net/packet/af_packet.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   4  *              operating system.  INET is implemented using the  BSD Socket
   5  *              interface as the means of communication with the user level.
   6  *
   7  *              PACKET - implements raw packet sockets.
   8  *
   9  * Authors:     Ross Biro
  10  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  11  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  12  *
  13  * Fixes:
  14  *              Alan Cox        :       verify_area() now used correctly
  15  *              Alan Cox        :       new skbuff lists, look ma no backlogs!
  16  *              Alan Cox        :       tidied skbuff lists.
  17  *              Alan Cox        :       Now uses generic datagram routines I
  18  *                                      added. Also fixed the peek/read crash
  19  *                                      from all old Linux datagram code.
  20  *              Alan Cox        :       Uses the improved datagram code.
  21  *              Alan Cox        :       Added NULL's for socket options.
  22  *              Alan Cox        :       Re-commented the code.
  23  *              Alan Cox        :       Use new kernel side addressing
  24  *              Rob Janssen     :       Correct MTU usage.
  25  *              Dave Platt      :       Counter leaks caused by incorrect
  26  *                                      interrupt locking and some slightly
  27  *                                      dubious gcc output. Can you read
  28  *                                      compiler: it said _VOLATILE_
  29  *      Richard Kooijman        :       Timestamp fixes.
  30  *              Alan Cox        :       New buffers. Use sk->mac.raw.
  31  *              Alan Cox        :       sendmsg/recvmsg support.
  32  *              Alan Cox        :       Protocol setting support
  33  *      Alexey Kuznetsov        :       Untied from IPv4 stack.
  34  *      Cyrus Durgin            :       Fixed kerneld for kmod.
  35  *      Michal Ostrowski        :       Module initialization cleanup.
  36  *         Ulises Alonso        :       Frame number limit removal and
  37  *                                      packet_set_ring memory leak.
  38  *              Eric Biederman  :       Allow for > 8 byte hardware addresses.
  39  *                                      The convention is that longer addresses
  40  *                                      will simply extend the hardware address
  41  *                                      byte arrays at the end of sockaddr_ll
  42  *                                      and packet_mreq.
  43  *              Johann Baudy    :       Added TX RING.
  44  *              Chetan Loke     :       Implemented TPACKET_V3 block abstraction
  45  *                                      layer.
  46  *                                      Copyright (C) 2011, <lokec@ccs.neu.edu>
  47  */
  48
  49 #include <linux/types.h>
  50 #include <linux/mm.h>
  51 #include <linux/capability.h>
  52 #include <linux/fcntl.h>
  53 #include <linux/socket.h>
  54 #include <linux/in.h>
  55 #include <linux/inet.h>
  56 #include <linux/netdevice.h>
  57 #include <linux/if_packet.h>
  58 #include <linux/wireless.h>
  59 #include <linux/kernel.h>
  60 #include <linux/kmod.h>
  61 #include <linux/slab.h>
  62 #include <linux/vmalloc.h>
  63 #include <net/net_namespace.h>
  64 #include <net/ip.h>
  65 #include <net/protocol.h>
  66 #include <linux/skbuff.h>
  67 #include <net/sock.h>
  68 #include <linux/errno.h>
  69 #include <linux/timer.h>
  70 #include <linux/uaccess.h>
  71 #include <asm/ioctls.h>
  72 #include <asm/page.h>
  73 #include <asm/cacheflush.h>
  74 #include <asm/io.h>
  75 #include <linux/proc_fs.h>
  76 #include <linux/seq_file.h>
  77 #include <linux/poll.h>
  78 #include <linux/module.h>
  79 #include <linux/init.h>
  80 #include <linux/mutex.h>
  81 #include <linux/if_vlan.h>
  82 #include <linux/virtio_net.h>
  83 #include <linux/errqueue.h>
  84 #include <linux/net_tstamp.h>
  85 #include <linux/percpu.h>
  86 #ifdef CONFIG_INET
  87 #include <net/inet_common.h>
  88 #endif
  89 #include <linux/bpf.h>
  90 #include <net/compat.h>
  91
  92 #include "internal.h"
  93
  94 /*
  95    Assumptions:
  96    - if device has no dev->hard_header routine, it adds and removes ll header
  97      inside itself. In this case ll header is invisible outside of device,
  98      but higher levels still should reserve dev->hard_header_len.
  99      Some devices are enough clever to reallocate skb, when header
 100      will not fit to reserved space (tunnel), another ones are silly
 101      (PPP).
 102    - packet socket receives packets with pulled ll header,
 103      so that SOCK_RAW should push it back.
 104
 105 On receive:
 106 -----------
 107
 108 Incoming, dev->hard_header!=NULL
 109    mac_header -> ll header
 110    data       -> data
 111
 112 Outgoing, dev->hard_header!=NULL
 113    mac_header -> ll header
 114    data       -> ll header
 115
 116 Incoming, dev->hard_header==NULL
 117    mac_header -> UNKNOWN position. It is very likely, that it points to ll
 118                  header.  PPP makes it, that is wrong, because introduce
 119                  assymetry between rx and tx paths.
 120    data       -> data
 121
 122 Outgoing, dev->hard_header==NULL
 123    mac_header -> data. ll header is still not built!
 124    data       -> data
 125
 126 Resume
 127   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 128
 129
 130 On transmit:
 131 ------------
 132
 133 dev->hard_header != NULL
 134    mac_header -> ll header
 135    data       -> ll header
 136
 137 dev->hard_header == NULL (ll header is added by device, we cannot control it)
 138    mac_header -> data
 139    data       -> data
 140
 141    We should set nh.raw on output to correct posistion,
 142    packet classifier depends on it.
 143  */
 144
 145 /* Private packet socket structures. */
 146
 147 /* identical to struct packet_mreq except it has
 148  * a longer address field.
