net/packet/af_packet.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              PACKET - implements raw packet sockets.
   7  *
   8  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  11  *
  12  * Fixes:
  13  *              Alan Cox        :       verify_area() now used correctly
  14  *              Alan Cox        :       new skbuff lists, look ma no backlogs!
  15  *              Alan Cox        :       tidied skbuff lists.
  16  *              Alan Cox        :       Now uses generic datagram routines I
  17  *                                      added. Also fixed the peek/read crash
  18  *                                      from all old Linux datagram code.
  19  *              Alan Cox        :       Uses the improved datagram code.
  20  *              Alan Cox        :       Added NULL's for socket options.
  21  *              Alan Cox        :       Re-commented the code.
  22  *              Alan Cox        :       Use new kernel side addressing
  23  *              Rob Janssen     :       Correct MTU usage.
  24  *              Dave Platt      :       Counter leaks caused by incorrect
  25  *                                      interrupt locking and some slightly
  26  *                                      dubious gcc output. Can you read
  27  *                                      compiler: it said _VOLATILE_
  28  *      Richard Kooijman        :       Timestamp fixes.
  29  *              Alan Cox        :       New buffers. Use sk->mac.raw.
  30  *              Alan Cox        :       sendmsg/recvmsg support.
  31  *              Alan Cox        :       Protocol setting support
  32  *      Alexey Kuznetsov        :       Untied from IPv4 stack.
  33  *      Cyrus Durgin            :       Fixed kerneld for kmod.
  34  *      Michal Ostrowski        :       Module initialization cleanup.
  35  *         Ulises Alonso        :       Frame number limit removal and
  36  *                                      packet_set_ring memory leak.
  37  *              Eric Biederman  :       Allow for > 8 byte hardware addresses.
  38  *                                      The convention is that longer addresses
  39  *                                      will simply extend the hardware address
  40  *                                      byte arrays at the end of sockaddr_ll
  41  *                                      and packet_mreq.
  42  *              Johann Baudy    :       Added TX RING.
  43  *              Chetan Loke     :       Implemented TPACKET_V3 block abstraction
  44  *                                      layer.
  45  *                                      Copyright (C) 2011, <lokec@ccs.neu.edu>
  46  *
  47  *
  48  *              This program is free software; you can redistribute it and/or
  49  *              modify it under the terms of the GNU General Public License
  50  *              as published by the Free Software Foundation; either version
  51  *              2 of the License, or (at your option) any later version.
  52  *
  53  */
  54
  55 #include <linux/types.h>
  56 #include <linux/mm.h>
  57 #include <linux/capability.h>
  58 #include <linux/fcntl.h>
  59 #include <linux/socket.h>
  60 #include <linux/in.h>
  61 #include <linux/inet.h>
  62 #include <linux/netdevice.h>
  63 #include <linux/if_packet.h>
  64 #include <linux/wireless.h>
  65 #include <linux/kernel.h>
  66 #include <linux/kmod.h>
  67 #include <linux/slab.h>
  68 #include <linux/vmalloc.h>
  69 #include <net/net_namespace.h>
  70 #include <net/ip.h>
  71 #include <net/protocol.h>
  72 #include <linux/skbuff.h>
  73 #include <net/sock.h>
  74 #include <linux/errno.h>
  75 #include <linux/timer.h>
  76 #include <asm/uaccess.h>
  77 #include <asm/ioctls.h>
  78 #include <asm/page.h>
  79 #include <asm/cacheflush.h>
  80 #include <asm/io.h>
  81 #include <linux/proc_fs.h>
  82 #include <linux/seq_file.h>
  83 #include <linux/poll.h>
  84 #include <linux/module.h>
  85 #include <linux/init.h>
  86 #include <linux/mutex.h>
  87 #include <linux/if_vlan.h>
  88 #include <linux/virtio_net.h>
  89 #include <linux/errqueue.h>
  90 #include <linux/net_tstamp.h>
  91 #include <linux/percpu.h>
  92 #ifdef CONFIG_INET
  93 #include <net/inet_common.h>
  94 #endif
  95
  96 #include "internal.h"
  97
  98 /*
  99    Assumptions:
 100    - if device has no dev->hard_header routine, it adds and removes ll header
 101      inside itself. In this case ll header is invisible outside of device,
 102      but higher levels still should reserve dev->hard_header_len.
 103      Some devices are enough clever to reallocate skb, when header
 104      will not fit to reserved space (tunnel), another ones are silly
 105      (PPP).
 106    - packet socket receives packets with pulled ll header,
 107      so that SOCK_RAW should push it back.
 108
 109 On receive:
 110 -----------
 111
 112 Incoming, dev->hard_header!=NULL
 113    mac_header -> ll header
 114    data       -> data
 115
 116 Outgoing, dev->hard_header!=NULL
 117    mac_header -> ll header
 118    data       -> ll header
 119
 120 Incoming, dev->hard_header==NULL
 121    mac_header -> UNKNOWN position. It is very likely, that it points to ll
 122                  header.  PPP makes it, that is wrong, because introduce
 123                  assymetry between rx and tx paths.
 124    data       -> data
 125
 126 Outgoing, dev->hard_header==NULL
 127    mac_header -> data. ll header is still not built!
 128    data       -> data
 129
 130 Resume
 131   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 132
 133
 134 On transmit:
 135 ------------
 136
 137 dev->hard_header != NULL
 138    mac_header -> ll header
 139    data       -> ll header
 140
 141 dev->hard_header == NULL (ll header is added by device, we cannot control it)
 142    mac_header -> data
 143    data       -> data
 144
 145    We should set nh.raw on output to correct posistion,
 146    packet classifier depends on it.
 147  */
 148
 149 /* Private packet socket structures. */
 150
 151 /* identical to struct packet_mreq except it has
 152  * a longer address field.
 153  */
 154 struct packet_mreq_max {
 155         int             mr_ifindex;
 156         unsigned short  mr_type;
 157         unsigned short  mr_alen;
 158         unsigned char   mr_address[MAX_ADDR_LEN];
 159 };
 160
 161 union tpacket_uhdr {
 162         struct tpacket_hdr  *h1;
 163         struct tpacket2_hdr *h2;
 164         struct tpacket3_hdr *h3;
 165         void *raw;
 166 };
 167
 168 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 169                 int closing, int tx_ring);
 170
 171 #define V3_ALIGNMENT    (8)
 172
 173 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 174
 175 #define BLK_PLUS_PRIV(sz_of_priv) \
 176         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 177
 178 #define PGV_FROM_VMALLOC 1
 179
 180 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
 181 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
 182 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
 183 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
 184 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
 185 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
 186 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
 187
 188 struct packet_sock;
 189 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
 190 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 191                        struct packet_type *pt, struct net_device *orig_dev);
 192
 193 static void *packet_previous_frame(struct packet_sock *po,
 194                 struct packet_ring_buffer *rb,
 195                 int status);
 196 static void packet_increment_head(struct packet_ring_buffer *buff);
 197 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
 198                         struct tpacket_block_desc *);
 199 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 200                         struct packet_sock *);
 201 static void prb_retire_current_block(struct tpacket_kbdq_core *,
 202                 struct packet_sock *, unsigned int status);
 203 static int prb_queue_frozen(struct tpacket_kbdq_core *);
 204 static void prb_open_block(struct tpacket_kbdq_core *,
 205                 struct tpacket_block_desc *);
 206 static void prb_retire_rx_blk_timer_expired(unsigned long);
 207 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 208 static void prb_init_blk_timer(struct packet_sock *,
 209                 struct tpacket_kbdq_core *,
 210                 void (*func) (unsigned long));
 211 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 212 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 213                 struct tpacket3_hdr *);
 214 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 215                 struct tpacket3_hdr *);
 216 static void packet_flush_mclist(struct sock *sk);
 217
 218 struct packet_skb_cb {
 219         union {
 220                 struct sockaddr_pkt pkt;
 221                 union {
 222                         /* Trick: alias skb original length with
 223                          * ll.sll_family and ll.protocol in order
 224                          * to save room.
 225                          */
 226                         unsigned int origlen;
 227                         struct sockaddr_ll ll;
 228                 };
 229         } sa;
 230 };
 231
 232 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 233
 234 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 235 #define GET_PBLOCK_DESC(x, bid) \
 236         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 237 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 238         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 239 #define GET_NEXT_PRB_BLK_NUM(x) \
 240         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 241         ((x)->kactive_blk_num+1) : 0)
 242
 243 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 244 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 245
 246 static int packet_direct_xmit(struct sk_buff *skb)
 247 {
 248         struct net_device *dev = skb->dev;
 249         netdev_features_t features;
 250         struct netdev_queue *txq;
 251         int ret = NETDEV_TX_BUSY;
 252
 253         if (unlikely(!netif_running(dev) ||
 254                      !netif_carrier_ok(dev)))
 255                 goto drop;
 256
 257         features = netif_skb_features(skb);
 258         if (skb_needs_linearize(skb, features) &&
 259             __skb_linearize(skb))
 260                 goto drop;
 261
 262         txq = skb_get_tx_queue(dev, skb);
 263
 264         local_bh_disable();
 265
 266         HARD_TX_LOCK(dev, txq, smp_processor_id());
 267         if (!netif_xmit_frozen_or_drv_stopped(txq))
 268                 ret = netdev_start_xmit(skb, dev, txq, false);
 269         HARD_TX_UNLOCK(dev, txq);
 270
 271         local_bh_enable();
 272
 273         if (!dev_xmit_complete(ret))
 274                 kfree_skb(skb);
 275
 276         return ret;
 277 drop:
 278         atomic_long_inc(&dev->tx_dropped);
 279         kfree_skb(skb);
 280         return NET_XMIT_DROP;
 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, struct sk_buff *skb)
 313 {
 314         return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
 315 }
 316
 317 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
 318 {
 319         const struct net_device_ops *ops = dev->netdev_ops;
 320         u16 queue_index;
 321
 322         if (ops->ndo_select_queue) {
 323                 queue_index = ops->ndo_select_queue(dev, skb, NULL,
 324                                                     __packet_pick_tx_queue);
 325                 queue_index = netdev_cap_txqueue(dev, queue_index);
 326         } else {
 327                 queue_index = __packet_pick_tx_queue(dev, skb);
 328         }
 329
 330         skb_set_queue_mapping(skb, queue_index);
 331 }
 332
 333 /* register_prot_hook must be invoked with the po->bind_lock held,
 334  * or from a context in which asynchronous accesses to the packet
 335  * socket is not possible (packet_create()).
 336  */
 337 static void register_prot_hook(struct sock *sk)
 338 {
 339         struct packet_sock *po = pkt_sk(sk);
 340
 341         if (!po->running) {
 342                 if (po->fanout)
 343                         __fanout_link(sk, po);
 344                 else
 345                         dev_add_pack(&po->prot_hook);
 346
 347                 sock_hold(sk);
 348                 po->running = 1;
 349         }
 350 }
 351
 352 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
 353  * held.   If the sync parameter is true, we will temporarily drop
 354  * the po->bind_lock and do a synchronize_net to make sure no
 355  * asynchronous packet processing paths still refer to the elements
 356  * of po->prot_hook.  If the sync parameter is false, it is the
 357  * callers responsibility to take care of this.
 358  */
 359 static void __unregister_prot_hook(struct sock *sk, bool sync)
 360 {
 361         struct packet_sock *po = pkt_sk(sk);
 362
 363         po->running = 0;
 364
 365         if (po->fanout)
 366                 __fanout_unlink(sk, po);
 367         else
 368                 __dev_remove_pack(&po->prot_hook);
 369
 370         __sock_put(sk);
 371
 372         if (sync) {
 373                 spin_unlock(&po->bind_lock);
 374                 synchronize_net();
 375                 spin_lock(&po->bind_lock);
 376         }
 377 }
 378
 379 static void unregister_prot_hook(struct sock *sk, bool sync)
 380 {
 381         struct packet_sock *po = pkt_sk(sk);
 382
 383         if (po->running)
 384                 __unregister_prot_hook(sk, sync);
 385 }
 386
 387 static inline struct page * __pure pgv_to_page(void *addr)
 388 {
 389         if (is_vmalloc_addr(addr))
 390                 return vmalloc_to_page(addr);
 391         return virt_to_page(addr);
 392 }
 393
 394 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 395 {
 396         union tpacket_uhdr h;
 397
 398         h.raw = frame;
 399         switch (po->tp_version) {
 400         case TPACKET_V1:
 401                 h.h1->tp_status = status;
 402                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 403                 break;
 404         case TPACKET_V2:
 405                 h.h2->tp_status = status;
 406                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 407                 break;
 408         case TPACKET_V3:
 409         default:
 410                 WARN(1, "TPACKET version not supported.\n");
 411                 BUG();
 412         }
 413
 414         smp_wmb();
 415 }
 416
 417 static int __packet_get_status(struct packet_sock *po, void *frame)
 418 {
 419         union tpacket_uhdr h;
 420
 421         smp_rmb();
 422
 423         h.raw = frame;
 424         switch (po->tp_version) {
 425         case TPACKET_V1:
 426                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 427                 return h.h1->tp_status;
 428         case TPACKET_V2:
 429                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 430                 return h.h2->tp_status;
 431         case TPACKET_V3:
 432         default:
 433                 WARN(1, "TPACKET version not supported.\n");
 434                 BUG();
 435                 return 0;
 436         }
 437 }
 438
 439 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
 440                                    unsigned int flags)
 441 {
 442         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 443
 444         if (shhwtstamps &&
 445             (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
 446             ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
 447                 return TP_STATUS_TS_RAW_HARDWARE;
 448
 449         if (ktime_to_timespec_cond(skb->tstamp, ts))
 450                 return TP_STATUS_TS_SOFTWARE;
 451
 452         return 0;
 453 }
 454
 455 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
 456                                     struct sk_buff *skb)
 457 {
 458         union tpacket_uhdr h;
 459         struct timespec ts;
 460         __u32 ts_status;
 461
 462         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
 463                 return 0;
 464
 465         h.raw = frame;
 466         switch (po->tp_version) {
 467         case TPACKET_V1:
 468                 h.h1->tp_sec = ts.tv_sec;
 469                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
 470                 break;
 471         case TPACKET_V2:
 472                 h.h2->tp_sec = ts.tv_sec;
 473                 h.h2->tp_nsec = ts.tv_nsec;
 474                 break;
 475         case TPACKET_V3:
 476         default:
 477                 WARN(1, "TPACKET version not supported.\n");
 478                 BUG();
 479         }
 480
 481         /* one flush is safe, as both fields always lie on the same cacheline */
 482         flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
 483         smp_wmb();
 484
 485         return ts_status;
 486 }
 487
 488 static void *packet_lookup_frame(struct packet_sock *po,
 489                 struct packet_ring_buffer *rb,
 490                 unsigned int position,
 491                 int status)
 492 {
 493         unsigned int pg_vec_pos, frame_offset;
 494         union tpacket_uhdr h;
 495
 496         pg_vec_pos = position / rb->frames_per_block;
 497         frame_offset = position % rb->frames_per_block;
 498
 499         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 500                 (frame_offset * rb->frame_size);
 501
 502         if (status != __packet_get_status(po, h.raw))
 503                 return NULL;
 504
 505         return h.raw;
 506 }
 507
 508 static void *packet_current_frame(struct packet_sock *po,
 509                 struct packet_ring_buffer *rb,
 510                 int status)
 511 {
 512         return packet_lookup_frame(po, rb, rb->head, status);
 513 }
 514
 515 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 516 {
 517         del_timer_sync(&pkc->retire_blk_timer);
 518 }
 519
 520 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 521                 int tx_ring,
 522                 struct sk_buff_head *rb_queue)
 523 {
 524         struct tpacket_kbdq_core *pkc;
 525
 526         pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
 527                         GET_PBDQC_FROM_RB(&po->rx_ring);
 528
 529         spin_lock_bh(&rb_queue->lock);
 530         pkc->delete_blk_timer = 1;
 531         spin_unlock_bh(&rb_queue->lock);
 532
 533         prb_del_retire_blk_timer(pkc);
 534 }
 535
 536 static void prb_init_blk_timer(struct packet_sock *po,
 537                 struct tpacket_kbdq_core *pkc,
 538                 void (*func) (unsigned long))
 539 {
 540         init_timer(&pkc->retire_blk_timer);
 541         pkc->retire_blk_timer.data = (long)po;
 542         pkc->retire_blk_timer.function = func;
 543         pkc->retire_blk_timer.expires = jiffies;
 544 }
 545
 546 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
 547 {
 548         struct tpacket_kbdq_core *pkc;
 549
 550         if (tx_ring)
 551                 BUG();
 552
 553         pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
 554                         GET_PBDQC_FROM_RB(&po->rx_ring);
 555         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
 556 }
 557
 558 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 559                                 int blk_size_in_bytes)
 560 {
 561         struct net_device *dev;
 562         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 563         struct ethtool_cmd ecmd;
 564         int err;
 565         u32 speed;
 566
 567         rtnl_lock();
 568         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 569         if (unlikely(!dev)) {
 570                 rtnl_unlock();
 571                 return DEFAULT_PRB_RETIRE_TOV;
 572         }
 573         err = __ethtool_get_settings(dev, &ecmd);
 574         speed = ethtool_cmd_speed(&ecmd);
 575         rtnl_unlock();
 576         if (!err) {
 577                 /*
 578                  * If the link speed is so slow you don't really
 579                  * need to worry about perf anyways
 580                  */
 581                 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
 582                         return DEFAULT_PRB_RETIRE_TOV;
 583                 } else {
 584                         msec = 1;
 585                         div = speed / 1000;
 586                 }
 587         }
 588
 589         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 590
 591         if (div)
 592                 mbits /= div;
 593
 594         tmo = mbits * msec;
 595
 596         if (div)
 597                 return tmo+1;
 598         return tmo;
 599 }
 600
 601 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 602                         union tpacket_req_u *req_u)
 603 {
 604         p1->feature_req_word = req_u->req3.tp_feature_req_word;
 605 }
 606
 607 static void init_prb_bdqc(struct packet_sock *po,
 608                         struct packet_ring_buffer *rb,
 609                         struct pgv *pg_vec,
 610                         union tpacket_req_u *req_u, int tx_ring)
 611 {
 612         struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
 613         struct tpacket_block_desc *pbd;
 614
 615         memset(p1, 0x0, sizeof(*p1));
 616
 617         p1->knxt_seq_num = 1;
 618         p1->pkbdq = pg_vec;
 619         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 620         p1->pkblk_start = pg_vec[0].buffer;
 621         p1->kblk_size = req_u->req3.tp_block_size;
 622         p1->knum_blocks = req_u->req3.tp_block_nr;
 623         p1->hdrlen = po->tp_hdrlen;
 624         p1->version = po->tp_version;
 625         p1->last_kactive_blk_num = 0;
 626         po->stats.stats3.tp_freeze_q_cnt = 0;
 627         if (req_u->req3.tp_retire_blk_tov)
 628                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 629         else
 630                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 631                                                 req_u->req3.tp_block_size);
 632         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 633         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 634
 635         p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
 636         prb_init_ft_ops(p1, req_u);
 637         prb_setup_retire_blk_timer(po, tx_ring);
 638         prb_open_block(p1, pbd);
 639 }
 640
 641 /*  Do NOT update the last_blk_num first.
