net/packet/af_packet.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              PACKET - implements raw packet sockets.
   7  *
   8  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  11  *
  12  * Fixes:
  13  *              Alan Cox        :       verify_area() now used correctly
  14  *              Alan Cox        :       new skbuff lists, look ma no backlogs!
  15  *              Alan Cox        :       tidied skbuff lists.
  16  *              Alan Cox        :       Now uses generic datagram routines I
  17  *                                      added. Also fixed the peek/read crash
  18  *                                      from all old Linux datagram code.
  19  *              Alan Cox        :       Uses the improved datagram code.
  20  *              Alan Cox        :       Added NULL's for socket options.
  21  *              Alan Cox        :       Re-commented the code.
  22  *              Alan Cox        :       Use new kernel side addressing
  23  *              Rob Janssen     :       Correct MTU usage.
  24  *              Dave Platt      :       Counter leaks caused by incorrect
  25  *                                      interrupt locking and some slightly
  26  *                                      dubious gcc output. Can you read
  27  *                                      compiler: it said _VOLATILE_
  28  *      Richard Kooijman        :       Timestamp fixes.
  29  *              Alan Cox        :       New buffers. Use sk->mac.raw.
  30  *              Alan Cox        :       sendmsg/recvmsg support.
  31  *              Alan Cox        :       Protocol setting support
  32  *      Alexey Kuznetsov        :       Untied from IPv4 stack.
  33  *      Cyrus Durgin            :       Fixed kerneld for kmod.
  34  *      Michal Ostrowski        :       Module initialization cleanup.
  35  *         Ulises Alonso        :       Frame number limit removal and
  36  *                                      packet_set_ring memory leak.
  37  *              Eric Biederman  :       Allow for > 8 byte hardware addresses.
  38  *                                      The convention is that longer addresses
  39  *                                      will simply extend the hardware address
  40  *                                      byte arrays at the end of sockaddr_ll
  41  *                                      and packet_mreq.
  42  *              Johann Baudy    :       Added TX RING.
  43  *              Chetan Loke     :       Implemented TPACKET_V3 block abstraction
  44  *                                      layer.
  45  *                                      Copyright (C) 2011, <lokec@ccs.neu.edu>
  46  *
  47  *
  48  *              This program is free software; you can redistribute it and/or
  49  *              modify it under the terms of the GNU General Public License
  50  *              as published by the Free Software Foundation; either version
  51  *              2 of the License, or (at your option) any later version.
  52  *
  53  */
  54
  55 #include <linux/types.h>
  56 #include <linux/mm.h>
  57 #include <linux/capability.h>
  58 #include <linux/fcntl.h>
  59 #include <linux/socket.h>
  60 #include <linux/in.h>
  61 #include <linux/inet.h>
  62 #include <linux/netdevice.h>
  63 #include <linux/if_packet.h>
  64 #include <linux/wireless.h>
  65 #include <linux/kernel.h>
  66 #include <linux/kmod.h>
  67 #include <linux/slab.h>
  68 #include <linux/vmalloc.h>
  69 #include <net/net_namespace.h>
  70 #include <net/ip.h>
  71 #include <net/protocol.h>
  72 #include <linux/skbuff.h>
  73 #include <net/sock.h>
  74 #include <linux/errno.h>
  75 #include <linux/timer.h>
  76 #include <asm/uaccess.h>
  77 #include <asm/ioctls.h>
  78 #include <asm/page.h>
  79 #include <asm/cacheflush.h>
  80 #include <asm/io.h>
  81 #include <linux/proc_fs.h>
  82 #include <linux/seq_file.h>
  83 #include <linux/poll.h>
  84 #include <linux/module.h>
  85 #include <linux/init.h>
  86 #include <linux/mutex.h>
  87 #include <linux/if_vlan.h>
  88 #include <linux/virtio_net.h>
  89 #include <linux/errqueue.h>
  90 #include <linux/net_tstamp.h>
  91 #include <linux/percpu.h>
  92 #ifdef CONFIG_INET
  93 #include <net/inet_common.h>
  94 #endif
  95
  96 #include "internal.h"
  97
  98 /*
  99    Assumptions:
 100    - if device has no dev->hard_header routine, it adds and removes ll header
 101      inside itself. In this case ll header is invisible outside of device,
 102      but higher levels still should reserve dev->hard_header_len.
 103      Some devices are enough clever to reallocate skb, when header
 104      will not fit to reserved space (tunnel), another ones are silly
 105      (PPP).
 106    - packet socket receives packets with pulled ll header,
 107      so that SOCK_RAW should push it back.
 108
 109 On receive:
 110 -----------
 111
 112 Incoming, dev->hard_header!=NULL
 113    mac_header -> ll header
 114    data       -> data
 115
 116 Outgoing, dev->hard_header!=NULL
 117    mac_header -> ll header
 118    data       -> ll header
 119
 120 Incoming, dev->hard_header==NULL
 121    mac_header -> UNKNOWN position. It is very likely, that it points to ll
 122                  header.  PPP makes it, that is wrong, because introduce
 123                  assymetry between rx and tx paths.
 124    data       -> data
 125
 126 Outgoing, dev->hard_header==NULL
 127    mac_header -> data. ll header is still not built!
 128    data       -> data
 129
 130 Resume
 131   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 132
 133
 134 On transmit:
 135 ------------
 136
 137 dev->hard_header != NULL
 138    mac_header -> ll header
 139    data       -> ll header
 140
 141 dev->hard_header == NULL (ll header is added by device, we cannot control it)
 142    mac_header -> data
 143    data       -> data
 144
 145    We should set nh.raw on output to correct posistion,
 146    packet classifier depends on it.
 147  */
 148
 149 /* Private packet socket structures. */
 150
 151 /* identical to struct packet_mreq except it has
 152  * a longer address field.
 153  */
 154 struct packet_mreq_max {
 155         int             mr_ifindex;
 156         unsigned short  mr_type;
 157         unsigned short  mr_alen;
 158         unsigned char   mr_address[MAX_ADDR_LEN];
 159 };
 160
 161 union tpacket_uhdr {
 162         struct tpacket_hdr  *h1;
 163         struct tpacket2_hdr *h2;
 164         struct tpacket3_hdr *h3;
 165         void *raw;
 166 };
 167
 168 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 169                 int closing, int tx_ring);
 170
 171 #define V3_ALIGNMENT    (8)
 172
 173 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 174
 175 #define BLK_PLUS_PRIV(sz_of_priv) \
 176         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 177
 178 #define PGV_FROM_VMALLOC 1
 179
 180 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
 181 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
 182 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
 183 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
 184 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
 185 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
 186 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
 187
 188 struct packet_sock;
 189 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
 190 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 191                        struct packet_type *pt, struct net_device *orig_dev);
 192
 193 static void *packet_previous_frame(struct packet_sock *po,
 194                 struct packet_ring_buffer *rb,
 195                 int status);
 196 static void packet_increment_head(struct packet_ring_buffer *buff);
 197 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
 198                         struct tpacket_block_desc *);
 199 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 200                         struct packet_sock *);
 201 static void prb_retire_current_block(struct tpacket_kbdq_core *,
 202                 struct packet_sock *, unsigned int status);
 203 static int prb_queue_frozen(struct tpacket_kbdq_core *);
 204 static void prb_open_block(struct tpacket_kbdq_core *,
 205                 struct tpacket_block_desc *);
 206 static void prb_retire_rx_blk_timer_expired(unsigned long);
 207 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 208 static void prb_init_blk_timer(struct packet_sock *,
 209                 struct tpacket_kbdq_core *,
 210                 void (*func) (unsigned long));
 211 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 212 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 213                 struct tpacket3_hdr *);
 214 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 215                 struct tpacket3_hdr *);
 216 static void packet_flush_mclist(struct sock *sk);
 217
 218 struct packet_skb_cb {
 219         unsigned int origlen;
 220         union {
 221                 struct sockaddr_pkt pkt;
 222                 struct sockaddr_ll ll;
 223         } sa;
 224 };
 225
 226 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 227
 228 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 229 #define GET_PBLOCK_DESC(x, bid) \
 230         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 231 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 232         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 233 #define GET_NEXT_PRB_BLK_NUM(x) \
 234         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 235         ((x)->kactive_blk_num+1) : 0)
 236
 237 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 238 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 239
 240 static int packet_direct_xmit(struct sk_buff *skb)
 241 {
 242         struct net_device *dev = skb->dev;
 243         const struct net_device_ops *ops = dev->netdev_ops;
 244         netdev_features_t features;
 245         struct netdev_queue *txq;
 246         int ret = NETDEV_TX_BUSY;
 247         u16 queue_map;
 248
 249         if (unlikely(!netif_running(dev) ||
 250                      !netif_carrier_ok(dev)))
 251                 goto drop;
 252
 253         features = netif_skb_features(skb);
 254         if (skb_needs_linearize(skb, features) &&
 255             __skb_linearize(skb))
 256                 goto drop;
 257
 258         queue_map = skb_get_queue_mapping(skb);
 259         txq = netdev_get_tx_queue(dev, queue_map);
 260
 261         local_bh_disable();
 262
 263         HARD_TX_LOCK(dev, txq, smp_processor_id());
 264         if (!netif_xmit_frozen_or_drv_stopped(txq)) {
 265                 ret = ops->ndo_start_xmit(skb, dev);
 266                 if (ret == NETDEV_TX_OK)
 267                         txq_trans_update(txq);
 268         }
 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 __pure struct page *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                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 708                         if (!prb_dispatch_next_block(pkc, po))
 709                                 goto refresh_timer;
 710                         else
 711                                 goto out;
 712                 } else {
 713                         /* Case 1. Queue was frozen because user-space was
 714                          *         lagging behind.
 715                          */
 716                         if (prb_curr_blk_in_use(pkc, pbd)) {
 717                                 /*
 718                                  * Ok, user-space is still behind.
 719                                  * So just refresh the timer.
 720                                  */
 721                                 goto refresh_timer;
 722                         } else {
 723                                /* Case 2. queue was frozen,user-space caught up,
 724                                 * now the link went idle && the timer fired.
 725                                 * We don't have a block to close.So we open this
 726                                 * block and restart the timer.
 727                                 * opening a block thaws the queue,restarts timer
 728                                 * Thawing/timer-refresh is a side effect.
 729                                 */
 730                                 prb_open_block(pkc, pbd);
 731                                 goto out;
 732                         }
 733                 }
 734         }
 735
 736 refresh_timer:
 737         _prb_refresh_rx_retire_blk_timer(pkc);
 738
 739 out:
 740         spin_unlock(&po->sk.sk_receive_queue.lock);
 741 }
 742
 743 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 744                 struct tpacket_block_desc *pbd1, __u32 status)
 745 {
 746         /* Flush everything minus the block header */
 747
 748 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 749         u8 *start, *end;
 750
 751         start = (u8 *)pbd1;
 752
 753         /* Skip the block header(we know header WILL fit in 4K) */
 754         start += PAGE_SIZE;
 755
 756         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 757         for (; start < end; start += PAGE_SIZE)
 758                 flush_dcache_page(pgv_to_page(start));
 759
 760         smp_wmb();
 761 #endif
 762
 763         /* Now update the block status. */
 764
 765         BLOCK_STATUS(pbd1) = status;
 766
 767         /* Flush the block header */
 768
 769 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 770         start = (u8 *)pbd1;
 771         flush_dcache_page(pgv_to_page(start));
 772
 773         smp_wmb();
 774 #endif
 775 }
 776
 777 /*
 778  * Side effect:
 779  *
 780  * 1) flush the block
 781  * 2) Increment active_blk_num
 782  *
 783  * Note:We DONT refresh the timer on purpose.
 784  *      Because almost always the next block will be opened.
 785  */
 786 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 787                 struct tpacket_block_desc *pbd1,
 788                 struct packet_sock *po, unsigned int stat)
 789 {
 790         __u32 status = TP_STATUS_USER | stat;
 791
 792         struct tpacket3_hdr *last_pkt;
 793         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 794
 795         if (po->stats.stats3.tp_drops)
 796                 status |= TP_STATUS_LOSING;
 797
 798         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 799         last_pkt->tp_next_offset = 0;
 800
 801         /* Get the ts of the last pkt */
 802         if (BLOCK_NUM_PKTS(pbd1)) {
 803                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 804                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 805         } else {
 806                 /* Ok, we tmo'd - so get the current time */
 807                 struct timespec ts;
 808                 getnstimeofday(&ts);
 809                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 810                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 811         }
 812
 813         smp_wmb();
 814
 815         /* Flush the block */
 816         prb_flush_block(pkc1, pbd1, status);
 817
 818         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 819 }
 820
 821 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 822 {
 823         pkc->reset_pending_on_curr_blk = 0;
 824 }
 825
 826 /*
 827  * Side effect of opening a block:
 828  *
 829  * 1) prb_queue is thawed.
 830  * 2) retire_blk_timer is refreshed.
 831  *
 832  */
 833 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 834         struct tpacket_block_desc *pbd1)
 835 {
 836         struct timespec ts;
 837         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 838
 839         smp_rmb();
 840
 841         /* We could have just memset this but we will lose the
 842          * flexibility of making the priv area sticky
 843          */
 844
 845         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 846         BLOCK_NUM_PKTS(pbd1) = 0;
 847         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 848
 849         getnstimeofday(&ts);
 850
 851         h1->ts_first_pkt.ts_sec = ts.tv_sec;
 852         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 853
 854         pkc1->pkblk_start = (char *)pbd1;
 855         pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 856
 857         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 858         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 859
 860         pbd1->version = pkc1->version;
 861         pkc1->prev = pkc1->nxt_offset;
 862         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 863
 864         prb_thaw_queue(pkc1);
 865         _prb_refresh_rx_retire_blk_timer(pkc1);
 866
 867         smp_wmb();
 868 }
 869
 870 /*
 871  * Queue freeze logic:
 872  * 1) Assume tp_block_nr = 8 blocks.
 873  * 2) At time 't0', user opens Rx ring.
 874  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 875  * 4) user-space is either sleeping or processing block '0'.
 876  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 877  *    it will close block-7,loop around and try to fill block '0'.
