecs, it is not too expensive
 655  * to refresh the timer, lets say every '8' msecs.
 656  * Either the user can set the 'tmo' or we can derive it based on
 657  * a) line-speed and b) block-size.
 658  * prb_calc_retire_blk_tmo() calculates the tmo.
 659  *
 660  */
 661 static void prb_retire_rx_blk_timer_expired(unsigned long data)
 662 {
 663         struct packet_sock *po = (struct packet_sock *)data;
 664         struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 665         unsigned int frozen;
 666         struct tpacket_block_desc *pbd;
 667
 668         spin_lock(&po->sk.sk_receive_queue.lock);
 669
 670         frozen = prb_queue_frozen(pkc);
 671         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 672
 673         if (unlikely(pkc->delete_blk_timer))
 674                 goto out;
 675
 676         /* We only need to plug the race when the block is partially filled.
 677          * tpacket_rcv:
 678          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 679          *              copy_bits() is in progress ...
 680          *              timer fires on other cpu:
 681          *              we can't retire the current block because copy_bits
 682          *              is in progress.
 683          *
 684          */
 685         if (BLOCK_NUM_PKTS(pbd)) {
 686                 while (atomic_read(&pkc->blk_fill_in_prog)) {
 687                         /* Waiting for skb_copy_bits to finish... */
 688                         cpu_relax();
 689                 }
 690         }
 691
 692         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 693                 if (!frozen) {
 694                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 695                         if (!prb_dispatch_next_block(pkc, po))
 696                                 goto refresh_timer;
 697                         else
 698                                 goto out;
 699                 } else {
 700                         /* Case 1. Queue was frozen because user-space was
 701                          *         lagging behind.
 702                          */
 703                         if (prb_curr_blk_in_use(pkc, pbd)) {
 704                                 /*
 705                                  * Ok, user-space is still behind.
 706                                  * So just refresh the timer.
 707                                  */
 708                                 goto refresh_timer;
 709                         } else {
 710                                /* Case 2. queue was frozen,user-space caught up,
 711                                 * now the link went idle && the timer fired.
 712                                 * We don't have a block to close.So we open this
 713                                 * block and restart the timer.
 714                                 * opening a block thaws the queue,restarts timer
 715                                 * Thawing/timer-refresh is a side effect.
 716                                 */
 717                                 prb_open_block(pkc, pbd);
 718                                 goto out;
 719                         }
 720                 }
 721         }
 722
 723 refresh_timer:
 724         _prb_refresh_rx_retire_blk_timer(pkc);
 725
 726 out:
 727         spin_unlock(&po->sk.sk_receive_queue.lock);
 728 }
 729
 730 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 731                 struct tpacket_block_desc *pbd1, __u32 status)
 732 {
 733         /* Flush everything minus the block header */
 734
 735 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 736         u8 *start, *end;
 737
 738         start = (u8 *)pbd1;
 739
 740         /* Skip the block header(we know header WILL fit in 4K) */
 741         start += PAGE_SIZE;
 742
 743         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 744         for (; start < end; start += PAGE_SIZE)
 745                 flush_dcache_page(pgv_to_page(start));
 746
 747         smp_wmb();
 748 #endif
 749
 750         /* Now update the block status. */
 751
 752         BLOCK_STATUS(pbd1) = status;
 753
 754         /* Flush the block header */
 755
 756 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 757         start = (u8 *)pbd1;
 758         flush_dcache_page(pgv_to_page(start));
 759
 760         smp_wmb();
 761 #endif
 762 }
 763
 764 /*
 765  * Side effect:
 766  *
 767  * 1) flush the block
 768  * 2) Increment active_blk_num
 769  *
 770  * Note:We DONT refresh the timer on purpose.
 771  *      Because almost always the next block will be opened.
 772  */
 773 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 774                 struct tpacket_block_desc *pbd1,
 775                 struct packet_sock *po, unsigned int stat)
 776 {
 777         __u32 status = TP_STATUS_USER | stat;
 778
 779         struct tpacket3_hdr *last_pkt;
 780         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 781
 782         if (po->stats.stats3.tp_drops)
 783                 status |= TP_STATUS_LOSING;
 784
 785         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 786         last_pkt->tp_next_offset = 0;
 787
 788         /* Get the ts of the last pkt */
 789         if (BLOCK_NUM_PKTS(pbd1)) {
 790                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 791                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 792         } else {
 793                 /* Ok, we tmo'd - so get the current time */
 794                 struct timespec ts;
 795                 getnstimeofday(&ts);
 796                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 797                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 798         }
 799
 800         smp_wmb();
 801
 802         /* Flush the block */
 803         prb_flush_block(pkc1, pbd1, status);
 804
 805         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 806 }
 807
 808 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 809 {
 810         pkc->reset_pending_on_curr_blk = 0;
 811 }
 812
 813 /*
 814  * Side effect of opening a block:
 815  *
 816  * 1) prb_queue is thawed.
 817  * 2) retire_blk_timer is refreshed.
 818  *
 819  */
 820 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 821         struct tpacket_block_desc *pbd1)
 822 {
 823         struct timespec ts;
 824         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 825
 826         smp_rmb();
 827
 828         /* We could have just memset this but we will lose the
 829          * flexibility of making the priv area sticky
 830          */
 831
 832         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 833         BLOCK_NUM_PKTS(pbd1) = 0;
 834         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 835
 836         getnstimeofday(&ts);
 837
 838         h1->ts_first_pkt.ts_sec = ts.tv_sec;
 839         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 840
 841         pkc1->pkblk_start = (char *)pbd1;
 842         pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 843
 844         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 845         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 846
 847         pbd1->version = pkc1->version;
 848         pkc1->prev = pkc1->nxt_offset;
 849         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 850
 851         prb_thaw_queue(pkc1);
 852         _prb_refresh_rx_retire_blk_timer(pkc1);
 853
 854         smp_wmb();
 855 }
 856
 857 /*
 858  * Queue freeze logic:
 859  * 1) Assume tp_block_nr = 8 blocks.
 860  * 2) At time 't0', user opens Rx ring.
 861  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 862  * 4) user-space is either sleeping or processing block '0'.
 863  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 864  *    it will close block-7,loop around and try to fill block '0'.
 865  *    call-flow:
 866  *    __packet_lookup_frame_in_block
 867  *      prb_retire_current_block()
 868  *      prb_dispatch_next_block()
 869  *        |->(BLOCK_STATUS == USER) evaluates to true
 870  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 871  * 6) Now there are two cases:
 872  *    6.1) Link goes idle right after the queue is frozen.
 873  *         But remember, the last open_block() refreshed the timer.
 874  *         When this timer expires,it will refresh itself so that we can
 875  *         re-open block-0 in near future.
 876  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 877  *         case and __packet_lookup_frame_in_block will check if block-0
 878  *         is free and can now be re-used.
 879  */
 880 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 881                                   struct packet_sock *po)
 882 {
 883         pkc->reset_pending_on_curr_blk = 1;
 884         po->stats.stats3.tp_freeze_q_cnt++;
 885 }
 886
 887 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 888
 889 /*
 890  * If the next block is free then we will dispatch it
 891  * and return a good offset.
 892  * Else, we will freeze the queue.
 893  * So, caller must check the return value.
 894  */
 895 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 896                 struct packet_sock *po)
 897 {
 898         struct tpacket_block_desc *pbd;
 899
 900         smp_rmb();
 901
 902         /* 1. Get current block num */
 903         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 904
 905         /* 2. If this block is currently in_use then freeze the queue */
 906         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 907                 prb_freeze_queue(pkc, po);
 908                 return NULL;
 909         }
 910
 911         /*
 912          * 3.
 913          * open this block and return the offset where the first packet
 914          * needs to get stored.
 915          */
 916         prb_open_block(pkc, pbd);
 917         return (void *)pkc->nxt_offset;
 918 }
 919
 920 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 921                 struct packet_sock *po, unsigned int status)
 922 {
 923         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 924
 925         /* retire/close the current block */
 926         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 927                 /*
 928                  * Plug the case where copy_bits() is in progress on
 929                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 930                  * have space to copy the pkt in the current block and
 931                  * called prb_retire_current_block()
 932                  *
 933                  * We don't need to worry about the TMO case because
 934                  * the timer-handler already handled this case.
 935                  */
 936                 if (!(status & TP_STATUS_BLK_TMO)) {
 937                         while (atomic_read(&pkc->blk_fill_in_prog)) {
 938                                 /* Waiting for skb_copy_bits to finish... */
 939                                 cpu_relax();
 940                         }
 941                 }
 942                 prb_close_block(pkc, pbd, po, status);
 943                 return;
 944         }
 945 }
 946
 947 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
 948                                       struct tpacket_block_desc *pbd)
 949 {
 950         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 951 }
 952
 953 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 954 {
 955         return pkc->reset_pending_on_curr_blk;
 956 }
 957
 958 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 959 {
 960         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 961         atomic_dec(&pkc->blk_fill_in_prog);
 962 }
 963
 964 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 965                         struct tpacket3_hdr *ppd)
 966 {
 967         ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
 968 }
 969
 970 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 971                         struct tpacket3_hdr *ppd)
 972 {
 973         ppd->hv1.tp_rxhash = 0;
 974 }
 975
 976 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
 977                         struct tpacket3_hdr *ppd)
 978 {
 979         if (vlan_tx_tag_present(pkc->skb)) {
 980                 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
 981                 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
 982                 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
 983         } else {
 984                 ppd->hv1.tp_vlan_tci = 0;
 985                 ppd->hv1.tp_vlan_tpid = 0;
 986                 ppd->tp_status = TP_STATUS_AVAILABLE;
 987         }
 988 }
 989
 990 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
 991                         struct tpacket3_hdr *ppd)
 992 {
 993         ppd->hv1.tp_padding = 0;
 994         prb_fill_vlan_info(pkc, ppd);
 995
 996         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
 997                 prb_fill_rxhash(pkc, ppd);
 998         else
 999                 prb_clear_rxhash(pkc, ppd);
1000 }
1001
1002 static void prb_fill_curr_block(char *curr,
1003                                 struct tpacket_kbdq_core *pkc,
1004                                 struct tpacket_block_desc *pbd,
1005                                 unsigned int len)
1006 {
1007         struct tpacket3_hdr *ppd;
1008
1009         ppd  = (struct tpacket3_hdr *)curr;
1010         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1011         pkc->prev = curr;
1012         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1013         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1014         BLOCK_NUM_PKTS(pbd) += 1;
1015         atomic_inc(&pkc->blk_fill_in_prog);
1016         prb_run_all_ft_ops(pkc, ppd);
1017 }
1018
1019 /* Assumes caller has the sk->rx_queue.lock */
1020 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1021                                             struct sk_buff *skb,
1022                                                 int status,
1023                                             unsigned int len
1024                                             )
1025 {
1026         struct tpacket_kbdq_core *pkc;
1027         struct tpacket_block_desc *pbd;
1028         char *curr, *end;
1029
1030         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1031         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1032
1033         /* Queue is frozen when user space is lagging behind */
1034         if (prb_queue_frozen(pkc)) {
1035                 /*
1036                  * Check if that last block which caused the queue to freeze,
1037                  * is still in_use by user-space.
1038                  */
1039                 if (prb_curr_blk_in_use(pkc, pbd)) {
1040                         /* Can't record this packet */
1041                         return NULL;
1042                 } else {
1043                         /*
1044                          * Ok, the block was released by user-space.
1045                          * Now let's open that block.
1046                          * opening a block also thaws the queue.
1047                          * Thawing is a side effect.
1048                          */
1049                         prb_open_block(pkc, pbd);
1050                 }
1051         }
1052
1053         smp_mb();
1054         curr = pkc->nxt_offset;
1055         pkc->skb = skb;
1056         end = (char *)pbd + pkc->kblk_size;
1057
1058         /* first try the current block */
1059         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1060                 prb_fill_curr_block(curr, pkc, pbd, len);
1061                 return (void *)curr;
1062         }
1063
1064         /* Ok, close the current block */
1065         prb_retire_current_block(pkc, po, 0);
1066
1067         /* Now, try to dispatch the next block */
1068         curr = (char *)prb_dispatch_next_block(pkc, po);
1069         if (curr) {
1070                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1071                 prb_fill_curr_block(curr, pkc, pbd, len);
1072                 return (void *)curr;
1073         }
1074
1075         /*
1076          * No free blocks are available.user_space hasn't caught up yet.
1077          * Queue was just frozen and now this packet will get dropped.