 642  *  Assumes sk_buff_head lock is held.
 643  */
 644 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 645 {
 646         mod_timer(&pkc->retire_blk_timer,
 647                         jiffies + pkc->tov_in_jiffies);
 648         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 649 }
 650
 651 /*
 652  * Timer logic:
 653  * 1) We refresh the timer only when we open a block.
 654  *    By doing this we don't waste cycles refreshing the timer
 655  *        on packet-by-packet basis.
 656  *
 657  * With a 1MB block-size, on a 1Gbps line, it will take
 658  * i) ~8 ms to fill a block + ii) memcpy etc.
 659  * In this cut we are not accounting for the memcpy time.
 660  *
 661  * So, if the user sets the 'tmo' to 10ms then the timer
 662  * will never fire while the block is still getting filled
 663  * (which is what we want). However, the user could choose
 664  * to close a block early and that's fine.
 665  *
 666  * But when the timer does fire, we check whether or not to refresh it.
 667  * Since the tmo granularity is in msecs, it is not too expensive
 668  * to refresh the timer, lets say every '8' msecs.
 669  * Either the user can set the 'tmo' or we can derive it based on
 670  * a) line-speed and b) block-size.
 671  * prb_calc_retire_blk_tmo() calculates the tmo.
 672  *
 673  */
 674 static void prb_retire_rx_blk_timer_expired(unsigned long data)
 675 {
 676         struct packet_sock *po = (struct packet_sock *)data;
 677         struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 678         unsigned int frozen;
 679         struct tpacket_block_desc *pbd;
 680
 681         spin_lock(&po->sk.sk_receive_queue.lock);
 682
 683         frozen = prb_queue_frozen(pkc);
 684         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 685
 686         if (unlikely(pkc->delete_blk_timer))
 687                 goto out;
 688
 689         /* We only need to plug the race when the block is partially filled.
 690          * tpacket_rcv:
 691          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 692          *              copy_bits() is in progress ...
 693          *              timer fires on other cpu:
 694          *              we can't retire the current block because copy_bits
 695          *              is in progress.
 696          *
 697          */
 698         if (BLOCK_NUM_PKTS(pbd)) {
 699                 while (atomic_read(&pkc->blk_fill_in_prog)) {
 700                         /* Waiting for skb_copy_bits to finish... */
 701                         cpu_relax();
 702                 }
 703         }
 704
 705         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 706                 if (!frozen) {
 707                         if (!BLOCK_NUM_PKTS(pbd)) {
 708                                 /* An empty block. Just refresh the timer. */
 709                                 goto refresh_timer;
 710                         }
 711                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 712                         if (!prb_dispatch_next_block(pkc, po))
 713                                 goto refresh_timer;
 714                         else
 715                                 goto out;
 716                 } else {
 717                         /* Case 1. Queue was frozen because user-space was
 718                          *         lagging behind.
 719                          */
 720                         if (prb_curr_blk_in_use(pkc, pbd)) {
 721                                 /*
 722                                  * Ok, user-space is still behind.
 723                                  * So just refresh the timer.
 724                                  */
 725                                 goto refresh_timer;
 726                         } else {
 727                                /* Case 2. queue was frozen,user-space caught up,
 728                                 * now the link went idle && the timer fired.
 729                                 * We don't have a block to close.So we open this
 730                                 * block and restart the timer.
 731                                 * opening a block thaws the queue,restarts timer
 732                                 * Thawing/timer-refresh is a side effect.
 733                                 */
 734                                 prb_open_block(pkc, pbd);
 735                                 goto out;
 736                         }
 737                 }
 738         }
 739
 740 refresh_timer:
 741         _prb_refresh_rx_retire_blk_timer(pkc);
 742
 743 out:
 744         spin_unlock(&po->sk.sk_receive_queue.lock);
 745 }
 746
 747 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 748                 struct tpacket_block_desc *pbd1, __u32 status)
 749 {
 750         /* Flush everything minus the block header */
 751
 752 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 753         u8 *start, *end;
 754
 755         start = (u8 *)pbd1;
 756
 757         /* Skip the block header(we know header WILL fit in 4K) */
 758         start += PAGE_SIZE;
 759
 760         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 761         for (; start < end; start += PAGE_SIZE)
 762                 flush_dcache_page(pgv_to_page(start));
 763
 764         smp_wmb();
 765 #endif
 766
 767         /* Now update the block status. */
 768
 769         BLOCK_STATUS(pbd1) = status;
 770
 771         /* Flush the block header */
 772
 773 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 774         start = (u8 *)pbd1;
 775         flush_dcache_page(pgv_to_page(start));
 776
 777         smp_wmb();
 778 #endif
 779 }
 780
 781 /*
 782  * Side effect:
 783  *
 784  * 1) flush the block
 785  * 2) Increment active_blk_num
 786  *
 787  * Note:We DONT refresh the timer on purpose.
 788  *      Because almost always the next block will be opened.
 789  */
 790 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 791                 struct tpacket_block_desc *pbd1,
 792                 struct packet_sock *po, unsigned int stat)
 793 {
 794         __u32 status = TP_STATUS_USER | stat;
 795
 796         struct tpacket3_hdr *last_pkt;
 797         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 798         struct sock *sk = &po->sk;
 799
 800         if (po->stats.stats3.tp_drops)
 801                 status |= TP_STATUS_LOSING;
 802
 803         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 804         last_pkt->tp_next_offset = 0;
 805
 806         /* Get the ts of the last pkt */
 807         if (BLOCK_NUM_PKTS(pbd1)) {
 808                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 809                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 810         } else {
 811                 /* Ok, we tmo'd - so get the current time.
 812                  *
 813                  * It shouldn't really happen as we don't close empty
 814                  * blocks. See prb_retire_rx_blk_timer_expired().
 815                  */
 816                 struct timespec ts;
 817                 getnstimeofday(&ts);
 818                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 819                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 820         }
 821
 822         smp_wmb();
 823
 824         /* Flush the block */
 825         prb_flush_block(pkc1, pbd1, status);
 826
 827         sk->sk_data_ready(sk);
 828
 829         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 830 }
 831
 832 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 833 {
 834         pkc->reset_pending_on_curr_blk = 0;
 835 }
 836
 837 /*
 838  * Side effect of opening a block:
 839  *
 840  * 1) prb_queue is thawed.
 841  * 2) retire_blk_timer is refreshed.
 842  *
 843  */
 844 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 845         struct tpacket_block_desc *pbd1)
 846 {
 847         struct timespec ts;
 848         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 849
 850         smp_rmb();
 851
 852         /* We could have just memset this but we will lose the
 853          * flexibility of making the priv area sticky
 854          */
 855
 856         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 857         BLOCK_NUM_PKTS(pbd1) = 0;
 858         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 859
 860         getnstimeofday(&ts);
 861
 862         h1->ts_first_pkt.ts_sec = ts.tv_sec;
 863         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 864
 865         pkc1->pkblk_start = (char *)pbd1;
 866         pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 867
 868         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 869         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 870
 871         pbd1->version = pkc1->version;
 872         pkc1->prev = pkc1->nxt_offset;
 873         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 874
 875         prb_thaw_queue(pkc1);
 876         _prb_refresh_rx_retire_blk_timer(pkc1);
 877
 878         smp_wmb();
 879 }
 880
 881 /*
 882  * Queue freeze logic:
 883  * 1) Assume tp_block_nr = 8 blocks.
 884  * 2) At time 't0', user opens Rx ring.
 885  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 886  * 4) user-space is either sleeping or processing block '0'.
 887  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 888  *    it will close block-7,loop around and try to fill block '0'.
 889  *    call-flow:
 890  *    __packet_lookup_frame_in_block
 891  *      prb_retire_current_block()
 892  *      prb_dispatch_next_block()
 893  *        |->(BLOCK_STATUS == USER) evaluates to true
 894  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 895  * 6) Now there are two cases:
 896  *    6.1) Link goes idle right after the queue is frozen.
 897  *         But remember, the last open_block() refreshed the timer.
 898  *         When this timer expires,it will refresh itself so that we can
 899  *         re-open block-0 in near future.
 900  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 901  *         case and __packet_lookup_frame_in_block will check if block-0
 902  *         is free and can now be re-used.
 903  */
 904 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 905                                   struct packet_sock *po)
 906 {
 907         pkc->reset_pending_on_curr_blk = 1;
 908         po->stats.stats3.tp_freeze_q_cnt++;
 909 }
 910
 911 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 912
 913 /*
 914  * If the next block is free then we will dispatch it
 915  * and return a good offset.
 916  * Else, we will freeze the queue.
 917  * So, caller must check the return value.
 918  */
 919 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 920                 struct packet_sock *po)
 921 {
 922         struct tpacket_block_desc *pbd;
 923
 924         smp_rmb();
 925
 926         /* 1. Get current block num */
 927         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 928
 929         /* 2. If this block is currently in_use then freeze the queue */
 930         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 931                 prb_freeze_queue(pkc, po);
 932                 return NULL;
 933         }
 934
 935         /*
 936          * 3.
 937          * open this block and return the offset where the first packet
 938          * needs to get stored.
 939          */
 940         prb_open_block(pkc, pbd);
 941         return (void *)pkc->nxt_offset;
 942 }
 943
 944 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 945                 struct packet_sock *po, unsigned int status)
 946 {
 947         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 948
 949         /* retire/close the current block */
 950         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 951                 /*
 952                  * Plug the case where copy_bits() is in progress on
 953                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 954                  * have space to copy the pkt in the current block and
 955                  * called prb_retire_current_block()
 956                  *
 957                  * We don't need to worry about the TMO case because
 958                  * the timer-handler already handled this case.
 959                  */
 960                 if (!(status & TP_STATUS_BLK_TMO)) {
 961                         while (atomic_read(&pkc->blk_fill_in_prog)) {
 962                                 /* Waiting for skb_copy_bits to finish... */
 963                                 cpu_relax();
 964                         }
 965                 }
 966                 prb_close_block(pkc, pbd, po, status);
 967                 return;
 968         }
 969 }
 970
 971 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
 972                                       struct tpacket_block_desc *pbd)
 973 {
 974         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 975 }
 976
 977 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 978 {
 979         return pkc->reset_pending_on_curr_blk;
 980 }
 981
 982 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 983 {
 984         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 985         atomic_dec(&pkc->blk_fill_in_prog);
 986 }
 987
 988 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 989                         struct tpacket3_hdr *ppd)
 990 {
 991         ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
 992 }
 993
 994 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 995                         struct tpacket3_hdr *ppd)
 996 {
 997         ppd->hv1.tp_rxhash = 0;
 998 }
 999
1000 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1001                         struct tpacket3_hdr *ppd)
1002 {
1003         if (skb_vlan_tag_present(pkc->skb)) {
1004                 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1005                 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1006                 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1007         } else {
1008                 ppd->hv1.tp_vlan_tci = 0;
1009                 ppd->hv1.tp_vlan_tpid = 0;
1010                 ppd->tp_status = TP_STATUS_AVAILABLE;
1011         }
1012 }
1013
1014 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1015                         struct tpacket3_hdr *ppd)
1016 {
1017         ppd->hv1.tp_padding = 0;
1018         prb_fill_vlan_info(pkc, ppd);
1019
1020         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1021                 prb_fill_rxhash(pkc, ppd);
1022         else
1023                 prb_clear_rxhash(pkc, ppd);
1024 }
1025
1026 static void prb_fill_curr_block(char *curr,
1027                                 struct tpacket_kbdq_core *pkc,
1028                                 struct tpacket_block_desc *pbd,
1029                                 unsigned int len)
1030 {
1031         struct tpacket3_hdr *ppd;
1032
1033         ppd  = (struct tpacket3_hdr *)curr;
1034         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1035         pkc->prev = curr;
1036         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1037         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1038         BLOCK_NUM_PKTS(pbd) += 1;
1039         atomic_inc(&pkc->blk_fill_in_prog);
1040         prb_run_all_ft_ops(pkc, ppd);
1041 }
1042
1043 /* Assumes caller has the sk->rx_queue.lock */
1044 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1045                                             struct sk_buff *skb,
1046                                                 int status,
1047                                             unsigned int len
1048                                             )
1049 {
1050         struct tpacket_kbdq_core *pkc;
1051         struct tpacket_block_desc *pbd;
1052         char *curr, *end;
1053
1054         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1055         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1056
1057         /* Queue is frozen when user space is lagging behind */
1058         if (prb_queue_frozen(pkc)) {
1059                 /*
1060                  * Check if that last block which caused the queue to freeze,
1061                  * is still in_use by user-space.
1062                  */
1063                 if (prb_curr_blk_in_use(pkc, pbd)) {
1064                         /* Can't record this packet */
1065                         return NULL;
1066                 } else {
1067                         /*
1068                          * Ok, the block was released by user-space.