 878  *    call-flow:
 879  *    __packet_lookup_frame_in_block
 880  *      prb_retire_current_block()
 881  *      prb_dispatch_next_block()
 882  *        |->(BLOCK_STATUS == USER) evaluates to true
 883  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 884  * 6) Now there are two cases:
 885  *    6.1) Link goes idle right after the queue is frozen.
 886  *         But remember, the last open_block() refreshed the timer.
 887  *         When this timer expires,it will refresh itself so that we can
 888  *         re-open block-0 in near future.
 889  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 890  *         case and __packet_lookup_frame_in_block will check if block-0
 891  *         is free and can now be re-used.
 892  */
 893 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 894                                   struct packet_sock *po)
 895 {
 896         pkc->reset_pending_on_curr_blk = 1;
 897         po->stats.stats3.tp_freeze_q_cnt++;
 898 }
 899
 900 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 901
 902 /*
 903  * If the next block is free then we will dispatch it
 904  * and return a good offset.
 905  * Else, we will freeze the queue.
 906  * So, caller must check the return value.
 907  */
 908 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 909                 struct packet_sock *po)
 910 {
 911         struct tpacket_block_desc *pbd;
 912
 913         smp_rmb();
 914
 915         /* 1. Get current block num */
 916         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 917
 918         /* 2. If this block is currently in_use then freeze the queue */
 919         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 920                 prb_freeze_queue(pkc, po);
 921                 return NULL;
 922         }
 923
 924         /*
 925          * 3.
 926          * open this block and return the offset where the first packet
 927          * needs to get stored.
 928          */
 929         prb_open_block(pkc, pbd);
 930         return (void *)pkc->nxt_offset;
 931 }
 932
 933 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 934                 struct packet_sock *po, unsigned int status)
 935 {
 936         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 937
 938         /* retire/close the current block */
 939         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 940                 /*
 941                  * Plug the case where copy_bits() is in progress on
 942                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 943                  * have space to copy the pkt in the current block and
 944                  * called prb_retire_current_block()
 945                  *
 946                  * We don't need to worry about the TMO case because
 947                  * the timer-handler already handled this case.
 948                  */
 949                 if (!(status & TP_STATUS_BLK_TMO)) {
 950                         while (atomic_read(&pkc->blk_fill_in_prog)) {
 951                                 /* Waiting for skb_copy_bits to finish... */
 952                                 cpu_relax();
 953                         }
 954                 }
 955                 prb_close_block(pkc, pbd, po, status);
 956                 return;
 957         }
 958 }
 959
 960 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
 961                                       struct tpacket_block_desc *pbd)
 962 {
 963         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 964 }
 965
 966 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 967 {
 968         return pkc->reset_pending_on_curr_blk;
 969 }
 970
 971 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 972 {
 973         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 974         atomic_dec(&pkc->blk_fill_in_prog);
 975 }
 976
 977 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 978                         struct tpacket3_hdr *ppd)
 979 {
 980         ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
 981 }
 982
 983 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 984                         struct tpacket3_hdr *ppd)
 985 {
 986         ppd->hv1.tp_rxhash = 0;
 987 }
 988
 989 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
 990                         struct tpacket3_hdr *ppd)
 991 {
 992         if (vlan_tx_tag_present(pkc->skb)) {
 993                 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
 994                 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
 995                 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
 996         } else {
 997                 ppd->hv1.tp_vlan_tci = 0;
 998                 ppd->hv1.tp_vlan_tpid = 0;
 999                 ppd->tp_status = TP_STATUS_AVAILABLE;
1000         }
1001 }
1002
1003 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1004                         struct tpacket3_hdr *ppd)
1005 {
1006         ppd->hv1.tp_padding = 0;
1007         prb_fill_vlan_info(pkc, ppd);
1008
1009         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1010                 prb_fill_rxhash(pkc, ppd);
1011         else
1012                 prb_clear_rxhash(pkc, ppd);
1013 }
1014
1015 static void prb_fill_curr_block(char *curr,
1016                                 struct tpacket_kbdq_core *pkc,
1017                                 struct tpacket_block_desc *pbd,
1018                                 unsigned int len)
1019 {
1020         struct tpacket3_hdr *ppd;
1021
1022         ppd  = (struct tpacket3_hdr *)curr;
1023         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1024         pkc->prev = curr;
1025         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1026         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1027         BLOCK_NUM_PKTS(pbd) += 1;
1028         atomic_inc(&pkc->blk_fill_in_prog);
1029         prb_run_all_ft_ops(pkc, ppd);
1030 }
1031
1032 /* Assumes caller has the sk->rx_queue.lock */
1033 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1034                                             struct sk_buff *skb,
1035                                                 int status,
1036                                             unsigned int len
1037                                             )
1038 {
1039         struct tpacket_kbdq_core *pkc;
1040         struct tpacket_block_desc *pbd;
1041         char *curr, *end;
1042
1043         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1044         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1045
1046         /* Queue is frozen when user space is lagging behind */
1047         if (prb_queue_frozen(pkc)) {
1048                 /*
1049                  * Check if that last block which caused the queue to freeze,
1050                  * is still in_use by user-space.
1051                  */
1052                 if (prb_curr_blk_in_use(pkc, pbd)) {
1053                         /* Can't record this packet */
1054                         return NULL;
1055                 } else {
1056                         /*
1057                          * Ok, the block was released by user-space.
1058                          * Now let's open that block.
1059                          * opening a block also thaws the queue.
1060                          * Thawing is a side effect.
1061                          */
1062                         prb_open_block(pkc, pbd);
1063                 }
1064         }
1065
1066         smp_mb();
1067         curr = pkc->nxt_offset;
1068         pkc->skb = skb;
1069         end = (char *)pbd + pkc->kblk_size;
1070
1071         /* first try the current block */
1072         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1073                 prb_fill_curr_block(curr, pkc, pbd, len);
1074                 return (void *)curr;
1075         }
1076
1077         /* Ok, close the current block */
1078         prb_retire_current_block(pkc, po, 0);
1079
1080         /* Now, try to dispatch the next block */
1081         curr = (char *)prb_dispatch_next_block(pkc, po);
1082         if (curr) {
1083                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1084                 prb_fill_curr_block(curr, pkc, pbd, len);
1085                 return (void *)curr;
1086         }
1087
1088         /*
1089          * No free blocks are available.user_space hasn't caught up yet.
1090          * Queue was just frozen and now this packet will get dropped.
1091          */
1092         return NULL;
1093 }
1094
1095 static void *packet_current_rx_frame(struct packet_sock *po,
1096                                             struct sk_buff *skb,
1097                                             int status, unsigned int len)
1098 {
1099         char *curr = NULL;
1100         switch (po->tp_version) {
1101         case TPACKET_V1:
1102         case TPACKET_V2:
1103                 curr = packet_lookup_frame(po, &po->rx_ring,
1104                                         po->rx_ring.head, status);
1105                 return curr;
1106         case TPACKET_V3:
1107                 return __packet_lookup_frame_in_block(po, skb, status, len);
1108         default:
1109                 WARN(1, "TPACKET version not supported\n");
1110                 BUG();
1111                 return NULL;
1112         }
1113 }
1114
1115 static void *prb_lookup_block(struct packet_sock *po,
1116                                      struct packet_ring_buffer *rb,
1117                                      unsigned int idx,
1118                                      int status)
1119 {
1120         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1121         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1122
1123         if (status != BLOCK_STATUS(pbd))
1124                 return NULL;
1125         return pbd;
1126 }
1127
1128 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1129 {
1130         unsigned int prev;
1131         if (rb->prb_bdqc.kactive_blk_num)
1132                 prev = rb->prb_bdqc.kactive_blk_num-1;
1133         else
1134                 prev = rb->prb_bdqc.knum_blocks-1;
1135         return prev;
1136 }
1137
1138 /* Assumes caller has held the rx_queue.lock */
1139 static void *__prb_previous_block(struct packet_sock *po,
1140                                          struct packet_ring_buffer *rb,
1141                                          int status)
1142 {
1143         unsigned int previous = prb_previous_blk_num(rb);
1144         return prb_lookup_block(po, rb, previous, status);
1145 }
1146
1147 static void *packet_previous_rx_frame(struct packet_sock *po,
1148                                              struct packet_ring_buffer *rb,
1149                                              int status)
1150 {
1151         if (po->tp_version <= TPACKET_V2)
1152                 return packet_previous_frame(po, rb, status);
1153
1154         return __prb_previous_block(po, rb, status);
1155 }
1156
1157 static void packet_increment_rx_head(struct packet_sock *po,
1158                                             struct packet_ring_buffer *rb)
1159 {
1160         switch (po->tp_version) {
1161         case TPACKET_V1:
1162         case TPACKET_V2:
1163                 return packet_increment_head(rb);
1164         case TPACKET_V3:
1165         default:
1166                 WARN(1, "TPACKET version not supported.\n");
1167                 BUG();
1168                 return;
1169         }
1170 }
1171
1172 static void *packet_previous_frame(struct packet_sock *po,
1173                 struct packet_ring_buffer *rb,
1174                 int status)
1175 {
1176         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1177         return packet_lookup_frame(po, rb, previous, status);
1178 }
1179
1180 static void packet_increment_head(struct packet_ring_buffer *buff)
1181 {
1182         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1183 }
1184
1185 static void packet_inc_pending(struct packet_ring_buffer *rb)
1186 {
1187         this_cpu_inc(*rb->pending_refcnt);
1188 }
1189
1190 static void packet_dec_pending(struct packet_ring_buffer *rb)
1191 {
1192         this_cpu_dec(*rb->pending_refcnt);
1193 }
1194
1195 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1196 {
1197         unsigned int refcnt = 0;
1198         int cpu;
1199
1200         /* We don't use pending refcount in rx_ring. */
1201         if (rb->pending_refcnt == NULL)
1202                 return 0;
1203
1204         for_each_possible_cpu(cpu)
1205                 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1206
1207         return refcnt;
1208 }
1209
1210 static int packet_alloc_pending(struct packet_sock *po)
1211 {
1212         po->rx_ring.pending_refcnt = NULL;
1213
1214         po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1215         if (unlikely(po->tx_ring.pending_refcnt == NULL))
1216                 return -ENOBUFS;
1217
1218         return 0;
1219 }
1220
1221 static void packet_free_pending(struct packet_sock *po)
1222 {
1223         free_percpu(po->tx_ring.pending_refcnt);
1224 }
1225
1226 static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1227 {
1228         struct sock *sk = &po->sk;
1229         bool has_room;
1230
1231         if (po->prot_hook.func != tpacket_rcv)
1232                 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1233                         <= sk->sk_rcvbuf;
1234
1235         spin_lock(&sk->sk_receive_queue.lock);
1236         if (po->tp_version == TPACKET_V3)
1237                 has_room = prb_lookup_block(po, &po->rx_ring,
1238                                             po->rx_ring.prb_bdqc.kactive_blk_num,
1239                                             TP_STATUS_KERNEL);
1240         else
1241                 has_room = packet_lookup_frame(po, &po->rx_ring,
1242                                                po->rx_ring.head,
1243                                                TP_STATUS_KERNEL);
1244         spin_unlock(&sk->sk_receive_queue.lock);
1245
1246         return has_room;
1247 }
1248
1249 static void packet_sock_destruct(struct sock *sk)
1250 {
1251         skb_queue_purge(&sk->sk_error_queue);
1252
1253         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1254         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1255
1256         if (!sock_flag(sk, SOCK_DEAD)) {
1257                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1258                 return;
1259         }
1260
1261         sk_refcnt_debug_dec(sk);
1262 }
1263
1264 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1265 {
1266         int x = atomic_read(&f->rr_cur) + 1;
1267
1268         if (x >= num)
1269                 x = 0;
1270
1271         return x;
1272 }
1273
1274 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1275                                       struct sk_buff *skb,
1276                                       unsigned int num)
1277 {
1278         return reciprocal_scale(skb_get_hash(skb), num);
1279 }
1280
1281 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1282                                     struct sk_buff *skb,
1283                                     unsigned int num)
1284 {
1285         int cur, old;
1286
1287         cur = atomic_read(&f->rr_cur);
1288         while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1289                                      fanout_rr_next(f, num))) != cur)
1290                 cur = old;
1291         return cur;
1292 }
1293
1294 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1295                                      struct sk_buff *skb,
1296                                      unsigned int num)
1297 {
1298         return smp_processor_id() % num;
1299 }
1300
1301 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1302                                      struct sk_buff *skb,
1303                                      unsigned int num)
1304 {
1305         return prandom_u32_max(num);
1306 }
1307
1308 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1309                                           struct sk_buff *skb,
1310                                           unsigned int idx, unsigned int skip,
1311                                           unsigned int num)
1312 {
1313         unsigned int i, j;
1314
1315         i = j = min_t(int, f->next[idx], num - 1);
1316         do {