1078          */
1079         return NULL;
1080 }
1081
1082 static void *packet_current_rx_frame(struct packet_sock *po,
1083                                             struct sk_buff *skb,
1084                                             int status, unsigned int len)
1085 {
1086         char *curr = NULL;
1087         switch (po->tp_version) {
1088         case TPACKET_V1:
1089         case TPACKET_V2:
1090                 curr = packet_lookup_frame(po, &po->rx_ring,
1091                                         po->rx_ring.head, status);
1092                 return curr;
1093         case TPACKET_V3:
1094                 return __packet_lookup_frame_in_block(po, skb, status, len);
1095         default:
1096                 WARN(1, "TPACKET version not supported\n");
1097                 BUG();
1098                 return NULL;
1099         }
1100 }
1101
1102 static void *prb_lookup_block(struct packet_sock *po,
1103                                      struct packet_ring_buffer *rb,
1104                                      unsigned int idx,
1105                                      int status)
1106 {
1107         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1108         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1109
1110         if (status != BLOCK_STATUS(pbd))
1111                 return NULL;
1112         return pbd;
1113 }
1114
1115 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1116 {
1117         unsigned int prev;
1118         if (rb->prb_bdqc.kactive_blk_num)
1119                 prev = rb->prb_bdqc.kactive_blk_num-1;
1120         else
1121                 prev = rb->prb_bdqc.knum_blocks-1;
1122         return prev;
1123 }
1124
1125 /* Assumes caller has held the rx_queue.lock */
1126 static void *__prb_previous_block(struct packet_sock *po,
1127                                          struct packet_ring_buffer *rb,
1128                                          int status)
1129 {
1130         unsigned int previous = prb_previous_blk_num(rb);
1131         return prb_lookup_block(po, rb, previous, status);
1132 }
1133
1134 static void *packet_previous_rx_frame(struct packet_sock *po,
1135                                              struct packet_ring_buffer *rb,
1136                                              int status)
1137 {
1138         if (po->tp_version <= TPACKET_V2)
1139                 return packet_previous_frame(po, rb, status);
1140
1141         return __prb_previous_block(po, rb, status);
1142 }
1143
1144 static void packet_increment_rx_head(struct packet_sock *po,
1145                                             struct packet_ring_buffer *rb)
1146 {
1147         switch (po->tp_version) {
1148         case TPACKET_V1:
1149         case TPACKET_V2:
1150                 return packet_increment_head(rb);
1151         case TPACKET_V3:
1152         default:
1153                 WARN(1, "TPACKET version not supported.\n");
1154                 BUG();
1155                 return;
1156         }
1157 }
1158
1159 static void *packet_previous_frame(struct packet_sock *po,
1160                 struct packet_ring_buffer *rb,
1161                 int status)
1162 {
1163         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1164         return packet_lookup_frame(po, rb, previous, status);
1165 }
1166
1167 static void packet_increment_head(struct packet_ring_buffer *buff)
1168 {
1169         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1170 }
1171
1172 static void packet_inc_pending(struct packet_ring_buffer *rb)
1173 {
1174         this_cpu_inc(*rb->pending_refcnt);
1175 }
1176
1177 static void packet_dec_pending(struct packet_ring_buffer *rb)
1178 {
1179         this_cpu_dec(*rb->pending_refcnt);
1180 }
1181
1182 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1183 {
1184         unsigned int refcnt = 0;
1185         int cpu;
1186
1187         /* We don't use pending refcount in rx_ring. */
1188         if (rb->pending_refcnt == NULL)
1189                 return 0;
1190
1191         for_each_possible_cpu(cpu)
1192                 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1193
1194         return refcnt;
1195 }
1196
1197 static int packet_alloc_pending(struct packet_sock *po)
1198 {
1199         po->rx_ring.pending_refcnt = NULL;
1200
1201         po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1202         if (unlikely(po->tx_ring.pending_refcnt == NULL))
1203                 return -ENOBUFS;
1204
1205         return 0;
1206 }
1207
1208 static void packet_free_pending(struct packet_sock *po)
1209 {
1210         free_percpu(po->tx_ring.pending_refcnt);
1211 }
1212
1213 static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1214 {
1215         struct sock *sk = &po->sk;
1216         bool has_room;
1217
1218         if (po->prot_hook.func != tpacket_rcv)
1219                 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1220                         <= sk->sk_rcvbuf;
1221
1222         spin_lock(&sk->sk_receive_queue.lock);
1223         if (po->tp_version == TPACKET_V3)
1224                 has_room = prb_lookup_block(po, &po->rx_ring,
1225                                             po->rx_ring.prb_bdqc.kactive_blk_num,
1226                                             TP_STATUS_KERNEL);
1227         else
1228                 has_room = packet_lookup_frame(po, &po->rx_ring,
1229                                                po->rx_ring.head,
1230                                                TP_STATUS_KERNEL);
1231         spin_unlock(&sk->sk_receive_queue.lock);
1232
1233         return has_room;
1234 }
1235
1236 static void packet_sock_destruct(struct sock *sk)
1237 {
1238         skb_queue_purge(&sk->sk_error_queue);
1239
1240         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1241         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1242
1243         if (!sock_flag(sk, SOCK_DEAD)) {
1244                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1245                 return;
1246         }
1247
1248         sk_refcnt_debug_dec(sk);
1249 }
1250
1251 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1252 {
1253         int x = atomic_read(&f->rr_cur) + 1;
1254
1255         if (x >= num)
1256                 x = 0;
1257
1258         return x;
1259 }
1260
1261 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1262                                       struct sk_buff *skb,
1263                                       unsigned int num)
1264 {
1265         return reciprocal_divide(skb->rxhash, num);
1266 }
1267
1268 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1269                                     struct sk_buff *skb,
1270                                     unsigned int num)
1271 {
1272         int cur, old;
1273
1274         cur = atomic_read(&f->rr_cur);
1275         while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1276                                      fanout_rr_next(f, num))) != cur)
1277                 cur = old;
1278         return cur;
1279 }
1280
1281 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1282                                      struct sk_buff *skb,
1283                                      unsigned int num)
1284 {
1285         return smp_processor_id() % num;
1286 }
1287
1288 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1289                                      struct sk_buff *skb,
1290                                      unsigned int num)
1291 {
1292         return reciprocal_divide(prandom_u32(), num);
1293 }
1294
1295 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1296                                           struct sk_buff *skb,
1297                                           unsigned int idx, unsigned int skip,
1298                                           unsigned int num)
1299 {
1300         unsigned int i, j;
1301
1302         i = j = min_t(int, f->next[idx], num - 1);
1303         do {
1304                 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1305                         if (i != j)
1306                                 f->next[idx] = i;
1307                         return i;
1308                 }
1309                 if (++i == num)
1310                         i = 0;
1311         } while (i != j);
1312
1313         return idx;
1314 }
1315
1316 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1317 {
1318         return f->flags & (flag >> 8);
1319 }
1320
1321 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1322                              struct packet_type *pt, struct net_device *orig_dev)
1323 {
1324         struct packet_fanout *f = pt->af_packet_priv;
1325         unsigned int num = f->num_members;
1326         struct packet_sock *po;
1327         unsigned int idx;
1328
1329         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1330             !num) {
1331                 kfree_skb(skb);
1332                 return 0;
1333         }
1334
1335         switch (f->type) {
1336         case PACKET_FANOUT_HASH:
1337         default:
1338                 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1339                         skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1340                         if (!skb)
1341                                 return 0;
1342                 }
1343                 skb_get_hash(skb);
1344                 idx = fanout_demux_hash(f, skb, num);
1345                 break;
1346         case PACKET_FANOUT_LB:
1347                 idx = fanout_demux_lb(f, skb, num);
1348                 break;
1349         case PACKET_FANOUT_CPU:
1350                 idx = fanout_demux_cpu(f, skb, num);
1351                 break;
1352         case PACKET_FANOUT_RND:
1353                 idx = fanout_demux_rnd(f, skb, num);
1354                 break;
1355         case PACKET_FANOUT_ROLLOVER:
1356                 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1357                 break;
1358         }
1359
1360         po = pkt_sk(f->arr[idx]);
1361         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1362             unlikely(!packet_rcv_has_room(po, skb))) {
1363                 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1364                 po = pkt_sk(f->arr[idx]);
1365         }
1366
1367         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1368 }
1369
1370 DEFINE_MUTEX(fanout_mutex);
1371 EXPORT_SYMBOL_GPL(fanout_mutex);
1372 static LIST_HEAD(fanout_list);
1373
1374 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1375 {
1376         struct packet_fanout *f = po->fanout;
1377
1378         spin_lock(&f->lock);
1379         f->arr[f->num_members] = sk;
1380         smp_wmb();
1381         f->num_members++;
1382         spin_unlock(&f->lock);
1383 }
1384
1385 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1386 {
1387         struct packet_fanout *f = po->fanout;
1388         int i;
1389
1390         spin_lock(&f->lock);
1391         for (i = 0; i < f->num_members; i++) {
1392                 if (f->arr[i] == sk)
1393                         break;
1394         }
1395         BUG_ON(i >= f->num_members);
1396         f->arr[i] = f->arr[f->num_members - 1];
1397         f->num_members--;
1398         spin_unlock(&f->lock);
1399 }
1400
1401 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1402 {
1403         if (ptype->af_packet_priv == (void *)((struct packet_sock *)sk)->fanout)
1404                 return true;
1405
1406         return false;
1407 }
1408
1409 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1410 {
1411         struct packet_sock *po = pkt_sk(sk);
1412         struct packet_fanout *f, *match;
1413         u8 type = type_flags & 0xff;
1414         u8 flags = type_flags >> 8;
1415         int err;
1416
1417         switch (type) {
1418         case PACKET_FANOUT_ROLLOVER:
1419                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1420                         return -EINVAL;
1421         case PACKET_FANOUT_HASH:
1422         case PACKET_FANOUT_LB:
1423         case PACKET_FANOUT_CPU:
1424         case PACKET_FANOUT_RND:
1425                 break;
1426         default:
1427                 return -EINVAL;
1428         }
1429
1430         if (!po->running)
1431                 return -EINVAL;
1432
1433         if (po->fanout)
1434                 return -EALREADY;
1435
1436         mutex_lock(&fanout_mutex);
1437         match = NULL;
1438         list_for_each_entry(f, &fanout_list, list) {
1439                 if (f->id == id &&
1440                     read_pnet(&f->net) == sock_net(sk)) {
1441                         match = f;
1442                         break;
1443                 }
1444         }
1445         err = -EINVAL;
1446         if (match && match->flags != flags)
1447                 goto out;
1448         if (!match) {
1449                 err = -ENOMEM;
1450                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1451                 if (!match)
1452                         goto out;
1453                 write_pnet(&match->net, sock_net(sk));
1454                 match->id = id;
1455                 match->type = type;
1456                 match->flags = flags;
1457                 atomic_set(&match->rr_cur, 0);
1458                 INIT_LIST_HEAD(&match->list);
1459                 spin_lock_init(&match->lock);
1460                 atomic_set(&match->sk_ref, 0);
1461                 match->prot_hook.type = po->prot_hook.type;
1462                 match->prot_hook.dev = po->prot_hook.dev;
1463                 match->prot_hook.func = packet_rcv_fanout;
1464                 match->prot_hook.af_packet_priv = match;
1465                 match->prot_hook.id_match = match_fanout_group;
1466                 dev_add_pack(&match->prot_hook);
1467                 list_add(&match->list, &fanout_list);
1468         }
1469         err = -EINVAL;
1470         if (match->type == type &&
1471             match->prot_hook.type == po->prot_hook.type &&
1472             match->prot_hook.dev == po->prot_hook.dev) {
1473                 err = -ENOSPC;
1474                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1475                         __dev_remove_pack(&po->prot_hook);
1476                         po->fanout = match;
1477                         atomic_inc(&match->sk_ref);
1478                         __fanout_link(sk, po);
1479                         err = 0;
1480                 }
1481         }
1482 out:
1483         mutex_unlock(&fanout_mutex);
1484         return err;
1485 }
1486
1487 static void fanout_release(struct sock *sk)
1488 {
1489         struct packet_sock *po = pkt_sk(sk);
1490         struct packet_fanout *f;
1491
1492         f = po->fanout;
1493         if (!f)
1494                 return;
1495
1496         mutex_lock(&fanout_mutex);
1497         po->fanout = NULL;
1498
1499         if (atomic_dec_and_test(&f->sk_ref)) {
1500                 list_del(&f->list);
1501                 dev_remove_pack(&f->prot_hook);
1502                 kfree(f);
1503         }
1504         mutex_unlock(&fanout_mutex);
1505 }
1506
1507 static const struct proto_ops packet_ops;
1508
1509 static const struct proto_ops packet_ops_spkt;
1510
1511 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1512                            struct packet_type *pt, struct net_device *orig_dev)
1513 {
1514         struct sock *sk;
1515         struct sockaddr_pkt *spkt;
1516
1517         /*
1518          *      When we registered the protocol we saved the socket in the data
1519          *      field for just this event.
1520          */
1521
1522         sk = pt->af_packet_priv;
1523
1524         /*
1525          *      Yank back the headers [hope the device set this
1526          *      right or kerboom...]
1527          *
1528          *      Incoming packets have ll header pulled,
1529          *      push it back.
1530          *
1531          *      For outgoing ones skb->data == skb_mac_header(skb)
1532          *      so that this procedure is noop.