1069                          * Now let's open that block.
1070                          * opening a block also thaws the queue.
1071                          * Thawing is a side effect.
1072                          */
1073                         prb_open_block(pkc, pbd);
1074                 }
1075         }
1076
1077         smp_mb();
1078         curr = pkc->nxt_offset;
1079         pkc->skb = skb;
1080         end = (char *)pbd + pkc->kblk_size;
1081
1082         /* first try the current block */
1083         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1084                 prb_fill_curr_block(curr, pkc, pbd, len);
1085                 return (void *)curr;
1086         }
1087
1088         /* Ok, close the current block */
1089         prb_retire_current_block(pkc, po, 0);
1090
1091         /* Now, try to dispatch the next block */
1092         curr = (char *)prb_dispatch_next_block(pkc, po);
1093         if (curr) {
1094                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1095                 prb_fill_curr_block(curr, pkc, pbd, len);
1096                 return (void *)curr;
1097         }
1098
1099         /*
1100          * No free blocks are available.user_space hasn't caught up yet.
1101          * Queue was just frozen and now this packet will get dropped.
1102          */
1103         return NULL;
1104 }
1105
1106 static void *packet_current_rx_frame(struct packet_sock *po,
1107                                             struct sk_buff *skb,
1108                                             int status, unsigned int len)
1109 {
1110         char *curr = NULL;
1111         switch (po->tp_version) {
1112         case TPACKET_V1:
1113         case TPACKET_V2:
1114                 curr = packet_lookup_frame(po, &po->rx_ring,
1115                                         po->rx_ring.head, status);
1116                 return curr;
1117         case TPACKET_V3:
1118                 return __packet_lookup_frame_in_block(po, skb, status, len);
1119         default:
1120                 WARN(1, "TPACKET version not supported\n");
1121                 BUG();
1122                 return NULL;
1123         }
1124 }
1125
1126 static void *prb_lookup_block(struct packet_sock *po,
1127                                      struct packet_ring_buffer *rb,
1128                                      unsigned int idx,
1129                                      int status)
1130 {
1131         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1132         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1133
1134         if (status != BLOCK_STATUS(pbd))
1135                 return NULL;
1136         return pbd;
1137 }
1138
1139 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1140 {
1141         unsigned int prev;
1142         if (rb->prb_bdqc.kactive_blk_num)
1143                 prev = rb->prb_bdqc.kactive_blk_num-1;
1144         else
1145                 prev = rb->prb_bdqc.knum_blocks-1;
1146         return prev;
1147 }
1148
1149 /* Assumes caller has held the rx_queue.lock */
1150 static void *__prb_previous_block(struct packet_sock *po,
1151                                          struct packet_ring_buffer *rb,
1152                                          int status)
1153 {
1154         unsigned int previous = prb_previous_blk_num(rb);
1155         return prb_lookup_block(po, rb, previous, status);
1156 }
1157
1158 static void *packet_previous_rx_frame(struct packet_sock *po,
1159                                              struct packet_ring_buffer *rb,
1160                                              int status)
1161 {
1162         if (po->tp_version <= TPACKET_V2)
1163                 return packet_previous_frame(po, rb, status);
1164
1165         return __prb_previous_block(po, rb, status);
1166 }
1167
1168 static void packet_increment_rx_head(struct packet_sock *po,
1169                                             struct packet_ring_buffer *rb)
1170 {
1171         switch (po->tp_version) {
1172         case TPACKET_V1:
1173         case TPACKET_V2:
1174                 return packet_increment_head(rb);
1175         case TPACKET_V3:
1176         default:
1177                 WARN(1, "TPACKET version not supported.\n");
1178                 BUG();
1179                 return;
1180         }
1181 }
1182
1183 static void *packet_previous_frame(struct packet_sock *po,
1184                 struct packet_ring_buffer *rb,
1185                 int status)
1186 {
1187         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1188         return packet_lookup_frame(po, rb, previous, status);
1189 }
1190
1191 static void packet_increment_head(struct packet_ring_buffer *buff)
1192 {
1193         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1194 }
1195
1196 static void packet_inc_pending(struct packet_ring_buffer *rb)
1197 {
1198         this_cpu_inc(*rb->pending_refcnt);
1199 }
1200
1201 static void packet_dec_pending(struct packet_ring_buffer *rb)
1202 {
1203         this_cpu_dec(*rb->pending_refcnt);
1204 }
1205
1206 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1207 {
1208         unsigned int refcnt = 0;
1209         int cpu;
1210
1211         /* We don't use pending refcount in rx_ring. */
1212         if (rb->pending_refcnt == NULL)
1213                 return 0;
1214
1215         for_each_possible_cpu(cpu)
1216                 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1217
1218         return refcnt;
1219 }
1220
1221 static int packet_alloc_pending(struct packet_sock *po)
1222 {
1223         po->rx_ring.pending_refcnt = NULL;
1224
1225         po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1226         if (unlikely(po->tx_ring.pending_refcnt == NULL))
1227                 return -ENOBUFS;
1228
1229         return 0;
1230 }
1231
1232 static void packet_free_pending(struct packet_sock *po)
1233 {
1234         free_percpu(po->tx_ring.pending_refcnt);
1235 }
1236
1237 static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1238 {
1239         struct sock *sk = &po->sk;
1240         bool has_room;
1241
1242         if (po->prot_hook.func != tpacket_rcv)
1243                 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1244                         <= sk->sk_rcvbuf;
1245
1246         spin_lock(&sk->sk_receive_queue.lock);
1247         if (po->tp_version == TPACKET_V3)
1248                 has_room = prb_lookup_block(po, &po->rx_ring,
1249                                             po->rx_ring.prb_bdqc.kactive_blk_num,
1250                                             TP_STATUS_KERNEL);
1251         else
1252                 has_room = packet_lookup_frame(po, &po->rx_ring,
1253                                                po->rx_ring.head,
1254                                                TP_STATUS_KERNEL);
1255         spin_unlock(&sk->sk_receive_queue.lock);
1256
1257         return has_room;
1258 }
1259
1260 static void packet_sock_destruct(struct sock *sk)
1261 {
1262         skb_queue_purge(&sk->sk_error_queue);
1263
1264         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1265         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1266
1267         if (!sock_flag(sk, SOCK_DEAD)) {
1268                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1269                 return;
1270         }
1271
1272         sk_refcnt_debug_dec(sk);
1273 }
1274
1275 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1276 {
1277         int x = atomic_read(&f->rr_cur) + 1;
1278
1279         if (x >= num)
1280                 x = 0;
1281
1282         return x;
1283 }
1284
1285 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1286                                       struct sk_buff *skb,
1287                                       unsigned int num)
1288 {
1289         return reciprocal_scale(skb_get_hash(skb), num);
1290 }
1291
1292 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1293                                     struct sk_buff *skb,
1294                                     unsigned int num)
1295 {
1296         int cur, old;
1297
1298         cur = atomic_read(&f->rr_cur);
1299         while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1300                                      fanout_rr_next(f, num))) != cur)
1301                 cur = old;
1302         return cur;
1303 }
1304
1305 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1306                                      struct sk_buff *skb,
1307                                      unsigned int num)
1308 {
1309         return smp_processor_id() % num;
1310 }
1311
1312 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1313                                      struct sk_buff *skb,
1314                                      unsigned int num)
1315 {
1316         return prandom_u32_max(num);
1317 }
1318
1319 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1320                                           struct sk_buff *skb,
1321                                           unsigned int idx, unsigned int skip,
1322                                           unsigned int num)
1323 {
1324         unsigned int i, j;
1325
1326         i = j = min_t(int, f->next[idx], num - 1);
1327         do {
1328                 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1329                         if (i != j)
1330                                 f->next[idx] = i;
1331                         return i;
1332                 }
1333                 if (++i == num)
1334                         i = 0;
1335         } while (i != j);
1336
1337         return idx;
1338 }
1339
1340 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1341                                     struct sk_buff *skb,
1342                                     unsigned int num)
1343 {
1344         return skb_get_queue_mapping(skb) % num;
1345 }
1346
1347 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1348 {
1349         return f->flags & (flag >> 8);
1350 }
1351
1352 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1353                              struct packet_type *pt, struct net_device *orig_dev)
1354 {
1355         struct packet_fanout *f = pt->af_packet_priv;
1356         unsigned int num = f->num_members;
1357         struct packet_sock *po;
1358         unsigned int idx;
1359
1360         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1361             !num) {
1362                 kfree_skb(skb);
1363                 return 0;
1364         }
1365
1366         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1367                 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1368                 if (!skb)
1369                         return 0;
1370         }
1371         switch (f->type) {
1372         case PACKET_FANOUT_HASH:
1373         default:
1374                 idx = fanout_demux_hash(f, skb, num);
1375                 break;
1376         case PACKET_FANOUT_LB:
1377                 idx = fanout_demux_lb(f, skb, num);
1378                 break;
1379         case PACKET_FANOUT_CPU:
1380                 idx = fanout_demux_cpu(f, skb, num);
1381                 break;
1382         case PACKET_FANOUT_RND:
1383                 idx = fanout_demux_rnd(f, skb, num);
1384                 break;
1385         case PACKET_FANOUT_QM:
1386                 idx = fanout_demux_qm(f, skb, num);
1387                 break;
1388         case PACKET_FANOUT_ROLLOVER:
1389                 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1390                 break;
1391         }
1392
1393         po = pkt_sk(f->arr[idx]);
1394         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1395             unlikely(!packet_rcv_has_room(po, skb))) {
1396                 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1397                 po = pkt_sk(f->arr[idx]);
1398         }
1399
1400         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1401 }
1402
1403 DEFINE_MUTEX(fanout_mutex);
1404 EXPORT_SYMBOL_GPL(fanout_mutex);
1405 static LIST_HEAD(fanout_list);
1406
1407 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1408 {
1409         struct packet_fanout *f = po->fanout;
1410
1411         spin_lock(&f->lock);
1412         f->arr[f->num_members] = sk;
1413         smp_wmb();
1414         f->num_members++;
1415         spin_unlock(&f->lock);
1416 }
1417
1418 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1419 {
1420         struct packet_fanout *f = po->fanout;
1421         int i;
1422
1423         spin_lock(&f->lock);
1424         for (i = 0; i < f->num_members; i++) {
1425                 if (f->arr[i] == sk)
1426                         break;
1427         }
1428         BUG_ON(i >= f->num_members);
1429         f->arr[i] = f->arr[f->num_members - 1];
1430         f->num_members--;
1431         spin_unlock(&f->lock);
1432 }
1433
1434 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1435 {
1436         if (ptype->af_packet_priv == (void *)((struct packet_sock *)sk)->fanout)
1437                 return true;
1438
1439         return false;
1440 }
1441
1442 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1443 {
1444         struct packet_sock *po = pkt_sk(sk);
1445         struct packet_fanout *f, *match;
1446         u8 type = type_flags & 0xff;
1447         u8 flags = type_flags >> 8;
1448         int err;
1449
1450         switch (type) {
1451         case PACKET_FANOUT_ROLLOVER:
1452                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1453                         return -EINVAL;
1454         case PACKET_FANOUT_HASH:
1455         case PACKET_FANOUT_LB:
1456         case PACKET_FANOUT_CPU:
1457         case PACKET_FANOUT_RND:
1458         case PACKET_FANOUT_QM:
1459                 break;
1460         default:
1461                 return -EINVAL;
1462         }
1463
1464         if (!po->running)
1465                 return -EINVAL;
1466
1467         if (po->fanout)
1468                 return -EALREADY;
1469
1470         mutex_lock(&fanout_mutex);
1471         match = NULL;
1472         list_for_each_entry(f, &fanout_list, list) {
1473                 if (f->id == id &&
1474                     read_pnet(&f->net) == sock_net(sk)) {
1475                         match = f;
1476                         break;
1477                 }
1478         }
1479         err = -EINVAL;
1480         if (match && match->flags != flags)
1481                 goto out;
1482         if (!match) {
1483                 err = -ENOMEM;
1484                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1485                 if (!match)
1486                         goto out;
1487                 write_pnet(&match->net, sock_net(sk));
1488                 match->id = id;
1489                 match->type = type;
1490                 match->flags = flags;
1491                 atomic_set(&match->rr_cur, 0);
1492                 INIT_LIST_HEAD(&match->list);
1493                 spin_lock_init(&match->lock);
1494                 atomic_set(&match->sk_ref, 0);
1495                 match->prot_hook.type = po->prot_hook.type;
1496                 match->prot_hook.dev = po->prot_hook.dev;
1497                 match->prot_hook.func = packet_rcv_fanout;
1498                 match->prot_hook.af_packet_priv = match;
1499                 match->prot_hook.id_match = match_fanout_group;
1500                 dev_add_pack(&match->prot_hook);
1501                 list_add(&match->list, &fanout_list);
1502         }
1503         err = -EINVAL;
1504         if (match->type == type &&
1505             match->prot_hook.type == po->prot_hook.type &&
1506             match->prot_hook.dev == po->prot_hook.dev) {
1507                 err = -ENOSPC;
1508                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1509                         __dev_remove_pack(&po->prot_hook);
1510                         po->fanout = match;
1511                         atomic_inc(&match->sk_ref);
1512                         __fanout_link(sk, po);
1513                         err = 0;
1514                 }
1515         }
1516 out:
1517         mutex_unlock(&fanout_mutex);
1518         return err;
1519 }
1520
1521 static void fanout_release(struct sock *sk)
1522 {
1523         struct packet_sock *po = pkt_sk(sk);
1524         struct packet_fanout *f;
1525
1526         f = po->fanout;
1527         if (!f)
1528                 return;
1529
1530         mutex_lock(&fanout_mutex);
1531         po->fanout = NULL;
1532
1533         if (atomic_dec_and_test(&f->sk_ref)) {
1534                 list_del(&f->list);
1535                 dev_remove_pack(&f->prot_hook);
1536                 kfree(f);
1537         }
1538         mutex_unlock(&fanout_mutex);
1539 }
1540
1541 static const struct proto_ops packet_ops;
1542
1543 static const struct proto_ops packet_ops_spkt;
1544
1545 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1546                            struct packet_type *pt, struct net_device *orig_dev)
1547 {
1548         struct sock *sk;
1549         struct sockaddr_pkt *spkt;
1550
1551         /*
1552          *      When we registered the protocol we saved the socket in the data
1553          *      field for just this event.
1554          */
1555
1556         sk = pt->af_packet_priv;
1557
1558         /*
1559          *      Yank back the headers [hope the device set this
1560          *      right or kerboom...]
1561          *
1562          *      Incoming packets have ll header pulled,
1563          *      push it back.
1564          *
1565          *      For outgoing ones skb->data == skb_mac_header(skb)
1566          *      so that this procedure is noop.
1567          */
1568
1569         if (skb->pkt_type == PACKET_LOOPBACK)
1570                 goto out;
1571
1572         if (!net_eq(dev_net(dev), sock_net(sk)))
1573                 goto out;
1574
1575         skb = skb_share_check(skb, GFP_ATOMIC);
1576         if (skb == NULL)
1577                 goto oom;
1578
1579         /* drop any routing info */
1580         skb_dst_drop(skb);
1581
1582         /* drop conntrack reference */
1583         nf_reset(skb);
1584
1585         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1586
1587         skb_push(skb, skb->data - skb_mac_header(skb));
1588
1589         /*
1590          *      The SOCK_PACKET socket receives _all_ frames.
1591          */
1592
1593         spkt->spkt_family = dev->type;
1594         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1595         spkt->spkt_protocol = skb->protocol;
1596
1597         /*
1598          *      Charge the memory to the socket. This is done specifically
1599          *      to prevent sockets using all the memory up.
1600          */
1601
1602         if (sock_queue_rcv_skb(sk, skb) == 0)
1603                 return 0;
1604
1605 out:
1606         kfree_skb(skb);
1607 oom:
1608         return 0;
1609 }
1610
1611
1612 /*
1613  *      Output a raw packet to a device layer. This bypasses all the other
1614  *      protocol layers and you must therefore supply it with a complete frame
1615  */
1616
1617 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1618                                size_t len)
1619 {
1620         struct sock *sk = sock->sk;
1621         DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1622         struct sk_buff *skb = NULL;
1623         struct net_device *dev;
1624         __be16 proto = 0;
1625         int err;
1626         int extra_len = 0;
1627
1628         /*
1629          *      Get and verify the address.