1317                 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1318                         if (i != j)
1319                                 f->next[idx] = i;
1320                         return i;
1321                 }
1322                 if (++i == num)
1323                         i = 0;
1324         } while (i != j);
1325
1326         return idx;
1327 }
1328
1329 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1330                                     struct sk_buff *skb,
1331                                     unsigned int num)
1332 {
1333         return skb_get_queue_mapping(skb) % num;
1334 }
1335
1336 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1337 {
1338         return f->flags & (flag >> 8);
1339 }
1340
1341 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1342                              struct packet_type *pt, struct net_device *orig_dev)
1343 {
1344         struct packet_fanout *f = pt->af_packet_priv;
1345         unsigned int num = f->num_members;
1346         struct packet_sock *po;
1347         unsigned int idx;
1348
1349         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1350             !num) {
1351                 kfree_skb(skb);
1352                 return 0;
1353         }
1354
1355         switch (f->type) {
1356         case PACKET_FANOUT_HASH:
1357         default:
1358                 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1359                         skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1360                         if (!skb)
1361                                 return 0;
1362                 }
1363                 idx = fanout_demux_hash(f, skb, num);
1364                 break;
1365         case PACKET_FANOUT_LB:
1366                 idx = fanout_demux_lb(f, skb, num);
1367                 break;
1368         case PACKET_FANOUT_CPU:
1369                 idx = fanout_demux_cpu(f, skb, num);
1370                 break;
1371         case PACKET_FANOUT_RND:
1372                 idx = fanout_demux_rnd(f, skb, num);
1373                 break;
1374         case PACKET_FANOUT_QM:
1375                 idx = fanout_demux_qm(f, skb, num);
1376                 break;
1377         case PACKET_FANOUT_ROLLOVER:
1378                 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1379                 break;
1380         }
1381
1382         po = pkt_sk(f->arr[idx]);
1383         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1384             unlikely(!packet_rcv_has_room(po, skb))) {
1385                 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1386                 po = pkt_sk(f->arr[idx]);
1387         }
1388
1389         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1390 }
1391
1392 DEFINE_MUTEX(fanout_mutex);
1393 EXPORT_SYMBOL_GPL(fanout_mutex);
1394 static LIST_HEAD(fanout_list);
1395
1396 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1397 {
1398         struct packet_fanout *f = po->fanout;
1399
1400         spin_lock(&f->lock);
1401         f->arr[f->num_members] = sk;
1402         smp_wmb();
1403         f->num_members++;
1404         spin_unlock(&f->lock);
1405 }
1406
1407 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1408 {
1409         struct packet_fanout *f = po->fanout;
1410         int i;
1411
1412         spin_lock(&f->lock);
1413         for (i = 0; i < f->num_members; i++) {
1414                 if (f->arr[i] == sk)
1415                         break;
1416         }
1417         BUG_ON(i >= f->num_members);
1418         f->arr[i] = f->arr[f->num_members - 1];
1419         f->num_members--;
1420         spin_unlock(&f->lock);
1421 }
1422
1423 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1424 {
1425         if (ptype->af_packet_priv == (void *)((struct packet_sock *)sk)->fanout)
1426                 return true;
1427
1428         return false;
1429 }
1430
1431 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1432 {
1433         struct packet_sock *po = pkt_sk(sk);
1434         struct packet_fanout *f, *match;
1435         u8 type = type_flags & 0xff;
1436         u8 flags = type_flags >> 8;
1437         int err;
1438
1439         switch (type) {
1440         case PACKET_FANOUT_ROLLOVER:
1441                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1442                         return -EINVAL;
1443         case PACKET_FANOUT_HASH:
1444         case PACKET_FANOUT_LB:
1445         case PACKET_FANOUT_CPU:
1446         case PACKET_FANOUT_RND:
1447         case PACKET_FANOUT_QM:
1448                 break;
1449         default:
1450                 return -EINVAL;
1451         }
1452
1453         if (!po->running)
1454                 return -EINVAL;
1455
1456         if (po->fanout)
1457                 return -EALREADY;
1458
1459         mutex_lock(&fanout_mutex);
1460         match = NULL;
1461         list_for_each_entry(f, &fanout_list, list) {
1462                 if (f->id == id &&
1463                     read_pnet(&f->net) == sock_net(sk)) {
1464                         match = f;
1465                         break;
1466                 }
1467         }
1468         err = -EINVAL;
1469         if (match && match->flags != flags)
1470                 goto out;
1471         if (!match) {
1472                 err = -ENOMEM;
1473                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1474                 if (!match)
1475                         goto out;
1476                 write_pnet(&match->net, sock_net(sk));
1477                 match->id = id;
1478                 match->type = type;
1479                 match->flags = flags;
1480                 atomic_set(&match->rr_cur, 0);
1481                 INIT_LIST_HEAD(&match->list);
1482                 spin_lock_init(&match->lock);
1483                 atomic_set(&match->sk_ref, 0);
1484                 match->prot_hook.type = po->prot_hook.type;
1485                 match->prot_hook.dev = po->prot_hook.dev;
1486                 match->prot_hook.func = packet_rcv_fanout;
1487                 match->prot_hook.af_packet_priv = match;
1488                 match->prot_hook.id_match = match_fanout_group;
1489                 dev_add_pack(&match->prot_hook);
1490                 list_add(&match->list, &fanout_list);
1491         }
1492         err = -EINVAL;
1493         if (match->type == type &&
1494             match->prot_hook.type == po->prot_hook.type &&
1495             match->prot_hook.dev == po->prot_hook.dev) {
1496                 err = -ENOSPC;
1497                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1498                         __dev_remove_pack(&po->prot_hook);
1499                         po->fanout = match;
1500                         atomic_inc(&match->sk_ref);
1501                         __fanout_link(sk, po);
1502                         err = 0;
1503                 }
1504         }
1505 out:
1506         mutex_unlock(&fanout_mutex);
1507         return err;
1508 }
1509
1510 static void fanout_release(struct sock *sk)
1511 {
1512         struct packet_sock *po = pkt_sk(sk);
1513         struct packet_fanout *f;
1514
1515         f = po->fanout;
1516         if (!f)
1517                 return;
1518
1519         mutex_lock(&fanout_mutex);
1520         po->fanout = NULL;
1521
1522         if (atomic_dec_and_test(&f->sk_ref)) {
1523                 list_del(&f->list);
1524                 dev_remove_pack(&f->prot_hook);
1525                 kfree(f);
1526         }
1527         mutex_unlock(&fanout_mutex);
1528 }
1529
1530 static const struct proto_ops packet_ops;
1531
1532 static const struct proto_ops packet_ops_spkt;
1533
1534 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1535                            struct packet_type *pt, struct net_device *orig_dev)
1536 {
1537         struct sock *sk;
1538         struct sockaddr_pkt *spkt;
1539
1540         /*
1541          *      When we registered the protocol we saved the socket in the data
1542          *      field for just this event.
1543          */
1544
1545         sk = pt->af_packet_priv;
1546
1547         /*
1548          *      Yank back the headers [hope the device set this
1549          *      right or kerboom...]
1550          *
1551          *      Incoming packets have ll header pulled,
1552          *      push it back.
1553          *
1554          *      For outgoing ones skb->data == skb_mac_header(skb)
1555          *      so that this procedure is noop.
1556          */
1557
1558         if (skb->pkt_type == PACKET_LOOPBACK)
1559                 goto out;
1560
1561         if (!net_eq(dev_net(dev), sock_net(sk)))
1562                 goto out;
1563
1564         skb = skb_share_check(skb, GFP_ATOMIC);
1565         if (skb == NULL)
1566                 goto oom;
1567
1568         /* drop any routing info */
1569         skb_dst_drop(skb);
1570
1571         /* drop conntrack reference */
1572         nf_reset(skb);
1573
1574         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1575
1576         skb_push(skb, skb->data - skb_mac_header(skb));
1577
1578         /*
1579          *      The SOCK_PACKET socket receives _all_ frames.
1580          */
1581
1582         spkt->spkt_family = dev->type;
1583         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1584         spkt->spkt_protocol = skb->protocol;
1585
1586         /*
1587          *      Charge the memory to the socket. This is done specifically
1588          *      to prevent sockets using all the memory up.
1589          */
1590
1591         if (sock_queue_rcv_skb(sk, skb) == 0)
1592                 return 0;
1593
1594 out:
1595         kfree_skb(skb);
1596 oom:
1597         return 0;
1598 }
1599
1600
1601 /*
1602  *      Output a raw packet to a device layer. This bypasses all the other
1603  *      protocol layers and you must therefore supply it with a complete frame
1604  */
1605
1606 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1607                                struct msghdr *msg, size_t len)
1608 {
1609         struct sock *sk = sock->sk;
1610         DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1611         struct sk_buff *skb = NULL;
1612         struct net_device *dev;
1613         __be16 proto = 0;
1614         int err;
1615         int extra_len = 0;
1616
1617         /*
1618          *      Get and verify the address.
1619          */
1620
1621         if (saddr) {
1622                 if (msg->msg_namelen < sizeof(struct sockaddr))
1623                         return -EINVAL;
1624                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1625                         proto = saddr->spkt_protocol;
1626         } else
1627                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1628
1629         /*
1630          *      Find the device first to size check it
1631          */
1632
1633         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1634 retry:
1635         rcu_read_lock();
1636         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1637         err = -ENODEV;
1638         if (dev == NULL)
1639                 goto out_unlock;
1640
1641         err = -ENETDOWN;
1642         if (!(dev->flags & IFF_UP))
1643                 goto out_unlock;
1644
1645         /*
1646          * You may not queue a frame bigger than the mtu. This is the lowest level
1647          * raw protocol and you must do your own fragmentation at this level.
1648          */
1649
1650         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1651                 if (!netif_supports_nofcs(dev)) {
1652                         err = -EPROTONOSUPPORT;
1653                         goto out_unlock;
1654                 }
1655                 extra_len = 4; /* We're doing our own CRC */
1656         }
1657
1658         err = -EMSGSIZE;
1659         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1660                 goto out_unlock;
1661
1662         if (!skb) {
1663                 size_t reserved = LL_RESERVED_SPACE(dev);
1664                 int tlen = dev->needed_tailroom;
1665                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1666
1667                 rcu_read_unlock();
1668                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1669                 if (skb == NULL)
1670                         return -ENOBUFS;
1671                 /* FIXME: Save some space for broken drivers that write a hard
1672                  * header at transmission time by themselves. PPP is the notable
1673                  * one here. This should really be fixed at the driver level.
1674                  */
1675                 skb_reserve(skb, reserved);
1676                 skb_reset_network_header(skb);
1677
1678                 /* Try to align data part correctly */
1679                 if (hhlen) {
1680                         skb->data -= hhlen;
1681                         skb->tail -= hhlen;
1682                         if (len < hhlen)
1683                                 skb_reset_network_header(skb);
1684                 }
1685                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1686                 if (err)
1687                         goto out_free;
1688                 goto retry;
1689         }
1690
1691         if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1692                 /* Earlier code assumed this would be a VLAN pkt,
1693                  * double-check this now that we have the actual
1694                  * packet in hand.
1695                  */
1696                 struct ethhdr *ehdr;
1697                 skb_reset_mac_header(skb);
1698                 ehdr = eth_hdr(skb);
1699                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1700                         err = -EMSGSIZE;
1701                         goto out_unlock;
1702                 }
1703         }
1704
1705         skb->protocol = proto;
1706         skb->dev = dev;
1707         skb->priority = sk->sk_priority;
1708         skb->mark = sk->sk_mark;
1709
1710         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1711
1712         if (unlikely(extra_len == 4))
1713                 skb->no_fcs = 1;
1714
1715         skb_probe_transport_header(skb, 0);
1716
1717         dev_queue_xmit(skb);
1718         rcu_read_unlock();
1719         return len;
1720
1721 out_unlock:
1722         rcu_read_unlock();
1723 out_free:
1724         kfree_skb(skb);
1725         return err;
1726 }
1727
1728 static unsigned int run_filter(const struct sk_buff *skb,
1729                                       const struct sock *sk,
1730                                       unsigned int res)
1731 {
1732         struct sk_filter *filter;
1733
1734         rcu_read_lock();
1735         filter = rcu_dereference(sk->sk_filter);
1736         if (filter != NULL)
1737                 res = SK_RUN_FILTER(filter, skb);
1738         rcu_read_unlock();
1739
1740         return res;
1741 }
1742
1743 /*
1744  * This function makes lazy skb cloning in hope that most of packets
1745  * are discarded by BPF.
1746  *
1747  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1748  * and skb->cb are mangled. It works because (and until) packets
1749  * falling here are owned by current CPU. Output packets are cloned
1750  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1751  * sequencially, so that if we return skb to original state on exit,
1752  * we will not harm anyone.
1753  */
1754
1755 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1756                       struct packet_type *pt, struct net_device *orig_dev)
1757 {
1758         struct sock *sk;
1759         struct sockaddr_ll *sll;
1760         struct packet_sock *po;
1761         u8 *skb_head = skb->data;
1762         int skb_len = skb->len;
1763         unsigned int snaplen, res;
1764
1765         if (skb->pkt_type == PACKET_LOOPBACK)
1766                 goto drop;
1767
1768         sk = pt->af_packet_priv;
1769         po = pkt_sk(sk);
1770
1771         if (!net_eq(dev_net(dev), sock_net(sk)))
1772                 goto drop;
1773
1774         skb->dev = dev;
1775
1776         if (dev->header_ops) {
1777                 /* The device has an explicit notion of ll header,
1778                  * exported to higher levels.
1779                  *
1780                  * Otherwise, the device hides details of its frame
1781                  * structure, so that corresponding packet head is
1782                  * never delivered to user.