1533          */
1534
1535         if (skb->pkt_type == PACKET_LOOPBACK)
1536                 goto out;
1537
1538         if (!net_eq(dev_net(dev), sock_net(sk)))
1539                 goto out;
1540
1541         skb = skb_share_check(skb, GFP_ATOMIC);
1542         if (skb == NULL)
1543                 goto oom;
1544
1545         /* drop any routing info */
1546         skb_dst_drop(skb);
1547
1548         /* drop conntrack reference */
1549         nf_reset(skb);
1550
1551         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1552
1553         skb_push(skb, skb->data - skb_mac_header(skb));
1554
1555         /*
1556          *      The SOCK_PACKET socket receives _all_ frames.
1557          */
1558
1559         spkt->spkt_family = dev->type;
1560         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1561         spkt->spkt_protocol = skb->protocol;
1562
1563         /*
1564          *      Charge the memory to the socket. This is done specifically
1565          *      to prevent sockets using all the memory up.
1566          */
1567
1568         if (sock_queue_rcv_skb(sk, skb) == 0)
1569                 return 0;
1570
1571 out:
1572         kfree_skb(skb);
1573 oom:
1574         return 0;
1575 }
1576
1577
1578 /*
1579  *      Output a raw packet to a device layer. This bypasses all the other
1580  *      protocol layers and you must therefore supply it with a complete frame
1581  */
1582
1583 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1584                                struct msghdr *msg, size_t len)
1585 {
1586         struct sock *sk = sock->sk;
1587         struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1588         struct sk_buff *skb = NULL;
1589         struct net_device *dev;
1590         __be16 proto = 0;
1591         int err;
1592         int extra_len = 0;
1593
1594         /*
1595          *      Get and verify the address.
1596          */
1597
1598         if (saddr) {
1599                 if (msg->msg_namelen < sizeof(struct sockaddr))
1600                         return -EINVAL;
1601                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1602                         proto = saddr->spkt_protocol;
1603         } else
1604                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1605
1606         /*
1607          *      Find the device first to size check it
1608          */
1609
1610         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1611 retry:
1612         rcu_read_lock();
1613         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1614         err = -ENODEV;
1615         if (dev == NULL)
1616                 goto out_unlock;
1617
1618         err = -ENETDOWN;
1619         if (!(dev->flags & IFF_UP))
1620                 goto out_unlock;
1621
1622         /*
1623          * You may not queue a frame bigger than the mtu. This is the lowest level
1624          * raw protocol and you must do your own fragmentation at this level.
1625          */
1626
1627         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1628                 if (!netif_supports_nofcs(dev)) {
1629                         err = -EPROTONOSUPPORT;
1630                         goto out_unlock;
1631                 }
1632                 extra_len = 4; /* We're doing our own CRC */
1633         }
1634
1635         err = -EMSGSIZE;
1636         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1637                 goto out_unlock;
1638
1639         if (!skb) {
1640                 size_t reserved = LL_RESERVED_SPACE(dev);
1641                 int tlen = dev->needed_tailroom;
1642                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1643
1644                 rcu_read_unlock();
1645                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1646                 if (skb == NULL)
1647                         return -ENOBUFS;
1648                 /* FIXME: Save some space for broken drivers that write a hard
1649                  * header at transmission time by themselves. PPP is the notable
1650                  * one here. This should really be fixed at the driver level.
1651                  */
1652                 skb_reserve(skb, reserved);
1653                 skb_reset_network_header(skb);
1654
1655                 /* Try to align data part correctly */
1656                 if (hhlen) {
1657                         skb->data -= hhlen;
1658                         skb->tail -= hhlen;
1659                         if (len < hhlen)
1660                                 skb_reset_network_header(skb);
1661                 }
1662                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1663                 if (err)
1664                         goto out_free;
1665                 goto retry;
1666         }
1667
1668         if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1669                 /* Earlier code assumed this would be a VLAN pkt,
1670                  * double-check this now that we have the actual
1671                  * packet in hand.
1672                  */
1673                 struct ethhdr *ehdr;
1674                 skb_reset_mac_header(skb);
1675                 ehdr = eth_hdr(skb);
1676                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1677                         err = -EMSGSIZE;
1678                         goto out_unlock;
1679                 }
1680         }
1681
1682         skb->protocol = proto;
1683         skb->dev = dev;
1684         skb->priority = sk->sk_priority;
1685         skb->mark = sk->sk_mark;
1686
1687         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1688
1689         if (unlikely(extra_len == 4))
1690                 skb->no_fcs = 1;
1691
1692         skb_probe_transport_header(skb, 0);
1693
1694         dev_queue_xmit(skb);
1695         rcu_read_unlock();
1696         return len;
1697
1698 out_unlock:
1699         rcu_read_unlock();
1700 out_free:
1701         kfree_skb(skb);
1702         return err;
1703 }
1704
1705 static unsigned int run_filter(const struct sk_buff *skb,
1706                                       const struct sock *sk,
1707                                       unsigned int res)
1708 {
1709         struct sk_filter *filter;
1710
1711         rcu_read_lock();
1712         filter = rcu_dereference(sk->sk_filter);
1713         if (filter != NULL)
1714                 res = SK_RUN_FILTER(filter, skb);
1715         rcu_read_unlock();
1716
1717         return res;
1718 }
1719
1720 /*
1721  * This function makes lazy skb cloning in hope that most of packets
1722  * are discarded by BPF.
1723  *
1724  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1725  * and skb->cb are mangled. It works because (and until) packets
1726  * falling here are owned by current CPU. Output packets are cloned
1727  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1728  * sequencially, so that if we return skb to original state on exit,
1729  * we will not harm anyone.
1730  */
1731
1732 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1733                       struct packet_type *pt, struct net_device *orig_dev)
1734 {
1735         struct sock *sk;
1736         struct sockaddr_ll *sll;
1737         struct packet_sock *po;
1738         u8 *skb_head = skb->data;
1739         int skb_len = skb->len;
1740         unsigned int snaplen, res;
1741
1742         if (skb->pkt_type == PACKET_LOOPBACK)
1743                 goto drop;
1744
1745         sk = pt->af_packet_priv;
1746         po = pkt_sk(sk);
1747
1748         if (!net_eq(dev_net(dev), sock_net(sk)))
1749                 goto drop;
1750
1751         skb->dev = dev;
1752
1753         if (dev->header_ops) {
1754                 /* The device has an explicit notion of ll header,
1755                  * exported to higher levels.
1756                  *
1757                  * Otherwise, the device hides details of its frame
1758                  * structure, so that corresponding packet head is
1759                  * never delivered to user.
1760                  */
1761                 if (sk->sk_type != SOCK_DGRAM)
1762                         skb_push(skb, skb->data - skb_mac_header(skb));
1763                 else if (skb->pkt_type == PACKET_OUTGOING) {
1764                         /* Special case: outgoing packets have ll header at head */
1765                         skb_pull(skb, skb_network_offset(skb));
1766                 }
1767         }
1768
1769         snaplen = skb->len;
1770
1771         res = run_filter(skb, sk, snaplen);
1772         if (!res)
1773                 goto drop_n_restore;
1774         if (snaplen > res)
1775                 snaplen = res;
1776
1777         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1778                 goto drop_n_acct;
1779
1780         if (skb_shared(skb)) {
1781                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1782                 if (nskb == NULL)
1783                         goto drop_n_acct;
1784
1785                 if (skb_head != skb->data) {
1786                         skb->data = skb_head;
1787                         skb->len = skb_len;
1788                 }
1789                 consume_skb(skb);
1790                 skb = nskb;
1791         }
1792
1793         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1794                      sizeof(skb->cb));
1795
1796         sll = &PACKET_SKB_CB(skb)->sa.ll;
1797         sll->sll_family = AF_PACKET;
1798         sll->sll_hatype = dev->type;
1799         sll->sll_protocol = skb->protocol;
1800         sll->sll_pkttype = skb->pkt_type;
1801         if (unlikely(po->origdev))
1802                 sll->sll_ifindex = orig_dev->ifindex;
1803         else
1804                 sll->sll_ifindex = dev->ifindex;
1805
1806         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1807
1808         PACKET_SKB_CB(skb)->origlen = skb->len;
1809
1810         if (pskb_trim(skb, snaplen))
1811                 goto drop_n_acct;
1812
1813         skb_set_owner_r(skb, sk);
1814         skb->dev = NULL;
1815         skb_dst_drop(skb);
1816
1817         /* drop conntrack reference */
1818         nf_reset(skb);
1819
1820         spin_lock(&sk->sk_receive_queue.lock);
1821         po->stats.stats1.tp_packets++;
1822         skb->dropcount = atomic_read(&sk->sk_drops);
1823         __skb_queue_tail(&sk->sk_receive_queue, skb);
1824         spin_unlock(&sk->sk_receive_queue.lock);
1825         sk->sk_data_ready(sk, skb->len);
1826         return 0;
1827
1828 drop_n_acct:
1829         spin_lock(&sk->sk_receive_queue.lock);
1830         po->stats.stats1.tp_drops++;
1831         atomic_inc(&sk->sk_drops);
1832         spin_unlock(&sk->sk_receive_queue.lock);
1833
1834 drop_n_restore:
1835         if (skb_head != skb->data && skb_shared(skb)) {
1836                 skb->data = skb_head;
1837                 skb->len = skb_len;
1838         }
1839 drop:
1840         consume_skb(skb);
1841         return 0;
1842 }
1843
1844 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1845                        struct packet_type *pt, struct net_device *orig_dev)
1846 {
1847         struct sock *sk;
1848         struct packet_sock *po;
1849         struct sockaddr_ll *sll;
1850         union tpacket_uhdr h;
1851         u8 *skb_head = skb->data;
1852         int skb_len = skb->len;
1853         unsigned int snaplen, res;
1854         unsigned long status = TP_STATUS_USER;
1855         unsigned short macoff, netoff, hdrlen;
1856         struct sk_buff *copy_skb = NULL;
1857         struct timespec ts;
1858         __u32 ts_status;
1859
1860         /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
1861          * We may add members to them until current aligned size without forcing
1862          * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
1863          */
1864         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
1865         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
1866
1867         if (skb->pkt_type == PACKET_LOOPBACK)
1868                 goto drop;
1869
1870         sk = pt->af_packet_priv;
1871         po = pkt_sk(sk);
1872
1873         if (!net_eq(dev_net(dev), sock_net(sk)))
1874                 goto drop;
1875
1876         if (dev->header_ops) {
1877                 if (sk->sk_type != SOCK_DGRAM)
1878                         skb_push(skb, skb->data - skb_mac_header(skb));
1879                 else if (skb->pkt_type == PACKET_OUTGOING) {
1880                         /* Special case: outgoing packets have ll header at head */
1881                         skb_pull(skb, skb_network_offset(skb));
1882                 }
1883         }
1884
1885         if (skb->ip_summed == CHECKSUM_PARTIAL)
1886                 status |= TP_STATUS_CSUMNOTREADY;
1887
1888         snaplen = skb->len;
1889
1890         res = run_filter(skb, sk, snaplen);
1891         if (!res)
1892                 goto drop_n_restore;
1893         if (snaplen > res)
1894                 snaplen = res;
1895
1896         if (sk->sk_type == SOCK_DGRAM) {
1897                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1898                                   po->tp_reserve;
1899         } else {
1900                 unsigned int maclen = skb_network_offset(skb);
1901                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1902                                        (maclen < 16 ? 16 : maclen)) +
1903                         po->tp_reserve;
1904                 macoff = netoff - maclen;
1905         }
1906         if (po->tp_version <= TPACKET_V2) {
1907                 if (macoff + snaplen > po->rx_ring.frame_size) {
1908                         if (po->copy_thresh &&
1909                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1910                                 if (skb_shared(skb)) {
1911                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1912                                 } else {
1913                                         copy_skb = skb_get(skb);
1914                                         skb_head = skb->data;
1915                                 }
1916                                 if (copy_skb)
1917                                         skb_set_owner_r(copy_skb, sk);
1918                         }
1919                         snaplen = po->rx_ring.frame_size - macoff;
1920                         if ((int)snaplen < 0)
1921                                 snaplen = 0;
1922                 }
1923         }
1924         spin_lock(&sk->sk_receive_queue.lock);
1925         h.raw = packet_current_rx_frame(po, skb,
1926                                         TP_STATUS_KERNEL, (macoff+snaplen));
1927         if (!h.raw)
1928                 goto ring_is_full;
1929         if (po->tp_version <= TPACKET_V2) {
1930                 packet_increment_rx_head(po, &po->rx_ring);
1931         /*
1932          * LOSING will be reported till you read the stats,
1933          * because it's COR - Clear On Read.
1934          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1935          * at packet level.