1630          */
1631
1632         if (saddr) {
1633                 if (msg->msg_namelen < sizeof(struct sockaddr))
1634                         return -EINVAL;
1635                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1636                         proto = saddr->spkt_protocol;
1637         } else
1638                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1639
1640         /*
1641          *      Find the device first to size check it
1642          */
1643
1644         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1645 retry:
1646         rcu_read_lock();
1647         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1648         err = -ENODEV;
1649         if (dev == NULL)
1650                 goto out_unlock;
1651
1652         err = -ENETDOWN;
1653         if (!(dev->flags & IFF_UP))
1654                 goto out_unlock;
1655
1656         /*
1657          * You may not queue a frame bigger than the mtu. This is the lowest level
1658          * raw protocol and you must do your own fragmentation at this level.
1659          */
1660
1661         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1662                 if (!netif_supports_nofcs(dev)) {
1663                         err = -EPROTONOSUPPORT;
1664                         goto out_unlock;
1665                 }
1666                 extra_len = 4; /* We're doing our own CRC */
1667         }
1668
1669         err = -EMSGSIZE;
1670         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1671                 goto out_unlock;
1672
1673         if (!skb) {
1674                 size_t reserved = LL_RESERVED_SPACE(dev);
1675                 int tlen = dev->needed_tailroom;
1676                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1677
1678                 rcu_read_unlock();
1679                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1680                 if (skb == NULL)
1681                         return -ENOBUFS;
1682                 /* FIXME: Save some space for broken drivers that write a hard
1683                  * header at transmission time by themselves. PPP is the notable
1684                  * one here. This should really be fixed at the driver level.
1685                  */
1686                 skb_reserve(skb, reserved);
1687                 skb_reset_network_header(skb);
1688
1689                 /* Try to align data part correctly */
1690                 if (hhlen) {
1691                         skb->data -= hhlen;
1692                         skb->tail -= hhlen;
1693                         if (len < hhlen)
1694                                 skb_reset_network_header(skb);
1695                 }
1696                 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1697                 if (err)
1698                         goto out_free;
1699                 goto retry;
1700         }
1701
1702         if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1703                 /* Earlier code assumed this would be a VLAN pkt,
1704                  * double-check this now that we have the actual
1705                  * packet in hand.
1706                  */
1707                 struct ethhdr *ehdr;
1708                 skb_reset_mac_header(skb);
1709                 ehdr = eth_hdr(skb);
1710                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1711                         err = -EMSGSIZE;
1712                         goto out_unlock;
1713                 }
1714         }
1715
1716         skb->protocol = proto;
1717         skb->dev = dev;
1718         skb->priority = sk->sk_priority;
1719         skb->mark = sk->sk_mark;
1720
1721         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1722
1723         if (unlikely(extra_len == 4))
1724                 skb->no_fcs = 1;
1725
1726         skb_probe_transport_header(skb, 0);
1727
1728         dev_queue_xmit(skb);
1729         rcu_read_unlock();
1730         return len;
1731
1732 out_unlock:
1733         rcu_read_unlock();
1734 out_free:
1735         kfree_skb(skb);
1736         return err;
1737 }
1738
1739 static unsigned int run_filter(const struct sk_buff *skb,
1740                                       const struct sock *sk,
1741                                       unsigned int res)
1742 {
1743         struct sk_filter *filter;
1744
1745         rcu_read_lock();
1746         filter = rcu_dereference(sk->sk_filter);
1747         if (filter != NULL)
1748                 res = SK_RUN_FILTER(filter, skb);
1749         rcu_read_unlock();
1750
1751         return res;
1752 }
1753
1754 /*
1755  * This function makes lazy skb cloning in hope that most of packets
1756  * are discarded by BPF.
1757  *
1758  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1759  * and skb->cb are mangled. It works because (and until) packets
1760  * falling here are owned by current CPU. Output packets are cloned
1761  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1762  * sequencially, so that if we return skb to original state on exit,
1763  * we will not harm anyone.
1764  */
1765
1766 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1767                       struct packet_type *pt, struct net_device *orig_dev)
1768 {
1769         struct sock *sk;
1770         struct sockaddr_ll *sll;
1771         struct packet_sock *po;
1772         u8 *skb_head = skb->data;
1773         int skb_len = skb->len;
1774         unsigned int snaplen, res;
1775
1776         if (skb->pkt_type == PACKET_LOOPBACK)
1777                 goto drop;
1778
1779         sk = pt->af_packet_priv;
1780         po = pkt_sk(sk);
1781
1782         if (!net_eq(dev_net(dev), sock_net(sk)))
1783                 goto drop;
1784
1785         skb->dev = dev;
1786
1787         if (dev->header_ops) {
1788                 /* The device has an explicit notion of ll header,
1789                  * exported to higher levels.
1790                  *
1791                  * Otherwise, the device hides details of its frame
1792                  * structure, so that corresponding packet head is
1793                  * never delivered to user.
1794                  */
1795                 if (sk->sk_type != SOCK_DGRAM)
1796                         skb_push(skb, skb->data - skb_mac_header(skb));
1797                 else if (skb->pkt_type == PACKET_OUTGOING) {
1798                         /* Special case: outgoing packets have ll header at head */
1799                         skb_pull(skb, skb_network_offset(skb));
1800                 }
1801         }
1802
1803         snaplen = skb->len;
1804
1805         res = run_filter(skb, sk, snaplen);
1806         if (!res)
1807                 goto drop_n_restore;
1808         if (snaplen > res)
1809                 snaplen = res;
1810
1811         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1812                 goto drop_n_acct;
1813
1814         if (skb_shared(skb)) {
1815                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1816                 if (nskb == NULL)
1817                         goto drop_n_acct;
1818
1819                 if (skb_head != skb->data) {
1820                         skb->data = skb_head;
1821                         skb->len = skb_len;
1822                 }
1823                 consume_skb(skb);
1824                 skb = nskb;
1825         }
1826
1827         sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
1828
1829         sll = &PACKET_SKB_CB(skb)->sa.ll;
1830         sll->sll_hatype = dev->type;
1831         sll->sll_pkttype = skb->pkt_type;
1832         if (unlikely(po->origdev))
1833                 sll->sll_ifindex = orig_dev->ifindex;
1834         else
1835                 sll->sll_ifindex = dev->ifindex;
1836
1837         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1838
1839         /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
1840          * Use their space for storing the original skb length.
1841          */
1842         PACKET_SKB_CB(skb)->sa.origlen = skb->len;
1843
1844         if (pskb_trim(skb, snaplen))
1845                 goto drop_n_acct;
1846
1847         skb_set_owner_r(skb, sk);
1848         skb->dev = NULL;
1849         skb_dst_drop(skb);
1850
1851         /* drop conntrack reference */
1852         nf_reset(skb);
1853
1854         spin_lock(&sk->sk_receive_queue.lock);
1855         po->stats.stats1.tp_packets++;
1856         sock_skb_set_dropcount(sk, skb);
1857         __skb_queue_tail(&sk->sk_receive_queue, skb);
1858         spin_unlock(&sk->sk_receive_queue.lock);
1859         sk->sk_data_ready(sk);
1860         return 0;
1861
1862 drop_n_acct:
1863         spin_lock(&sk->sk_receive_queue.lock);
1864         po->stats.stats1.tp_drops++;
1865         atomic_inc(&sk->sk_drops);
1866         spin_unlock(&sk->sk_receive_queue.lock);
1867
1868 drop_n_restore:
1869         if (skb_head != skb->data && skb_shared(skb)) {
1870                 skb->data = skb_head;
1871                 skb->len = skb_len;
1872         }
1873 drop:
1874         consume_skb(skb);
1875         return 0;
1876 }
1877
1878 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1879                        struct packet_type *pt, struct net_device *orig_dev)
1880 {
1881         struct sock *sk;
1882         struct packet_sock *po;
1883         struct sockaddr_ll *sll;
1884         union tpacket_uhdr h;
1885         u8 *skb_head = skb->data;
1886         int skb_len = skb->len;
1887         unsigned int snaplen, res;
1888         unsigned long status = TP_STATUS_USER;
1889         unsigned short macoff, netoff, hdrlen;
1890         struct sk_buff *copy_skb = NULL;
1891         struct timespec ts;
1892         __u32 ts_status;
1893
1894         /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
1895          * We may add members to them until current aligned size without forcing
1896          * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
1897          */
1898         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
1899         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
1900
1901         if (skb->pkt_type == PACKET_LOOPBACK)
1902                 goto drop;
1903
1904         sk = pt->af_packet_priv;
1905         po = pkt_sk(sk);
1906
1907         if (!net_eq(dev_net(dev), sock_net(sk)))
1908                 goto drop;
1909
1910         if (dev->header_ops) {
1911                 if (sk->sk_type != SOCK_DGRAM)
1912                         skb_push(skb, skb->data - skb_mac_header(skb));
1913                 else if (skb->pkt_type == PACKET_OUTGOING) {
1914                         /* Special case: outgoing packets have ll header at head */
1915                         skb_pull(skb, skb_network_offset(skb));
1916                 }
1917         }
1918
1919         snaplen = skb->len;
1920
1921         res = run_filter(skb, sk, snaplen);
1922         if (!res)
1923                 goto drop_n_restore;
1924
1925         if (skb->ip_summed == CHECKSUM_PARTIAL)
1926                 status |= TP_STATUS_CSUMNOTREADY;
1927         else if (skb->pkt_type != PACKET_OUTGOING &&
1928                  (skb->ip_summed == CHECKSUM_COMPLETE ||
1929                   skb_csum_unnecessary(skb)))
1930                 status |= TP_STATUS_CSUM_VALID;
1931
1932         if (snaplen > res)
1933                 snaplen = res;
1934
1935         if (sk->sk_type == SOCK_DGRAM) {
1936                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1937                                   po->tp_reserve;
1938         } else {
1939                 unsigned int maclen = skb_network_offset(skb);
1940                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1941                                        (maclen < 16 ? 16 : maclen)) +
1942                         po->tp_reserve;
1943                 macoff = netoff - maclen;
1944         }
1945         if (po->tp_version <= TPACKET_V2) {
1946                 if (macoff + snaplen > po->rx_ring.frame_size) {
1947                         if (po->copy_thresh &&
1948                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1949                                 if (skb_shared(skb)) {
1950                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1951                                 } else {
1952                                         copy_skb = skb_get(skb);
1953                                         skb_head = skb->data;
1954                                 }
1955                                 if (copy_skb)
1956                                         skb_set_owner_r(copy_skb, sk);
1957                         }
1958                         snaplen = po->rx_ring.frame_size - macoff;
1959                         if ((int)snaplen < 0)
1960                                 snaplen = 0;
1961                 }
1962         } else if (unlikely(macoff + snaplen >
1963                             GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
1964                 u32 nval;
1965
1966                 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
1967                 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
1968                             snaplen, nval, macoff);
1969                 snaplen = nval;
1970                 if (unlikely((int)snaplen < 0)) {
1971                         snaplen = 0;
1972                         macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
1973                 }
1974         }
1975         spin_lock(&sk->sk_receive_queue.lock);
1976         h.raw = packet_current_rx_frame(po, skb,
1977                                         TP_STATUS_KERNEL, (macoff+snaplen));
1978         if (!h.raw)
1979                 goto ring_is_full;
1980         if (po->tp_version <= TPACKET_V2) {
1981                 packet_increment_rx_head(po, &po->rx_ring);
1982         /*
1983          * LOSING will be reported till you read the stats,
1984          * because it's COR - Clear On Read.
1985          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1986          * at packet level.
1987          */
1988                 if (po->stats.stats1.tp_drops)
1989                         status |= TP_STATUS_LOSING;
1990         }
1991         po->stats.stats1.tp_packets++;
1992         if (copy_skb) {
1993                 status |= TP_STATUS_COPY;
1994                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1995         }
1996         spin_unlock(&sk->sk_receive_queue.lock);
1997
1998         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1999
2000         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2001                 getnstimeofday(&ts);
2002
2003         status |= ts_status;
2004
2005         switch (po->tp_version) {
2006         case TPACKET_V1:
2007                 h.h1->tp_len = skb->len;
2008                 h.h1->tp_snaplen = snaplen;
2009                 h.h1->tp_mac = macoff;
2010                 h.h1->tp_net = netoff;
2011                 h.h1->tp_sec = ts.tv_sec;
2012                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2013                 hdrlen = sizeof(*h.h1);
2014                 break;
2015         case TPACKET_V2:
2016                 h.h2->tp_len = skb->len;
2017                 h.h2->tp_snaplen = snaplen;
2018                 h.h2->tp_mac = macoff;
2019                 h.h2->tp_net = netoff;
2020                 h.h2->tp_sec = ts.tv_sec;
2021                 h.h2->tp_nsec = ts.tv_nsec;
2022                 if (skb_vlan_tag_present(skb)) {
2023                         h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2024                         h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2025                         status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2026                 } else {
2027                         h.h2->tp_vlan_tci = 0;
2028                         h.h2->tp_vlan_tpid = 0;
2029                 }
2030                 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2031                 hdrlen = sizeof(*h.h2);
2032                 break;
2033         case TPACKET_V3:
2034                 /* tp_nxt_offset,vlan are already populated above.