1783                  */
1784                 if (sk->sk_type != SOCK_DGRAM)
1785                         skb_push(skb, skb->data - skb_mac_header(skb));
1786                 else if (skb->pkt_type == PACKET_OUTGOING) {
1787                         /* Special case: outgoing packets have ll header at head */
1788                         skb_pull(skb, skb_network_offset(skb));
1789                 }
1790         }
1791
1792         snaplen = skb->len;
1793
1794         res = run_filter(skb, sk, snaplen);
1795         if (!res)
1796                 goto drop_n_restore;
1797         if (snaplen > res)
1798                 snaplen = res;
1799
1800         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1801                 goto drop_n_acct;
1802
1803         if (skb_shared(skb)) {
1804                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1805                 if (nskb == NULL)
1806                         goto drop_n_acct;
1807
1808                 if (skb_head != skb->data) {
1809                         skb->data = skb_head;
1810                         skb->len = skb_len;
1811                 }
1812                 consume_skb(skb);
1813                 skb = nskb;
1814         }
1815
1816         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1817                      sizeof(skb->cb));
1818
1819         sll = &PACKET_SKB_CB(skb)->sa.ll;
1820         sll->sll_family = AF_PACKET;
1821         sll->sll_hatype = dev->type;
1822         sll->sll_protocol = skb->protocol;
1823         sll->sll_pkttype = skb->pkt_type;
1824         if (unlikely(po->origdev))
1825                 sll->sll_ifindex = orig_dev->ifindex;
1826         else
1827                 sll->sll_ifindex = dev->ifindex;
1828
1829         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1830
1831         PACKET_SKB_CB(skb)->origlen = skb->len;
1832
1833         if (pskb_trim(skb, snaplen))
1834                 goto drop_n_acct;
1835
1836         skb_set_owner_r(skb, sk);
1837         skb->dev = NULL;
1838         skb_dst_drop(skb);
1839
1840         /* drop conntrack reference */
1841         nf_reset(skb);
1842
1843         spin_lock(&sk->sk_receive_queue.lock);
1844         po->stats.stats1.tp_packets++;
1845         skb->dropcount = atomic_read(&sk->sk_drops);
1846         __skb_queue_tail(&sk->sk_receive_queue, skb);
1847         spin_unlock(&sk->sk_receive_queue.lock);
1848         sk->sk_data_ready(sk);
1849         return 0;
1850
1851 drop_n_acct:
1852         spin_lock(&sk->sk_receive_queue.lock);
1853         po->stats.stats1.tp_drops++;
1854         atomic_inc(&sk->sk_drops);
1855         spin_unlock(&sk->sk_receive_queue.lock);
1856
1857 drop_n_restore:
1858         if (skb_head != skb->data && skb_shared(skb)) {
1859                 skb->data = skb_head;
1860                 skb->len = skb_len;
1861         }
1862 drop:
1863         consume_skb(skb);
1864         return 0;
1865 }
1866
1867 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1868                        struct packet_type *pt, struct net_device *orig_dev)
1869 {
1870         struct sock *sk;
1871         struct packet_sock *po;
1872         struct sockaddr_ll *sll;
1873         union tpacket_uhdr h;
1874         u8 *skb_head = skb->data;
1875         int skb_len = skb->len;
1876         unsigned int snaplen, res;
1877         unsigned long status = TP_STATUS_USER;
1878         unsigned short macoff, netoff, hdrlen;
1879         struct sk_buff *copy_skb = NULL;
1880         struct timespec ts;
1881         __u32 ts_status;
1882
1883         /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
1884          * We may add members to them until current aligned size without forcing
1885          * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
1886          */
1887         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
1888         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
1889
1890         if (skb->pkt_type == PACKET_LOOPBACK)
1891                 goto drop;
1892
1893         sk = pt->af_packet_priv;
1894         po = pkt_sk(sk);
1895
1896         if (!net_eq(dev_net(dev), sock_net(sk)))
1897                 goto drop;
1898
1899         if (dev->header_ops) {
1900                 if (sk->sk_type != SOCK_DGRAM)
1901                         skb_push(skb, skb->data - skb_mac_header(skb));
1902                 else if (skb->pkt_type == PACKET_OUTGOING) {
1903                         /* Special case: outgoing packets have ll header at head */
1904                         skb_pull(skb, skb_network_offset(skb));
1905                 }
1906         }
1907
1908         if (skb->ip_summed == CHECKSUM_PARTIAL)
1909                 status |= TP_STATUS_CSUMNOTREADY;
1910
1911         snaplen = skb->len;
1912
1913         res = run_filter(skb, sk, snaplen);
1914         if (!res)
1915                 goto drop_n_restore;
1916         if (snaplen > res)
1917                 snaplen = res;
1918
1919         if (sk->sk_type == SOCK_DGRAM) {
1920                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1921                                   po->tp_reserve;
1922         } else {
1923                 unsigned int maclen = skb_network_offset(skb);
1924                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1925                                        (maclen < 16 ? 16 : maclen)) +
1926                         po->tp_reserve;
1927                 macoff = netoff - maclen;
1928         }
1929         if (po->tp_version <= TPACKET_V2) {
1930                 if (macoff + snaplen > po->rx_ring.frame_size) {
1931                         if (po->copy_thresh &&
1932                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1933                                 if (skb_shared(skb)) {
1934                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1935                                 } else {
1936                                         copy_skb = skb_get(skb);
1937                                         skb_head = skb->data;
1938                                 }
1939                                 if (copy_skb)
1940                                         skb_set_owner_r(copy_skb, sk);
1941                         }
1942                         snaplen = po->rx_ring.frame_size - macoff;
1943                         if ((int)snaplen < 0)
1944                                 snaplen = 0;
1945                 }
1946         } else if (unlikely(macoff + snaplen >
1947                             GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
1948                 u32 nval;
1949
1950                 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
1951                 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
1952                             snaplen, nval, macoff);
1953                 snaplen = nval;
1954                 if (unlikely((int)snaplen < 0)) {
1955                         snaplen = 0;
1956                         macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
1957                 }
1958         }
1959         spin_lock(&sk->sk_receive_queue.lock);
1960         h.raw = packet_current_rx_frame(po, skb,
1961                                         TP_STATUS_KERNEL, (macoff+snaplen));
1962         if (!h.raw)
1963                 goto ring_is_full;
1964         if (po->tp_version <= TPACKET_V2) {
1965                 packet_increment_rx_head(po, &po->rx_ring);
1966         /*
1967          * LOSING will be reported till you read the stats,
1968          * because it's COR - Clear On Read.
1969          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1970          * at packet level.
1971          */
1972                 if (po->stats.stats1.tp_drops)
1973                         status |= TP_STATUS_LOSING;
1974         }
1975         po->stats.stats1.tp_packets++;
1976         if (copy_skb) {
1977                 status |= TP_STATUS_COPY;
1978                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1979         }
1980         spin_unlock(&sk->sk_receive_queue.lock);
1981
1982         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1983
1984         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
1985                 getnstimeofday(&ts);
1986
1987         status |= ts_status;
1988
1989         switch (po->tp_version) {
1990         case TPACKET_V1:
1991                 h.h1->tp_len = skb->len;
1992                 h.h1->tp_snaplen = snaplen;
1993                 h.h1->tp_mac = macoff;
1994                 h.h1->tp_net = netoff;
1995                 h.h1->tp_sec = ts.tv_sec;
1996                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
1997                 hdrlen = sizeof(*h.h1);
1998                 break;
1999         case TPACKET_V2:
2000                 h.h2->tp_len = skb->len;
2001                 h.h2->tp_snaplen = snaplen;
2002                 h.h2->tp_mac = macoff;
2003                 h.h2->tp_net = netoff;
2004                 h.h2->tp_sec = ts.tv_sec;
2005                 h.h2->tp_nsec = ts.tv_nsec;
2006                 if (vlan_tx_tag_present(skb)) {
2007                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
2008                         h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2009                         status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2010                 } else {
2011                         h.h2->tp_vlan_tci = 0;
2012                         h.h2->tp_vlan_tpid = 0;
2013                 }
2014                 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2015                 hdrlen = sizeof(*h.h2);
2016                 break;
2017         case TPACKET_V3:
2018                 /* tp_nxt_offset,vlan are already populated above.
2019                  * So DONT clear those fields here
2020                  */
2021                 h.h3->tp_status |= status;
2022                 h.h3->tp_len = skb->len;
2023                 h.h3->tp_snaplen = snaplen;
2024                 h.h3->tp_mac = macoff;
2025                 h.h3->tp_net = netoff;
2026                 h.h3->tp_sec  = ts.tv_sec;
2027                 h.h3->tp_nsec = ts.tv_nsec;
2028                 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2029                 hdrlen = sizeof(*h.h3);
2030                 break;
2031         default:
2032                 BUG();
2033         }
2034
2035         sll = h.raw + TPACKET_ALIGN(hdrlen);
2036         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2037         sll->sll_family = AF_PACKET;
2038         sll->sll_hatype = dev->type;
2039         sll->sll_protocol = skb->protocol;
2040         sll->sll_pkttype = skb->pkt_type;
2041         if (unlikely(po->origdev))
2042                 sll->sll_ifindex = orig_dev->ifindex;
2043         else
2044                 sll->sll_ifindex = dev->ifindex;
2045
2046         smp_mb();
2047
2048 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2049         if (po->tp_version <= TPACKET_V2) {
2050                 u8 *start, *end;
2051
2052                 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2053                                         macoff + snaplen);
2054
2055                 for (start = h.raw; start < end; start += PAGE_SIZE)
2056                         flush_dcache_page(pgv_to_page(start));
2057         }
2058         smp_wmb();
2059 #endif
2060
2061         if (po->tp_version <= TPACKET_V2)
2062                 __packet_set_status(po, h.raw, status);
2063         else
2064                 prb_clear_blk_fill_status(&po->rx_ring);
2065
2066         sk->sk_data_ready(sk);
2067
2068 drop_n_restore:
2069         if (skb_head != skb->data && skb_shared(skb)) {
2070                 skb->data = skb_head;
2071                 skb->len = skb_len;
2072         }
2073 drop:
2074         kfree_skb(skb);
2075         return 0;
2076
2077 ring_is_full:
2078         po->stats.stats1.tp_drops++;
2079         spin_unlock(&sk->sk_receive_queue.lock);
2080
2081         sk->sk_data_ready(sk);
2082         kfree_skb(copy_skb);
2083         goto drop_n_restore;
2084 }
2085
2086 static void tpacket_destruct_skb(struct sk_buff *skb)
2087 {
2088         struct packet_sock *po = pkt_sk(skb->sk);
2089
2090         if (likely(po->tx_ring.pg_vec)) {
2091                 void *ph;
2092                 __u32 ts;
2093
2094                 ph = skb_shinfo(skb)->destructor_arg;
2095                 packet_dec_pending(&po->tx_ring);
2096
2097                 ts = __packet_set_timestamp(po, ph, skb);
2098                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2099         }
2100
2101         sock_wfree(skb);
2102 }
2103
2104 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2105                 void *frame, struct net_device *dev, int size_max,
2106                 __be16 proto, unsigned char *addr, int hlen)
2107 {
2108         union tpacket_uhdr ph;
2109         int to_write, offset, len, tp_len, nr_frags, len_max;
2110         struct socket *sock = po->sk.sk_socket;
2111         struct page *page;
2112         void *data;
2113         int err;
2114
2115         ph.raw = frame;
2116
2117         skb->protocol = proto;
2118         skb->dev = dev;
2119         skb->priority = po->sk.sk_priority;
2120         skb->mark = po->sk.sk_mark;
2121         sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2122         skb_shinfo(skb)->destructor_arg = ph.raw;
2123
2124         switch (po->tp_version) {
2125         case TPACKET_V2:
2126                 tp_len = ph.h2->tp_len;
2127                 break;
2128         default:
2129                 tp_len = ph.h1->tp_len;
2130                 break;
2131         }
2132         if (unlikely(tp_len > size_max)) {
2133                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2134                 return -EMSGSIZE;
2135         }
2136
2137         skb_reserve(skb, hlen);
2138         skb_reset_network_header(skb);
2139
2140         if (!packet_use_direct_xmit(po))
2141                 skb_probe_transport_header(skb, 0);
2142         if (unlikely(po->tp_tx_has_off)) {
2143                 int off_min, off_max, off;
2144                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2145                 off_max = po->tx_ring.frame_size - tp_len;
2146                 if (sock->type == SOCK_DGRAM) {
2147                         switch (po->tp_version) {
2148                         case TPACKET_V2:
2149                                 off = ph.h2->tp_net;
2150                                 break;
2151                         default:
2152                                 off = ph.h1->tp_net;
2153                                 break;
2154                         }
2155                 } else {
2156                         switch (po->tp_version) {
2157                         case TPACKET_V2:
2158                                 off = ph.h2->tp_mac;
2159                                 break;
2160                         default:
2161                                 off = ph.h1->tp_mac;
2162                                 break;
2163                         }
2164                 }
2165                 if (unlikely((off < off_min) || (off_max < off)))
2166                         return -EINVAL;
2167                 data = ph.raw + off;
2168         } else {
2169                 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2170         }
2171         to_write = tp_len;
2172
2173         if (sock->type == SOCK_DGRAM) {
2174                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2175                                 NULL, tp_len);
2176                 if (unlikely(err < 0))
2177                         return -EINVAL;
2178         } else if (dev->hard_header_len) {
2179                 /* net device doesn't like empty head */
2180                 if (unlikely(tp_len <= dev->hard_header_len)) {
2181                         pr_err("packet size is too short (%d < %d)\n",
2182                                tp_len, dev->hard_header_len);
2183                         return -EINVAL;
2184                 }
2185
2186                 skb_push(skb, dev->hard_header_len);
2187                 err = skb_store_bits(skb, 0, data,
2188                                 dev->hard_header_len);
2189                 if (unlikely(err))
2190                         return err;
2191
2192                 data += dev->hard_header_len;
2193                 to_write -= dev->hard_header_len;
2194         }
2195
2196         offset = offset_in_page(data);
2197         len_max = PAGE_SIZE - offset;
2198         len = ((to_write > len_max) ? len_max : to_write);
2199
2200         skb->data_len = to_write;
2201         skb->len += to_write;
2202         skb->truesize += to_write;
2203         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2204
2205         while (likely(to_write)) {
2206                 nr_frags = skb_shinfo(skb)->nr_frags;
2207
2208                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2209                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2210                                MAX_SKB_FRAGS);
2211                         return -EFAULT;
2212                 }
2213
2214                 page = pgv_to_page(data);
2215                 data += len;
2216                 flush_dcache_page(page);
2217                 get_page(page);
2218                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2219                 to_write -= len;
2220                 offset = 0;
2221                 len_max = PAGE_SIZE;
2222                 len = ((to_write > len_max) ? len_max : to_write);
2223         }
2224
2225         return tp_len;
2226 }
2227
2228 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2229 {
2230         struct sk_buff *skb;
2231         struct net_device *dev;
2232         __be16 proto;
2233         int err, reserve = 0;
2234         void *ph;
2235         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2236         bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2237         int tp_len, size_max;
2238         unsigned char *addr;
2239         int len_sum = 0;
2240         int status = TP_STATUS_AVAILABLE;
2241         int hlen, tlen;
2242
2243         mutex_lock(&po->pg_vec_lock);
2244
2245         if (likely(saddr == NULL)) {
2246                 dev     = packet_cached_dev_get(po);
2247                 proto   = po->num;
2248                 addr    = NULL;
2249         } else {
2250                 err = -EINVAL;
2251                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2252                         goto out;
2253                 if (msg->msg_namelen < (saddr->sll_halen
2254                                         + offsetof(struct sockaddr_ll,
2255                                                 sll_addr)))
2256                         goto out;
2257                 proto   = saddr->sll_protocol;
2258                 addr    = saddr->sll_addr;
2259                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2260         }
2261
2262         err = -ENXIO;
2263         if (unlikely(dev == NULL))
2264                 goto out;
2265         err = -ENETDOWN;
2266         if (unlikely(!(dev->flags & IFF_UP)))
2267                 goto out_put;
2268
2269         reserve = dev->hard_header_len + VLAN_HLEN;
2270         size_max = po->tx_ring.frame_size
2271                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2272
2273         if (size_max > dev->mtu + reserve)
2274                 size_max = dev->mtu + reserve;
2275
2276         do {
2277                 ph = packet_current_frame(po, &po->tx_ring,
2278                                           TP_STATUS_SEND_REQUEST);
2279                 if (unlikely(ph == NULL)) {
2280                         if (need_wait && need_resched())
2281                                 schedule();
2282                         continue;
2283                 }
2284
2285                 status = TP_STATUS_SEND_REQUEST;
2286                 hlen = LL_RESERVED_SPACE(dev);
2287                 tlen = dev->needed_tailroom;
2288                 skb = sock_alloc_send_skb(&po->sk,
2289                                 hlen + tlen + sizeof(struct sockaddr_ll),
2290                                 0, &err);
2291
2292                 if (unlikely(skb == NULL))
2293                         goto out_status;
2294
2295                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2296                                           addr, hlen);
2297                 if (tp_len > dev->mtu + dev->hard_header_len) {
2298                         struct ethhdr *ehdr;
2299                         /* Earlier code assumed this would be a VLAN pkt,
2300                          * double-check this now that we have the actual
2301                          * packet in hand.