1936          */
1937                 if (po->stats.stats1.tp_drops)
1938                         status |= TP_STATUS_LOSING;
1939         }
1940         po->stats.stats1.tp_packets++;
1941         if (copy_skb) {
1942                 status |= TP_STATUS_COPY;
1943                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1944         }
1945         spin_unlock(&sk->sk_receive_queue.lock);
1946
1947         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1948
1949         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
1950                 getnstimeofday(&ts);
1951
1952         status |= ts_status;
1953
1954         switch (po->tp_version) {
1955         case TPACKET_V1:
1956                 h.h1->tp_len = skb->len;
1957                 h.h1->tp_snaplen = snaplen;
1958                 h.h1->tp_mac = macoff;
1959                 h.h1->tp_net = netoff;
1960                 h.h1->tp_sec = ts.tv_sec;
1961                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
1962                 hdrlen = sizeof(*h.h1);
1963                 break;
1964         case TPACKET_V2:
1965                 h.h2->tp_len = skb->len;
1966                 h.h2->tp_snaplen = snaplen;
1967                 h.h2->tp_mac = macoff;
1968                 h.h2->tp_net = netoff;
1969                 h.h2->tp_sec = ts.tv_sec;
1970                 h.h2->tp_nsec = ts.tv_nsec;
1971                 if (vlan_tx_tag_present(skb)) {
1972                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1973                         h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
1974                         status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1975                 } else {
1976                         h.h2->tp_vlan_tci = 0;
1977                         h.h2->tp_vlan_tpid = 0;
1978                 }
1979                 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
1980                 hdrlen = sizeof(*h.h2);
1981                 break;
1982         case TPACKET_V3:
1983                 /* tp_nxt_offset,vlan are already populated above.
1984                  * So DONT clear those fields here
1985                  */
1986                 h.h3->tp_status |= status;
1987                 h.h3->tp_len = skb->len;
1988                 h.h3->tp_snaplen = snaplen;
1989                 h.h3->tp_mac = macoff;
1990                 h.h3->tp_net = netoff;
1991                 h.h3->tp_sec  = ts.tv_sec;
1992                 h.h3->tp_nsec = ts.tv_nsec;
1993                 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
1994                 hdrlen = sizeof(*h.h3);
1995                 break;
1996         default:
1997                 BUG();
1998         }
1999
2000         sll = h.raw + TPACKET_ALIGN(hdrlen);
2001         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2002         sll->sll_family = AF_PACKET;
2003         sll->sll_hatype = dev->type;
2004         sll->sll_protocol = skb->protocol;
2005         sll->sll_pkttype = skb->pkt_type;
2006         if (unlikely(po->origdev))
2007                 sll->sll_ifindex = orig_dev->ifindex;
2008         else
2009                 sll->sll_ifindex = dev->ifindex;
2010
2011         smp_mb();
2012 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2013         {
2014                 u8 *start, *end;
2015
2016                 if (po->tp_version <= TPACKET_V2) {
2017                         end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
2018                                 + macoff + snaplen);
2019                         for (start = h.raw; start < end; start += PAGE_SIZE)
2020                                 flush_dcache_page(pgv_to_page(start));
2021                 }
2022                 smp_wmb();
2023         }
2024 #endif
2025         if (po->tp_version <= TPACKET_V2)
2026                 __packet_set_status(po, h.raw, status);
2027         else
2028                 prb_clear_blk_fill_status(&po->rx_ring);
2029
2030         sk->sk_data_ready(sk, 0);
2031
2032 drop_n_restore:
2033         if (skb_head != skb->data && skb_shared(skb)) {
2034                 skb->data = skb_head;
2035                 skb->len = skb_len;
2036         }
2037 drop:
2038         kfree_skb(skb);
2039         return 0;
2040
2041 ring_is_full:
2042         po->stats.stats1.tp_drops++;
2043         spin_unlock(&sk->sk_receive_queue.lock);
2044
2045         sk->sk_data_ready(sk, 0);
2046         kfree_skb(copy_skb);
2047         goto drop_n_restore;
2048 }
2049
2050 static void tpacket_destruct_skb(struct sk_buff *skb)
2051 {
2052         struct packet_sock *po = pkt_sk(skb->sk);
2053         void *ph;
2054
2055         if (likely(po->tx_ring.pg_vec)) {
2056                 __u32 ts;
2057
2058                 ph = skb_shinfo(skb)->destructor_arg;
2059                 packet_dec_pending(&po->tx_ring);
2060
2061                 ts = __packet_set_timestamp(po, ph, skb);
2062                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2063         }
2064
2065         sock_wfree(skb);
2066 }
2067
2068 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2069                 void *frame, struct net_device *dev, int size_max,
2070                 __be16 proto, unsigned char *addr, int hlen)
2071 {
2072         union tpacket_uhdr ph;
2073         int to_write, offset, len, tp_len, nr_frags, len_max;
2074         struct socket *sock = po->sk.sk_socket;
2075         struct page *page;
2076         void *data;
2077         int err;
2078
2079         ph.raw = frame;
2080
2081         skb->protocol = proto;
2082         skb->dev = dev;
2083         skb->priority = po->sk.sk_priority;
2084         skb->mark = po->sk.sk_mark;
2085         sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2086         skb_shinfo(skb)->destructor_arg = ph.raw;
2087
2088         switch (po->tp_version) {
2089         case TPACKET_V2:
2090                 tp_len = ph.h2->tp_len;
2091                 break;
2092         default:
2093                 tp_len = ph.h1->tp_len;
2094                 break;
2095         }
2096         if (unlikely(tp_len > size_max)) {
2097                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2098                 return -EMSGSIZE;
2099         }
2100
2101         skb_reserve(skb, hlen);
2102         skb_reset_network_header(skb);
2103
2104         if (!packet_use_direct_xmit(po))
2105                 skb_probe_transport_header(skb, 0);
2106         if (unlikely(po->tp_tx_has_off)) {
2107                 int off_min, off_max, off;
2108                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2109                 off_max = po->tx_ring.frame_size - tp_len;
2110                 if (sock->type == SOCK_DGRAM) {
2111                         switch (po->tp_version) {
2112                         case TPACKET_V2:
2113                                 off = ph.h2->tp_net;
2114                                 break;
2115                         default:
2116                                 off = ph.h1->tp_net;
2117                                 break;
2118                         }
2119                 } else {
2120                         switch (po->tp_version) {
2121                         case TPACKET_V2:
2122                                 off = ph.h2->tp_mac;
2123                                 break;
2124                         default:
2125                                 off = ph.h1->tp_mac;
2126                                 break;
2127                         }
2128                 }
2129                 if (unlikely((off < off_min) || (off_max < off)))
2130                         return -EINVAL;
2131                 data = ph.raw + off;
2132         } else {
2133                 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2134         }
2135         to_write = tp_len;
2136
2137         if (sock->type == SOCK_DGRAM) {
2138                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2139                                 NULL, tp_len);
2140                 if (unlikely(err < 0))
2141                         return -EINVAL;
2142         } else if (dev->hard_header_len) {
2143                 /* net device doesn't like empty head */
2144                 if (unlikely(tp_len <= dev->hard_header_len)) {
2145                         pr_err("packet size is too short (%d < %d)\n",
2146                                tp_len, dev->hard_header_len);
2147                         return -EINVAL;
2148                 }
2149
2150                 skb_push(skb, dev->hard_header_len);
2151                 err = skb_store_bits(skb, 0, data,
2152                                 dev->hard_header_len);
2153                 if (unlikely(err))
2154                         return err;
2155
2156                 data += dev->hard_header_len;
2157                 to_write -= dev->hard_header_len;
2158         }
2159
2160         offset = offset_in_page(data);
2161         len_max = PAGE_SIZE - offset;
2162         len = ((to_write > len_max) ? len_max : to_write);
2163
2164         skb->data_len = to_write;
2165         skb->len += to_write;
2166         skb->truesize += to_write;
2167         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2168
2169         while (likely(to_write)) {
2170                 nr_frags = skb_shinfo(skb)->nr_frags;
2171
2172                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2173                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2174                                MAX_SKB_FRAGS);
2175                         return -EFAULT;
2176                 }
2177
2178                 page = pgv_to_page(data);
2179                 data += len;
2180                 flush_dcache_page(page);
2181                 get_page(page);
2182                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2183                 to_write -= len;
2184                 offset = 0;
2185                 len_max = PAGE_SIZE;
2186                 len = ((to_write > len_max) ? len_max : to_write);
2187         }
2188
2189         return tp_len;
2190 }
2191
2192 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2193 {
2194         struct sk_buff *skb;
2195         struct net_device *dev;
2196         __be16 proto;
2197         int err, reserve = 0;
2198         void *ph;
2199         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2200         bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2201         int tp_len, size_max;
2202         unsigned char *addr;
2203         int len_sum = 0;
2204         int status = TP_STATUS_AVAILABLE;
2205         int hlen, tlen;
2206
2207         mutex_lock(&po->pg_vec_lock);
2208
2209         if (likely(saddr == NULL)) {
2210                 dev     = packet_cached_dev_get(po);
2211                 proto   = po->num;
2212                 addr    = NULL;
2213         } else {
2214                 err = -EINVAL;
2215                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2216                         goto out;
2217                 if (msg->msg_namelen < (saddr->sll_halen
2218                                         + offsetof(struct sockaddr_ll,
2219                                                 sll_addr)))
2220                         goto out;
2221                 proto   = saddr->sll_protocol;
2222                 addr    = saddr->sll_addr;
2223                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2224         }
2225
2226         err = -ENXIO;
2227         if (unlikely(dev == NULL))
2228                 goto out;
2229         err = -ENETDOWN;
2230         if (unlikely(!(dev->flags & IFF_UP)))
2231                 goto out_put;
2232
2233         reserve = dev->hard_header_len;
2234
2235         size_max = po->tx_ring.frame_size
2236                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2237
2238         if (size_max > dev->mtu + reserve)
2239                 size_max = dev->mtu + reserve;
2240
2241         do {
2242                 ph = packet_current_frame(po, &po->tx_ring,
2243                                           TP_STATUS_SEND_REQUEST);
2244                 if (unlikely(ph == NULL)) {
2245                         if (need_wait && need_resched())
2246                                 schedule();
2247                         continue;
2248                 }
2249
2250                 status = TP_STATUS_SEND_REQUEST;
2251                 hlen = LL_RESERVED_SPACE(dev);
2252                 tlen = dev->needed_tailroom;
2253                 skb = sock_alloc_send_skb(&po->sk,
2254                                 hlen + tlen + sizeof(struct sockaddr_ll),
2255                                 0, &err);
2256
2257                 if (unlikely(skb == NULL))
2258                         goto out_status;
2259
2260                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2261                                 addr, hlen);
2262
2263                 if (unlikely(tp_len < 0)) {
2264                         if (po->tp_loss) {
2265                                 __packet_set_status(po, ph,
2266                                                 TP_STATUS_AVAILABLE);
2267                                 packet_increment_head(&po->tx_ring);
2268                                 kfree_skb(skb);
2269                                 continue;
2270                         } else {
2271                                 status = TP_STATUS_WRONG_FORMAT;
2272                                 err = tp_len;
2273                                 goto out_status;
2274                         }
2275                 }
2276
2277                 skb_set_queue_mapping(skb, packet_pick_tx_queue(dev));
2278                 skb->destructor = tpacket_destruct_skb;
2279                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2280                 packet_inc_pending(&po->tx_ring);
2281
2282                 status = TP_STATUS_SEND_REQUEST;
2283                 err = po->xmit(skb);
2284                 if (unlikely(err > 0)) {
2285                         err = net_xmit_errno(err);
2286                         if (err && __packet_get_status(po, ph) ==
2287                                    TP_STATUS_AVAILABLE) {
2288                                 /* skb was destructed already */
2289                                 skb = NULL;
2290                                 goto out_status;
2291                         }
2292                         /*
2293                          * skb was dropped but not destructed yet;
2294                          * let's treat it like congestion or err < 0
2295                          */
2296                         err = 0;
2297                 }
2298                 packet_increment_head(&po->tx_ring);
2299                 len_sum += tp_len;
2300         } while (likely((ph != NULL) ||
2301                 /* Note: packet_read_pending() might be slow if we have
2302                  * to call it as it's per_cpu variable, but in fast-path
2303                  * we already short-circuit the loop with the first
2304                  * condition, and luckily don't have to go that path
2305                  * anyway.
2306                  */
2307                  (need_wait && packet_read_pending(&po->tx_ring))));
2308
2309         err = len_sum;
2310         goto out_put;
2311
2312 out_status:
2313         __packet_set_status(po, ph, status);
2314         kfree_skb(skb);
2315 out_put:
2316         dev_put(dev);
2317 out:
2318         mutex_unlock(&po->pg_vec_lock);
2319         return err;
2320 }
2321
2322 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2323                                         size_t reserve, size_t len,
2324                                         size_t linear, int noblock,
2325                                         int *err)
2326 {
2327         struct sk_buff *skb;
2328
2329         /* Under a page?  Don't bother with paged skb. */
2330         if (prepad + len < PAGE_SIZE || !linear)
2331                 linear = len;
2332
2333         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2334                                    err, 0);
2335         if (!skb)
2336                 return NULL;
2337
2338         skb_reserve(skb, reserve);
2339         skb_put(skb, linear);
2340         skb->data_len = len - linear;
2341         skb->len += len - linear;
2342
2343         return skb;
2344 }
2345
2346 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2347 {
2348         struct sock *sk = sock->sk;
2349         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2350         struct sk_buff *skb;
2351         struct net_device *dev;
2352         __be16 proto;
2353         unsigned char *addr;
2354         int err, reserve = 0;
2355         struct virtio_net_hdr vnet_hdr = { 0 };
2356         int offset = 0;
2357         int vnet_hdr_len;
2358         struct packet_sock *po = pkt_sk(sk);
2359         unsigned short gso_type = 0;
2360         int hlen, tlen;
2361         int extra_len = 0;
2362
2363         /*
2364          *      Get and verify the address.