2035                  * So DONT clear those fields here
2036                  */
2037                 h.h3->tp_status |= status;
2038                 h.h3->tp_len = skb->len;
2039                 h.h3->tp_snaplen = snaplen;
2040                 h.h3->tp_mac = macoff;
2041                 h.h3->tp_net = netoff;
2042                 h.h3->tp_sec  = ts.tv_sec;
2043                 h.h3->tp_nsec = ts.tv_nsec;
2044                 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2045                 hdrlen = sizeof(*h.h3);
2046                 break;
2047         default:
2048                 BUG();
2049         }
2050
2051         sll = h.raw + TPACKET_ALIGN(hdrlen);
2052         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2053         sll->sll_family = AF_PACKET;
2054         sll->sll_hatype = dev->type;
2055         sll->sll_protocol = skb->protocol;
2056         sll->sll_pkttype = skb->pkt_type;
2057         if (unlikely(po->origdev))
2058                 sll->sll_ifindex = orig_dev->ifindex;
2059         else
2060                 sll->sll_ifindex = dev->ifindex;
2061
2062         smp_mb();
2063
2064 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2065         if (po->tp_version <= TPACKET_V2) {
2066                 u8 *start, *end;
2067
2068                 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2069                                         macoff + snaplen);
2070
2071                 for (start = h.raw; start < end; start += PAGE_SIZE)
2072                         flush_dcache_page(pgv_to_page(start));
2073         }
2074         smp_wmb();
2075 #endif
2076
2077         if (po->tp_version <= TPACKET_V2) {
2078                 __packet_set_status(po, h.raw, status);
2079                 sk->sk_data_ready(sk);
2080         } else {
2081                 prb_clear_blk_fill_status(&po->rx_ring);
2082         }
2083
2084 drop_n_restore:
2085         if (skb_head != skb->data && skb_shared(skb)) {
2086                 skb->data = skb_head;
2087                 skb->len = skb_len;
2088         }
2089 drop:
2090         kfree_skb(skb);
2091         return 0;
2092
2093 ring_is_full:
2094         po->stats.stats1.tp_drops++;
2095         spin_unlock(&sk->sk_receive_queue.lock);
2096
2097         sk->sk_data_ready(sk);
2098         kfree_skb(copy_skb);
2099         goto drop_n_restore;
2100 }
2101
2102 static void tpacket_destruct_skb(struct sk_buff *skb)
2103 {
2104         struct packet_sock *po = pkt_sk(skb->sk);
2105
2106         if (likely(po->tx_ring.pg_vec)) {
2107                 void *ph;
2108                 __u32 ts;
2109
2110                 ph = skb_shinfo(skb)->destructor_arg;
2111                 packet_dec_pending(&po->tx_ring);
2112
2113                 ts = __packet_set_timestamp(po, ph, skb);
2114                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2115         }
2116
2117         sock_wfree(skb);
2118 }
2119
2120 static bool ll_header_truncated(const struct net_device *dev, int len)
2121 {
2122         /* net device doesn't like empty head */
2123         if (unlikely(len <= dev->hard_header_len)) {
2124                 net_warn_ratelimited("%s: packet size is too short (%d <= %d)\n",
2125                                      current->comm, len, dev->hard_header_len);
2126                 return true;
2127         }
2128
2129         return false;
2130 }
2131
2132 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2133                 void *frame, struct net_device *dev, int size_max,
2134                 __be16 proto, unsigned char *addr, int hlen)
2135 {
2136         union tpacket_uhdr ph;
2137         int to_write, offset, len, tp_len, nr_frags, len_max;
2138         struct socket *sock = po->sk.sk_socket;
2139         struct page *page;
2140         void *data;
2141         int err;
2142
2143         ph.raw = frame;
2144
2145         skb->protocol = proto;
2146         skb->dev = dev;
2147         skb->priority = po->sk.sk_priority;
2148         skb->mark = po->sk.sk_mark;
2149         sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2150         skb_shinfo(skb)->destructor_arg = ph.raw;
2151
2152         switch (po->tp_version) {
2153         case TPACKET_V2:
2154                 tp_len = ph.h2->tp_len;
2155                 break;
2156         default:
2157                 tp_len = ph.h1->tp_len;
2158                 break;
2159         }
2160         if (unlikely(tp_len > size_max)) {
2161                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2162                 return -EMSGSIZE;
2163         }
2164
2165         skb_reserve(skb, hlen);
2166         skb_reset_network_header(skb);
2167
2168         if (!packet_use_direct_xmit(po))
2169                 skb_probe_transport_header(skb, 0);
2170         if (unlikely(po->tp_tx_has_off)) {
2171                 int off_min, off_max, off;
2172                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2173                 off_max = po->tx_ring.frame_size - tp_len;
2174                 if (sock->type == SOCK_DGRAM) {
2175                         switch (po->tp_version) {
2176                         case TPACKET_V2:
2177                                 off = ph.h2->tp_net;
2178                                 break;
2179                         default:
2180                                 off = ph.h1->tp_net;
2181                                 break;
2182                         }
2183                 } else {
2184                         switch (po->tp_version) {
2185                         case TPACKET_V2:
2186                                 off = ph.h2->tp_mac;
2187                                 break;
2188                         default:
2189                                 off = ph.h1->tp_mac;
2190                                 break;
2191                         }
2192                 }
2193                 if (unlikely((off < off_min) || (off_max < off)))
2194                         return -EINVAL;
2195                 data = ph.raw + off;
2196         } else {
2197                 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2198         }
2199         to_write = tp_len;
2200
2201         if (sock->type == SOCK_DGRAM) {
2202                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2203                                 NULL, tp_len);
2204                 if (unlikely(err < 0))
2205                         return -EINVAL;
2206         } else if (dev->hard_header_len) {
2207                 if (ll_header_truncated(dev, tp_len))
2208                         return -EINVAL;
2209
2210                 skb_push(skb, dev->hard_header_len);
2211                 err = skb_store_bits(skb, 0, data,
2212                                 dev->hard_header_len);
2213                 if (unlikely(err))
2214                         return err;
2215
2216                 data += dev->hard_header_len;
2217                 to_write -= dev->hard_header_len;
2218         }
2219
2220         offset = offset_in_page(data);
2221         len_max = PAGE_SIZE - offset;
2222         len = ((to_write > len_max) ? len_max : to_write);
2223
2224         skb->data_len = to_write;
2225         skb->len += to_write;
2226         skb->truesize += to_write;
2227         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2228
2229         while (likely(to_write)) {
2230                 nr_frags = skb_shinfo(skb)->nr_frags;
2231
2232                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2233                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2234                                MAX_SKB_FRAGS);
2235                         return -EFAULT;
2236                 }
2237
2238                 page = pgv_to_page(data);
2239                 data += len;
2240                 flush_dcache_page(page);
2241                 get_page(page);
2242                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2243                 to_write -= len;
2244                 offset = 0;
2245                 len_max = PAGE_SIZE;
2246                 len = ((to_write > len_max) ? len_max : to_write);
2247         }
2248
2249         return tp_len;
2250 }
2251
2252 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2253 {
2254         struct sk_buff *skb;
2255         struct net_device *dev;
2256         __be16 proto;
2257         int err, reserve = 0;
2258         void *ph;
2259         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2260         bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2261         int tp_len, size_max;
2262         unsigned char *addr;
2263         int len_sum = 0;
2264         int status = TP_STATUS_AVAILABLE;
2265         int hlen, tlen;
2266
2267         mutex_lock(&po->pg_vec_lock);
2268
2269         if (likely(saddr == NULL)) {
2270                 dev     = packet_cached_dev_get(po);
2271                 proto   = po->num;
2272                 addr    = NULL;
2273         } else {
2274                 err = -EINVAL;
2275                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2276                         goto out;
2277                 if (msg->msg_namelen < (saddr->sll_halen
2278                                         + offsetof(struct sockaddr_ll,
2279                                                 sll_addr)))
2280                         goto out;
2281                 proto   = saddr->sll_protocol;
2282                 addr    = saddr->sll_addr;
2283                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2284         }
2285
2286         err = -ENXIO;
2287         if (unlikely(dev == NULL))
2288                 goto out;
2289         err = -ENETDOWN;
2290         if (unlikely(!(dev->flags & IFF_UP)))
2291                 goto out_put;
2292
2293         reserve = dev->hard_header_len + VLAN_HLEN;
2294         size_max = po->tx_ring.frame_size
2295                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2296
2297         if (size_max > dev->mtu + reserve)
2298                 size_max = dev->mtu + reserve;
2299
2300         do {
2301                 ph = packet_current_frame(po, &po->tx_ring,
2302                                           TP_STATUS_SEND_REQUEST);
2303                 if (unlikely(ph == NULL)) {
2304                         if (need_wait && need_resched())
2305                                 schedule();
2306                         continue;
2307                 }
2308
2309                 status = TP_STATUS_SEND_REQUEST;
2310                 hlen = LL_RESERVED_SPACE(dev);
2311                 tlen = dev->needed_tailroom;
2312                 skb = sock_alloc_send_skb(&po->sk,
2313                                 hlen + tlen + sizeof(struct sockaddr_ll),
2314                                 !need_wait, &err);
2315
2316                 if (unlikely(skb == NULL)) {
2317                         /* we assume the socket was initially writeable ... */
2318                         if (likely(len_sum > 0))
2319                                 err = len_sum;
2320                         goto out_status;
2321                 }
2322                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2323                                           addr, hlen);
2324                 if (tp_len > dev->mtu + dev->hard_header_len) {
2325                         struct ethhdr *ehdr;
2326                         /* Earlier code assumed this would be a VLAN pkt,
2327                          * double-check this now that we have the actual
2328                          * packet in hand.
2329                          */
2330
2331                         skb_reset_mac_header(skb);
2332                         ehdr = eth_hdr(skb);
2333                         if (ehdr->h_proto != htons(ETH_P_8021Q))
2334                                 tp_len = -EMSGSIZE;
2335                 }
2336                 if (unlikely(tp_len < 0)) {
2337                         if (po->tp_loss) {
2338                                 __packet_set_status(po, ph,
2339                                                 TP_STATUS_AVAILABLE);
2340                                 packet_increment_head(&po->tx_ring);
2341                                 kfree_skb(skb);
2342                                 continue;
2343                         } else {
2344                                 status = TP_STATUS_WRONG_FORMAT;
2345                                 err = tp_len;
2346                                 goto out_status;
2347                         }
2348                 }
2349
2350                 packet_pick_tx_queue(dev, skb);
2351
2352                 skb->destructor = tpacket_destruct_skb;
2353                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2354                 packet_inc_pending(&po->tx_ring);
2355
2356                 status = TP_STATUS_SEND_REQUEST;
2357                 err = po->xmit(skb);
2358                 if (unlikely(err > 0)) {
2359                         err = net_xmit_errno(err);
2360                         if (err && __packet_get_status(po, ph) ==
2361                                    TP_STATUS_AVAILABLE) {
2362                                 /* skb was destructed already */
2363                                 skb = NULL;
2364                                 goto out_status;
2365                         }
2366                         /*
2367                          * skb was dropped but not destructed yet;
2368                          * let's treat it like congestion or err < 0
2369                          */
2370                         err = 0;
2371                 }
2372                 packet_increment_head(&po->tx_ring);
2373                 len_sum += tp_len;
2374         } while (likely((ph != NULL) ||
2375                 /* Note: packet_read_pending() might be slow if we have
2376                  * to call it as it's per_cpu variable, but in fast-path
2377                  * we already short-circuit the loop with the first
2378                  * condition, and luckily don't have to go that path
2379                  * anyway.
2380                  */
2381                  (need_wait && packet_read_pending(&po->tx_ring))));
2382
2383         err = len_sum;
2384         goto out_put;
2385
2386 out_status:
2387         __packet_set_status(po, ph, status);
2388         kfree_skb(skb);
2389 out_put:
2390         dev_put(dev);
2391 out:
2392         mutex_unlock(&po->pg_vec_lock);
2393         return err;
2394 }
2395
2396 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2397                                         size_t reserve, size_t len,
2398                                         size_t linear, int noblock,
2399                                         int *err)
2400 {
2401         struct sk_buff *skb;
2402
2403         /* Under a page?  Don't bother with paged skb. */
2404         if (prepad + len < PAGE_SIZE || !linear)
2405                 linear = len;
2406
2407         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2408                                    err, 0);
2409         if (!skb)
2410                 return NULL;
2411
2412         skb_reserve(skb, reserve);
2413         skb_put(skb, linear);
2414         skb->data_len = len - linear;
2415         skb->len += len - linear;
2416
2417         return skb;
2418 }
2419
2420 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2421 {
2422         struct sock *sk = sock->sk;
2423         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2424         struct sk_buff *skb;
2425         struct net_device *dev;
2426         __be16 proto;
2427         unsigned char *addr;
2428         int err, reserve = 0;
2429         struct virtio_net_hdr vnet_hdr = { 0 };
2430         int offset = 0;
2431         int vnet_hdr_len;
2432         struct packet_sock *po = pkt_sk(sk);
2433         unsigned short gso_type = 0;
2434         int hlen, tlen;
2435         int extra_len = 0;
2436         ssize_t n;
2437
2438         /*
2439          *      Get and verify the address.
2440          */
2441
2442         if (likely(saddr == NULL)) {
2443                 dev     = packet_cached_dev_get(po);
2444                 proto   = po->num;
2445                 addr    = NULL;
2446         } else {
2447                 err = -EINVAL;
2448                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2449                         goto out;
2450                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2451                         goto out;
2452                 proto   = saddr->sll_protocol;
2453                 addr    = saddr->sll_addr;
2454                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2455         }
2456
2457         err = -ENXIO;
2458         if (unlikely(dev == NULL))
2459                 goto out_unlock;
2460         err = -ENETDOWN;
2461         if (unlikely(!(dev->flags & IFF_UP)))
2462                 goto out_unlock;
2463
2464         if (sock->type == SOCK_RAW)
2465                 reserve = dev->hard_header_len;
2466         if (po->has_vnet_hdr) {
2467                 vnet_hdr_len = sizeof(vnet_hdr);
2468
2469                 err = -EINVAL;
2470                 if (len < vnet_hdr_len)
2471                         goto out_unlock;
2472
2473                 len -= vnet_hdr_len;
2474
2475                 err = -EFAULT;
2476                 n = copy_from_iter(&vnet_hdr, vnet_hdr_len, &msg->msg_iter);
2477                 if (n != vnet_hdr_len)
2478                         goto out_unlock;
2479
2480                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2481                     (__virtio16_to_cpu(false, vnet_hdr.csum_start) +
2482                      __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2 >
2483                       __virtio16_to_cpu(false, vnet_hdr.hdr_len)))
2484                         vnet_hdr.hdr_len = __cpu_to_virtio16(false,
2485                                  __virtio16_to_cpu(false, vnet_hdr.csum_start) +
2486                                 __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2);
2487
2488                 err = -EINVAL;
2489                 if (__virtio16_to_cpu(false, vnet_hdr.hdr_len) > len)
2490                         goto out_unlock;
2491
2492                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2493                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2494                         case VIRTIO_NET_HDR_GSO_TCPV4:
2495                                 gso_type = SKB_GSO_TCPV4;
2496                                 break;
2497                         case VIRTIO_NET_HDR_GSO_TCPV6:
2498                                 gso_type = SKB_GSO_TCPV6;
2499                                 break;
2500                         case VIRTIO_NET_HDR_GSO_UDP:
2501                                 gso_type = SKB_GSO_UDP;
2502                                 break;
2503                         default:
2504                                 goto out_unlock;
2505                         }
2506
2507                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2508                                 gso_type |= SKB_GSO_TCP_ECN;
2509
2510                         if (vnet_hdr.gso_size == 0)
2511                                 goto out_unlock;
2512
2513                 }
2514         }
2515
2516         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2517                 if (!netif_supports_nofcs(dev)) {
2518                         err = -EPROTONOSUPPORT;
2519                         goto out_unlock;
2520                 }
2521                 extra_len = 4; /* We're doing our own CRC */
2522         }
2523
2524         err = -EMSGSIZE;
2525         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2526                 goto out_unlock;
2527
2528         err = -ENOBUFS;
2529         hlen = LL_RESERVED_SPACE(dev);
2530         tlen = dev->needed_tailroom;
2531         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
2532                                __virtio16_to_cpu(false, vnet_hdr.hdr_len),
2533                                msg->msg_flags & MSG_DONTWAIT, &err);
2534         if (skb == NULL)
2535                 goto out_unlock;
2536
2537         skb_set_network_header(skb, reserve);
2538
2539         err = -EINVAL;
2540         if (sock->type == SOCK_DGRAM) {
2541                 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2542                 if (unlikely(offset < 0))
2543                         goto out_free;
2544         } else {
2545                 if (ll_header_truncated(dev, len))
2546                         goto out_free;
2547         }
2548
2549         /* Returns -EFAULT on error */
2550         err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2551         if (err)
2552                 goto out_free;
2553
2554         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2555
2556         if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2557                 /* Earlier code assumed this would be a VLAN pkt,
2558                  * double-check this now that we have the actual
2559                  * packet in hand.
2560                  */
2561                 struct ethhdr *ehdr;
2562                 skb_reset_mac_header(skb);
2563                 ehdr = eth_hdr(skb);
2564                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2565                         err = -EMSGSIZE;
2566                         goto out_free;
2567                 }
2568         }
2569
2570         skb->protocol = proto;
2571         skb->dev = dev;
2572         skb->priority = sk->sk_priority;
2573         skb->mark = sk->sk_mark;
2574
2575         packet_pick_tx_queue(dev, skb);
2576
2577         if (po->has_vnet_hdr) {
2578                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2579                         u16 s = __virtio16_to_cpu(false, vnet_hdr.csum_start);
2580                         u16 o = __virtio16_to_cpu(false, vnet_hdr.csum_offset);
2581                         if (!skb_partial_csum_set(skb, s, o)) {
2582                                 err = -EINVAL;
2583                                 goto out_free;
2584                         }
2585                 }
2586
2587                 skb_shinfo(skb)->gso_size =
2588                         __virtio16_to_cpu(false, vnet_hdr.gso_size);
2589                 skb_shinfo(skb)->gso_type = gso_type;
2590
2591                 /* Header must be checked, and gso_segs computed. */
2592                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2593                 skb_shinfo(skb)->gso_segs = 0;
2594
2595                 len += vnet_hdr_len;
2596         }
2597
2598         if (!packet_use_direct_xmit(po))
2599                 skb_probe_transport_header(skb, reserve);
2600         if (unlikely(extra_len == 4))
2601                 skb->no_fcs = 1;
2602
2603         err = po->xmit(skb);
2604         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2605                 goto out_unlock;
2606
2607         dev_put(dev);
2608
2609         return len;
2610
2611 out_free:
2612         kfree_skb(skb);
2613 out_unlock:
2614         if (dev)
2615                 dev_put(dev);
2616 out:
2617         return err;
2618 }
2619
2620 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2621 {
2622         struct sock *sk = sock->sk;
2623         struct packet_sock *po = pkt_sk(sk);
2624
2625         if (po->tx_ring.pg_vec)
2626                 return tpacket_snd(po, msg);
2627         else
2628                 return packet_snd(sock, msg, len);
2629 }
2630
2631 /*
2632  *      Close a PACKET socket. This is fairly simple. We immediately go
2633  *      to 'closed' state and remove our protocol entry in the device list.