2302                          */
2303
2304                         skb_reset_mac_header(skb);
2305                         ehdr = eth_hdr(skb);
2306                         if (ehdr->h_proto != htons(ETH_P_8021Q))
2307                                 tp_len = -EMSGSIZE;
2308                 }
2309                 if (unlikely(tp_len < 0)) {
2310                         if (po->tp_loss) {
2311                                 __packet_set_status(po, ph,
2312                                                 TP_STATUS_AVAILABLE);
2313                                 packet_increment_head(&po->tx_ring);
2314                                 kfree_skb(skb);
2315                                 continue;
2316                         } else {
2317                                 status = TP_STATUS_WRONG_FORMAT;
2318                                 err = tp_len;
2319                                 goto out_status;
2320                         }
2321                 }
2322
2323                 packet_pick_tx_queue(dev, skb);
2324
2325                 skb->destructor = tpacket_destruct_skb;
2326                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2327                 packet_inc_pending(&po->tx_ring);
2328
2329                 status = TP_STATUS_SEND_REQUEST;
2330                 err = po->xmit(skb);
2331                 if (unlikely(err > 0)) {
2332                         err = net_xmit_errno(err);
2333                         if (err && __packet_get_status(po, ph) ==
2334                                    TP_STATUS_AVAILABLE) {
2335                                 /* skb was destructed already */
2336                                 skb = NULL;
2337                                 goto out_status;
2338                         }
2339                         /*
2340                          * skb was dropped but not destructed yet;
2341                          * let's treat it like congestion or err < 0
2342                          */
2343                         err = 0;
2344                 }
2345                 packet_increment_head(&po->tx_ring);
2346                 len_sum += tp_len;
2347         } while (likely((ph != NULL) ||
2348                 /* Note: packet_read_pending() might be slow if we have
2349                  * to call it as it's per_cpu variable, but in fast-path
2350                  * we already short-circuit the loop with the first
2351                  * condition, and luckily don't have to go that path
2352                  * anyway.
2353                  */
2354                  (need_wait && packet_read_pending(&po->tx_ring))));
2355
2356         err = len_sum;
2357         goto out_put;
2358
2359 out_status:
2360         __packet_set_status(po, ph, status);
2361         kfree_skb(skb);
2362 out_put:
2363         dev_put(dev);
2364 out:
2365         mutex_unlock(&po->pg_vec_lock);
2366         return err;
2367 }
2368
2369 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2370                                         size_t reserve, size_t len,
2371                                         size_t linear, int noblock,
2372                                         int *err)
2373 {
2374         struct sk_buff *skb;
2375
2376         /* Under a page?  Don't bother with paged skb. */
2377         if (prepad + len < PAGE_SIZE || !linear)
2378                 linear = len;
2379
2380         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2381                                    err, 0);
2382         if (!skb)
2383                 return NULL;
2384
2385         skb_reserve(skb, reserve);
2386         skb_put(skb, linear);
2387         skb->data_len = len - linear;
2388         skb->len += len - linear;
2389
2390         return skb;
2391 }
2392
2393 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2394 {
2395         struct sock *sk = sock->sk;
2396         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2397         struct sk_buff *skb;
2398         struct net_device *dev;
2399         __be16 proto;
2400         unsigned char *addr;
2401         int err, reserve = 0;
2402         struct virtio_net_hdr vnet_hdr = { 0 };
2403         int offset = 0;
2404         int vnet_hdr_len;
2405         struct packet_sock *po = pkt_sk(sk);
2406         unsigned short gso_type = 0;
2407         int hlen, tlen;
2408         int extra_len = 0;
2409
2410         /*
2411          *      Get and verify the address.
2412          */
2413
2414         if (likely(saddr == NULL)) {
2415                 dev     = packet_cached_dev_get(po);
2416                 proto   = po->num;
2417                 addr    = NULL;
2418         } else {
2419                 err = -EINVAL;
2420                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2421                         goto out;
2422                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2423                         goto out;
2424                 proto   = saddr->sll_protocol;
2425                 addr    = saddr->sll_addr;
2426                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2427         }
2428
2429         err = -ENXIO;
2430         if (unlikely(dev == NULL))
2431                 goto out_unlock;
2432         err = -ENETDOWN;
2433         if (unlikely(!(dev->flags & IFF_UP)))
2434                 goto out_unlock;
2435
2436         if (sock->type == SOCK_RAW)
2437                 reserve = dev->hard_header_len;
2438         if (po->has_vnet_hdr) {
2439                 vnet_hdr_len = sizeof(vnet_hdr);
2440
2441                 err = -EINVAL;
2442                 if (len < vnet_hdr_len)
2443                         goto out_unlock;
2444
2445                 len -= vnet_hdr_len;
2446
2447                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2448                                        vnet_hdr_len);
2449                 if (err < 0)
2450                         goto out_unlock;
2451
2452                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2453                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2454                       vnet_hdr.hdr_len))
2455                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
2456                                                  vnet_hdr.csum_offset + 2;
2457
2458                 err = -EINVAL;
2459                 if (vnet_hdr.hdr_len > len)
2460                         goto out_unlock;
2461
2462                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2463                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2464                         case VIRTIO_NET_HDR_GSO_TCPV4:
2465                                 gso_type = SKB_GSO_TCPV4;
2466                                 break;
2467                         case VIRTIO_NET_HDR_GSO_TCPV6:
2468                                 gso_type = SKB_GSO_TCPV6;
2469                                 break;
2470                         case VIRTIO_NET_HDR_GSO_UDP:
2471                                 gso_type = SKB_GSO_UDP;
2472                                 break;
2473                         default:
2474                                 goto out_unlock;
2475                         }
2476
2477                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2478                                 gso_type |= SKB_GSO_TCP_ECN;
2479
2480                         if (vnet_hdr.gso_size == 0)
2481                                 goto out_unlock;
2482
2483                 }
2484         }
2485
2486         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2487                 if (!netif_supports_nofcs(dev)) {
2488                         err = -EPROTONOSUPPORT;
2489                         goto out_unlock;
2490                 }
2491                 extra_len = 4; /* We're doing our own CRC */
2492         }
2493
2494         err = -EMSGSIZE;
2495         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2496                 goto out_unlock;
2497
2498         err = -ENOBUFS;
2499         hlen = LL_RESERVED_SPACE(dev);
2500         tlen = dev->needed_tailroom;
2501         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2502                                msg->msg_flags & MSG_DONTWAIT, &err);
2503         if (skb == NULL)
2504                 goto out_unlock;
2505
2506         skb_set_network_header(skb, reserve);
2507
2508         err = -EINVAL;
2509         if (sock->type == SOCK_DGRAM &&
2510             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2511                 goto out_free;
2512
2513         /* Returns -EFAULT on error */
2514         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2515         if (err)
2516                 goto out_free;
2517
2518         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2519
2520         if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2521                 /* Earlier code assumed this would be a VLAN pkt,
2522                  * double-check this now that we have the actual
2523                  * packet in hand.
2524                  */
2525                 struct ethhdr *ehdr;
2526                 skb_reset_mac_header(skb);
2527                 ehdr = eth_hdr(skb);
2528                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2529                         err = -EMSGSIZE;
2530                         goto out_free;
2531                 }
2532         }
2533
2534         skb->protocol = proto;
2535         skb->dev = dev;
2536         skb->priority = sk->sk_priority;
2537         skb->mark = sk->sk_mark;
2538
2539         packet_pick_tx_queue(dev, skb);
2540
2541         if (po->has_vnet_hdr) {
2542                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2543                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2544                                                   vnet_hdr.csum_offset)) {
2545                                 err = -EINVAL;
2546                                 goto out_free;
2547                         }
2548                 }
2549
2550                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2551                 skb_shinfo(skb)->gso_type = gso_type;
2552
2553                 /* Header must be checked, and gso_segs computed. */
2554                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2555                 skb_shinfo(skb)->gso_segs = 0;
2556
2557                 len += vnet_hdr_len;
2558         }
2559
2560         if (!packet_use_direct_xmit(po))
2561                 skb_probe_transport_header(skb, reserve);
2562         if (unlikely(extra_len == 4))
2563                 skb->no_fcs = 1;
2564
2565         err = po->xmit(skb);
2566         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2567                 goto out_unlock;
2568
2569         dev_put(dev);
2570
2571         return len;
2572
2573 out_free:
2574         kfree_skb(skb);
2575 out_unlock:
2576         if (dev)
2577                 dev_put(dev);
2578 out:
2579         return err;
2580 }
2581
2582 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2583                 struct msghdr *msg, size_t len)
2584 {
2585         struct sock *sk = sock->sk;
2586         struct packet_sock *po = pkt_sk(sk);
2587
2588         if (po->tx_ring.pg_vec)
2589                 return tpacket_snd(po, msg);
2590         else
2591                 return packet_snd(sock, msg, len);
2592 }
2593
2594 /*
2595  *      Close a PACKET socket. This is fairly simple. We immediately go
2596  *      to 'closed' state and remove our protocol entry in the device list.
2597  */
2598
2599 static int packet_release(struct socket *sock)
2600 {
2601         struct sock *sk = sock->sk;
2602         struct packet_sock *po;
2603         struct net *net;
2604         union tpacket_req_u req_u;
2605
2606         if (!sk)
2607                 return 0;
2608
2609         net = sock_net(sk);
2610         po = pkt_sk(sk);
2611
2612         mutex_lock(&net->packet.sklist_lock);
2613         sk_del_node_init_rcu(sk);
2614         mutex_unlock(&net->packet.sklist_lock);
2615
2616         preempt_disable();
2617         sock_prot_inuse_add(net, sk->sk_prot, -1);
2618         preempt_enable();
2619
2620         spin_lock(&po->bind_lock);
2621         unregister_prot_hook(sk, false);
2622         packet_cached_dev_reset(po);
2623
2624         if (po->prot_hook.dev) {
2625                 dev_put(po->prot_hook.dev);
2626                 po->prot_hook.dev = NULL;
2627         }
2628         spin_unlock(&po->bind_lock);
2629
2630         packet_flush_mclist(sk);
2631
2632         if (po->rx_ring.pg_vec) {
2633                 memset(&req_u, 0, sizeof(req_u));
2634                 packet_set_ring(sk, &req_u, 1, 0);
2635         }
2636
2637         if (po->tx_ring.pg_vec) {
2638                 memset(&req_u, 0, sizeof(req_u));
2639                 packet_set_ring(sk, &req_u, 1, 1);
2640         }
2641
2642         fanout_release(sk);
2643
2644         synchronize_net();
2645         /*
2646          *      Now the socket is dead. No more input will appear.
2647          */
2648         sock_orphan(sk);
2649         sock->sk = NULL;
2650
2651         /* Purge queues */
2652
2653         skb_queue_purge(&sk->sk_receive_queue);
2654         packet_free_pending(po);
2655         sk_refcnt_debug_release(sk);
2656
2657         sock_put(sk);
2658         return 0;
2659 }
2660
2661 /*
2662  *      Attach a packet hook.
2663  */
2664
2665 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 proto)
2666 {
2667         struct packet_sock *po = pkt_sk(sk);
2668         const struct net_device *dev_curr;
2669         __be16 proto_curr;
2670         bool need_rehook;
2671
2672         if (po->fanout) {
2673                 if (dev)
2674                         dev_put(dev);
2675
2676                 return -EINVAL;
2677         }
2678
2679         lock_sock(sk);
2680         spin_lock(&po->bind_lock);
2681
2682         proto_curr = po->prot_hook.type;
2683         dev_curr = po->prot_hook.dev;
2684
2685         need_rehook = proto_curr != proto || dev_curr != dev;
2686
2687         if (need_rehook) {
2688                 unregister_prot_hook(sk, true);
2689
2690                 po->num = proto;
2691                 po->prot_hook.type = proto;
2692
2693                 if (po->prot_hook.dev)
2694                         dev_put(po->prot_hook.dev);
2695
2696                 po->prot_hook.dev = dev;
2697
2698                 po->ifindex = dev ? dev->ifindex : 0;
2699                 packet_cached_dev_assign(po, dev);
2700         }
2701
2702         if (proto == 0 || !need_rehook)
2703                 goto out_unlock;
2704
2705         if (!dev || (dev->flags & IFF_UP)) {
2706                 register_prot_hook(sk);
2707         } else {
2708                 sk->sk_err = ENETDOWN;
2709                 if (!sock_flag(sk, SOCK_DEAD))
2710                         sk->sk_error_report(sk);
2711         }
2712
2713 out_unlock:
2714         spin_unlock(&po->bind_lock);
2715         release_sock(sk);
2716         return 0;
2717 }
2718
2719 /*
2720  *      Bind a packet socket to a device
2721  */
2722
2723 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2724                             int addr_len)
2725 {
2726         struct sock *sk = sock->sk;
2727         char name[15];
2728         struct net_device *dev;
2729         int err = -ENODEV;
2730
2731         /*
2732          *      Check legality
2733          */
2734
2735         if (addr_len != sizeof(struct sockaddr))
2736                 return -EINVAL;
2737         strlcpy(name, uaddr->sa_data, sizeof(name));
2738
2739         dev = dev_get_by_name(sock_net(sk), name);
2740         if (dev)
2741                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2742         return err;
2743 }
2744
2745 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2746 {
2747         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2748         struct sock *sk = sock->sk;
2749         struct net_device *dev = NULL;
2750         int err;
2751
2752
2753         /*
2754          *      Check legality
2755          */
2756
2757         if (addr_len < sizeof(struct sockaddr_ll))
2758                 return -EINVAL;
2759         if (sll->sll_family != AF_PACKET)
2760                 return -EINVAL;
2761
2762         if (sll->sll_ifindex) {
2763                 err = -ENODEV;
2764                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2765                 if (dev == NULL)
2766                         goto out;
2767         }
2768         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2769
2770 out:
2771         return err;
2772 }
2773
2774 static struct proto packet_proto = {
2775         .name     = "PACKET",
2776         .owner    = THIS_MODULE,
2777         .obj_size = sizeof(struct packet_sock),
2778 };
2779
2780 /*
2781  *      Create a packet of type SOCK_PACKET.