 149  */
 150 struct packet_mreq_max {
 151         int             mr_ifindex;
 152         unsigned short  mr_type;
 153         unsigned short  mr_alen;
 154         unsigned char   mr_address[MAX_ADDR_LEN];
 155 };
 156
 157 union tpacket_uhdr {
 158         struct tpacket_hdr  *h1;
 159         struct tpacket2_hdr *h2;
 160         struct tpacket3_hdr *h3;
 161         void *raw;
 162 };
 163
 164 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 165                 int closing, int tx_ring);
 166
 167 #define V3_ALIGNMENT    (8)
 168
 169 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 170
 171 #define BLK_PLUS_PRIV(sz_of_priv) \
 172         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 173
 174 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
 175 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
 176 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
 177 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
 178 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
 179 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
 180 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
 181
 182 struct packet_sock;
 183 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 184                        struct packet_type *pt, struct net_device *orig_dev);
 185
 186 static void *packet_previous_frame(struct packet_sock *po,
 187                 struct packet_ring_buffer *rb,
 188                 int status);
 189 static void packet_increment_head(struct packet_ring_buffer *buff);
 190 static int prb_curr_blk_in_use(struct tpacket_block_desc *);
 191 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 192                         struct packet_sock *);
 193 static void prb_retire_current_block(struct tpacket_kbdq_core *,
 194                 struct packet_sock *, unsigned int status);
 195 static int prb_queue_frozen(struct tpacket_kbdq_core *);
 196 static void prb_open_block(struct tpacket_kbdq_core *,
 197                 struct tpacket_block_desc *);
 198 static void prb_retire_rx_blk_timer_expired(struct timer_list *);
 199 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 200 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 201 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 202                 struct tpacket3_hdr *);
 203 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 204                 struct tpacket3_hdr *);
 205 static void packet_flush_mclist(struct sock *sk);
 206 static u16 packet_pick_tx_queue(struct sk_buff *skb);
 207
 208 struct packet_skb_cb {
 209         union {
 210                 struct sockaddr_pkt pkt;
 211                 union {
 212                         /* Trick: alias skb original length with
 213                          * ll.sll_family and ll.protocol in order
 214                          * to save room.
 215                          */
 216                         unsigned int origlen;
 217                         struct sockaddr_ll ll;
 218                 };
 219         } sa;
 220 };
 221
 222 #define vio_le() virtio_legacy_is_little_endian()
 223
 224 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 225
 226 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 227 #define GET_PBLOCK_DESC(x, bid) \
 228         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 229 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 230         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 231 #define GET_NEXT_PRB_BLK_NUM(x) \
 232         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 233         ((x)->kactive_blk_num+1) : 0)
 234
 235 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 236 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 237
 238 static int packet_direct_xmit(struct sk_buff *skb)
 239 {
 240         return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
 241 }
 242
 243 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
 244 {
 245         struct net_device *dev;
 246
 247         rcu_read_lock();
 248         dev = rcu_dereference(po->cached_dev);
 249         if (likely(dev))
 250                 dev_hold(dev);
 251         rcu_read_unlock();
 252
 253         return dev;
 254 }
 255
 256 static void packet_cached_dev_assign(struct packet_sock *po,
 257                                      struct net_device *dev)
 258 {
 259         rcu_assign_pointer(po->cached_dev, dev);
 260 }
 261
 262 static void packet_cached_dev_reset(struct packet_sock *po)
 263 {
 264         RCU_INIT_POINTER(po->cached_dev, NULL);
 265 }
 266
 267 static bool packet_use_direct_xmit(const struct packet_sock *po)
 268 {
 269         return po->xmit == packet_direct_xmit;
 270 }
 271
 272 static u16 packet_pick_tx_queue(struct sk_buff *skb)
 273 {
 274         struct net_device *dev = skb->dev;
 275         const struct net_device_ops *ops = dev->netdev_ops;
 276         int cpu = raw_smp_processor_id();
 277         u16 queue_index;
 278
 279 #ifdef CONFIG_XPS
 280         skb->sender_cpu = cpu + 1;
 281 #endif
 282         skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues);
 283         if (ops->ndo_select_queue) {
 284                 queue_index = ops->ndo_select_queue(dev, skb, NULL);
 285                 queue_index = netdev_cap_txqueue(dev, queue_index);
 286         } else {
 287                 queue_index = netdev_pick_tx(dev, skb, NULL);
 288         }
 289
 290         return queue_index;
 291 }
 292
 293 /* __register_prot_hook must be invoked through register_prot_hook
 294  * or from a context in which asynchronous accesses to the packet
 295  * socket is not possible (packet_create()).
 296  */
 297 static void __register_prot_hook(struct sock *sk)
 298 {
 299         struct packet_sock *po = pkt_sk(sk);
 300
 301         if (!po->running) {
 302                 if (po->fanout)
 303                         __fanout_link(sk, po);
 304                 else
 305                         dev_add_pack(&po->prot_hook);
 306
 307                 sock_hold(sk);
 308                 po->running = 1;
 309         }
 310 }
 311
 312 static void register_prot_hook(struct sock *sk)
 313 {
 314         lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
 315         __register_prot_hook(sk);
 316 }
 317
 318 /* If the sync parameter is true, we will temporarily drop
 319  * the po->bind_lock and do a synchronize_net to make sure no
 320  * asynchronous packet processing paths still refer to the elements
 321  * of po->prot_hook.  If the sync parameter is false, it is the
 322  * callers responsibility to take care of this.