2634  */
2635
2636 static int packet_release(struct socket *sock)
2637 {
2638         struct sock *sk = sock->sk;
2639         struct packet_sock *po;
2640         struct net *net;
2641         union tpacket_req_u req_u;
2642
2643         if (!sk)
2644                 return 0;
2645
2646         net = sock_net(sk);
2647         po = pkt_sk(sk);
2648
2649         mutex_lock(&net->packet.sklist_lock);
2650         sk_del_node_init_rcu(sk);
2651         mutex_unlock(&net->packet.sklist_lock);
2652
2653         preempt_disable();
2654         sock_prot_inuse_add(net, sk->sk_prot, -1);
2655         preempt_enable();
2656
2657         spin_lock(&po->bind_lock);
2658         unregister_prot_hook(sk, false);
2659         packet_cached_dev_reset(po);
2660
2661         if (po->prot_hook.dev) {
2662                 dev_put(po->prot_hook.dev);
2663                 po->prot_hook.dev = NULL;
2664         }
2665         spin_unlock(&po->bind_lock);
2666
2667         packet_flush_mclist(sk);
2668
2669         if (po->rx_ring.pg_vec) {
2670                 memset(&req_u, 0, sizeof(req_u));
2671                 packet_set_ring(sk, &req_u, 1, 0);
2672         }
2673
2674         if (po->tx_ring.pg_vec) {
2675                 memset(&req_u, 0, sizeof(req_u));
2676                 packet_set_ring(sk, &req_u, 1, 1);
2677         }
2678
2679         fanout_release(sk);
2680
2681         synchronize_net();
2682         /*
2683          *      Now the socket is dead. No more input will appear.
2684          */
2685         sock_orphan(sk);
2686         sock->sk = NULL;
2687
2688         /* Purge queues */
2689
2690         skb_queue_purge(&sk->sk_receive_queue);
2691         packet_free_pending(po);
2692         sk_refcnt_debug_release(sk);
2693
2694         sock_put(sk);
2695         return 0;
2696 }
2697
2698 /*
2699  *      Attach a packet hook.
2700  */
2701
2702 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 proto)
2703 {
2704         struct packet_sock *po = pkt_sk(sk);
2705         const struct net_device *dev_curr;
2706         __be16 proto_curr;
2707         bool need_rehook;
2708
2709         if (po->fanout) {
2710                 if (dev)
2711                         dev_put(dev);
2712
2713                 return -EINVAL;
2714         }
2715
2716         lock_sock(sk);
2717         spin_lock(&po->bind_lock);
2718
2719         proto_curr = po->prot_hook.type;
2720         dev_curr = po->prot_hook.dev;
2721
2722         need_rehook = proto_curr != proto || dev_curr != dev;
2723
2724         if (need_rehook) {
2725                 unregister_prot_hook(sk, true);
2726
2727                 po->num = proto;
2728                 po->prot_hook.type = proto;
2729
2730                 if (po->prot_hook.dev)
2731                         dev_put(po->prot_hook.dev);
2732
2733                 po->prot_hook.dev = dev;
2734
2735                 po->ifindex = dev ? dev->ifindex : 0;
2736                 packet_cached_dev_assign(po, dev);
2737         }
2738
2739         if (proto == 0 || !need_rehook)
2740                 goto out_unlock;
2741
2742         if (!dev || (dev->flags & IFF_UP)) {
2743                 register_prot_hook(sk);
2744         } else {
2745                 sk->sk_err = ENETDOWN;
2746                 if (!sock_flag(sk, SOCK_DEAD))
2747                         sk->sk_error_report(sk);
2748         }
2749
2750 out_unlock:
2751         spin_unlock(&po->bind_lock);
2752         release_sock(sk);
2753         return 0;
2754 }
2755
2756 /*
2757  *      Bind a packet socket to a device
2758  */
2759
2760 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2761                             int addr_len)
2762 {
2763         struct sock *sk = sock->sk;
2764         char name[15];
2765         struct net_device *dev;
2766         int err = -ENODEV;
2767
2768         /*
2769          *      Check legality
2770          */
2771
2772         if (addr_len != sizeof(struct sockaddr))
2773                 return -EINVAL;
2774         strlcpy(name, uaddr->sa_data, sizeof(name));
2775
2776         dev = dev_get_by_name(sock_net(sk), name);
2777         if (dev)
2778                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2779         return err;
2780 }
2781
2782 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2783 {
2784         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2785         struct sock *sk = sock->sk;
2786         struct net_device *dev = NULL;
2787         int err;
2788
2789
2790         /*
2791          *      Check legality
2792          */
2793
2794         if (addr_len < sizeof(struct sockaddr_ll))
2795                 return -EINVAL;
2796         if (sll->sll_family != AF_PACKET)
2797                 return -EINVAL;
2798
2799         if (sll->sll_ifindex) {
2800                 err = -ENODEV;
2801                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2802                 if (dev == NULL)
2803                         goto out;
2804         }
2805         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2806
2807 out:
2808         return err;
2809 }
2810
2811 static struct proto packet_proto = {
2812         .name     = "PACKET",
2813         .owner    = THIS_MODULE,
2814         .obj_size = sizeof(struct packet_sock),
2815 };
2816
2817 /*
2818  *      Create a packet of type SOCK_PACKET.
2819  */
2820
2821 static int packet_create(struct net *net, struct socket *sock, int protocol,
2822                          int kern)
2823 {
2824         struct sock *sk;
2825         struct packet_sock *po;
2826         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2827         int err;
2828
2829         if (!ns_capable(net->user_ns, CAP_NET_RAW))
2830                 return -EPERM;
2831         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2832             sock->type != SOCK_PACKET)
2833                 return -ESOCKTNOSUPPORT;
2834
2835         sock->state = SS_UNCONNECTED;
2836
2837         err = -ENOBUFS;
2838         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2839         if (sk == NULL)
2840                 goto out;
2841
2842         sock->ops = &packet_ops;
2843         if (sock->type == SOCK_PACKET)
2844                 sock->ops = &packet_ops_spkt;
2845
2846         sock_init_data(sock, sk);
2847
2848         po = pkt_sk(sk);
2849         sk->sk_family = PF_PACKET;
2850         po->num = proto;
2851         po->xmit = dev_queue_xmit;
2852
2853         err = packet_alloc_pending(po);
2854         if (err)
2855                 goto out2;
2856
2857         packet_cached_dev_reset(po);
2858
2859         sk->sk_destruct = packet_sock_destruct;
2860         sk_refcnt_debug_inc(sk);
2861
2862         /*
2863          *      Attach a protocol block
2864          */
2865
2866         spin_lock_init(&po->bind_lock);
2867         mutex_init(&po->pg_vec_lock);
2868         po->prot_hook.func = packet_rcv;
2869
2870         if (sock->type == SOCK_PACKET)
2871                 po->prot_hook.func = packet_rcv_spkt;
2872
2873         po->prot_hook.af_packet_priv = sk;
2874
2875         if (proto) {
2876                 po->prot_hook.type = proto;
2877                 register_prot_hook(sk);
2878         }
2879
2880         mutex_lock(&net->packet.sklist_lock);
2881         sk_add_node_rcu(sk, &net->packet.sklist);
2882         mutex_unlock(&net->packet.sklist_lock);
2883
2884         preempt_disable();
2885         sock_prot_inuse_add(net, &packet_proto, 1);
2886         preempt_enable();
2887
2888         return 0;
2889 out2:
2890         sk_free(sk);
2891 out:
2892         return err;
2893 }
2894
2895 /*
2896  *      Pull a packet from our receive queue and hand it to the user.
2897  *      If necessary we block.
2898  */
2899
2900 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2901                           int flags)
2902 {
2903         struct sock *sk = sock->sk;
2904         struct sk_buff *skb;
2905         int copied, err;
2906         int vnet_hdr_len = 0;
2907         unsigned int origlen = 0;
2908
2909         err = -EINVAL;
2910         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2911                 goto out;
2912
2913 #if 0
2914         /* What error should we return now? EUNATTACH? */
2915         if (pkt_sk(sk)->ifindex < 0)
2916                 return -ENODEV;
2917 #endif
2918
2919         if (flags & MSG_ERRQUEUE) {
2920                 err = sock_recv_errqueue(sk, msg, len,
2921                                          SOL_PACKET, PACKET_TX_TIMESTAMP);
2922                 goto out;
2923         }
2924
2925         /*
2926          *      Call the generic datagram receiver. This handles all sorts
2927          *      of horrible races and re-entrancy so we can forget about it
2928          *      in the protocol layers.
2929          *
2930          *      Now it will return ENETDOWN, if device have just gone down,
2931          *      but then it will block.
2932          */
2933
2934         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2935
2936         /*
2937          *      An error occurred so return it. Because skb_recv_datagram()
2938          *      handles the blocking we don't see and worry about blocking
2939          *      retries.
2940          */
2941
2942         if (skb == NULL)
2943                 goto out;
2944
2945         if (pkt_sk(sk)->has_vnet_hdr) {
2946                 struct virtio_net_hdr vnet_hdr = { 0 };
2947
2948                 err = -EINVAL;
2949                 vnet_hdr_len = sizeof(vnet_hdr);
2950                 if (len < vnet_hdr_len)
2951                         goto out_free;
2952
2953                 len -= vnet_hdr_len;
2954
2955                 if (skb_is_gso(skb)) {
2956                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2957
2958                         /* This is a hint as to how much should be linear. */
2959                         vnet_hdr.hdr_len =
2960                                 __cpu_to_virtio16(false, skb_headlen(skb));
2961                         vnet_hdr.gso_size =
2962                                 __cpu_to_virtio16(false, sinfo->gso_size);
2963                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2964                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2965                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2966                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2967                         else if (sinfo->gso_type & SKB_GSO_UDP)
2968                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2969                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2970                                 goto out_free;
2971                         else
2972                                 BUG();
2973                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2974                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2975                 } else
2976                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2977
2978                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2979                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2980                         vnet_hdr.csum_start = __cpu_to_virtio16(false,
2981                                           skb_checksum_start_offset(skb));
2982                         vnet_hdr.csum_offset = __cpu_to_virtio16(false,
2983                                                          skb->csum_offset);
2984                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2985                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2986                 } /* else everything is zero */
2987
2988                 err = memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_len);
2989                 if (err < 0)
2990                         goto out_free;
2991         }
2992
2993         /* You lose any data beyond the buffer you gave. If it worries
2994          * a user program they can ask the device for its MTU
2995          * anyway.
2996          */
2997         copied = skb->len;
2998         if (copied > len) {
2999                 copied = len;
3000                 msg->msg_flags |= MSG_TRUNC;
3001         }
3002
3003         err = skb_copy_datagram_msg(skb, 0, msg, copied);
3004         if (err)
3005                 goto out_free;
3006
3007         if (sock->type != SOCK_PACKET) {
3008                 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3009
3010                 /* Original length was stored in sockaddr_ll fields */
3011                 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3012                 sll->sll_family = AF_PACKET;
3013                 sll->sll_protocol = skb->protocol;
3014         }
3015
3016         sock_recv_ts_and_drops(msg, sk, skb);
3017
3018         if (msg->msg_name) {
3019                 /* If the address length field is there to be filled
3020                  * in, we fill it in now.
3021                  */
3022                 if (sock->type == SOCK_PACKET) {
3023                         __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3024                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
3025                 } else {
3026                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3027
3028                         msg->msg_namelen = sll->sll_halen +
3029                                 offsetof(struct sockaddr_ll, sll_addr);
3030                 }
3031                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3032                        msg->msg_namelen);
3033         }
3034
3035         if (pkt_sk(sk)->auxdata) {
3036                 struct tpacket_auxdata aux;
3037
3038                 aux.tp_status = TP_STATUS_USER;
3039                 if (skb->ip_summed == CHECKSUM_PARTIAL)
3040                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3041                 else if (skb->pkt_type != PACKET_OUTGOING &&
3042                          (skb->ip_summed == CHECKSUM_COMPLETE ||
3043                           skb_csum_unnecessary(skb)))
3044                         aux.tp_status |= TP_STATUS_CSUM_VALID;
3045
3046                 aux.tp_len = origlen;
3047                 aux.tp_snaplen = skb->len;
3048                 aux.tp_mac = 0;
3049                 aux.tp_net = skb_network_offset(skb);
3050                 if (skb_vlan_tag_present(skb)) {
3051                         aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3052                         aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3053                         aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3054                 } else {
3055                         aux.tp_vlan_tci = 0;
3056                         aux.tp_vlan_tpid = 0;
3057                 }
3058                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3059         }
3060
3061         /*
3062          *      Free or return the buffer as appropriate. Again this
3063          *      hides all the races and re-entrancy issues from us.