2365          */
2366
2367         if (likely(saddr == NULL)) {
2368                 dev     = packet_cached_dev_get(po);
2369                 proto   = po->num;
2370                 addr    = NULL;
2371         } else {
2372                 err = -EINVAL;
2373                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2374                         goto out;
2375                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2376                         goto out;
2377                 proto   = saddr->sll_protocol;
2378                 addr    = saddr->sll_addr;
2379                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2380         }
2381
2382         err = -ENXIO;
2383         if (unlikely(dev == NULL))
2384                 goto out_unlock;
2385         err = -ENETDOWN;
2386         if (unlikely(!(dev->flags & IFF_UP)))
2387                 goto out_unlock;
2388
2389         if (sock->type == SOCK_RAW)
2390                 reserve = dev->hard_header_len;
2391         if (po->has_vnet_hdr) {
2392                 vnet_hdr_len = sizeof(vnet_hdr);
2393
2394                 err = -EINVAL;
2395                 if (len < vnet_hdr_len)
2396                         goto out_unlock;
2397
2398                 len -= vnet_hdr_len;
2399
2400                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2401                                        vnet_hdr_len);
2402                 if (err < 0)
2403                         goto out_unlock;
2404
2405                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2406                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2407                       vnet_hdr.hdr_len))
2408                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
2409                                                  vnet_hdr.csum_offset + 2;
2410
2411                 err = -EINVAL;
2412                 if (vnet_hdr.hdr_len > len)
2413                         goto out_unlock;
2414
2415                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2416                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2417                         case VIRTIO_NET_HDR_GSO_TCPV4:
2418                                 gso_type = SKB_GSO_TCPV4;
2419                                 break;
2420                         case VIRTIO_NET_HDR_GSO_TCPV6:
2421                                 gso_type = SKB_GSO_TCPV6;
2422                                 break;
2423                         case VIRTIO_NET_HDR_GSO_UDP:
2424                                 gso_type = SKB_GSO_UDP;
2425                                 break;
2426                         default:
2427                                 goto out_unlock;
2428                         }
2429
2430                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2431                                 gso_type |= SKB_GSO_TCP_ECN;
2432
2433                         if (vnet_hdr.gso_size == 0)
2434                                 goto out_unlock;
2435
2436                 }
2437         }
2438
2439         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2440                 if (!netif_supports_nofcs(dev)) {
2441                         err = -EPROTONOSUPPORT;
2442                         goto out_unlock;
2443                 }
2444                 extra_len = 4; /* We're doing our own CRC */
2445         }
2446
2447         err = -EMSGSIZE;
2448         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2449                 goto out_unlock;
2450
2451         err = -ENOBUFS;
2452         hlen = LL_RESERVED_SPACE(dev);
2453         tlen = dev->needed_tailroom;
2454         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2455                                msg->msg_flags & MSG_DONTWAIT, &err);
2456         if (skb == NULL)
2457                 goto out_unlock;
2458
2459         skb_set_network_header(skb, reserve);
2460
2461         err = -EINVAL;
2462         if (sock->type == SOCK_DGRAM &&
2463             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2464                 goto out_free;
2465
2466         /* Returns -EFAULT on error */
2467         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2468         if (err)
2469                 goto out_free;
2470
2471         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2472
2473         if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2474                 /* Earlier code assumed this would be a VLAN pkt,
2475                  * double-check this now that we have the actual
2476                  * packet in hand.
2477                  */
2478                 struct ethhdr *ehdr;
2479                 skb_reset_mac_header(skb);
2480                 ehdr = eth_hdr(skb);
2481                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2482                         err = -EMSGSIZE;
2483                         goto out_free;
2484                 }
2485         }
2486
2487         skb->protocol = proto;
2488         skb->dev = dev;
2489         skb->priority = sk->sk_priority;
2490         skb->mark = sk->sk_mark;
2491         skb_set_queue_mapping(skb, packet_pick_tx_queue(dev));
2492
2493         if (po->has_vnet_hdr) {
2494                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2495                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2496                                                   vnet_hdr.csum_offset)) {
2497                                 err = -EINVAL;
2498                                 goto out_free;
2499                         }
2500                 }
2501
2502                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2503                 skb_shinfo(skb)->gso_type = gso_type;
2504
2505                 /* Header must be checked, and gso_segs computed. */
2506                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2507                 skb_shinfo(skb)->gso_segs = 0;
2508
2509                 len += vnet_hdr_len;
2510         }
2511
2512         if (!packet_use_direct_xmit(po))
2513                 skb_probe_transport_header(skb, reserve);
2514         if (unlikely(extra_len == 4))
2515                 skb->no_fcs = 1;
2516
2517         err = po->xmit(skb);
2518         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2519                 goto out_unlock;
2520
2521         dev_put(dev);
2522
2523         return len;
2524
2525 out_free:
2526         kfree_skb(skb);
2527 out_unlock:
2528         if (dev)
2529                 dev_put(dev);
2530 out:
2531         return err;
2532 }
2533
2534 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2535                 struct msghdr *msg, size_t len)
2536 {
2537         struct sock *sk = sock->sk;
2538         struct packet_sock *po = pkt_sk(sk);
2539
2540         if (po->tx_ring.pg_vec)
2541                 return tpacket_snd(po, msg);
2542         else
2543                 return packet_snd(sock, msg, len);
2544 }
2545
2546 /*
2547  *      Close a PACKET socket. This is fairly simple. We immediately go
2548  *      to 'closed' state and remove our protocol entry in the device list.
2549  */
2550
2551 static int packet_release(struct socket *sock)
2552 {
2553         struct sock *sk = sock->sk;
2554         struct packet_sock *po;
2555         struct net *net;
2556         union tpacket_req_u req_u;
2557
2558         if (!sk)
2559                 return 0;
2560
2561         net = sock_net(sk);
2562         po = pkt_sk(sk);
2563
2564         mutex_lock(&net->packet.sklist_lock);
2565         sk_del_node_init_rcu(sk);
2566         mutex_unlock(&net->packet.sklist_lock);
2567
2568         preempt_disable();
2569         sock_prot_inuse_add(net, sk->sk_prot, -1);
2570         preempt_enable();
2571
2572         spin_lock(&po->bind_lock);
2573         unregister_prot_hook(sk, false);
2574         packet_cached_dev_reset(po);
2575
2576         if (po->prot_hook.dev) {
2577                 dev_put(po->prot_hook.dev);
2578                 po->prot_hook.dev = NULL;
2579         }
2580         spin_unlock(&po->bind_lock);
2581
2582         packet_flush_mclist(sk);
2583
2584         if (po->rx_ring.pg_vec) {
2585                 memset(&req_u, 0, sizeof(req_u));
2586                 packet_set_ring(sk, &req_u, 1, 0);
2587         }
2588
2589         if (po->tx_ring.pg_vec) {
2590                 memset(&req_u, 0, sizeof(req_u));
2591                 packet_set_ring(sk, &req_u, 1, 1);
2592         }
2593
2594         fanout_release(sk);
2595
2596         synchronize_net();
2597         /*
2598          *      Now the socket is dead. No more input will appear.
2599          */
2600         sock_orphan(sk);
2601         sock->sk = NULL;
2602
2603         /* Purge queues */
2604
2605         skb_queue_purge(&sk->sk_receive_queue);
2606         packet_free_pending(po);
2607         sk_refcnt_debug_release(sk);
2608
2609         sock_put(sk);
2610         return 0;
2611 }
2612
2613 /*
2614  *      Attach a packet hook.
2615  */
2616
2617 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 proto)
2618 {
2619         struct packet_sock *po = pkt_sk(sk);
2620         const struct net_device *dev_curr;
2621         __be16 proto_curr;
2622         bool need_rehook;
2623
2624         if (po->fanout) {
2625                 if (dev)
2626                         dev_put(dev);
2627
2628                 return -EINVAL;
2629         }
2630
2631         lock_sock(sk);
2632         spin_lock(&po->bind_lock);
2633
2634         proto_curr = po->prot_hook.type;
2635         dev_curr = po->prot_hook.dev;
2636
2637         need_rehook = proto_curr != proto || dev_curr != dev;
2638
2639         if (need_rehook) {
2640                 unregister_prot_hook(sk, true);
2641
2642                 po->num = proto;
2643                 po->prot_hook.type = proto;
2644
2645                 if (po->prot_hook.dev)
2646                         dev_put(po->prot_hook.dev);
2647
2648                 po->prot_hook.dev = dev;
2649
2650                 po->ifindex = dev ? dev->ifindex : 0;
2651                 packet_cached_dev_assign(po, dev);
2652         }
2653
2654         if (proto == 0 || !need_rehook)
2655                 goto out_unlock;
2656
2657         if (!dev || (dev->flags & IFF_UP)) {
2658                 register_prot_hook(sk);
2659         } else {
2660                 sk->sk_err = ENETDOWN;
2661                 if (!sock_flag(sk, SOCK_DEAD))
2662                         sk->sk_error_report(sk);
2663         }
2664
2665 out_unlock:
2666         spin_unlock(&po->bind_lock);
2667         release_sock(sk);
2668         return 0;
2669 }
2670
2671 /*
2672  *      Bind a packet socket to a device
2673  */
2674
2675 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2676                             int addr_len)
2677 {
2678         struct sock *sk = sock->sk;
2679         char name[15];
2680         struct net_device *dev;
2681         int err = -ENODEV;
2682
2683         /*
2684          *      Check legality
2685          */
2686
2687         if (addr_len != sizeof(struct sockaddr))
2688                 return -EINVAL;
2689         strlcpy(name, uaddr->sa_data, sizeof(name));
2690
2691         dev = dev_get_by_name(sock_net(sk), name);
2692         if (dev)
2693                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2694         return err;
2695 }
2696
2697 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2698 {
2699         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2700         struct sock *sk = sock->sk;
2701         struct net_device *dev = NULL;
2702         int err;
2703
2704
2705         /*
2706          *      Check legality
2707          */
2708
2709         if (addr_len < sizeof(struct sockaddr_ll))
2710                 return -EINVAL;
2711         if (sll->sll_family != AF_PACKET)
2712                 return -EINVAL;
2713
2714         if (sll->sll_ifindex) {
2715                 err = -ENODEV;
2716                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2717                 if (dev == NULL)
2718                         goto out;
2719         }
2720         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2721
2722 out:
2723         return err;
2724 }
2725
2726 static struct proto packet_proto = {
2727         .name     = "PACKET",
2728         .owner    = THIS_MODULE,
2729         .obj_size = sizeof(struct packet_sock),
2730 };
2731
2732 /*
2733  *      Create a packet of type SOCK_PACKET.
2734  */
2735
2736 static int packet_create(struct net *net, struct socket *sock, int protocol,
2737                          int kern)
2738 {
2739         struct sock *sk;
2740         struct packet_sock *po;
2741         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2742         int err;
2743
2744         if (!ns_capable(net->user_ns, CAP_NET_RAW))
2745                 return -EPERM;
2746         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2747             sock->type != SOCK_PACKET)
2748                 return -ESOCKTNOSUPPORT;
2749
2750         sock->state = SS_UNCONNECTED;
2751
2752         err = -ENOBUFS;
2753         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2754         if (sk == NULL)
2755                 goto out;
2756
2757         sock->ops = &packet_ops;
2758         if (sock->type == SOCK_PACKET)
2759                 sock->ops = &packet_ops_spkt;
2760
2761         sock_init_data(sock, sk);
2762
2763         po = pkt_sk(sk);
2764         sk->sk_family = PF_PACKET;
2765         po->num = proto;
2766         po->xmit = dev_queue_xmit;
2767
2768         err = packet_alloc_pending(po);
2769         if (err)
2770                 goto out2;
2771
2772         packet_cached_dev_reset(po);
2773
2774         sk->sk_destruct = packet_sock_destruct;
2775         sk_refcnt_debug_inc(sk);
2776
2777         /*
2778          *      Attach a protocol block
2779          */
2780
2781         spin_lock_init(&po->bind_lock);
2782         mutex_init(&po->pg_vec_lock);
2783         po->prot_hook.func = packet_rcv;
2784
2785         if (sock->type == SOCK_PACKET)
2786                 po->prot_hook.func = packet_rcv_spkt;
2787
2788         po->prot_hook.af_packet_priv = sk;
2789
2790         if (proto) {
2791                 po->prot_hook.type = proto;
2792                 register_prot_hook(sk);
2793         }
2794
2795         mutex_lock(&net->packet.sklist_lock);
2796         sk_add_node_rcu(sk, &net->packet.sklist);
2797         mutex_unlock(&net->packet.sklist_lock);
2798
2799         preempt_disable();
2800         sock_prot_inuse_add(net, &packet_proto, 1);
2801         preempt_enable();
2802
2803         return 0;
2804 out2:
2805         sk_free(sk);
2806 out:
2807         return err;
2808 }
2809
2810 /*
2811  *      Pull a packet from our receive queue and hand it to the user.