 323  */
 324 static void __unregister_prot_hook(struct sock *sk, bool sync)
 325 {
 326         struct packet_sock *po = pkt_sk(sk);
 327
 328         lockdep_assert_held_once(&po->bind_lock);
 329
 330         po->running = 0;
 331
 332         if (po->fanout)
 333                 __fanout_unlink(sk, po);
 334         else
 335                 __dev_remove_pack(&po->prot_hook);
 336
 337         __sock_put(sk);
 338
 339         if (sync) {
 340                 spin_unlock(&po->bind_lock);
 341                 synchronize_net();
 342                 spin_lock(&po->bind_lock);
 343         }
 344 }
 345
 346 static void unregister_prot_hook(struct sock *sk, bool sync)
 347 {
 348         struct packet_sock *po = pkt_sk(sk);
 349
 350         if (po->running)
 351                 __unregister_prot_hook(sk, sync);
 352 }
 353
 354 static inline struct page * __pure pgv_to_page(void *addr)
 355 {
 356         if (is_vmalloc_addr(addr))
 357                 return vmalloc_to_page(addr);
 358         return virt_to_page(addr);
 359 }
 360
 361 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 362 {
 363         union tpacket_uhdr h;
 364
 365         h.raw = frame;
 366         switch (po->tp_version) {
 367         case TPACKET_V1:
 368                 h.h1->tp_status = status;
 369                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 370                 break;
 371         case TPACKET_V2:
 372                 h.h2->tp_status = status;
 373                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 374                 break;
 375         case TPACKET_V3:
 376                 h.h3->tp_status = status;
 377                 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
 378                 break;
 379         default:
 380                 WARN(1, "TPACKET version not supported.\n");
 381                 BUG();
 382         }
 383
 384         smp_wmb();
 385 }
 386
 387 static int __packet_get_status(const struct packet_sock *po, void *frame)
 388 {
 389         union tpacket_uhdr h;
 390
 391         smp_rmb();
 392
 393         h.raw = frame;
 394         switch (po->tp_version) {
 395         case TPACKET_V1:
 396                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 397                 return h.h1->tp_status;
 398         case TPACKET_V2:
 399                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 400                 return h.h2->tp_status;
 401         case TPACKET_V3:
 402                 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
 403                 return h.h3->tp_status;
 404         default:
 405                 WARN(1, "TPACKET version not supported.\n");
 406                 BUG();
 407                 return 0;
 408         }
 409 }
 410
 411 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
 412                                    unsigned int flags)
 413 {
 414         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 415
 416         if (shhwtstamps &&
 417             (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
 418             ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
 419                 return TP_STATUS_TS_RAW_HARDWARE;
 420
 421         if (ktime_to_timespec_cond(skb->tstamp, ts))
 422                 return TP_STATUS_TS_SOFTWARE;
 423
 424         return 0;
 425 }
 426
 427 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
 428                                     struct sk_buff *skb)
 429 {
 430         union tpacket_uhdr h;
 431         struct timespec ts;
 432         __u32 ts_status;
 433
 434         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
 435                 return 0;
 436
 437         h.raw = frame;
 438         switch (po->tp_version) {
 439         case TPACKET_V1:
 440                 h.h1->tp_sec = ts.tv_sec;
 441                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
 442                 break;
 443         case TPACKET_V2:
 444                 h.h2->tp_sec = ts.tv_sec;
 445                 h.h2->tp_nsec = ts.tv_nsec;
 446                 break;
 447         case TPACKET_V3:
 448                 h.h3->tp_sec = ts.tv_sec;
 449                 h.h3->tp_nsec = ts.tv_nsec;
 450                 break;
 451         default:
 452                 WARN(1, "TPACKET version not supported.\n");
 453                 BUG();
 454         }
 455
 456         /* one flush is safe, as both fields always lie on the same cacheline */
 457         flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
 458         smp_wmb();
 459
 460         return ts_status;
 461 }
 462
 463 static void *packet_lookup_frame(const struct packet_sock *po,
 464                                  const struct packet_ring_buffer *rb,
 465                                  unsigned int position,
 466                                  int status)
 467 {
 468         unsigned int pg_vec_pos, frame_offset;
 469         union tpacket_uhdr h;
 470
 471         pg_vec_pos = position / rb->frames_per_block;
 472         frame_offset = position % rb->frames_per_block;
 473
 474         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 475                 (frame_offset * rb->frame_size);
 476
 477         if (status != __packet_get_status(po, h.raw))
 478                 return NULL;
 479
 480         return h.raw;
 481 }
 482
 483 static void *packet_current_frame(struct packet_sock *po,
 484                 struct packet_ring_buffer *rb,
 485                 int status)
 486 {
 487         return packet_lookup_frame(po, rb, rb->head, status);
 488 }
 489
 490 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 491 {
 492         del_timer_sync(&pkc->retire_blk_timer);
 493 }
 494
 495 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 496                 struct sk_buff_head *rb_queue)
 497 {
 498         struct tpacket_kbdq_core *pkc;
 499
 500         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 501
 502         spin_lock_bh(&rb_queue->lock);
 503         pkc->delete_blk_timer = 1;
 504         spin_unlock_bh(&rb_queue->lock);
 505
 506         prb_del_retire_blk_timer(pkc);
 507 }
 508
 509 static void prb_setup_retire_blk_timer(struct packet_sock *po)
 510 {
 511         struct tpacket_kbdq_core *pkc;
 512
 513         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 514         timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
 515                     0);
 516         pkc->retire_blk_timer.expires = jiffies;
 517 }
 518
 519 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 520                                 int blk_size_in_bytes)
 521 {
 522         struct net_device *dev;
 523         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 524         struct ethtool_link_ksettings ecmd;
 525         int err;
 526
 527         rtnl_lock();
 528         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 529         if (unlikely(!dev)) {
 530                 rtnl_unlock();
 531                 return DEFAULT_PRB_RETIRE_TOV;