3064          */
3065         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3066
3067 out_free:
3068         skb_free_datagram(sk, skb);
3069 out:
3070         return err;
3071 }
3072
3073 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3074                                int *uaddr_len, int peer)
3075 {
3076         struct net_device *dev;
3077         struct sock *sk = sock->sk;
3078
3079         if (peer)
3080                 return -EOPNOTSUPP;
3081
3082         uaddr->sa_family = AF_PACKET;
3083         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3084         rcu_read_lock();
3085         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3086         if (dev)
3087                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3088         rcu_read_unlock();
3089         *uaddr_len = sizeof(*uaddr);
3090
3091         return 0;
3092 }
3093
3094 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3095                           int *uaddr_len, int peer)
3096 {
3097         struct net_device *dev;
3098         struct sock *sk = sock->sk;
3099         struct packet_sock *po = pkt_sk(sk);
3100         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3101
3102         if (peer)
3103                 return -EOPNOTSUPP;
3104
3105         sll->sll_family = AF_PACKET;
3106         sll->sll_ifindex = po->ifindex;
3107         sll->sll_protocol = po->num;
3108         sll->sll_pkttype = 0;
3109         rcu_read_lock();
3110         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3111         if (dev) {
3112                 sll->sll_hatype = dev->type;
3113                 sll->sll_halen = dev->addr_len;
3114                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3115         } else {
3116                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
3117                 sll->sll_halen = 0;
3118         }
3119         rcu_read_unlock();
3120         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3121
3122         return 0;
3123 }
3124
3125 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3126                          int what)
3127 {
3128         switch (i->type) {
3129         case PACKET_MR_MULTICAST:
3130                 if (i->alen != dev->addr_len)
3131                         return -EINVAL;
3132                 if (what > 0)
3133                         return dev_mc_add(dev, i->addr);
3134                 else
3135                         return dev_mc_del(dev, i->addr);
3136                 break;
3137         case PACKET_MR_PROMISC:
3138                 return dev_set_promiscuity(dev, what);
3139         case PACKET_MR_ALLMULTI:
3140                 return dev_set_allmulti(dev, what);
3141         case PACKET_MR_UNICAST:
3142                 if (i->alen != dev->addr_len)
3143                         return -EINVAL;
3144                 if (what > 0)
3145                         return dev_uc_add(dev, i->addr);
3146                 else
3147                         return dev_uc_del(dev, i->addr);
3148                 break;
3149         default:
3150                 break;
3151         }
3152         return 0;
3153 }
3154
3155 static void packet_dev_mclist_delete(struct net_device *dev,
3156                                      struct packet_mclist **mlp)
3157 {
3158         struct packet_mclist *ml;
3159
3160         while ((ml = *mlp) != NULL) {
3161                 if (ml->ifindex == dev->ifindex) {
3162                         packet_dev_mc(dev, ml, -1);
3163                         *mlp = ml->next;
3164                         kfree(ml);
3165                 } else
3166                         mlp = &ml->next;
3167         }
3168 }
3169
3170 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3171 {
3172         struct packet_sock *po = pkt_sk(sk);
3173         struct packet_mclist *ml, *i;
3174         struct net_device *dev;
3175         int err;
3176
3177         rtnl_lock();
3178
3179         err = -ENODEV;
3180         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3181         if (!dev)
3182                 goto done;
3183
3184         err = -EINVAL;
3185         if (mreq->mr_alen > dev->addr_len)
3186                 goto done;
3187
3188         err = -ENOBUFS;
3189         i = kmalloc(sizeof(*i), GFP_KERNEL);
3190         if (i == NULL)
3191                 goto done;
3192
3193         err = 0;
3194         for (ml = po->mclist; ml; ml = ml->next) {
3195                 if (ml->ifindex == mreq->mr_ifindex &&
3196                     ml->type == mreq->mr_type &&
3197                     ml->alen == mreq->mr_alen &&
3198                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3199                         ml->count++;
3200                         /* Free the new element ... */
3201                         kfree(i);
3202                         goto done;
3203                 }
3204         }
3205
3206         i->type = mreq->mr_type;
3207         i->ifindex = mreq->mr_ifindex;
3208         i->alen = mreq->mr_alen;
3209         memcpy(i->addr, mreq->mr_address, i->alen);
3210         i->count = 1;
3211         i->next = po->mclist;
3212         po->mclist = i;
3213         err = packet_dev_mc(dev, i, 1);
3214         if (err) {
3215                 po->mclist = i->next;
3216                 kfree(i);
3217         }
3218
3219 done:
3220         rtnl_unlock();
3221         return err;
3222 }
3223
3224 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3225 {
3226         struct packet_mclist *ml, **mlp;
3227
3228         rtnl_lock();
3229
3230         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3231                 if (ml->ifindex == mreq->mr_ifindex &&
3232                     ml->type == mreq->mr_type &&
3233                     ml->alen == mreq->mr_alen &&
3234                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3235                         if (--ml->count == 0) {
3236                                 struct net_device *dev;
3237                                 *mlp = ml->next;
3238                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3239                                 if (dev)
3240                                         packet_dev_mc(dev, ml, -1);
3241                                 kfree(ml);
3242                         }
3243                         break;
3244                 }
3245         }
3246         rtnl_unlock();
3247         return 0;
3248 }
3249
3250 static void packet_flush_mclist(struct sock *sk)
3251 {
3252         struct packet_sock *po = pkt_sk(sk);
3253         struct packet_mclist *ml;
3254
3255         if (!po->mclist)
3256                 return;
3257
3258         rtnl_lock();
3259         while ((ml = po->mclist) != NULL) {
3260                 struct net_device *dev;
3261
3262                 po->mclist = ml->next;
3263                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3264                 if (dev != NULL)
3265                         packet_dev_mc(dev, ml, -1);
3266                 kfree(ml);
3267         }
3268         rtnl_unlock();
3269 }
3270
3271 static int
3272 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3273 {
3274         struct sock *sk = sock->sk;
3275         struct packet_sock *po = pkt_sk(sk);
3276         int ret;
3277
3278         if (level != SOL_PACKET)
3279                 return -ENOPROTOOPT;
3280
3281         switch (optname) {
3282         case PACKET_ADD_MEMBERSHIP:
3283         case PACKET_DROP_MEMBERSHIP:
3284         {
3285                 struct packet_mreq_max mreq;
3286                 int len = optlen;
3287                 memset(&mreq, 0, sizeof(mreq));
3288                 if (len < sizeof(struct packet_mreq))
3289                         return -EINVAL;
3290                 if (len > sizeof(mreq))
3291                         len = sizeof(mreq);
3292                 if (copy_from_user(&mreq, optval, len))
3293                         return -EFAULT;
3294                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3295                         return -EINVAL;
3296                 if (optname == PACKET_ADD_MEMBERSHIP)
3297                         ret = packet_mc_add(sk, &mreq);
3298                 else
3299                         ret = packet_mc_drop(sk, &mreq);
3300                 return ret;
3301         }
3302
3303         case PACKET_RX_RING:
3304         case PACKET_TX_RING:
3305         {
3306                 union tpacket_req_u req_u;
3307                 int len;
3308
3309                 switch (po->tp_version) {
3310                 case TPACKET_V1:
3311                 case TPACKET_V2:
3312                         len = sizeof(req_u.req);
3313                         break;
3314                 case TPACKET_V3:
3315                 default:
3316                         len = sizeof(req_u.req3);
3317                         break;
3318                 }
3319                 if (optlen < len)
3320                         return -EINVAL;
3321                 if (pkt_sk(sk)->has_vnet_hdr)
3322                         return -EINVAL;
3323                 if (copy_from_user(&req_u.req, optval, len))
3324                         return -EFAULT;
3325                 return packet_set_ring(sk, &req_u, 0,
3326                         optname == PACKET_TX_RING);
3327         }
3328         case PACKET_COPY_THRESH:
3329         {
3330                 int val;
3331
3332                 if (optlen != sizeof(val))
3333                         return -EINVAL;
3334                 if (copy_from_user(&val, optval, sizeof(val)))
3335                         return -EFAULT;
3336
3337                 pkt_sk(sk)->copy_thresh = val;
3338                 return 0;
3339         }
3340         case PACKET_VERSION:
3341         {
3342                 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                 switch (val) {
3351                 case TPACKET_V1:
3352                 case TPACKET_V2:
3353                 case TPACKET_V3:
3354                         po->tp_version = val;
3355                         return 0;
3356                 default:
3357                         return -EINVAL;
3358                 }
3359         }
3360         case PACKET_RESERVE:
3361         {
3362                 unsigned int val;
3363
3364                 if (optlen != sizeof(val))
3365                         return -EINVAL;
3366                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3367                         return -EBUSY;
3368                 if (copy_from_user(&val, optval, sizeof(val)))
3369                         return -EFAULT;
3370                 po->tp_reserve = val;
3371                 return 0;
3372         }
3373         case PACKET_LOSS:
3374         {
3375                 unsigned int val;
3376
3377                 if (optlen != sizeof(val))
3378                         return -EINVAL;
3379                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3380                         return -EBUSY;
3381                 if (copy_from_user(&val, optval, sizeof(val)))
3382                         return -EFAULT;
3383                 po->tp_loss = !!val;
3384                 return 0;
3385         }
3386         case PACKET_AUXDATA:
3387         {
3388                 int val;
3389
3390                 if (optlen < sizeof(val))
3391                         return -EINVAL;
3392                 if (copy_from_user(&val, optval, sizeof(val)))
3393                         return -EFAULT;
3394
3395                 po->auxdata = !!val;
3396                 return 0;
3397         }
3398         case PACKET_ORIGDEV:
3399         {
3400                 int val;
3401
3402                 if (optlen < sizeof(val))
3403                         return -EINVAL;
3404                 if (copy_from_user(&val, optval, sizeof(val)))
3405                         return -EFAULT;
3406
3407                 po->origdev = !!val;
3408                 return 0;
3409         }
3410         case PACKET_VNET_HDR:
3411         {
3412                 int val;
3413
3414                 if (sock->type != SOCK_RAW)
3415                         return -EINVAL;
3416                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3417                         return -EBUSY;
3418                 if (optlen < sizeof(val))
3419                         return -EINVAL;
3420                 if (copy_from_user(&val, optval, sizeof(val)))
3421                         return -EFAULT;
3422
3423                 po->has_vnet_hdr = !!val;
3424                 return 0;
3425         }
3426         case PACKET_TIMESTAMP:
3427         {
3428                 int val;
3429
3430                 if (optlen != sizeof(val))
3431                         return -EINVAL;
3432                 if (copy_from_user(&val, optval, sizeof(val)))
3433                         return -EFAULT;
3434
3435                 po->tp_tstamp = val;
3436                 return 0;
3437         }
3438         case PACKET_FANOUT:
3439         {
3440                 int val;
3441
3442                 if (optlen != sizeof(val))
3443                         return -EINVAL;
3444                 if (copy_from_user(&val, optval, sizeof(val)))
3445                         return -EFAULT;
3446
3447                 return fanout_add(sk, val & 0xffff, val >> 16);
3448         }
3449         case PACKET_TX_HAS_OFF:
3450         {
3451                 unsigned int val;
3452
3453                 if (optlen != sizeof(val))
3454                         return -EINVAL;
3455                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3456                         return -EBUSY;
3457                 if (copy_from_user(&val, optval, sizeof(val)))
3458                         return -EFAULT;
3459                 po->tp_tx_has_off = !!val;
3460                 return 0;
3461         }
3462         case PACKET_QDISC_BYPASS:
3463         {
3464                 int val;
3465
3466                 if (optlen != sizeof(val))
3467                         return -EINVAL;
3468                 if (copy_from_user(&val, optval, sizeof(val)))
3469                         return -EFAULT;
3470
3471                 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3472                 return 0;
3473         }
3474         default:
3475                 return -ENOPROTOOPT;
3476         }
3477 }
3478
3479 static int packet_getsockopt(struct socket *sock, int level, int optname,
3480                              char __user *optval, int __user *optlen)
3481 {
3482         int len;
3483         int val, lv = sizeof(val);
3484         struct sock *sk = sock->sk;
3485         struct packet_sock *po = pkt_sk(sk);
3486         void *data = &val;
3487         union tpacket_stats_u st;
3488
3489         if (level != SOL_PACKET)
3490                 return -ENOPROTOOPT;
3491
3492         if (get_user(len, optlen))
3493                 return -EFAULT;
3494
3495         if (len < 0)
3496                 return -EINVAL;
3497
3498         switch (optname) {
3499         case PACKET_STATISTICS:
3500                 spin_lock_bh(&sk->sk_receive_queue.lock);
3501                 memcpy(&st, &po->stats, sizeof(st));
3502                 memset(&po->stats, 0, sizeof(po->stats));
3503                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3504
3505                 if (po->tp_version == TPACKET_V3) {
3506                         lv = sizeof(struct tpacket_stats_v3);
3507                         st.stats3.tp_packets += st.stats3.tp_drops;
3508                         data = &st.stats3;
3509                 } else {
3510                         lv = sizeof(struct tpacket_stats);
3511                         st.stats1.tp_packets += st.stats1.tp_drops;
3512                         data = &st.stats1;
3513                 }
3514
3515                 break;
3516         case PACKET_AUXDATA:
3517                 val = po->auxdata;
3518                 break;
3519         case PACKET_ORIGDEV:
3520                 val = po->origdev;
3521                 break;
3522         case PACKET_VNET_HDR:
3523                 val = po->has_vnet_hdr;
3524                 break;
3525         case PACKET_VERSION:
3526                 val = po->tp_version;
3527                 break;
3528         case PACKET_HDRLEN:
3529                 if (len > sizeof(int))
3530                         len = sizeof(int);
3531                 if (copy_from_user(&val, optval, len))
3532                         return -EFAULT;
3533                 switch (val) {
3534                 case TPACKET_V1:
3535                         val = sizeof(struct tpacket_hdr);
3536                         break;
3537                 case TPACKET_V2:
3538                         val = sizeof(struct tpacket2_hdr);
3539                         break;
3540                 case TPACKET_V3:
3541                         val = sizeof(struct tpacket3_hdr);
3542                         break;
3543                 default:
3544                         return -EINVAL;
3545                 }
3546                 break;
3547         case PACKET_RESERVE:
3548                 val = po->tp_reserve;
3549                 break;
3550         case PACKET_LOSS:
3551                 val = po->tp_loss;
3552                 break;
3553         case PACKET_TIMESTAMP:
3554                 val = po->tp_tstamp;
3555                 break;
3556         case PACKET_FANOUT:
3557                 val = (po->fanout ?
3558                        ((u32)po->fanout->id |
3559                         ((u32)po->fanout->type << 16) |
3560                         ((u32)po->fanout->flags << 24)) :
3561                        0);
3562                 break;
3563         case PACKET_TX_HAS_OFF:
3564                 val = po->tp_tx_has_off;
3565                 break;
3566         case PACKET_QDISC_BYPASS:
3567                 val = packet_use_direct_xmit(po);
3568                 break;
3569         default:
3570                 return -ENOPROTOOPT;
3571         }
3572
3573         if (len > lv)
3574                 len = lv;
3575         if (put_user(len, optlen))
3576                 return -EFAULT;
3577         if (copy_to_user(optval, data, len))
3578                 return -EFAULT;
3579         return 0;
3580 }
3581
3582
3583 static int packet_notifier(struct notifier_block *this,
3584                            unsigned long msg, void *ptr)
3585 {
3586         struct sock *sk;
3587         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3588         struct net *net = dev_net(dev);
3589
3590         rcu_read_lock();
3591         sk_for_each_rcu(sk, &net->packet.sklist) {
3592                 struct packet_sock *po = pkt_sk(sk);
3593
3594                 switch (msg) {
3595                 case NETDEV_UNREGISTER:
3596                         if (po->mclist)
3597                                 packet_dev_mclist_delete(dev, &po->mclist);
3598                         /* fallthrough */
3599
3600                 case NETDEV_DOWN:
3601                         if (dev->ifindex == po->ifindex) {
3602                                 spin_lock(&po->bind_lock);
3603                                 if (po->running) {
3604                                         __unregister_prot_hook(sk, false);
3605                                         sk->sk_err = ENETDOWN;
3606                                         if (!sock_flag(sk, SOCK_DEAD))
3607                                                 sk->sk_error_report(sk);
3608                                 }
3609                                 if (msg == NETDEV_UNREGISTER) {
3610                                         packet_cached_dev_reset(po);
3611                                         po->ifindex = -1;
3612                                         if (po->prot_hook.dev)
3613                                                 dev_put(po->prot_hook.dev);
3614                                         po->prot_hook.dev = NULL;
3615                                 }
3616                                 spin_unlock(&po->bind_lock);
3617                         }
3618                         break;
3619                 case NETDEV_UP:
3620                         if (dev->ifindex == po->ifindex) {
3621                                 spin_lock(&po->bind_lock);
3622                                 if (po->num)
3623                                         register_prot_hook(sk);
3624                                 spin_unlock(&po->bind_lock);
3625                         }
3626                         break;
3627                 }
3628         }
3629         rcu_read_unlock();
3630         return NOTIFY_DONE;
3631 }
3632
3633
3634 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3635                         unsigned long arg)
3636 {
3637         struct sock *sk = sock->sk;
3638
3639         switch (cmd) {
3640         case SIOCOUTQ:
3641         {
3642                 int amount = sk_wmem_alloc_get(sk);
3643
3644                 return put_user(amount, (int __user *)arg);
3645         }
3646         case SIOCINQ:
3647         {
3648                 struct sk_buff *skb;
3649                 int amount = 0;
3650
3651                 spin_lock_bh(&sk->sk_receive_queue.lock);
3652                 skb = skb_peek(&sk->sk_receive_queue);
3653                 if (skb)
3654                         amount = skb->len;
3655                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3656                 return put_user(amount, (int __user *)arg);
3657         }
3658         case SIOCGSTAMP:
3659                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3660         case SIOCGSTAMPNS:
3661                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3662
3663 #ifdef CONFIG_INET
3664         case SIOCADDRT:
3665         case SIOCDELRT:
3666         case SIOCDARP:
3667         case SIOCGARP:
3668         case SIOCSARP:
3669         case SIOCGIFADDR:
3670         case SIOCSIFADDR:
3671         case SIOCGIFBRDADDR:
3672         case SIOCSIFBRDADDR:
3673         case SIOCGIFNETMASK:
3674         case SIOCSIFNETMASK:
3675         case SIOCGIFDSTADDR:
3676         case SIOCSIFDSTADDR:
3677         case SIOCSIFFLAGS:
3678                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3679 #endif
3680
3681         default:
3682                 return -ENOIOCTLCMD;
3683         }
3684         return 0;
3685 }
3686
3687 static unsigned int packet_poll(struct file *file, struct socket *sock,
3688                                 poll_table *wait)
3689 {
3690         struct sock *sk = sock->sk;
3691         struct packet_sock *po = pkt_sk(sk);
3692         unsigned int mask = datagram_poll(file, sock, wait);
3693
3694         spin_lock_bh(&sk->sk_receive_queue.lock);
3695         if (po->rx_ring.pg_vec) {
3696                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3697                         TP_STATUS_KERNEL))
3698                         mask |= POLLIN | POLLRDNORM;
3699         }
3700         spin_unlock_bh(&sk->sk_receive_queue.lock);
3701         spin_lock_bh(&sk->sk_write_queue.lock);
3702         if (po->tx_ring.pg_vec) {
3703                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3704                         mask |= POLLOUT | POLLWRNORM;
3705         }
3706         spin_unlock_bh(&sk->sk_write_queue.lock);
3707         return mask;
3708 }
3709
3710
3711 /* Dirty? Well, I still did not learn better way to account
3712  * for user mmaps.