2812  *      If necessary we block.
2813  */
2814
2815 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2816                           struct msghdr *msg, size_t len, int flags)
2817 {
2818         struct sock *sk = sock->sk;
2819         struct sk_buff *skb;
2820         int copied, err;
2821         int vnet_hdr_len = 0;
2822
2823         err = -EINVAL;
2824         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2825                 goto out;
2826
2827 #if 0
2828         /* What error should we return now? EUNATTACH? */
2829         if (pkt_sk(sk)->ifindex < 0)
2830                 return -ENODEV;
2831 #endif
2832
2833         if (flags & MSG_ERRQUEUE) {
2834                 err = sock_recv_errqueue(sk, msg, len,
2835                                          SOL_PACKET, PACKET_TX_TIMESTAMP);
2836                 goto out;
2837         }
2838
2839         /*
2840          *      Call the generic datagram receiver. This handles all sorts
2841          *      of horrible races and re-entrancy so we can forget about it
2842          *      in the protocol layers.
2843          *
2844          *      Now it will return ENETDOWN, if device have just gone down,
2845          *      but then it will block.
2846          */
2847
2848         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2849
2850         /*
2851          *      An error occurred so return it. Because skb_recv_datagram()
2852          *      handles the blocking we don't see and worry about blocking
2853          *      retries.
2854          */
2855
2856         if (skb == NULL)
2857                 goto out;
2858
2859         if (pkt_sk(sk)->has_vnet_hdr) {
2860                 struct virtio_net_hdr vnet_hdr = { 0 };
2861
2862                 err = -EINVAL;
2863                 vnet_hdr_len = sizeof(vnet_hdr);
2864                 if (len < vnet_hdr_len)
2865                         goto out_free;
2866
2867                 len -= vnet_hdr_len;
2868
2869                 if (skb_is_gso(skb)) {
2870                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2871
2872                         /* This is a hint as to how much should be linear. */
2873                         vnet_hdr.hdr_len = skb_headlen(skb);
2874                         vnet_hdr.gso_size = sinfo->gso_size;
2875                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2876                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2877                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2878                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2879                         else if (sinfo->gso_type & SKB_GSO_UDP)
2880                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2881                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2882                                 goto out_free;
2883                         else
2884                                 BUG();
2885                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2886                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2887                 } else
2888                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2889
2890                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2891                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2892                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2893                         vnet_hdr.csum_offset = skb->csum_offset;
2894                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2895                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2896                 } /* else everything is zero */
2897
2898                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2899                                      vnet_hdr_len);
2900                 if (err < 0)
2901                         goto out_free;
2902         }
2903
2904         /* You lose any data beyond the buffer you gave. If it worries
2905          * a user program they can ask the device for its MTU
2906          * anyway.
2907          */
2908         copied = skb->len;
2909         if (copied > len) {
2910                 copied = len;
2911                 msg->msg_flags |= MSG_TRUNC;
2912         }
2913
2914         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2915         if (err)
2916                 goto out_free;
2917
2918         sock_recv_ts_and_drops(msg, sk, skb);
2919
2920         if (msg->msg_name) {
2921                 /* If the address length field is there to be filled
2922                  * in, we fill it in now.
2923                  */
2924                 if (sock->type == SOCK_PACKET) {
2925                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
2926                 } else {
2927                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2928                         msg->msg_namelen = sll->sll_halen +
2929                                 offsetof(struct sockaddr_ll, sll_addr);
2930                 }
2931                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2932                        msg->msg_namelen);
2933         }
2934
2935         if (pkt_sk(sk)->auxdata) {
2936                 struct tpacket_auxdata aux;
2937
2938                 aux.tp_status = TP_STATUS_USER;
2939                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2940                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2941                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2942                 aux.tp_snaplen = skb->len;
2943                 aux.tp_mac = 0;
2944                 aux.tp_net = skb_network_offset(skb);
2945                 if (vlan_tx_tag_present(skb)) {
2946                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2947                         aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
2948                         aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2949                 } else {
2950                         aux.tp_vlan_tci = 0;
2951                         aux.tp_vlan_tpid = 0;
2952                 }
2953                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2954         }
2955
2956         /*
2957          *      Free or return the buffer as appropriate. Again this
2958          *      hides all the races and re-entrancy issues from us.
2959          */
2960         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2961
2962 out_free:
2963         skb_free_datagram(sk, skb);
2964 out:
2965         return err;
2966 }
2967
2968 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2969                                int *uaddr_len, int peer)
2970 {
2971         struct net_device *dev;
2972         struct sock *sk = sock->sk;
2973
2974         if (peer)
2975                 return -EOPNOTSUPP;
2976
2977         uaddr->sa_family = AF_PACKET;
2978         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
2979         rcu_read_lock();
2980         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2981         if (dev)
2982                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
2983         rcu_read_unlock();
2984         *uaddr_len = sizeof(*uaddr);
2985
2986         return 0;
2987 }
2988
2989 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2990                           int *uaddr_len, int peer)
2991 {
2992         struct net_device *dev;
2993         struct sock *sk = sock->sk;
2994         struct packet_sock *po = pkt_sk(sk);
2995         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2996
2997         if (peer)
2998                 return -EOPNOTSUPP;
2999
3000         sll->sll_family = AF_PACKET;
3001         sll->sll_ifindex = po->ifindex;
3002         sll->sll_protocol = po->num;
3003         sll->sll_pkttype = 0;
3004         rcu_read_lock();
3005         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3006         if (dev) {
3007                 sll->sll_hatype = dev->type;
3008                 sll->sll_halen = dev->addr_len;
3009                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3010         } else {
3011                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
3012                 sll->sll_halen = 0;
3013         }
3014         rcu_read_unlock();
3015         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3016
3017         return 0;
3018 }
3019
3020 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3021                          int what)
3022 {
3023         switch (i->type) {
3024         case PACKET_MR_MULTICAST:
3025                 if (i->alen != dev->addr_len)
3026                         return -EINVAL;
3027                 if (what > 0)
3028                         return dev_mc_add(dev, i->addr);
3029                 else
3030                         return dev_mc_del(dev, i->addr);
3031                 break;
3032         case PACKET_MR_PROMISC:
3033                 return dev_set_promiscuity(dev, what);
3034                 break;
3035         case PACKET_MR_ALLMULTI:
3036                 return dev_set_allmulti(dev, what);
3037                 break;
3038         case PACKET_MR_UNICAST:
3039                 if (i->alen != dev->addr_len)
3040                         return -EINVAL;
3041                 if (what > 0)
3042                         return dev_uc_add(dev, i->addr);
3043                 else
3044                         return dev_uc_del(dev, i->addr);
3045                 break;
3046         default:
3047                 break;
3048         }
3049         return 0;
3050 }
3051
3052 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
3053 {
3054         for ( ; i; i = i->next) {
3055                 if (i->ifindex == dev->ifindex)
3056                         packet_dev_mc(dev, i, what);
3057         }
3058 }
3059
3060 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3061 {
3062         struct packet_sock *po = pkt_sk(sk);
3063         struct packet_mclist *ml, *i;
3064         struct net_device *dev;
3065         int err;
3066
3067         rtnl_lock();
3068
3069         err = -ENODEV;
3070         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3071         if (!dev)
3072                 goto done;
3073
3074         err = -EINVAL;
3075         if (mreq->mr_alen > dev->addr_len)
3076                 goto done;
3077
3078         err = -ENOBUFS;
3079         i = kmalloc(sizeof(*i), GFP_KERNEL);
3080         if (i == NULL)
3081                 goto done;
3082
3083         err = 0;
3084         for (ml = po->mclist; ml; ml = ml->next) {
3085                 if (ml->ifindex == mreq->mr_ifindex &&
3086                     ml->type == mreq->mr_type &&
3087                     ml->alen == mreq->mr_alen &&
3088                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3089                         ml->count++;
3090                         /* Free the new element ... */
3091                         kfree(i);
3092                         goto done;
3093                 }
3094         }
3095
3096         i->type = mreq->mr_type;
3097         i->ifindex = mreq->mr_ifindex;
3098         i->alen = mreq->mr_alen;
3099         memcpy(i->addr, mreq->mr_address, i->alen);
3100         i->count = 1;
3101         i->next = po->mclist;
3102         po->mclist = i;
3103         err = packet_dev_mc(dev, i, 1);
3104         if (err) {
3105                 po->mclist = i->next;
3106                 kfree(i);
3107         }
3108
3109 done:
3110         rtnl_unlock();
3111         return err;
3112 }
3113
3114 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3115 {
3116         struct packet_mclist *ml, **mlp;
3117
3118         rtnl_lock();
3119
3120         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3121                 if (ml->ifindex == mreq->mr_ifindex &&
3122                     ml->type == mreq->mr_type &&
3123                     ml->alen == mreq->mr_alen &&
3124                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3125                         if (--ml->count == 0) {
3126                                 struct net_device *dev;
3127                                 *mlp = ml->next;
3128                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3129                                 if (dev)
3130                                         packet_dev_mc(dev, ml, -1);
3131                                 kfree(ml);
3132                         }
3133                         rtnl_unlock();
3134                         return 0;
3135                 }
3136         }
3137         rtnl_unlock();
3138         return -EADDRNOTAVAIL;
3139 }
3140
3141 static void packet_flush_mclist(struct sock *sk)
3142 {
3143         struct packet_sock *po = pkt_sk(sk);
3144         struct packet_mclist *ml;
3145
3146         if (!po->mclist)
3147                 return;
3148
3149         rtnl_lock();
3150         while ((ml = po->mclist) != NULL) {
3151                 struct net_device *dev;
3152
3153                 po->mclist = ml->next;
3154                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3155                 if (dev != NULL)
3156                         packet_dev_mc(dev, ml, -1);
3157                 kfree(ml);
3158         }
3159         rtnl_unlock();
3160 }
3161
3162 static int
3163 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3164 {
3165         struct sock *sk = sock->sk;
3166         struct packet_sock *po = pkt_sk(sk);
3167         int ret;
3168
3169         if (level != SOL_PACKET)
3170                 return -ENOPROTOOPT;
3171
3172         switch (optname) {
3173         case PACKET_ADD_MEMBERSHIP:
3174         case PACKET_DROP_MEMBERSHIP:
3175         {
3176                 struct packet_mreq_max mreq;
3177                 int len = optlen;
3178                 memset(&mreq, 0, sizeof(mreq));
3179                 if (len < sizeof(struct packet_mreq))
3180                         return -EINVAL;
3181                 if (len > sizeof(mreq))
3182                         len = sizeof(mreq);
3183                 if (copy_from_user(&mreq, optval, len))
3184                         return -EFAULT;
3185                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3186                         return -EINVAL;
3187                 if (optname == PACKET_ADD_MEMBERSHIP)
3188                         ret = packet_mc_add(sk, &mreq);
3189                 else
3190                         ret = packet_mc_drop(sk, &mreq);
3191                 return ret;
3192         }
3193
3194         case PACKET_RX_RING:
3195         case PACKET_TX_RING:
3196         {
3197                 union tpacket_req_u req_u;
3198                 int len;
3199
3200                 switch (po->tp_version) {