 532         }
 533         err = __ethtool_get_link_ksettings(dev, &ecmd);
 534         rtnl_unlock();
 535         if (!err) {
 536                 /*
 537                  * If the link speed is so slow you don't really
 538                  * need to worry about perf anyways
 539                  */
 540                 if (ecmd.base.speed < SPEED_1000 ||
 541                     ecmd.base.speed == SPEED_UNKNOWN) {
 542                         return DEFAULT_PRB_RETIRE_TOV;
 543                 } else {
 544                         msec = 1;
 545                         div = ecmd.base.speed / 1000;
 546                 }
 547         }
 548
 549         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 550
 551         if (div)
 552                 mbits /= div;
 553
 554         tmo = mbits * msec;
 555
 556         if (div)
 557                 return tmo+1;
 558         return tmo;
 559 }
 560
 561 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 562                         union tpacket_req_u *req_u)
 563 {
 564         p1->feature_req_word = req_u->req3.tp_feature_req_word;
 565 }
 566
 567 static void init_prb_bdqc(struct packet_sock *po,
 568                         struct packet_ring_buffer *rb,
 569                         struct pgv *pg_vec,
 570                         union tpacket_req_u *req_u)
 571 {
 572         struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
 573         struct tpacket_block_desc *pbd;
 574
 575         memset(p1, 0x0, sizeof(*p1));
 576
 577         p1->knxt_seq_num = 1;
 578         p1->pkbdq = pg_vec;
 579         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 580         p1->pkblk_start = pg_vec[0].buffer;
 581         p1->kblk_size = req_u->req3.tp_block_size;
 582         p1->knum_blocks = req_u->req3.tp_block_nr;
 583         p1->hdrlen = po->tp_hdrlen;
 584         p1->version = po->tp_version;
 585         p1->last_kactive_blk_num = 0;
 586         po->stats.stats3.tp_freeze_q_cnt = 0;
 587         if (req_u->req3.tp_retire_blk_tov)
 588                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 589         else
 590                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 591                                                 req_u->req3.tp_block_size);
 592         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 593         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 594
 595         p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
 596         prb_init_ft_ops(p1, req_u);
 597         prb_setup_retire_blk_timer(po);
 598         prb_open_block(p1, pbd);
 599 }
 600
 601 /*  Do NOT update the last_blk_num first.
 602  *  Assumes sk_buff_head lock is held.
 603  */
 604 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 605 {
 606         mod_timer(&pkc->retire_blk_timer,
 607                         jiffies + pkc->tov_in_jiffies);
 608         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 609 }
 610
 611 /*
 612  * Timer logic:
 613  * 1) We refresh the timer only when we open a block.
 614  *    By doing this we don't waste cycles refreshing the timer
 615  *        on packet-by-packet basis.
 616  *
 617  * With a 1MB block-size, on a 1Gbps line, it will take
 618  * i) ~8 ms to fill a block + ii) memcpy etc.
 619  * In this cut we are not accounting for the memcpy time.
 620  *
 621  * So, if the user sets the 'tmo' to 10ms then the timer
 622  * will never fire while the block is still getting filled
 623  * (which is what we want). However, the user could choose
 624  * to close a block early and that's fine.
 625  *
 626  * But when the timer does fire, we check whether or not to refresh it.
 627  * Since the tmo granularity is in msecs, it is not too expensive
 628  * to refresh the timer, lets say every '8' msecs.
 629  * Either the user can set the 'tmo' or we can derive it based on
 630  * a) line-speed and b) block-size.
 631  * prb_calc_retire_blk_tmo() calculates the tmo.
 632  *
 633  */
 634 static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
 635 {
 636         struct packet_sock *po =
 637                 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
 638         struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 639         unsigned int frozen;
 640         struct tpacket_block_desc *pbd;
 641
 642         spin_lock(&po->sk.sk_receive_queue.lock);
 643
 644         frozen = prb_queue_frozen(pkc);
 645         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 646
 647         if (unlikely(pkc->delete_blk_timer))
 648                 goto out;
 649
 650         /* We only need to plug the race when the block is partially filled.
 651          * tpacket_rcv:
 652          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 653          *              copy_bits() is in progress ...
 654          *              timer fires on other cpu:
 655          *              we can't retire the current block because copy_bits
 656          *              is in progress.
 657          *
 658          */
 659         if (BLOCK_NUM_PKTS(pbd)) {
 660                 while (atomic_read(&pkc->blk_fill_in_prog)) {
 661                         /* Waiting for skb_copy_bits to finish... */
 662                         cpu_relax();
 663                 }
 664         }
 665
 666         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 667                 if (!frozen) {
 668                         if (!BLOCK_NUM_PKTS(pbd)) {
 669                                 /* An empty block. Just refresh the timer. */
 670                                 goto refresh_timer;
 671                         }
 672                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 673                         if (!prb_dispatch_next_block(pkc, po))
 674                                 goto refresh_timer;
 675                         else
 676                                 goto out;
 677                 } else {
 678                         /* Case 1. Queue was frozen because user-space was
 679                          *         lagging behind.
 680                          */
 681                         if (prb_curr_blk_in_use(pbd)) {
 682                                 /*
 683                                  * Ok, user-space is still behind.
 684                                  * So just refresh the timer.
 685                                  */
 686                                 goto refresh_timer;
 687                         } else {
 688                                /* Case 2. queue was frozen,user-space caught up,
 689                                 * now the link went idle && the timer fired.
 690                                 * We don't have a block to close.So we open this
 691                                 * block and restart the timer.
 692                                 * opening a block thaws the queue,restarts timer
 693                                 * Thawing/timer-refresh is a side effect.