2782  */
2783
2784 static int packet_create(struct net *net, struct socket *sock, int protocol,
2785                          int kern)
2786 {
2787         struct sock *sk;
2788         struct packet_sock *po;
2789         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2790         int err;
2791
2792         if (!ns_capable(net->user_ns, CAP_NET_RAW))
2793                 return -EPERM;
2794         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2795             sock->type != SOCK_PACKET)
2796                 return -ESOCKTNOSUPPORT;
2797
2798         sock->state = SS_UNCONNECTED;
2799
2800         err = -ENOBUFS;
2801         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2802         if (sk == NULL)
2803                 goto out;
2804
2805         sock->ops = &packet_ops;
2806         if (sock->type == SOCK_PACKET)
2807                 sock->ops = &packet_ops_spkt;
2808
2809         sock_init_data(sock, sk);
2810
2811         po = pkt_sk(sk);
2812         sk->sk_family = PF_PACKET;
2813         po->num = proto;
2814         po->xmit = dev_queue_xmit;
2815
2816         err = packet_alloc_pending(po);
2817         if (err)
2818                 goto out2;
2819
2820         packet_cached_dev_reset(po);
2821
2822         sk->sk_destruct = packet_sock_destruct;
2823         sk_refcnt_debug_inc(sk);
2824
2825         /*
2826          *      Attach a protocol block
2827          */
2828
2829         spin_lock_init(&po->bind_lock);
2830         mutex_init(&po->pg_vec_lock);
2831         po->prot_hook.func = packet_rcv;
2832
2833         if (sock->type == SOCK_PACKET)
2834                 po->prot_hook.func = packet_rcv_spkt;
2835
2836         po->prot_hook.af_packet_priv = sk;
2837
2838         if (proto) {
2839                 po->prot_hook.type = proto;
2840                 register_prot_hook(sk);
2841         }
2842
2843         mutex_lock(&net->packet.sklist_lock);
2844         sk_add_node_rcu(sk, &net->packet.sklist);
2845         mutex_unlock(&net->packet.sklist_lock);
2846
2847         preempt_disable();
2848         sock_prot_inuse_add(net, &packet_proto, 1);
2849         preempt_enable();
2850
2851         return 0;
2852 out2:
2853         sk_free(sk);
2854 out:
2855         return err;
2856 }
2857
2858 /*
2859  *      Pull a packet from our receive queue and hand it to the user.
2860  *      If necessary we block.
2861  */
2862
2863 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2864                           struct msghdr *msg, size_t len, int flags)
2865 {
2866         struct sock *sk = sock->sk;
2867         struct sk_buff *skb;
2868         int copied, err;
2869         int vnet_hdr_len = 0;
2870
2871         err = -EINVAL;
2872         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2873                 goto out;
2874
2875 #if 0
2876         /* What error should we return now? EUNATTACH? */
2877         if (pkt_sk(sk)->ifindex < 0)
2878                 return -ENODEV;
2879 #endif
2880
2881         if (flags & MSG_ERRQUEUE) {
2882                 err = sock_recv_errqueue(sk, msg, len,
2883                                          SOL_PACKET, PACKET_TX_TIMESTAMP);
2884                 goto out;
2885         }
2886
2887         /*
2888          *      Call the generic datagram receiver. This handles all sorts
2889          *      of horrible races and re-entrancy so we can forget about it
2890          *      in the protocol layers.
2891          *
2892          *      Now it will return ENETDOWN, if device have just gone down,
2893          *      but then it will block.
2894          */
2895
2896         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2897
2898         /*
2899          *      An error occurred so return it. Because skb_recv_datagram()
2900          *      handles the blocking we don't see and worry about blocking
2901          *      retries.
2902          */
2903
2904         if (skb == NULL)
2905                 goto out;
2906
2907         if (pkt_sk(sk)->has_vnet_hdr) {
2908                 struct virtio_net_hdr vnet_hdr = { 0 };
2909
2910                 err = -EINVAL;
2911                 vnet_hdr_len = sizeof(vnet_hdr);
2912                 if (len < vnet_hdr_len)
2913                         goto out_free;
2914
2915                 len -= vnet_hdr_len;
2916
2917                 if (skb_is_gso(skb)) {
2918                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2919
2920                         /* This is a hint as to how much should be linear. */
2921                         vnet_hdr.hdr_len = skb_headlen(skb);
2922                         vnet_hdr.gso_size = sinfo->gso_size;
2923                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2924                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2925                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2926                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2927                         else if (sinfo->gso_type & SKB_GSO_UDP)
2928                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2929                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2930                                 goto out_free;
2931                         else
2932                                 BUG();
2933                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2934                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2935                 } else
2936                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2937
2938                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2939                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2940                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2941                         vnet_hdr.csum_offset = skb->csum_offset;
2942                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2943                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2944                 } /* else everything is zero */
2945
2946                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2947                                      vnet_hdr_len);
2948                 if (err < 0)
2949                         goto out_free;
2950         }
2951
2952         /* You lose any data beyond the buffer you gave. If it worries
2953          * a user program they can ask the device for its MTU
2954          * anyway.
2955          */
2956         copied = skb->len;
2957         if (copied > len) {
2958                 copied = len;
2959                 msg->msg_flags |= MSG_TRUNC;
2960         }
2961
2962         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2963         if (err)
2964                 goto out_free;
2965
2966         sock_recv_ts_and_drops(msg, sk, skb);
2967
2968         if (msg->msg_name) {
2969                 /* If the address length field is there to be filled
2970                  * in, we fill it in now.
2971                  */
2972                 if (sock->type == SOCK_PACKET) {
2973                         __sockaddr_check_size(sizeof(struct sockaddr_pkt));
2974                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
2975                 } else {
2976                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2977                         msg->msg_namelen = sll->sll_halen +
2978                                 offsetof(struct sockaddr_ll, sll_addr);
2979                 }
2980                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2981                        msg->msg_namelen);
2982         }
2983
2984         if (pkt_sk(sk)->auxdata) {
2985                 struct tpacket_auxdata aux;
2986
2987                 aux.tp_status = TP_STATUS_USER;
2988                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2989                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2990                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2991                 aux.tp_snaplen = skb->len;
2992                 aux.tp_mac = 0;
2993                 aux.tp_net = skb_network_offset(skb);
2994                 if (vlan_tx_tag_present(skb)) {
2995                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2996                         aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
2997                         aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2998                 } else {
2999                         aux.tp_vlan_tci = 0;
3000                         aux.tp_vlan_tpid = 0;
3001                 }
3002                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3003         }
3004
3005         /*
3006          *      Free or return the buffer as appropriate. Again this
3007          *      hides all the races and re-entrancy issues from us.
3008          */
3009         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3010
3011 out_free:
3012         skb_free_datagram(sk, skb);
3013 out:
3014         return err;
3015 }
3016
3017 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3018                                int *uaddr_len, int peer)
3019 {
3020         struct net_device *dev;
3021         struct sock *sk = sock->sk;
3022
3023         if (peer)
3024                 return -EOPNOTSUPP;
3025
3026         uaddr->sa_family = AF_PACKET;
3027         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3028         rcu_read_lock();
3029         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3030         if (dev)
3031                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3032         rcu_read_unlock();
3033         *uaddr_len = sizeof(*uaddr);
3034
3035         return 0;
3036 }
3037
3038 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3039                           int *uaddr_len, int peer)
3040 {
3041         struct net_device *dev;
3042         struct sock *sk = sock->sk;
3043         struct packet_sock *po = pkt_sk(sk);
3044         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3045
3046         if (peer)
3047                 return -EOPNOTSUPP;
3048
3049         sll->sll_family = AF_PACKET;
3050         sll->sll_ifindex = po->ifindex;
3051         sll->sll_protocol = po->num;
3052         sll->sll_pkttype = 0;
3053         rcu_read_lock();
3054         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3055         if (dev) {
3056                 sll->sll_hatype = dev->type;
3057                 sll->sll_halen = dev->addr_len;
3058                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3059         } else {
3060                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
3061                 sll->sll_halen = 0;
3062         }
3063         rcu_read_unlock();
3064         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3065
3066         return 0;
3067 }
3068
3069 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3070                          int what)
3071 {
3072         switch (i->type) {
3073         case PACKET_MR_MULTICAST:
3074                 if (i->alen != dev->addr_len)
3075                         return -EINVAL;
3076                 if (what > 0)
3077                         return dev_mc_add(dev, i->addr);
3078                 else
3079                         return dev_mc_del(dev, i->addr);
3080                 break;
3081         case PACKET_MR_PROMISC:
3082                 return dev_set_promiscuity(dev, what);
3083         case PACKET_MR_ALLMULTI:
3084                 return dev_set_allmulti(dev, what);
3085         case PACKET_MR_UNICAST:
3086                 if (i->alen != dev->addr_len)
3087                         return -EINVAL;
3088                 if (what > 0)
3089                         return dev_uc_add(dev, i->addr);
3090                 else
3091                         return dev_uc_del(dev, i->addr);
3092                 break;
3093         default:
3094                 break;
3095         }
3096         return 0;
3097 }
3098
3099 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
3100 {
3101         for ( ; i; i = i->next) {
3102                 if (i->ifindex == dev->ifindex)
3103                         packet_dev_mc(dev, i, what);
3104         }
3105 }
3106
3107 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3108 {
3109         struct packet_sock *po = pkt_sk(sk);
3110         struct packet_mclist *ml, *i;
3111         struct net_device *dev;
3112         int err;
3113
3114         rtnl_lock();
3115
3116         err = -ENODEV;
3117         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3118         if (!dev)
3119                 goto done;
3120
3121         err = -EINVAL;
3122         if (mreq->mr_alen > dev->addr_len)
3123                 goto done;
3124
3125         err = -ENOBUFS;
3126         i = kmalloc(sizeof(*i), GFP_KERNEL);
3127         if (i == NULL)
3128                 goto done;
3129
3130         err = 0;
3131         for (ml = po->mclist; ml; ml = ml->next) {
3132                 if (ml->ifindex == mreq->mr_ifindex &&
3133                     ml->type == mreq->mr_type &&
3134                     ml->alen == mreq->mr_alen &&
3135                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3136                         ml->count++;
3137                         /* Free the new element ... */
3138                         kfree(i);
3139                         goto done;
3140                 }
3141         }
3142
3143         i->type = mreq->mr_type;
3144         i->ifindex = mreq->mr_ifindex;
3145         i->alen = mreq->mr_alen;
3146         memcpy(i->addr, mreq->mr_address, i->alen);
3147         i->count = 1;
3148         i->next = po->mclist;
3149         po->mclist = i;
3150         err = packet_dev_mc(dev, i, 1);
3151         if (err) {
3152                 po->mclist = i->next;
3153                 kfree(i);
3154         }
3155
3156 done:
3157         rtnl_unlock();
3158         return err;
3159 }
3160
3161 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3162 {
3163         struct packet_mclist *ml, **mlp;
3164
3165         rtnl_lock();
3166
3167         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3168                 if (ml->ifindex == mreq->mr_ifindex &&
3169                     ml->type == mreq->mr_type &&
3170                     ml->alen == mreq->mr_alen &&
3171                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3172                         if (--ml->count == 0) {
3173                                 struct net_device *dev;
3174                                 *mlp = ml->next;
3175                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3176                                 if (dev)
3177                                         packet_dev_mc(dev, ml, -1);
3178                                 kfree(ml);
3179                         }
3180                         rtnl_unlock();
3181                         return 0;
3182                 }
3183         }
3184         rtnl_unlock();
3185         return -EADDRNOTAVAIL;
3186 }
3187
3188 static void packet_flush_mclist(struct sock *sk)
3189 {
3190         struct packet_sock *po = pkt_sk(sk);
3191         struct packet_mclist *ml;
3192
3193         if (!po->mclist)
3194                 return;
3195
3196         rtnl_lock();
3197         while ((ml = po->mclist) != NULL) {
3198                 struct net_device *dev;
3199
3200                 po->mclist = ml->next;
3201                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3202                 if (dev != NULL)
3203                         packet_dev_mc(dev, ml, -1);
3204                 kfree(ml);
3205         }
3206         rtnl_unlock();
3207 }
3208
3209 static int
3210 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3211 {
3212         struct sock *sk = sock->sk;
3213         struct packet_sock *po = pkt_sk(sk);
3214         int ret;
3215
3216         if (level != SOL_PACKET)
3217                 return -ENOPROTOOPT;
3218
3219         switch (optname) {
3220         case PACKET_ADD_MEMBERSHIP:
3221         case PACKET_DROP_MEMBERSHIP:
3222         {
3223                 struct packet_mreq_max mreq;
3224                 int len = optlen;