3201                 case TPACKET_V1:
3202                 case TPACKET_V2:
3203                         len = sizeof(req_u.req);
3204                         break;
3205                 case TPACKET_V3:
3206                 default:
3207                         len = sizeof(req_u.req3);
3208                         break;
3209                 }
3210                 if (optlen < len)
3211                         return -EINVAL;
3212                 if (pkt_sk(sk)->has_vnet_hdr)
3213                         return -EINVAL;
3214                 if (copy_from_user(&req_u.req, optval, len))
3215                         return -EFAULT;
3216                 return packet_set_ring(sk, &req_u, 0,
3217                         optname == PACKET_TX_RING);
3218         }
3219         case PACKET_COPY_THRESH:
3220         {
3221                 int val;
3222
3223                 if (optlen != sizeof(val))
3224                         return -EINVAL;
3225                 if (copy_from_user(&val, optval, sizeof(val)))
3226                         return -EFAULT;
3227
3228                 pkt_sk(sk)->copy_thresh = val;
3229                 return 0;
3230         }
3231         case PACKET_VERSION:
3232         {
3233                 int val;
3234
3235                 if (optlen != sizeof(val))
3236                         return -EINVAL;
3237                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3238                         return -EBUSY;
3239                 if (copy_from_user(&val, optval, sizeof(val)))
3240                         return -EFAULT;
3241                 switch (val) {
3242                 case TPACKET_V1:
3243                 case TPACKET_V2:
3244                 case TPACKET_V3:
3245                         po->tp_version = val;
3246                         return 0;
3247                 default:
3248                         return -EINVAL;
3249                 }
3250         }
3251         case PACKET_RESERVE:
3252         {
3253                 unsigned int val;
3254
3255                 if (optlen != sizeof(val))
3256                         return -EINVAL;
3257                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3258                         return -EBUSY;
3259                 if (copy_from_user(&val, optval, sizeof(val)))
3260                         return -EFAULT;
3261                 po->tp_reserve = val;
3262                 return 0;
3263         }
3264         case PACKET_LOSS:
3265         {
3266                 unsigned int val;
3267
3268                 if (optlen != sizeof(val))
3269                         return -EINVAL;
3270                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3271                         return -EBUSY;
3272                 if (copy_from_user(&val, optval, sizeof(val)))
3273                         return -EFAULT;
3274                 po->tp_loss = !!val;
3275                 return 0;
3276         }
3277         case PACKET_AUXDATA:
3278         {
3279                 int val;
3280
3281                 if (optlen < sizeof(val))
3282                         return -EINVAL;
3283                 if (copy_from_user(&val, optval, sizeof(val)))
3284                         return -EFAULT;
3285
3286                 po->auxdata = !!val;
3287                 return 0;
3288         }
3289         case PACKET_ORIGDEV:
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->origdev = !!val;
3299                 return 0;
3300         }
3301         case PACKET_VNET_HDR:
3302         {
3303                 int val;
3304
3305                 if (sock->type != SOCK_RAW)
3306                         return -EINVAL;
3307                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3308                         return -EBUSY;
3309                 if (optlen < sizeof(val))
3310                         return -EINVAL;
3311                 if (copy_from_user(&val, optval, sizeof(val)))
3312                         return -EFAULT;
3313
3314                 po->has_vnet_hdr = !!val;
3315                 return 0;
3316         }
3317         case PACKET_TIMESTAMP:
3318         {
3319                 int val;
3320
3321                 if (optlen != sizeof(val))
3322                         return -EINVAL;
3323                 if (copy_from_user(&val, optval, sizeof(val)))
3324                         return -EFAULT;
3325
3326                 po->tp_tstamp = val;
3327                 return 0;
3328         }
3329         case PACKET_FANOUT:
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                 return fanout_add(sk, val & 0xffff, val >> 16);
3339         }
3340         case PACKET_TX_HAS_OFF:
3341         {
3342                 unsigned int val;
3343
3344                 if (optlen != sizeof(val))
3345                         return -EINVAL;
3346                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3347                         return -EBUSY;
3348                 if (copy_from_user(&val, optval, sizeof(val)))
3349                         return -EFAULT;
3350                 po->tp_tx_has_off = !!val;
3351                 return 0;
3352         }
3353         case PACKET_QDISC_BYPASS:
3354         {
3355                 int val;
3356
3357                 if (optlen != sizeof(val))
3358                         return -EINVAL;
3359                 if (copy_from_user(&val, optval, sizeof(val)))
3360                         return -EFAULT;
3361
3362                 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3363                 return 0;
3364         }
3365         default:
3366                 return -ENOPROTOOPT;
3367         }
3368 }
3369
3370 static int packet_getsockopt(struct socket *sock, int level, int optname,
3371                              char __user *optval, int __user *optlen)
3372 {
3373         int len;
3374         int val, lv = sizeof(val);
3375         struct sock *sk = sock->sk;
3376         struct packet_sock *po = pkt_sk(sk);
3377         void *data = &val;
3378         union tpacket_stats_u st;
3379
3380         if (level != SOL_PACKET)
3381                 return -ENOPROTOOPT;
3382
3383         if (get_user(len, optlen))
3384                 return -EFAULT;
3385
3386         if (len < 0)
3387                 return -EINVAL;
3388
3389         switch (optname) {
3390         case PACKET_STATISTICS:
3391                 spin_lock_bh(&sk->sk_receive_queue.lock);
3392                 memcpy(&st, &po->stats, sizeof(st));
3393                 memset(&po->stats, 0, sizeof(po->stats));
3394                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3395
3396                 if (po->tp_version == TPACKET_V3) {
3397                         lv = sizeof(struct tpacket_stats_v3);
3398                         st.stats3.tp_packets += st.stats3.tp_drops;
3399                         data = &st.stats3;
3400                 } else {
3401                         lv = sizeof(struct tpacket_stats);
3402                         st.stats1.tp_packets += st.stats1.tp_drops;
3403                         data = &st.stats1;
3404                 }
3405
3406                 break;
3407         case PACKET_AUXDATA:
3408                 val = po->auxdata;
3409                 break;
3410         case PACKET_ORIGDEV:
3411                 val = po->origdev;
3412                 break;
3413         case PACKET_VNET_HDR:
3414                 val = po->has_vnet_hdr;
3415                 break;
3416         case PACKET_VERSION:
3417                 val = po->tp_version;
3418                 break;
3419         case PACKET_HDRLEN:
3420                 if (len > sizeof(int))
3421                         len = sizeof(int);
3422                 if (copy_from_user(&val, optval, len))
3423                         return -EFAULT;
3424                 switch (val) {
3425                 case TPACKET_V1:
3426                         val = sizeof(struct tpacket_hdr);
3427                         break;
3428                 case TPACKET_V2:
3429                         val = sizeof(struct tpacket2_hdr);
3430                         break;
3431                 case TPACKET_V3:
3432                         val = sizeof(struct tpacket3_hdr);
3433                         break;
3434                 default:
3435                         return -EINVAL;
3436                 }
3437                 break;
3438         case PACKET_RESERVE:
3439                 val = po->tp_reserve;
3440                 break;
3441         case PACKET_LOSS:
3442                 val = po->tp_loss;
3443                 break;
3444         case PACKET_TIMESTAMP:
3445                 val = po->tp_tstamp;
3446                 break;
3447         case PACKET_FANOUT:
3448                 val = (po->fanout ?
3449                        ((u32)po->fanout->id |
3450                         ((u32)po->fanout->type << 16) |
3451                         ((u32)po->fanout->flags << 24)) :
3452                        0);
3453                 break;
3454         case PACKET_TX_HAS_OFF:
3455                 val = po->tp_tx_has_off;
3456                 break;
3457         case PACKET_QDISC_BYPASS:
3458                 val = packet_use_direct_xmit(po);
3459                 break;
3460         default:
3461                 return -ENOPROTOOPT;
3462         }
3463
3464         if (len > lv)
3465                 len = lv;
3466         if (put_user(len, optlen))
3467                 return -EFAULT;
3468         if (copy_to_user(optval, data, len))
3469                 return -EFAULT;
3470         return 0;
3471 }
3472
3473
3474 static int packet_notifier(struct notifier_block *this,
3475                            unsigned long msg, void *ptr)
3476 {
3477         struct sock *sk;
3478         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3479         struct net *net = dev_net(dev);
3480
3481         rcu_read_lock();
3482         sk_for_each_rcu(sk, &net->packet.sklist) {
3483                 struct packet_sock *po = pkt_sk(sk);
3484
3485                 switch (msg) {
3486                 case NETDEV_UNREGISTER:
3487                         if (po->mclist)
3488                                 packet_dev_mclist(dev, po->mclist, -1);
3489                         /* fallthrough */
3490
3491                 case NETDEV_DOWN:
3492                         if (dev->ifindex == po->ifindex) {
3493                                 spin_lock(&po->bind_lock);
3494                                 if (po->running) {
3495                                         __unregister_prot_hook(sk, false);
3496                                         sk->sk_err = ENETDOWN;
3497                                         if (!sock_flag(sk, SOCK_DEAD))
3498                                                 sk->sk_error_report(sk);
3499                                 }
3500                                 if (msg == NETDEV_UNREGISTER) {
3501                                         packet_cached_dev_reset(po);
3502                                         po->ifindex = -1;
3503                                         if (po->prot_hook.dev)
3504                                                 dev_put(po->prot_hook.dev);
3505                                         po->prot_hook.dev = NULL;
3506                                 }
3507                                 spin_unlock(&po->bind_lock);
3508                         }
3509                         break;
3510                 case NETDEV_UP:
3511                         if (dev->ifindex == po->ifindex) {
3512                                 spin_lock(&po->bind_lock);
3513                                 if (po->num)
3514                                         register_prot_hook(sk);
3515                                 spin_unlock(&po->bind_lock);
3516                         }
3517                         break;
3518                 }
3519         }
3520         rcu_read_unlock();
3521         return NOTIFY_DONE;
3522 }
3523
3524
3525 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3526                         unsigned long arg)
3527 {
3528         struct sock *sk = sock->sk;
3529
3530         switch (cmd) {
3531         case SIOCOUTQ:
3532         {
3533                 int amount = sk_wmem_alloc_get(sk);
3534
3535                 return put_user(amount, (int __user *)arg);
3536         }
3537         case SIOCINQ:
3538         {
3539                 struct sk_buff *skb;
3540                 int amount = 0;
3541
3542                 spin_lock_bh(&sk->sk_receive_queue.lock);
3543                 skb = skb_peek(&sk->sk_receive_queue);
3544                 if (skb)
3545                         amount = skb->len;
3546                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3547                 return put_user(amount, (int __user *)arg);
3548         }
3549         case SIOCGSTAMP:
3550                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3551         case SIOCGSTAMPNS:
3552                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3553
3554 #ifdef CONFIG_INET
3555         case SIOCADDRT:
3556         case SIOCDELRT:
3557         case SIOCDARP:
3558         case SIOCGARP:
3559         case SIOCSARP:
3560         case SIOCGIFADDR:
3561         case SIOCSIFADDR:
3562         case SIOCGIFBRDADDR:
3563         case SIOCSIFBRDADDR:
3564         case SIOCGIFNETMASK:
3565         case SIOCSIFNETMASK:
3566         case SIOCGIFDSTADDR:
3567         case SIOCSIFDSTADDR:
3568         case SIOCSIFFLAGS:
3569                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3570 #endif
3571
3572         default:
3573                 return -ENOIOCTLCMD;
3574         }
3575         return 0;
3576 }
3577
3578 static unsigned int packet_poll(struct file *file, struct socket *sock,
3579                                 poll_table *wait)
3580 {
3581         struct sock *sk = sock->sk;
3582         struct packet_sock *po = pkt_sk(sk);
3583         unsigned int mask = datagram_poll(file, sock, wait);
3584
3585         spin_lock_bh(&sk->sk_receive_queue.lock);
3586         if (po->rx_ring.pg_vec) {
3587                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3588                         TP_STATUS_KERNEL))
3589                         mask |= POLLIN | POLLRDNORM;
3590         }
3591         spin_unlock_bh(&sk->sk_receive_queue.lock);
3592         spin_lock_bh(&sk->sk_write_queue.lock);
3593         if (po->tx_ring.pg_vec) {
3594                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3595                         mask |= POLLOUT | POLLWRNORM;
3596         }
3597         spin_unlock_bh(&sk->sk_write_queue.lock);
3598         return mask;
3599 }
3600
3601
3602 /* Dirty? Well, I still did not learn better way to account
3603  * for user mmaps.