 694                                 */
 695                                 prb_open_block(pkc, pbd);
 696                                 goto out;
 697                         }
 698                 }
 699         }
 700
 701 refresh_timer:
 702         _prb_refresh_rx_retire_blk_timer(pkc);
 703
 704 out:
 705         spin_unlock(&po->sk.sk_receive_queue.lock);
 706 }
 707
 708 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 709                 struct tpacket_block_desc *pbd1, __u32 status)
 710 {
 711         /* Flush everything minus the block header */
 712
 713 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 714         u8 *start, *end;
 715
 716         start = (u8 *)pbd1;
 717
 718         /* Skip the block header(we know header WILL fit in 4K) */
 719         start += PAGE_SIZE;
 720
 721         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 722         for (; start < end; start += PAGE_SIZE)
 723                 flush_dcache_page(pgv_to_page(start));
 724
 725         smp_wmb();
 726 #endif
 727
 728         /* Now update the block status. */
 729
 730         BLOCK_STATUS(pbd1) = status;
 731
 732         /* Flush the block header */
 733
 734 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 735         start = (u8 *)pbd1;
 736         flush_dcache_page(pgv_to_page(start));
 737
 738         smp_wmb();
 739 #endif
 740 }
 741
 742 /*
 743  * Side effect:
 744  *
 745  * 1) flush the block
 746  * 2) Increment active_blk_num
 747  *
 748  * Note:We DONT refresh the timer on purpose.
 749  *      Because almost always the next block will be opened.
 750  */
 751 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 752                 struct tpacket_block_desc *pbd1,
 753                 struct packet_sock *po, unsigned int stat)
 754 {
 755         __u32 status = TP_STATUS_USER | stat;
 756
 757         struct tpacket3_hdr *last_pkt;
 758         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 759         struct sock *sk = &po->sk;
 760
 761         if (atomic_read(&po->tp_drops))
 762                 status |= TP_STATUS_LOSING;
 763
 764         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 765         last_pkt->tp_next_offset = 0;
 766
 767         /* Get the ts of the last pkt */
 768         if (BLOCK_NUM_PKTS(pbd1)) {
 769                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 770                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 771         } else {
 772                 /* Ok, we tmo'd - so get the current time.
 773                  *
 774                  * It shouldn't really happen as we don't close empty
 775                  * blocks. See prb_retire_rx_blk_timer_expired().
 776                  */
 777                 struct timespec ts;
 778                 getnstimeofday(&ts);
 779                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 780                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 781         }
 782
 783         smp_wmb();
 784
 785         /* Flush the block */
 786         prb_flush_block(pkc1, pbd1, status);
 787
 788         sk->sk_data_ready(sk);
 789
 790         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 791 }
 792
 793 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 794 {
 795         pkc->reset_pending_on_curr_blk = 0;
 796 }
 797
 798 /*
 799  * Side effect of opening a block:
 800  *
 801  * 1) prb_queue is thawed.
 802  * 2) retire_blk_timer is refreshed.
 803  *
 804  */
 805 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 806         struct tpacket_block_desc *pbd1)
 807 {
 808         struct timespec ts;
 809         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 810
 811         smp_rmb();
 812
 813         /* We could have just memset this but we will lose the
 814          * flexibility of making the priv area sticky
 815          */
 816
 817         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 818         BLOCK_NUM_PKTS(pbd1) = 0;
 819         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 820
 821         getnstimeofday(&ts);
 822
 823         h1->ts_first_pkt.ts_sec = ts.tv_sec;
 824         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 825
 826         pkc1->pkblk_start = (char *)pbd1;
 827         pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 828
 829         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 830         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 831
 832         pbd1->version = pkc1->version;
 833         pkc1->prev = pkc1->nxt_offset;
 834         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 835
 836         prb_thaw_queue(pkc1);
 837         _prb_refresh_rx_retire_blk_timer(pkc1);
 838
 839         smp_wmb();
 840 }
 841
 842 /*
 843  * Queue freeze logic:
 844  * 1) Assume tp_block_nr = 8 blocks.
 845  * 2) At time 't0', user opens Rx ring.
 846  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 847  * 4) user-space is either sleeping or processing block '0'.
 848  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 849  *    it will close block-7,loop around and try to fill block '0'.
 850  *    call-flow:
 851  *    __packet_lookup_frame_in_block
 852  *      prb_retire_current_block()
 853  *      prb_dispatch_next_block()
 854  *        |->(BLOCK_STATUS == USER) evaluates to true
 855  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 856  * 6) Now there are two cases:
 857  *    6.1) Link goes idle right after the queue is frozen.
 858  *         But remember, the last open_block() refreshed the timer.
 859  *         When this timer expires,it will refresh itself so that we can
 860  *         re-open block-0 in near future.
 861  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 862  *         case and __packet_lookup_frame_in_block will check if block-0
 863  *         is free and can now be re-used.
 864  */
 865 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 866                                   struct packet_sock *po)
 867 {
 868         pkc->reset_pending_on_curr_blk = 1;
 869         po->stats.stats3.tp_freeze_q_cnt++;
 870 }
 871
 872 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 873
 874 /*
 875  * If the next block is free then we will dispatch it
 876  * and return a good offset.
 877  * Else, we will freeze the queue.
 878  * So, caller must check the return value.
 879  */
 880 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 881                 struct packet_sock *po)
 882 {
 883         struct tpacket_block_desc *pbd;
 884
 885         smp_rmb();
 886
 887         /* 1. Get current block num */
 888         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 889
 890         /* 2. If this block is currently in_use then freeze the queue */
 891         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 892                 prb_freeze_queue(pkc, po);
 893                 return NULL;
 894         }
 895
 896         /*
 897          * 3.
 898          * open this block and return the offset where the first packet
 899          * needs to get stored.
 900          */
 901         prb_open_block(pkc, pbd);
 902         return (void *)pkc->nxt_offset;
 903 }
 904
 905 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 906                 struct packet_sock *po, unsigned int status)
 907 {
 908         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 909
 910         /* retire/close the current block */
 911         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 912                 /*
 913                  * Plug the case where copy_bits() is in progress on
 914                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 915                  * have space to copy the pkt in the current block and
 916                  * called prb_retire_current_block()
 917                  *
 918                  * We don't need to worry about the TMO case because
 919                  * the timer-handler already handled this case.