3225                 memset(&mreq, 0, sizeof(mreq));
3226                 if (len < sizeof(struct packet_mreq))
3227                         return -EINVAL;
3228                 if (len > sizeof(mreq))
3229                         len = sizeof(mreq);
3230                 if (copy_from_user(&mreq, optval, len))
3231                         return -EFAULT;
3232                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3233                         return -EINVAL;
3234                 if (optname == PACKET_ADD_MEMBERSHIP)
3235                         ret = packet_mc_add(sk, &mreq);
3236                 else
3237                         ret = packet_mc_drop(sk, &mreq);
3238                 return ret;
3239         }
3240
3241         case PACKET_RX_RING:
3242         case PACKET_TX_RING:
3243         {
3244                 union tpacket_req_u req_u;
3245                 int len;
3246
3247                 switch (po->tp_version) {
3248                 case TPACKET_V1:
3249                 case TPACKET_V2:
3250                         len = sizeof(req_u.req);
3251                         break;
3252                 case TPACKET_V3:
3253                 default:
3254                         len = sizeof(req_u.req3);
3255                         break;
3256                 }
3257                 if (optlen < len)
3258                         return -EINVAL;
3259                 if (pkt_sk(sk)->has_vnet_hdr)
3260                         return -EINVAL;
3261                 if (copy_from_user(&req_u.req, optval, len))
3262                         return -EFAULT;
3263                 return packet_set_ring(sk, &req_u, 0,
3264                         optname == PACKET_TX_RING);
3265         }
3266         case PACKET_COPY_THRESH:
3267         {
3268                 int val;
3269
3270                 if (optlen != sizeof(val))
3271                         return -EINVAL;
3272                 if (copy_from_user(&val, optval, sizeof(val)))
3273                         return -EFAULT;
3274
3275                 pkt_sk(sk)->copy_thresh = val;
3276                 return 0;
3277         }
3278         case PACKET_VERSION:
3279         {
3280                 int val;
3281
3282                 if (optlen != sizeof(val))
3283                         return -EINVAL;
3284                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3285                         return -EBUSY;
3286                 if (copy_from_user(&val, optval, sizeof(val)))
3287                         return -EFAULT;
3288                 switch (val) {
3289                 case TPACKET_V1:
3290                 case TPACKET_V2:
3291                 case TPACKET_V3:
3292                         po->tp_version = val;
3293                         return 0;
3294                 default:
3295                         return -EINVAL;
3296                 }
3297         }
3298         case PACKET_RESERVE:
3299         {
3300                 unsigned int val;
3301
3302                 if (optlen != sizeof(val))
3303                         return -EINVAL;
3304                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3305                         return -EBUSY;
3306                 if (copy_from_user(&val, optval, sizeof(val)))
3307                         return -EFAULT;
3308                 po->tp_reserve = val;
3309                 return 0;
3310         }
3311         case PACKET_LOSS:
3312         {
3313                 unsigned int val;
3314
3315                 if (optlen != sizeof(val))
3316                         return -EINVAL;
3317                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3318                         return -EBUSY;
3319                 if (copy_from_user(&val, optval, sizeof(val)))
3320                         return -EFAULT;
3321                 po->tp_loss = !!val;
3322                 return 0;
3323         }
3324         case PACKET_AUXDATA:
3325         {
3326                 int val;
3327
3328                 if (optlen < sizeof(val))
3329                         return -EINVAL;
3330                 if (copy_from_user(&val, optval, sizeof(val)))
3331                         return -EFAULT;
3332
3333                 po->auxdata = !!val;
3334                 return 0;
3335         }
3336         case PACKET_ORIGDEV:
3337         {
3338                 int val;
3339
3340                 if (optlen < sizeof(val))
3341                         return -EINVAL;
3342                 if (copy_from_user(&val, optval, sizeof(val)))
3343                         return -EFAULT;
3344
3345                 po->origdev = !!val;
3346                 return 0;
3347         }
3348         case PACKET_VNET_HDR:
3349         {
3350                 int val;
3351
3352                 if (sock->type != SOCK_RAW)
3353                         return -EINVAL;
3354                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3355                         return -EBUSY;
3356                 if (optlen < sizeof(val))
3357                         return -EINVAL;
3358                 if (copy_from_user(&val, optval, sizeof(val)))
3359                         return -EFAULT;
3360
3361                 po->has_vnet_hdr = !!val;
3362                 return 0;
3363         }
3364         case PACKET_TIMESTAMP:
3365         {
3366                 int val;
3367
3368                 if (optlen != sizeof(val))
3369                         return -EINVAL;
3370                 if (copy_from_user(&val, optval, sizeof(val)))
3371                         return -EFAULT;
3372
3373                 po->tp_tstamp = val;
3374                 return 0;
3375         }
3376         case PACKET_FANOUT:
3377         {
3378                 int val;
3379
3380                 if (optlen != sizeof(val))
3381                         return -EINVAL;
3382                 if (copy_from_user(&val, optval, sizeof(val)))
3383                         return -EFAULT;
3384
3385                 return fanout_add(sk, val & 0xffff, val >> 16);
3386         }
3387         case PACKET_TX_HAS_OFF:
3388         {
3389                 unsigned int val;
3390
3391                 if (optlen != sizeof(val))
3392                         return -EINVAL;
3393                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3394                         return -EBUSY;
3395                 if (copy_from_user(&val, optval, sizeof(val)))
3396                         return -EFAULT;
3397                 po->tp_tx_has_off = !!val;
3398                 return 0;
3399         }
3400         case PACKET_QDISC_BYPASS:
3401         {
3402                 int val;
3403
3404                 if (optlen != sizeof(val))
3405                         return -EINVAL;
3406                 if (copy_from_user(&val, optval, sizeof(val)))
3407                         return -EFAULT;
3408
3409                 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3410                 return 0;
3411         }
3412         default:
3413                 return -ENOPROTOOPT;
3414         }
3415 }
3416
3417 static int packet_getsockopt(struct socket *sock, int level, int optname,
3418                              char __user *optval, int __user *optlen)
3419 {
3420         int len;
3421         int val, lv = sizeof(val);
3422         struct sock *sk = sock->sk;
3423         struct packet_sock *po = pkt_sk(sk);
3424         void *data = &val;
3425         union tpacket_stats_u st;
3426
3427         if (level != SOL_PACKET)
3428                 return -ENOPROTOOPT;
3429
3430         if (get_user(len, optlen))
3431                 return -EFAULT;
3432
3433         if (len < 0)
3434                 return -EINVAL;
3435
3436         switch (optname) {
3437         case PACKET_STATISTICS:
3438                 spin_lock_bh(&sk->sk_receive_queue.lock);
3439                 memcpy(&st, &po->stats, sizeof(st));
3440                 memset(&po->stats, 0, sizeof(po->stats));
3441                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3442
3443                 if (po->tp_version == TPACKET_V3) {
3444                         lv = sizeof(struct tpacket_stats_v3);
3445                         st.stats3.tp_packets += st.stats3.tp_drops;
3446                         data = &st.stats3;
3447                 } else {
3448                         lv = sizeof(struct tpacket_stats);
3449                         st.stats1.tp_packets += st.stats1.tp_drops;
3450                         data = &st.stats1;
3451                 }
3452
3453                 break;
3454         case PACKET_AUXDATA:
3455                 val = po->auxdata;
3456                 break;
3457         case PACKET_ORIGDEV:
3458                 val = po->origdev;
3459                 break;
3460         case PACKET_VNET_HDR:
3461                 val = po->has_vnet_hdr;
3462                 break;
3463         case PACKET_VERSION:
3464                 val = po->tp_version;
3465                 break;
3466         case PACKET_HDRLEN:
3467                 if (len > sizeof(int))
3468                         len = sizeof(int);
3469                 if (copy_from_user(&val, optval, len))
3470                         return -EFAULT;
3471                 switch (val) {
3472                 case TPACKET_V1:
3473                         val = sizeof(struct tpacket_hdr);
3474                         break;
3475                 case TPACKET_V2:
3476                         val = sizeof(struct tpacket2_hdr);
3477                         break;
3478                 case TPACKET_V3:
3479                         val = sizeof(struct tpacket3_hdr);
3480                         break;
3481                 default:
3482                         return -EINVAL;
3483                 }
3484                 break;
3485         case PACKET_RESERVE:
3486                 val = po->tp_reserve;
3487                 break;
3488         case PACKET_LOSS:
3489                 val = po->tp_loss;
3490                 break;
3491         case PACKET_TIMESTAMP:
3492                 val = po->tp_tstamp;
3493                 break;
3494         case PACKET_FANOUT:
3495                 val = (po->fanout ?
3496                        ((u32)po->fanout->id |
3497                         ((u32)po->fanout->type << 16) |
3498                         ((u32)po->fanout->flags << 24)) :
3499                        0);
3500                 break;
3501         case PACKET_TX_HAS_OFF:
3502                 val = po->tp_tx_has_off;
3503                 break;
3504         case PACKET_QDISC_BYPASS:
3505                 val = packet_use_direct_xmit(po);
3506                 break;
3507         default:
3508                 return -ENOPROTOOPT;
3509         }
3510
3511         if (len > lv)
3512                 len = lv;
3513         if (put_user(len, optlen))
3514                 return -EFAULT;
3515         if (copy_to_user(optval, data, len))
3516                 return -EFAULT;
3517         return 0;
3518 }
3519
3520
3521 static int packet_notifier(struct notifier_block *this,
3522                            unsigned long msg, void *ptr)
3523 {
3524         struct sock *sk;
3525         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3526         struct net *net = dev_net(dev);
3527
3528         rcu_read_lock();
3529         sk_for_each_rcu(sk, &net->packet.sklist) {
3530                 struct packet_sock *po = pkt_sk(sk);
3531
3532                 switch (msg) {
3533                 case NETDEV_UNREGISTER:
3534                         if (po->mclist)
3535                                 packet_dev_mclist(dev, po->mclist, -1);
3536                         /* fallthrough */
3537
3538                 case NETDEV_DOWN:
3539                         if (dev->ifindex == po->ifindex) {
3540                                 spin_lock(&po->bind_lock);
3541                                 if (po->running) {
3542                                         __unregister_prot_hook(sk, false);
3543                                         sk->sk_err = ENETDOWN;
3544                                         if (!sock_flag(sk, SOCK_DEAD))
3545                                                 sk->sk_error_report(sk);
3546                                 }
3547                                 if (msg == NETDEV_UNREGISTER) {
3548                                         packet_cached_dev_reset(po);
3549                                         po->ifindex = -1;
3550                                         if (po->prot_hook.dev)
3551                                                 dev_put(po->prot_hook.dev);
3552                                         po->prot_hook.dev = NULL;
3553                                 }
3554                                 spin_unlock(&po->bind_lock);
3555                         }
3556                         break;
3557                 case NETDEV_UP:
3558                         if (dev->ifindex == po->ifindex) {
3559                                 spin_lock(&po->bind_lock);
3560                                 if (po->num)
3561                                         register_prot_hook(sk);
3562                                 spin_unlock(&po->bind_lock);
3563                         }
3564                         break;
3565                 }
3566         }
3567         rcu_read_unlock();
3568         return NOTIFY_DONE;
3569 }
3570
3571
3572 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3573                         unsigned long arg)
3574 {
3575         struct sock *sk = sock->sk;
3576
3577         switch (cmd) {
3578         case SIOCOUTQ:
3579         {
3580                 int amount = sk_wmem_alloc_get(sk);
3581
3582                 return put_user(amount, (int __user *)arg);
3583         }
3584         case SIOCINQ:
3585         {
3586                 struct sk_buff *skb;
3587                 int amount = 0;
3588
3589                 spin_lock_bh(&sk->sk_receive_queue.lock);
3590                 skb = skb_peek(&sk->sk_receive_queue);
3591                 if (skb)
3592                         amount = skb->len;
3593                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3594                 return put_user(amount, (int __user *)arg);
3595         }
3596         case SIOCGSTAMP:
3597                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3598         case SIOCGSTAMPNS:
3599                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3600
3601 #ifdef CONFIG_INET
3602         case SIOCADDRT:
3603         case SIOCDELRT:
3604         case SIOCDARP:
3605         case SIOCGARP:
3606         case SIOCSARP:
3607         case SIOCGIFADDR:
3608         case SIOCSIFADDR:
3609         case SIOCGIFBRDADDR:
3610         case SIOCSIFBRDADDR:
3611         case SIOCGIFNETMASK:
3612         case SIOCSIFNETMASK:
3613         case SIOCGIFDSTADDR:
3614         case SIOCSIFDSTADDR:
3615         case SIOCSIFFLAGS:
3616                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3617 #endif
3618
3619         default:
3620                 return -ENOIOCTLCMD;
3621         }
3622         return 0;
3623 }
3624
3625 static unsigned int packet_poll(struct file *file, struct socket *sock,
3626                                 poll_table *wait)
3627 {
3628         struct sock *sk = sock->sk;
3629         struct packet_sock *po = pkt_sk(sk);
3630         unsigned int mask = datagram_poll(file, sock, wait);
3631
3632         spin_lock_bh(&sk->sk_receive_queue.lock);
3633         if (po->rx_ring.pg_vec) {
3634                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3635                         TP_STATUS_KERNEL))
3636                         mask |= POLLIN | POLLRDNORM;
3637         }
3638         spin_unlock_bh(&sk->sk_receive_queue.lock);
3639         spin_lock_bh(&sk->sk_write_queue.lock);
3640         if (po->tx_ring.pg_vec) {
3641                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3642                         mask |= POLLOUT | POLLWRNORM;
3643         }
3644         spin_unlock_bh(&sk->sk_write_queue.lock);
3645         return mask;
3646 }
3647
3648
3649 /* Dirty? Well, I still did not learn better way to account
3650  * for user mmaps.