3604  */
3605
3606 static void packet_mm_open(struct vm_area_struct *vma)
3607 {
3608         struct file *file = vma->vm_file;
3609         struct socket *sock = file->private_data;
3610         struct sock *sk = sock->sk;
3611
3612         if (sk)
3613                 atomic_inc(&pkt_sk(sk)->mapped);
3614 }
3615
3616 static void packet_mm_close(struct vm_area_struct *vma)
3617 {
3618         struct file *file = vma->vm_file;
3619         struct socket *sock = file->private_data;
3620         struct sock *sk = sock->sk;
3621
3622         if (sk)
3623                 atomic_dec(&pkt_sk(sk)->mapped);
3624 }
3625
3626 static const struct vm_operations_struct packet_mmap_ops = {
3627         .open   =       packet_mm_open,
3628         .close  =       packet_mm_close,
3629 };
3630
3631 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3632                         unsigned int len)
3633 {
3634         int i;
3635
3636         for (i = 0; i < len; i++) {
3637                 if (likely(pg_vec[i].buffer)) {
3638                         if (is_vmalloc_addr(pg_vec[i].buffer))
3639                                 vfree(pg_vec[i].buffer);
3640                         else
3641                                 free_pages((unsigned long)pg_vec[i].buffer,
3642                                            order);
3643                         pg_vec[i].buffer = NULL;
3644                 }
3645         }
3646         kfree(pg_vec);
3647 }
3648
3649 static char *alloc_one_pg_vec_page(unsigned long order)
3650 {
3651         char *buffer = NULL;
3652         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3653                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3654
3655         buffer = (char *) __get_free_pages(gfp_flags, order);
3656
3657         if (buffer)
3658                 return buffer;
3659
3660         /*
3661          * __get_free_pages failed, fall back to vmalloc
3662          */
3663         buffer = vzalloc((1 << order) * PAGE_SIZE);
3664
3665         if (buffer)
3666                 return buffer;
3667
3668         /*
3669          * vmalloc failed, lets dig into swap here
3670          */
3671         gfp_flags &= ~__GFP_NORETRY;
3672         buffer = (char *)__get_free_pages(gfp_flags, order);
3673         if (buffer)
3674                 return buffer;
3675
3676         /*
3677          * complete and utter failure
3678          */
3679         return NULL;
3680 }
3681
3682 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3683 {
3684         unsigned int block_nr = req->tp_block_nr;
3685         struct pgv *pg_vec;
3686         int i;
3687
3688         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3689         if (unlikely(!pg_vec))
3690                 goto out;
3691
3692         for (i = 0; i < block_nr; i++) {
3693                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3694                 if (unlikely(!pg_vec[i].buffer))
3695                         goto out_free_pgvec;
3696         }
3697
3698 out:
3699         return pg_vec;
3700
3701 out_free_pgvec:
3702         free_pg_vec(pg_vec, order, block_nr);
3703         pg_vec = NULL;
3704         goto out;
3705 }
3706
3707 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3708                 int closing, int tx_ring)
3709 {
3710         struct pgv *pg_vec = NULL;
3711         struct packet_sock *po = pkt_sk(sk);
3712         int was_running, order = 0;
3713         struct packet_ring_buffer *rb;
3714         struct sk_buff_head *rb_queue;
3715         __be16 num;
3716         int err = -EINVAL;
3717         /* Added to avoid minimal code churn */
3718         struct tpacket_req *req = &req_u->req;
3719
3720         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3721         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3722                 WARN(1, "Tx-ring is not supported.\n");
3723                 goto out;
3724         }
3725
3726         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3727         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3728
3729         err = -EBUSY;
3730         if (!closing) {
3731                 if (atomic_read(&po->mapped))
3732                         goto out;
3733                 if (packet_read_pending(rb))
3734                         goto out;
3735         }
3736
3737         if (req->tp_block_nr) {
3738                 /* Sanity tests and some calculations */
3739                 err = -EBUSY;
3740                 if (unlikely(rb->pg_vec))
3741                         goto out;
3742
3743                 switch (po->tp_version) {
3744                 case TPACKET_V1:
3745                         po->tp_hdrlen = TPACKET_HDRLEN;
3746                         break;
3747                 case TPACKET_V2:
3748                         po->tp_hdrlen = TPACKET2_HDRLEN;
3749                         break;
3750                 case TPACKET_V3:
3751                         po->tp_hdrlen = TPACKET3_HDRLEN;
3752                         break;
3753                 }
3754
3755                 err = -EINVAL;
3756                 if (unlikely((int)req->tp_block_size <= 0))
3757                         goto out;
3758                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3759                         goto out;
3760                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3761                                         po->tp_reserve))
3762                         goto out;
3763                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3764                         goto out;
3765
3766                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3767                 if (unlikely(rb->frames_per_block <= 0))
3768                         goto out;
3769                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3770                                         req->tp_frame_nr))
3771                         goto out;
3772
3773                 err = -ENOMEM;
3774                 order = get_order(req->tp_block_size);
3775                 pg_vec = alloc_pg_vec(req, order);
3776                 if (unlikely(!pg_vec))
3777                         goto out;
3778                 switch (po->tp_version) {
3779                 case TPACKET_V3:
3780                 /* Transmit path is not supported. We checked
3781                  * it above but just being paranoid
3782                  */
3783                         if (!tx_ring)
3784                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3785                                 break;
3786                 default:
3787                         break;
3788                 }
3789         }
3790         /* Done */
3791         else {
3792                 err = -EINVAL;
3793                 if (unlikely(req->tp_frame_nr))
3794                         goto out;
3795         }
3796
3797         lock_sock(sk);
3798
3799         /* Detach socket from network */
3800         spin_lock(&po->bind_lock);
3801         was_running = po->running;
3802         num = po->num;
3803         if (was_running) {
3804                 po->num = 0;
3805                 __unregister_prot_hook(sk, false);
3806         }
3807         spin_unlock(&po->bind_lock);
3808
3809         synchronize_net();
3810
3811         err = -EBUSY;
3812         mutex_lock(&po->pg_vec_lock);
3813         if (closing || atomic_read(&po->mapped) == 0) {
3814                 err = 0;
3815                 spin_lock_bh(&rb_queue->lock);
3816                 swap(rb->pg_vec, pg_vec);
3817                 rb->frame_max = (req->tp_frame_nr - 1);
3818                 rb->head = 0;
3819                 rb->frame_size = req->tp_frame_size;
3820                 spin_unlock_bh(&rb_queue->lock);
3821
3822                 swap(rb->pg_vec_order, order);
3823                 swap(rb->pg_vec_len, req->tp_block_nr);
3824
3825                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3826                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3827                                                 tpacket_rcv : packet_rcv;
3828                 skb_queue_purge(rb_queue);
3829                 if (atomic_read(&po->mapped))
3830                         pr_err("packet_mmap: vma is busy: %d\n",
3831                                atomic_read(&po->mapped));
3832         }
3833         mutex_unlock(&po->pg_vec_lock);
3834
3835         spin_lock(&po->bind_lock);
3836         if (was_running) {
3837                 po->num = num;
3838                 register_prot_hook(sk);
3839         }
3840         spin_unlock(&po->bind_lock);
3841         if (closing && (po->tp_version > TPACKET_V2)) {
3842                 /* Because we don't support block-based V3 on tx-ring */
3843                 if (!tx_ring)
3844                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3845         }
3846         release_sock(sk);
3847
3848         if (pg_vec)
3849                 free_pg_vec(pg_vec, order, req->tp_block_nr);
3850 out:
3851         return err;
3852 }
3853
3854 static int packet_mmap(struct file *file, struct socket *sock,
3855                 struct vm_area_struct *vma)
3856 {
3857         struct sock *sk = sock->sk;
3858         struct packet_sock *po = pkt_sk(sk);
3859         unsigned long size, expected_size;
3860         struct packet_ring_buffer *rb;
3861         unsigned long start;
3862         int err = -EINVAL;
3863         int i;
3864
3865         if (vma->vm_pgoff)
3866                 return -EINVAL;
3867
3868         mutex_lock(&po->pg_vec_lock);
3869
3870         expected_size = 0;
3871         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3872                 if (rb->pg_vec) {
3873                         expected_size += rb->pg_vec_len
3874                                                 * rb->pg_vec_pages
3875                                                 * PAGE_SIZE;
3876                 }
3877         }
3878
3879         if (expected_size == 0)
3880                 goto out;
3881
3882         size = vma->vm_end - vma->vm_start;
3883         if (size != expected_size)
3884                 goto out;
3885
3886         start = vma->vm_start;
3887         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3888                 if (rb->pg_vec == NULL)
3889                         continue;
3890
3891                 for (i = 0; i < rb->pg_vec_len; i++) {
3892                         struct page *page;
3893                         void *kaddr = rb->pg_vec[i].buffer;
3894                         int pg_num;
3895
3896                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3897                                 page = pgv_to_page(kaddr);
3898                                 err = vm_insert_page(vma, start, page);
3899                                 if (unlikely(err))
3900                                         goto out;
3901                                 start += PAGE_SIZE;
3902                                 kaddr += PAGE_SIZE;
3903                         }
3904                 }
3905         }
3906
3907         atomic_inc(&po->mapped);
3908         vma->vm_ops = &packet_mmap_ops;
3909         err = 0;
3910
3911 out:
3912         mutex_unlock(&po->pg_vec_lock);
3913         return err;
3914 }
3915
3916 static const struct proto_ops packet_ops_spkt = {
3917         .family =       PF_PACKET,
3918         .owner =        THIS_MODULE,
3919         .release =      packet_release,
3920         .bind =         packet_bind_spkt,
3921         .connect =      sock_no_connect,
3922         .socketpair =   sock_no_socketpair,
3923         .accept =       sock_no_accept,
3924         .getname =      packet_getname_spkt,
3925         .poll =         datagram_poll,
3926         .ioctl =        packet_ioctl,
3927         .listen =       sock_no_listen,
3928         .shutdown =     sock_no_shutdown,
3929         .setsockopt =   sock_no_setsockopt,
3930         .getsockopt =   sock_no_getsockopt,
3931         .sendmsg =      packet_sendmsg_spkt,
3932         .recvmsg =      packet_recvmsg,
3933         .mmap =         sock_no_mmap,
3934         .sendpage =     sock_no_sendpage,
3935 };
3936
3937 static const struct proto_ops packet_ops = {
3938         .family =       PF_PACKET,
3939         .owner =        THIS_MODULE,
3940         .release =      packet_release,
3941         .bind =         packet_bind,
3942         .connect =      sock_no_connect,
3943         .socketpair =   sock_no_socketpair,
3944         .accept =       sock_no_accept,
3945         .getname =      packet_getname,
3946         .poll =         packet_poll,
3947         .ioctl =        packet_ioctl,
3948         .listen =       sock_no_listen,
3949         .shutdown =     sock_no_shutdown,
3950         .setsockopt =   packet_setsockopt,
3951         .getsockopt =   packet_getsockopt,
3952         .sendmsg =      packet_sendmsg,
3953         .recvmsg =      packet_recvmsg,
3954         .mmap =         packet_mmap,
3955         .sendpage =     sock_no_sendpage,
3956 };
3957
3958 static const struct net_proto_family packet_family_ops = {
3959         .family =       PF_PACKET,
3960         .create =       packet_create,
3961         .owner  =       THIS_MODULE,
3962 };
3963
3964 static struct notifier_block packet_netdev_notifier = {
3965         .notifier_call =        packet_notifier,
3966 };
3967
3968 #ifdef CONFIG_PROC_FS
3969
3970 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3971         __acquires(RCU)
3972 {
3973         struct net *net = seq_file_net(seq);
3974
3975         rcu_read_lock();
3976         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3977 }
3978
3979 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3980 {
3981         struct net *net = seq_file_net(seq);
3982         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3983 }
3984
3985 static void packet_seq_stop(struct seq_file *seq, void *v)
3986         __releases(RCU)
3987 {
3988         rcu_read_unlock();
3989 }
3990
3991 static int packet_seq_show(struct seq_file *seq, void *v)
3992 {
3993         if (v == SEQ_START_TOKEN)
3994                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
3995         else {
3996                 struct sock *s = sk_entry(v);
3997                 const struct packet_sock *po = pkt_sk(s);
3998
3999                 seq_printf(seq,
4000                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4001                            s,
4002                            atomic_read(&s->sk_refcnt),
4003                            s->sk_type,
4004                            ntohs(po->num),
4005                            po->ifindex,
4006                            po->running,
4007                            atomic_read(&s->sk_rmem_alloc),
4008                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4009                            sock_i_ino(s));
4010         }
4011
4012         return 0;
4013 }
4014
4015 static const struct seq_operations packet_seq_ops = {
4016         .start  = packet_seq_start,
4017         .next   = packet_seq_next,
4018         .stop   = packet_seq_stop,
4019         .show   = packet_seq_show,
4020 };
4021
4022 static int packet_seq_open(struct inode *inode, struct file *file)
4023 {
4024         return seq_open_net(inode, file, &packet_seq_ops,
4025                             sizeof(struct seq_net_private));
4026 }
4027
4028 static const struct file_operations packet_seq_fops = {
4029         .owner          = THIS_MODULE,
4030         .open           = packet_seq_open,
4031         .read           = seq_read,
4032         .llseek         = seq_lseek,
4033         .release        = seq_release_net,
4034 };
4035
4036 #endif
4037
4038 static int __net_init packet_net_init(struct net *net)
4039 {
4040         mutex_init(&net->packet.sklist_lock);
4041         INIT_HLIST_HEAD(&net->packet.sklist);
4042
4043         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4044                 return -ENOMEM;
4045
4046         return 0;
4047 }
4048
4049 static void __net_exit packet_net_exit(struct net *net)
4050 {
4051         remove_proc_entry("packet", net->proc_net);
4052 }
4053
4054 static struct pernet_operations packet_net_ops = {
4055         .init = packet_net_init,
4056         .exit = packet_net_exit,
4057 };
4058
4059
4060 static void __exit packet_exit(void)
4061 {
4062         unregister_netdevice_notifier(&packet_netdev_notifier);
4063         unregister_pernet_subsys(&packet_net_ops);
4064         sock_unregister(PF_PACKET);
4065         proto_unregister(&packet_proto);
4066 }
4067
4068 static int __init packet_init(void)
4069 {
4070         int rc = proto_register(&packet_proto, 0);
4071
4072         if (rc != 0)
4073                 goto out;
4074
4075         sock_register(&packet_family_ops);
4076         register_pernet_subsys(&packet_net_ops);
4077         register_netdevice_notifier(&packet_netdev_notifier);
4078 out:
4079         return rc;
4080 }
4081
4082 module_init(packet_init);
4083 module_exit(packet_exit);
4084 MODULE_LICENSE("GPL");
4085 MODULE_ALIAS_NETPROTO(PF_PACKET);