 920                  */
 921                 if (!(status & TP_STATUS_BLK_TMO)) {
 922                         while (atomic_read(&pkc->blk_fill_in_prog)) {
 923                                 /* Waiting for skb_copy_bits to finish... */
 924                                 cpu_relax();
 925                         }
 926                 }
 927                 prb_close_block(pkc, pbd, po, status);
 928                 return;
 929         }
 930 }
 931
 932 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
 933 {
 934         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 935 }
 936
 937 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 938 {
 939         return pkc->reset_pending_on_curr_blk;
 940 }
 941
 942 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 943 {
 944         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 945         atomic_dec(&pkc->blk_fill_in_prog);
 946 }
 947
 948 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 949                         struct tpacket3_hdr *ppd)
 950 {
 951         ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
 952 }
 953
 954 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 955                         struct tpacket3_hdr *ppd)
 956 {
 957         ppd->hv1.tp_rxhash = 0;
 958 }
 959
 960 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
 961                         struct tpacket3_hdr *ppd)
 962 {
 963         if (skb_vlan_tag_present(pkc->skb)) {
 964                 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
 965                 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
 966                 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
 967         } else {
 968                 ppd->hv1.tp_vlan_tci = 0;
 969                 ppd->hv1.tp_vlan_tpid = 0;
 970                 ppd->tp_status = TP_STATUS_AVAILABLE;
 971         }
 972 }
 973
 974 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
 975                         struct tpacket3_hdr *ppd)
 976 {
 977         ppd->hv1.tp_padding = 0;
 978         prb_fill_vlan_info(pkc, ppd);
 979
 980         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
 981                 prb_fill_rxhash(pkc, ppd);
 982         else
 983                 prb_clear_rxhash(pkc, ppd);
 984 }
 985
 986 static void prb_fill_curr_block(char *curr,
 987                                 struct tpacket_kbdq_core *pkc,
 988                                 struct tpacket_block_desc *pbd,
 989                                 unsigned int len)
 990 {
 991         struct tpacket3_hdr *ppd;
 992
 993         ppd  = (struct tpacket3_hdr *)curr;
 994         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
 995         pkc->prev = curr;
 996         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
 997         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
 998         BLOCK_NUM_PKTS(pbd) += 1;
 999         atomic_inc(&pkc->blk_fill_in_prog);
1000         prb_run_all_ft_ops(pkc, ppd);
1001 }
1002
1003 /* Assumes caller has the sk->rx_queue.lock */
1004 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1005                                             struct sk_buff *skb,
1006                                             unsigned int len
1007                                             )
1008 {
1009         struct tpacket_kbdq_core *pkc;
1010         struct tpacket_block_desc *pbd;
1011         char *curr, *end;
1012
1013         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1014         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1015
1016         /* Queue is frozen when user space is lagging behind */
1017         if (prb_queue_frozen(pkc)) {
1018                 /*
1019                  * Check if that last block which caused the queue to freeze,
1020                  * is still in_use by user-space.
1021                  */
1022                 if (prb_curr_blk_in_use(pbd)) {
1023                         /* Can't record this packet */
1024                         return NULL;
1025                 } else {
1026                         /*
1027                          * Ok, the block was released by user-space.
1028                          * Now let's open that block.
1029                          * opening a block also thaws the queue.
1030                          * Thawing is a side effect.
1031                          */
1032                         prb_open_block(pkc, pbd);
1033                 }
1034         }
1035
1036         smp_mb();
1037         curr = pkc->nxt_offset;
1038         pkc->skb = skb;
1039         end = (char *)pbd + pkc->kblk_size;
1040
1041         /* first try the current block */
1042         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1043                 prb_fill_curr_block(curr, pkc, pbd, len);
1044                 return (void *)curr;
1045         }
1046
1047         /* Ok, close the current block */
1048         prb_retire_current_block(pkc, po, 0);
1049
1050         /* Now, try to dispatch the next block */
1051         curr = (char *)prb_dispatch_next_block(pkc, po);
1052         if (curr) {
1053                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1054                 prb_fill_curr_block(curr, pkc, pbd, len);
1055                 return (void *)curr;
1056         }
1057
1058         /*
1059          * No free blocks are available.user_space hasn't caught up yet.
1060          * Queue was just frozen and now this packet will get dropped.