3651  */
3652
3653 static void packet_mm_open(struct vm_area_struct *vma)
3654 {
3655         struct file *file = vma->vm_file;
3656         struct socket *sock = file->private_data;
3657         struct sock *sk = sock->sk;
3658
3659         if (sk)
3660                 atomic_inc(&pkt_sk(sk)->mapped);
3661 }
3662
3663 static void packet_mm_close(struct vm_area_struct *vma)
3664 {
3665         struct file *file = vma->vm_file;
3666         struct socket *sock = file->private_data;
3667         struct sock *sk = sock->sk;
3668
3669         if (sk)
3670                 atomic_dec(&pkt_sk(sk)->mapped);
3671 }
3672
3673 static const struct vm_operations_struct packet_mmap_ops = {
3674         .open   =       packet_mm_open,
3675         .close  =       packet_mm_close,
3676 };
3677
3678 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3679                         unsigned int len)
3680 {
3681         int i;
3682
3683         for (i = 0; i < len; i++) {
3684                 if (likely(pg_vec[i].buffer)) {
3685                         if (is_vmalloc_addr(pg_vec[i].buffer))
3686                                 vfree(pg_vec[i].buffer);
3687                         else
3688                                 free_pages((unsigned long)pg_vec[i].buffer,
3689                                            order);
3690                         pg_vec[i].buffer = NULL;
3691                 }
3692         }
3693         kfree(pg_vec);
3694 }
3695
3696 static char *alloc_one_pg_vec_page(unsigned long order)
3697 {
3698         char *buffer;
3699         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3700                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3701
3702         buffer = (char *) __get_free_pages(gfp_flags, order);
3703         if (buffer)
3704                 return buffer;
3705
3706         /* __get_free_pages failed, fall back to vmalloc */
3707         buffer = vzalloc((1 << order) * PAGE_SIZE);
3708         if (buffer)
3709                 return buffer;
3710
3711         /* vmalloc failed, lets dig into swap here */
3712         gfp_flags &= ~__GFP_NORETRY;
3713         buffer = (char *) __get_free_pages(gfp_flags, order);
3714         if (buffer)
3715                 return buffer;
3716
3717         /* complete and utter failure */
3718         return NULL;
3719 }
3720
3721 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3722 {
3723         unsigned int block_nr = req->tp_block_nr;
3724         struct pgv *pg_vec;
3725         int i;
3726
3727         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3728         if (unlikely(!pg_vec))
3729                 goto out;
3730
3731         for (i = 0; i < block_nr; i++) {
3732                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3733                 if (unlikely(!pg_vec[i].buffer))
3734                         goto out_free_pgvec;
3735         }
3736
3737 out:
3738         return pg_vec;
3739
3740 out_free_pgvec:
3741         free_pg_vec(pg_vec, order, block_nr);
3742         pg_vec = NULL;
3743         goto out;
3744 }
3745
3746 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3747                 int closing, int tx_ring)
3748 {
3749         struct pgv *pg_vec = NULL;
3750         struct packet_sock *po = pkt_sk(sk);
3751         int was_running, order = 0;
3752         struct packet_ring_buffer *rb;
3753         struct sk_buff_head *rb_queue;
3754         __be16 num;
3755         int err = -EINVAL;
3756         /* Added to avoid minimal code churn */
3757         struct tpacket_req *req = &req_u->req;
3758
3759         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3760         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3761                 WARN(1, "Tx-ring is not supported.\n");
3762                 goto out;
3763         }
3764
3765         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3766         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3767
3768         err = -EBUSY;
3769         if (!closing) {
3770                 if (atomic_read(&po->mapped))
3771                         goto out;
3772                 if (packet_read_pending(rb))
3773                         goto out;
3774         }
3775
3776         if (req->tp_block_nr) {
3777                 /* Sanity tests and some calculations */
3778                 err = -EBUSY;
3779                 if (unlikely(rb->pg_vec))
3780                         goto out;
3781
3782                 switch (po->tp_version) {
3783                 case TPACKET_V1:
3784                         po->tp_hdrlen = TPACKET_HDRLEN;
3785                         break;
3786                 case TPACKET_V2:
3787                         po->tp_hdrlen = TPACKET2_HDRLEN;
3788                         break;
3789                 case TPACKET_V3:
3790                         po->tp_hdrlen = TPACKET3_HDRLEN;
3791                         break;
3792                 }
3793
3794                 err = -EINVAL;
3795                 if (unlikely((int)req->tp_block_size <= 0))
3796                         goto out;
3797                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3798                         goto out;
3799                 if (po->tp_version >= TPACKET_V3 &&
3800                     (int)(req->tp_block_size -
3801                           BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
3802                         goto out;
3803                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3804                                         po->tp_reserve))
3805                         goto out;
3806                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3807                         goto out;
3808
3809                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3810                 if (unlikely(rb->frames_per_block <= 0))
3811                         goto out;
3812                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3813                                         req->tp_frame_nr))
3814                         goto out;
3815
3816                 err = -ENOMEM;
3817                 order = get_order(req->tp_block_size);
3818                 pg_vec = alloc_pg_vec(req, order);
3819                 if (unlikely(!pg_vec))
3820                         goto out;
3821                 switch (po->tp_version) {
3822                 case TPACKET_V3:
3823                 /* Transmit path is not supported. We checked
3824                  * it above but just being paranoid
3825                  */
3826                         if (!tx_ring)
3827                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3828                         break;
3829                 default:
3830                         break;
3831                 }
3832         }
3833         /* Done */
3834         else {
3835                 err = -EINVAL;
3836                 if (unlikely(req->tp_frame_nr))
3837                         goto out;
3838         }
3839
3840         lock_sock(sk);
3841
3842         /* Detach socket from network */
3843         spin_lock(&po->bind_lock);
3844         was_running = po->running;
3845         num = po->num;
3846         if (was_running) {
3847                 po->num = 0;
3848                 __unregister_prot_hook(sk, false);
3849         }
3850         spin_unlock(&po->bind_lock);
3851
3852         synchronize_net();
3853
3854         err = -EBUSY;
3855         mutex_lock(&po->pg_vec_lock);
3856         if (closing || atomic_read(&po->mapped) == 0) {
3857                 err = 0;
3858                 spin_lock_bh(&rb_queue->lock);
3859                 swap(rb->pg_vec, pg_vec);
3860                 rb->frame_max = (req->tp_frame_nr - 1);
3861                 rb->head = 0;
3862                 rb->frame_size = req->tp_frame_size;
3863                 spin_unlock_bh(&rb_queue->lock);
3864
3865                 swap(rb->pg_vec_order, order);
3866                 swap(rb->pg_vec_len, req->tp_block_nr);
3867
3868                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3869                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3870                                                 tpacket_rcv : packet_rcv;
3871                 skb_queue_purge(rb_queue);
3872                 if (atomic_read(&po->mapped))
3873                         pr_err("packet_mmap: vma is busy: %d\n",
3874                                atomic_read(&po->mapped));
3875         }
3876         mutex_unlock(&po->pg_vec_lock);
3877
3878         spin_lock(&po->bind_lock);
3879         if (was_running) {
3880                 po->num = num;
3881                 register_prot_hook(sk);
3882         }
3883         spin_unlock(&po->bind_lock);
3884         if (closing && (po->tp_version > TPACKET_V2)) {
3885                 /* Because we don't support block-based V3 on tx-ring */
3886                 if (!tx_ring)
3887                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3888         }
3889         release_sock(sk);
3890
3891         if (pg_vec)
3892                 free_pg_vec(pg_vec, order, req->tp_block_nr);
3893 out:
3894         return err;
3895 }
3896
3897 static int packet_mmap(struct file *file, struct socket *sock,
3898                 struct vm_area_struct *vma)
3899 {
3900         struct sock *sk = sock->sk;
3901         struct packet_sock *po = pkt_sk(sk);
3902         unsigned long size, expected_size;
3903         struct packet_ring_buffer *rb;
3904         unsigned long start;
3905         int err = -EINVAL;
3906         int i;
3907
3908         if (vma->vm_pgoff)
3909                 return -EINVAL;
3910
3911         mutex_lock(&po->pg_vec_lock);
3912
3913         expected_size = 0;
3914         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3915                 if (rb->pg_vec) {
3916                         expected_size += rb->pg_vec_len
3917                                                 * rb->pg_vec_pages
3918                                                 * PAGE_SIZE;
3919                 }
3920         }
3921
3922         if (expected_size == 0)
3923                 goto out;
3924
3925         size = vma->vm_end - vma->vm_start;
3926         if (size != expected_size)
3927                 goto out;
3928
3929         start = vma->vm_start;
3930         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3931                 if (rb->pg_vec == NULL)
3932                         continue;
3933
3934                 for (i = 0; i < rb->pg_vec_len; i++) {
3935                         struct page *page;
3936                         void *kaddr = rb->pg_vec[i].buffer;
3937                         int pg_num;
3938
3939                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3940                                 page = pgv_to_page(kaddr);
3941                                 err = vm_insert_page(vma, start, page);
3942                                 if (unlikely(err))
3943                                         goto out;
3944                                 start += PAGE_SIZE;
3945                                 kaddr += PAGE_SIZE;
3946                         }
3947                 }
3948         }
3949
3950         atomic_inc(&po->mapped);
3951         vma->vm_ops = &packet_mmap_ops;
3952         err = 0;
3953
3954 out:
3955         mutex_unlock(&po->pg_vec_lock);
3956         return err;
3957 }
3958
3959 static const struct proto_ops packet_ops_spkt = {
3960         .family =       PF_PACKET,
3961         .owner =        THIS_MODULE,
3962         .release =      packet_release,
3963         .bind =         packet_bind_spkt,
3964         .connect =      sock_no_connect,
3965         .socketpair =   sock_no_socketpair,
3966         .accept =       sock_no_accept,
3967         .getname =      packet_getname_spkt,
3968         .poll =         datagram_poll,
3969         .ioctl =        packet_ioctl,
3970         .listen =       sock_no_listen,
3971         .shutdown =     sock_no_shutdown,
3972         .setsockopt =   sock_no_setsockopt,
3973         .getsockopt =   sock_no_getsockopt,
3974         .sendmsg =      packet_sendmsg_spkt,
3975         .recvmsg =      packet_recvmsg,
3976         .mmap =         sock_no_mmap,
3977         .sendpage =     sock_no_sendpage,
3978 };
3979
3980 static const struct proto_ops packet_ops = {
3981         .family =       PF_PACKET,
3982         .owner =        THIS_MODULE,
3983         .release =      packet_release,
3984         .bind =         packet_bind,
3985         .connect =      sock_no_connect,
3986         .socketpair =   sock_no_socketpair,
3987         .accept =       sock_no_accept,
3988         .getname =      packet_getname,
3989         .poll =         packet_poll,
3990         .ioctl =        packet_ioctl,
3991         .listen =       sock_no_listen,
3992         .shutdown =     sock_no_shutdown,
3993         .setsockopt =   packet_setsockopt,
3994         .getsockopt =   packet_getsockopt,
3995         .sendmsg =      packet_sendmsg,
3996         .recvmsg =      packet_recvmsg,
3997         .mmap =         packet_mmap,
3998         .sendpage =     sock_no_sendpage,
3999 };
4000
4001 static const struct net_proto_family packet_family_ops = {
4002         .family =       PF_PACKET,
4003         .create =       packet_create,
4004         .owner  =       THIS_MODULE,
4005 };
4006
4007 static struct notifier_block packet_netdev_notifier = {
4008         .notifier_call =        packet_notifier,
4009 };
4010
4011 #ifdef CONFIG_PROC_FS
4012
4013 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4014         __acquires(RCU)
4015 {
4016         struct net *net = seq_file_net(seq);
4017
4018         rcu_read_lock();
4019         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4020 }
4021
4022 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4023 {
4024         struct net *net = seq_file_net(seq);
4025         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4026 }
4027
4028 static void packet_seq_stop(struct seq_file *seq, void *v)
4029         __releases(RCU)
4030 {
4031         rcu_read_unlock();
4032 }
4033
4034 static int packet_seq_show(struct seq_file *seq, void *v)
4035 {
4036         if (v == SEQ_START_TOKEN)
4037                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
4038         else {
4039                 struct sock *s = sk_entry(v);
4040                 const struct packet_sock *po = pkt_sk(s);
4041
4042                 seq_printf(seq,
4043                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4044                            s,
4045                            atomic_read(&s->sk_refcnt),
4046                            s->sk_type,
4047                            ntohs(po->num),
4048                            po->ifindex,
4049                            po->running,
4050                            atomic_read(&s->sk_rmem_alloc),
4051                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4052                            sock_i_ino(s));
4053         }
4054
4055         return 0;
4056 }
4057
4058 static const struct seq_operations packet_seq_ops = {
4059         .start  = packet_seq_start,
4060         .next   = packet_seq_next,
4061         .stop   = packet_seq_stop,
4062         .show   = packet_seq_show,
4063 };
4064
4065 static int packet_seq_open(struct inode *inode, struct file *file)
4066 {
4067         return seq_open_net(inode, file, &packet_seq_ops,
4068                             sizeof(struct seq_net_private));
4069 }
4070
4071 static const struct file_operations packet_seq_fops = {
4072         .owner          = THIS_MODULE,
4073         .open           = packet_seq_open,
4074         .read           = seq_read,
4075         .llseek         = seq_lseek,
4076         .release        = seq_release_net,
4077 };
4078
4079 #endif
4080
4081 static int __net_init packet_net_init(struct net *net)
4082 {
4083         mutex_init(&net->packet.sklist_lock);
4084         INIT_HLIST_HEAD(&net->packet.sklist);
4085
4086         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4087                 return -ENOMEM;
4088
4089         return 0;
4090 }
4091
4092 static void __net_exit packet_net_exit(struct net *net)
4093 {
4094         remove_proc_entry("packet", net->proc_net);
4095 }
4096
4097 static struct pernet_operations packet_net_ops = {
4098         .init = packet_net_init,
4099         .exit = packet_net_exit,
4100 };
4101
4102
4103 static void __exit packet_exit(void)
4104 {
4105         unregister_netdevice_notifier(&packet_netdev_notifier);
4106         unregister_pernet_subsys(&packet_net_ops);
4107         sock_unregister(PF_PACKET);
4108         proto_unregister(&packet_proto);
4109 }
4110
4111 static int __init packet_init(void)
4112 {
4113         int rc = proto_register(&packet_proto, 0);
4114
4115         if (rc != 0)
4116                 goto out;
4117
4118         sock_register(&packet_family_ops);
4119         register_pernet_subsys(&packet_net_ops);
4120         register_netdevice_notifier(&packet_netdev_notifier);
4121 out:
4122         return rc;
4123 }
4124
4125 module_init(packet_init);
4126 module_exit(packet_exit);
4127 MODULE_LICENSE("GPL");
4128 MODULE_ALIAS_NETPROTO(PF_PACKET);