3713  */
3714
3715 static void packet_mm_open(struct vm_area_struct *vma)
3716 {
3717         struct file *file = vma->vm_file;
3718         struct socket *sock = file->private_data;
3719         struct sock *sk = sock->sk;
3720
3721         if (sk)
3722                 atomic_inc(&pkt_sk(sk)->mapped);
3723 }
3724
3725 static void packet_mm_close(struct vm_area_struct *vma)
3726 {
3727         struct file *file = vma->vm_file;
3728         struct socket *sock = file->private_data;
3729         struct sock *sk = sock->sk;
3730
3731         if (sk)
3732                 atomic_dec(&pkt_sk(sk)->mapped);
3733 }
3734
3735 static const struct vm_operations_struct packet_mmap_ops = {
3736         .open   =       packet_mm_open,
3737         .close  =       packet_mm_close,
3738 };
3739
3740 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3741                         unsigned int len)
3742 {
3743         int i;
3744
3745         for (i = 0; i < len; i++) {
3746                 if (likely(pg_vec[i].buffer)) {
3747                         if (is_vmalloc_addr(pg_vec[i].buffer))
3748                                 vfree(pg_vec[i].buffer);
3749                         else
3750                                 free_pages((unsigned long)pg_vec[i].buffer,
3751                                            order);
3752                         pg_vec[i].buffer = NULL;
3753                 }
3754         }
3755         kfree(pg_vec);
3756 }
3757
3758 static char *alloc_one_pg_vec_page(unsigned long order)
3759 {
3760         char *buffer;
3761         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3762                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3763
3764         buffer = (char *) __get_free_pages(gfp_flags, order);
3765         if (buffer)
3766                 return buffer;
3767
3768         /* __get_free_pages failed, fall back to vmalloc */
3769         buffer = vzalloc((1 << order) * PAGE_SIZE);
3770         if (buffer)
3771                 return buffer;
3772
3773         /* vmalloc failed, lets dig into swap here */
3774         gfp_flags &= ~__GFP_NORETRY;
3775         buffer = (char *) __get_free_pages(gfp_flags, order);
3776         if (buffer)
3777                 return buffer;
3778
3779         /* complete and utter failure */
3780         return NULL;
3781 }
3782
3783 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3784 {
3785         unsigned int block_nr = req->tp_block_nr;
3786         struct pgv *pg_vec;
3787         int i;
3788
3789         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3790         if (unlikely(!pg_vec))
3791                 goto out;
3792
3793         for (i = 0; i < block_nr; i++) {
3794                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3795                 if (unlikely(!pg_vec[i].buffer))
3796                         goto out_free_pgvec;
3797         }
3798
3799 out:
3800         return pg_vec;
3801
3802 out_free_pgvec:
3803         free_pg_vec(pg_vec, order, block_nr);
3804         pg_vec = NULL;
3805         goto out;
3806 }
3807
3808 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3809                 int closing, int tx_ring)
3810 {
3811         struct pgv *pg_vec = NULL;
3812         struct packet_sock *po = pkt_sk(sk);
3813         int was_running, order = 0;
3814         struct packet_ring_buffer *rb;
3815         struct sk_buff_head *rb_queue;
3816         __be16 num;
3817         int err = -EINVAL;
3818         /* Added to avoid minimal code churn */
3819         struct tpacket_req *req = &req_u->req;
3820
3821         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3822         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3823                 WARN(1, "Tx-ring is not supported.\n");
3824                 goto out;
3825         }
3826
3827         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3828         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3829
3830         err = -EBUSY;
3831         if (!closing) {
3832                 if (atomic_read(&po->mapped))
3833                         goto out;
3834                 if (packet_read_pending(rb))
3835                         goto out;
3836         }
3837
3838         if (req->tp_block_nr) {
3839                 /* Sanity tests and some calculations */
3840                 err = -EBUSY;
3841                 if (unlikely(rb->pg_vec))
3842                         goto out;
3843
3844                 switch (po->tp_version) {
3845                 case TPACKET_V1:
3846                         po->tp_hdrlen = TPACKET_HDRLEN;
3847                         break;
3848                 case TPACKET_V2:
3849                         po->tp_hdrlen = TPACKET2_HDRLEN;
3850                         break;
3851                 case TPACKET_V3:
3852                         po->tp_hdrlen = TPACKET3_HDRLEN;
3853                         break;
3854                 }
3855
3856                 err = -EINVAL;
3857                 if (unlikely((int)req->tp_block_size <= 0))
3858                         goto out;
3859                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3860                         goto out;
3861                 if (po->tp_version >= TPACKET_V3 &&
3862                     (int)(req->tp_block_size -
3863                           BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
3864                         goto out;
3865                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3866                                         po->tp_reserve))
3867                         goto out;
3868                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3869                         goto out;
3870
3871                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3872                 if (unlikely(rb->frames_per_block <= 0))
3873                         goto out;
3874                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3875                                         req->tp_frame_nr))
3876                         goto out;
3877
3878                 err = -ENOMEM;
3879                 order = get_order(req->tp_block_size);
3880                 pg_vec = alloc_pg_vec(req, order);
3881                 if (unlikely(!pg_vec))
3882                         goto out;
3883                 switch (po->tp_version) {
3884                 case TPACKET_V3:
3885                 /* Transmit path is not supported. We checked
3886                  * it above but just being paranoid
3887                  */
3888                         if (!tx_ring)
3889                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3890                         break;
3891                 default:
3892                         break;
3893                 }
3894         }
3895         /* Done */
3896         else {
3897                 err = -EINVAL;
3898                 if (unlikely(req->tp_frame_nr))
3899                         goto out;
3900         }
3901
3902         lock_sock(sk);
3903
3904         /* Detach socket from network */
3905         spin_lock(&po->bind_lock);
3906         was_running = po->running;
3907         num = po->num;
3908         if (was_running) {
3909                 po->num = 0;
3910                 __unregister_prot_hook(sk, false);
3911         }
3912         spin_unlock(&po->bind_lock);
3913
3914         synchronize_net();
3915
3916         err = -EBUSY;
3917         mutex_lock(&po->pg_vec_lock);
3918         if (closing || atomic_read(&po->mapped) == 0) {
3919                 err = 0;
3920                 spin_lock_bh(&rb_queue->lock);
3921                 swap(rb->pg_vec, pg_vec);
3922                 rb->frame_max = (req->tp_frame_nr - 1);
3923                 rb->head = 0;
3924                 rb->frame_size = req->tp_frame_size;
3925                 spin_unlock_bh(&rb_queue->lock);
3926
3927                 swap(rb->pg_vec_order, order);
3928                 swap(rb->pg_vec_len, req->tp_block_nr);
3929
3930                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3931                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3932                                                 tpacket_rcv : packet_rcv;
3933                 skb_queue_purge(rb_queue);
3934                 if (atomic_read(&po->mapped))
3935                         pr_err("packet_mmap: vma is busy: %d\n",
3936                                atomic_read(&po->mapped));
3937         }
3938         mutex_unlock(&po->pg_vec_lock);
3939
3940         spin_lock(&po->bind_lock);
3941         if (was_running) {
3942                 po->num = num;
3943                 register_prot_hook(sk);
3944         }
3945         spin_unlock(&po->bind_lock);
3946         if (closing && (po->tp_version > TPACKET_V2)) {
3947                 /* Because we don't support block-based V3 on tx-ring */
3948                 if (!tx_ring)
3949                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3950         }
3951         release_sock(sk);
3952
3953         if (pg_vec)
3954                 free_pg_vec(pg_vec, order, req->tp_block_nr);
3955 out:
3956         return err;
3957 }
3958
3959 static int packet_mmap(struct file *file, struct socket *sock,
3960                 struct vm_area_struct *vma)
3961 {
3962         struct sock *sk = sock->sk;
3963         struct packet_sock *po = pkt_sk(sk);
3964         unsigned long size, expected_size;
3965         struct packet_ring_buffer *rb;
3966         unsigned long start;
3967         int err = -EINVAL;
3968         int i;
3969
3970         if (vma->vm_pgoff)
3971                 return -EINVAL;
3972
3973         mutex_lock(&po->pg_vec_lock);
3974
3975         expected_size = 0;
3976         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3977                 if (rb->pg_vec) {
3978                         expected_size += rb->pg_vec_len
3979                                                 * rb->pg_vec_pages
3980                                                 * PAGE_SIZE;
3981                 }
3982         }
3983
3984         if (expected_size == 0)
3985                 goto out;
3986
3987         size = vma->vm_end - vma->vm_start;
3988         if (size != expected_size)
3989                 goto out;
3990
3991         start = vma->vm_start;
3992         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3993                 if (rb->pg_vec == NULL)
3994                         continue;
3995
3996                 for (i = 0; i < rb->pg_vec_len; i++) {
3997                         struct page *page;
3998                         void *kaddr = rb->pg_vec[i].buffer;
3999                         int pg_num;
4000
4001                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4002                                 page = pgv_to_page(kaddr);
4003                                 err = vm_insert_page(vma, start, page);
4004                                 if (unlikely(err))
4005                                         goto out;
4006                                 start += PAGE_SIZE;
4007                                 kaddr += PAGE_SIZE;
4008                         }
4009                 }
4010         }
4011
4012         atomic_inc(&po->mapped);
4013         vma->vm_ops = &packet_mmap_ops;
4014         err = 0;
4015
4016 out:
4017         mutex_unlock(&po->pg_vec_lock);
4018         return err;
4019 }
4020
4021 static const struct proto_ops packet_ops_spkt = {
4022         .family =       PF_PACKET,
4023         .owner =        THIS_MODULE,
4024         .release =      packet_release,
4025         .bind =         packet_bind_spkt,
4026         .connect =      sock_no_connect,
4027         .socketpair =   sock_no_socketpair,
4028         .accept =       sock_no_accept,
4029         .getname =      packet_getname_spkt,
4030         .poll =         datagram_poll,
4031         .ioctl =        packet_ioctl,
4032         .listen =       sock_no_listen,
4033         .shutdown =     sock_no_shutdown,
4034         .setsockopt =   sock_no_setsockopt,
4035         .getsockopt =   sock_no_getsockopt,
4036         .sendmsg =      packet_sendmsg_spkt,
4037         .recvmsg =      packet_recvmsg,
4038         .mmap =         sock_no_mmap,
4039         .sendpage =     sock_no_sendpage,
4040 };
4041
4042 static const struct proto_ops packet_ops = {
4043         .family =       PF_PACKET,
4044         .owner =        THIS_MODULE,
4045         .release =      packet_release,
4046         .bind =         packet_bind,
4047         .connect =      sock_no_connect,
4048         .socketpair =   sock_no_socketpair,
4049         .accept =       sock_no_accept,
4050         .getname =      packet_getname,
4051         .poll =         packet_poll,
4052         .ioctl =        packet_ioctl,
4053         .listen =       sock_no_listen,
4054         .shutdown =     sock_no_shutdown,
4055         .setsockopt =   packet_setsockopt,
4056         .getsockopt =   packet_getsockopt,
4057         .sendmsg =      packet_sendmsg,
4058         .recvmsg =      packet_recvmsg,
4059         .mmap =         packet_mmap,
4060         .sendpage =     sock_no_sendpage,
4061 };
4062
4063 static const struct net_proto_family packet_family_ops = {
4064         .family =       PF_PACKET,
4065         .create =       packet_create,
4066         .owner  =       THIS_MODULE,
4067 };
4068
4069 static struct notifier_block packet_netdev_notifier = {
4070         .notifier_call =        packet_notifier,
4071 };
4072
4073 #ifdef CONFIG_PROC_FS
4074
4075 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4076         __acquires(RCU)
4077 {
4078         struct net *net = seq_file_net(seq);
4079
4080         rcu_read_lock();
4081         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4082 }
4083
4084 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4085 {
4086         struct net *net = seq_file_net(seq);
4087         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4088 }
4089
4090 static void packet_seq_stop(struct seq_file *seq, void *v)
4091         __releases(RCU)
4092 {
4093         rcu_read_unlock();
4094 }
4095
4096 static int packet_seq_show(struct seq_file *seq, void *v)
4097 {
4098         if (v == SEQ_START_TOKEN)
4099                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
4100         else {
4101                 struct sock *s = sk_entry(v);
4102                 const struct packet_sock *po = pkt_sk(s);
4103
4104                 seq_printf(seq,
4105                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4106                            s,
4107                            atomic_read(&s->sk_refcnt),
4108                            s->sk_type,
4109                            ntohs(po->num),
4110                            po->ifindex,
4111                            po->running,
4112                            atomic_read(&s->sk_rmem_alloc),
4113                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4114                            sock_i_ino(s));
4115         }
4116
4117         return 0;
4118 }
4119
4120 static const struct seq_operations packet_seq_ops = {
4121         .start  = packet_seq_start,
4122         .next   = packet_seq_next,
4123         .stop   = packet_seq_stop,
4124         .show   = packet_seq_show,
4125 };
4126
4127 static int packet_seq_open(struct inode *inode, struct file *file)
4128 {
4129         return seq_open_net(inode, file, &packet_seq_ops,
4130                             sizeof(struct seq_net_private));
4131 }
4132
4133 static const struct file_operations packet_seq_fops = {
4134         .owner          = THIS_MODULE,
4135         .open           = packet_seq_open,
4136         .read           = seq_read,
4137         .llseek         = seq_lseek,
4138         .release        = seq_release_net,
4139 };
4140
4141 #endif
4142
4143 static int __net_init packet_net_init(struct net *net)
4144 {
4145         mutex_init(&net->packet.sklist_lock);
4146         INIT_HLIST_HEAD(&net->packet.sklist);
4147
4148         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4149                 return -ENOMEM;
4150
4151         return 0;
4152 }
4153
4154 static void __net_exit packet_net_exit(struct net *net)
4155 {
4156         remove_proc_entry("packet", net->proc_net);
4157 }
4158
4159 static struct pernet_operations packet_net_ops = {
4160         .init = packet_net_init,
4161         .exit = packet_net_exit,
4162 };
4163
4164
4165 static void __exit packet_exit(void)
4166 {
4167         unregister_netdevice_notifier(&packet_netdev_notifier);
4168         unregister_pernet_subsys(&packet_net_ops);
4169         sock_unregister(PF_PACKET);
4170         proto_unregister(&packet_proto);
4171 }
4172
4173 static int __init packet_init(void)
4174 {
4175         int rc = proto_register(&packet_proto, 0);
4176
4177         if (rc != 0)
4178                 goto out;
4179
4180         sock_register(&packet_family_ops);
4181         register_pernet_subsys(&packet_net_ops);
4182         register_netdevice_notifier(&packet_netdev_notifier);
4183 out:
4184         return rc;
4185 }
4186
4187 module_init(packet_init);
4188 module_exit(packet_exit);
4189 MODULE_LICENSE("GPL");
4190 MODULE_ALIAS_NETPROTO(PF_PACKET);