1061          */
1062         return NULL;
1063 }
1064
1065 static void *packet_current_rx_frame(struct packet_sock *po,
1066                                             struct sk_buff *skb,
1067                                             int status, unsigned int len)
1068 {
1069         char *curr = NULL;
1070         switch (po->tp_version) {
1071         case TPACKET_V1:
1072         case TPACKET_V2:
1073                 curr = packet_lookup_frame(po, &po->rx_ring,
1074                                         po->rx_ring.head, status);
1075                 return curr;
1076         case TPACKET_V3:
1077                 return __packet_lookup_frame_in_block(po, skb, len);
1078         default:
1079                 WARN(1, "TPACKET version not supported\n");
1080                 BUG();
1081                 return NULL;
1082         }
1083 }
1084
1085 static void *prb_lookup_block(const struct packet_sock *po,
1086                               const struct packet_ring_buffer *rb,
1087                               unsigned int idx,
1088                               int status)
1089 {
1090         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1091         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1092
1093         if (status != BLOCK_STATUS(pbd))
1094                 return NULL;
1095         return pbd;
1096 }
1097
1098 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1099 {
1100         unsigned int prev;
1101         if (rb->prb_bdqc.kactive_blk_num)
1102                 prev = rb->prb_bdqc.kactive_blk_num-1;
1103         else
1104                 prev = rb->prb_bdqc.knum_blocks-1;
1105         return prev;
1106 }
1107
1108 /* Assumes caller has held the rx_queue.lock */
1109 static void *__prb_previous_block(struct packet_sock *po,
1110                                          struct packet_ring_buffer *rb,
1111                                          int status)
1112 {
1113         unsigned int previous = prb_previous_blk_num(rb);
1114         return prb_lookup_block(po, rb, previous, status);
1115 }
1116
1117 static void *packet_previous_rx_frame(struct packet_sock *po,
1118                                              struct packet_ring_buffer *rb,
1119                                              int status)
1120 {
1121         if (po->tp_version <= TPACKET_V2)
1122                 return packet_previous_frame(po, rb, status);
1123
1124         return __prb_previous_block(po, rb, status);
1125 }
1126
1127 static void packet_increment_rx_head(struct packet_sock *po,
1128                                             struct packet_ring_buffer *rb)
1129 {
1130         switch (po->tp_version) {
1131         case TPACKET_V1:
1132         case TPACKET_V2:
1133                 return packet_increment_head(rb);
1134         case TPACKET_V3:
1135         default:
1136                 WARN(1, "TPACKET version not supported.\n");
1137                 BUG();
1138                 return;
1139         }
1140 }
1141
1142 static void *packet_previous_frame(struct packet_sock *po,
1143                 struct packet_ring_buffer *rb,
1144                 int status)
1145 {
1146         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1147         return packet_lookup_frame(po, rb, previous, status);
1148 }
1149
1150 static void packet_increment_head(struct packet_ring_buffer *buff)
1151 {
1152         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1153 }
1154
1155 static void packet_inc_pending(struct packet_ring_buffer *rb)
1156 {
1157         this_cpu_inc(*rb->pending_refcnt);
1158 }
1159
1160 static void packet_dec_pending(struct packet_ring_buffer *rb)
1161 {
1162         this_cpu_dec(*rb->pending_refcnt);
1163 }
1164
1165 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1166 {
1167         unsigned int refcnt = 0;
1168         int cpu;
1169
1170         /* We don't use pending refcount in rx_ring. */
1171         if (rb->pending_refcnt == NULL)
1172                 return 0;
1173
1174         for_each_possible_cpu(cpu)
1175                 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1176
1177         return refcnt;
1178 }
1179
1180 static int packet_alloc_pending(struct packet_sock *po)
1181 {
1182         po->rx_ring.pending_refcnt = NULL;
1183
1184         po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1185         if (unlikely(po->tx_ring.pending_refcnt == NULL))
1186                 return -ENOBUFS;
1187
1188         return 0;
1189 }
1190
1191 static void packet_free_pending(struct packet_sock *po)
1192 {
1193         free_percpu(po->tx_ring.pending_refcnt);
1194 }
1195
1196 #define ROOM_POW_OFF    2
1197 #define ROOM_NONE       0x0
1198 #define ROOM_LOW        0x1
1199 #define ROOM_NORMAL     0x2
1200
1201 static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1202 {
1203         int idx, len;
1204
1205         len = READ_ONCE(po->rx_ring.frame_max) + 1;
1206         idx = READ_ONCE(po->rx_ring.head);
1207         if (pow_off)
1208                 idx += len >> pow_off;
1209         if (idx >= len)
1210                 idx -= len;
1211         return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1212 }
1213
1214 static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1215 {
1216         int idx, len;
1217
1218         len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1219         idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1220         if (pow_off)
1221                 idx += len >> pow_off;
1222         if (idx >= len)
1223                 idx -= len;
1224         return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1225 }
1226
1227 static int __packet_rcv_has_room(const struct packet_sock *po,
1228                                  const struct sk_buff *skb)
1229 {
1230         const struct sock *sk = &po->sk;
1231         int ret = ROOM_NONE;
1232
1233         if (po->prot_hook.func != tpacket_rcv) {
1234                 int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1235                 int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1236                                    - (skb ? skb->truesize : 0);
1237
1238                 if (avail > (rcvbuf >> ROOM_POW_OFF))
1239                         return ROOM_NORMAL;
1240                 else if (avail > 0)
1241                         return ROOM_LOW;
1242                 else
1243                         return ROOM_NONE;
1244         }
1245
1246         if (po->tp_version == TPACKET_V3) {
1247                 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1248                         ret = ROOM_NORMAL;
1249                 else if (__tpacket_v3_has_room(po, 0))
1250                         ret = ROOM_LOW;
1251         } else {
1252                 if (__tpacket_has_room(po, ROOM_POW_OFF))
1253                         ret = ROOM_NORMAL;
1254                 else if (__tpacket_has_room(po, 0))
1255                         ret = ROOM_LOW;
1256         }
1257
1258         return ret;
1259 }
1260
1261 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1262 {
1263         int pressure, ret;
1264
1265         ret = __packet_rcv_has_room(po, skb);
1266         pressure = ret != ROOM_NORMAL;
1267
1268         if (READ_ONCE(po->pressure) != pressure)
1269                 WRITE_ONCE(po->pressure, pressure);
1270
1271         return ret;
1272 }
1273
1274 static void packet_sock_destruct(struct sock *sk)
1275 {
1276         skb_queue_purge(&sk->sk_error_queue);
1277
1278         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1279         WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1280
1281         if (!sock_flag(sk, SOCK_DEAD)) {
1282                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1283                 return;
1284         }
1285
1286         sk_refcnt_debug_dec(sk);
1287 }
1288
1289 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1290 {
1291         u32 rxhash;
1292         int i, count = 0;
1293
1294         rxhash = skb_get_hash(skb);
1295         for (i = 0; i < ROLLOVER_HLEN; i++)
1296                 if (po->rollover->history[i] == rxhash)
1297                         count++;
1298
1299         po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1300         return count > (ROLLOVER_HLEN >> 1);
1301 }
1302
1303 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1304                                       struct sk_buff *skb,
1305                                       unsigned int num)
1306 {
1307         return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1308 }
1309
1310 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1311                                     struct sk_buff *skb,
1312                                     unsigned int num)
1313 {
1314         unsigned int val = atomic_inc_return(&f->rr_cur);
1315
1316         return val % num;
1317 }
1318
1319 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1320                                      struct sk_buff *skb,
1321                                      unsigned int num)
1322 {
1323         return smp_processor_id() % num;
1324 }
1325
1326 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1327                                      struct sk_buff *skb,
1328                                      unsigned int num)
1329 {
1330         return prandom_u32_max(num);
1331 }
1332
1333 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1334                                           struct sk_buff *skb,
1335                                           unsigned int idx, bool try_self,
1336                                           unsigned int num)
1337 {
1338         struct packet_sock *po, *po_next, *po_skip = NULL;
1339         unsigned int i, j, room = ROOM_NONE;
1340
1341         po = pkt_sk(f->arr[idx]);
1342
1343         if (try_self) {
1344                 room = packet_rcv_has_room(po, skb);
1345                 if (room == ROOM_NORMAL ||
1346                     (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1347                         return idx;
1348                 po_skip = po;
1349         }
1350
1351         i = j = min_t(int, po->rollover->sock, num - 1);
1352         do {
1353                 po_next = pkt_sk(f->arr[i]);
1354                 if (po_next != po_skip && !READ_ONCE(po_next->pressure) &&
1355                     packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1356                         if (i != j)
1357                                 po->rollover->sock = i;
1358                         atomic_long_inc(&po->rollover->num);
1359                         if (room == ROOM_LOW)
1360                                 atomic_long_inc(&po->rollover->num_huge);
1361                         return i;
1362                 }
1363
1364                 if (++i == num)
1365                         i = 0;
1366         } while (i != j);
1367
1368         atomic_long_inc(&po->rollover->num_failed);
1369         return idx;
1370 }
1371
1372 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1373                                     struct sk_buff *skb,
1374                                     unsigned int num)
1375 {
1376         return skb_get_queue_mapping(skb) % num;
1377 }
1378
1379 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1380                                      struct sk_buff *skb,
1381                                      unsigned int num)
1382 {
1383         struct bpf_prog *prog;
1384         unsigned int ret = 0;
1385
1386         rcu_read_lock();
1387         prog = rcu_dereference(f->bpf_prog);
1388         if (prog)
1389                 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1390         rcu_read_unlock();
1391
1392         return ret;
1393 }
1394
1395 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1396 {
1397         return f->flags & (flag >> 8);
1398 }
1399
1400 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1401                              struct packet_type *pt, struct net_device *orig_dev)
1402 {
1403         struct packet_fanout *f = pt->af_packet_priv;
1404         unsigned int num = READ_ONCE(f->num_members);
1405         struct net *net = read_pnet(&f->net);
1406         struct packet_sock *po;
1407         unsigned int idx;
1408
1409         if (!net_eq(dev_net(dev), net) || !num) {
1410                 kfree_skb(skb);
1411                 return 0;
1412         }
1413
1414         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1415                 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1416                 if (!skb)
1417                         return 0;
1418         }
1419         switch (f->type) {
1420         case PACKET_FANOUT_HASH:
1421         default:
1422                 idx = fanout_demux_hash(f, skb, num);
1423                 break;
1424         case PACKET_FANOUT_LB:
1425                 idx = fanout_demux_lb(f, skb, num);
1426                 break;
1427         case PACKET_FANOUT_CPU:
1428                 idx = fanout_demux_cpu(f, skb, num);
1429                 break;
1430         case PACKET_FANOUT_RND:
1431                 idx = fanout_demux_rnd(f, skb, num);
1432                 break;
1433         case PACKET_FANOUT_QM:
1434                 idx = fanout_demux_qm(f, skb, num);
1435                 break;
1436         case PACKET_FANOUT_ROLLOVER:
1437                 idx = fanout_demux_rollover(f, skb, 0, false, num);
1438                 break;
1439         case PACKET_FANOUT_CBPF:
1440         case PACKET_FANOUT_EBPF:
1441                 idx = fanout_demux_bpf(f, skb, num);
1442                 break;
1443         }
1444
1445         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1446                 idx = fanout_demux_rollover(f, skb, idx, true, num);
1447
1448         po = pkt_sk(f->arr[idx]);
1449         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1450 }
1451
1452 DEFINE_MUTEX(fanout_mutex);
1453 EXPORT_SYMBOL_GPL(fanout_mutex);
1454 static LIST_HEAD(fanout_list);
1455 static u16 fanout_next_id;
1456
1457 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1458 {
1459         struct packet_fanout *f = po->fanout;
1460
1461         spin_lock(&f->lock);
1462         f->arr[f->num_members] = sk;
1463         smp_wmb();
1464         f->num_members++;
1465         if (f->num_members == 1)
1466                 dev_add_pack(&f->prot_hook);
1467         spin_unlock(&f->lock);
1468 }
1469
1470 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1471 {
1472         struct packet_fanout *f = po->fanout;
1473         int i;
1474
1475         spin_lock(&f->lock);
1476         for (i = 0; i < f->num_members; i++) {
1477                 if (f->arr[i] == sk)
1478                         break;
1479         }
1480         BUG_ON(i >= f->num_members);
1481         f->arr[i] = f->arr[f->num_members - 1];
1482         f->num_members--;
1483         if (f->num_members == 0)
1484                 __dev_remove_pack(&f->prot_hook);
1485         spin_unlock(&f->lock);
1486 }
1487
1488 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1489 {
1490         if (sk->sk_family != PF_PACKET)
1491                 return false;
1492
1493         return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1494 }
1495
1496 static void fanout_init_data(struct packet_fanout *f)
1497 {
1498         switch (f->type) {
1499         case PACKET_FANOUT_LB:
1500                 atomic_set(&f->rr_cur, 0);
1501                 break;
1502         case PACKET_FANOUT_CBPF:
1503         case PACKET_FANOUT_EBPF:
1504                 RCU_INIT_POINTER(f->bpf_prog, NULL);
1505                 break;
1506         }
1507 }
1508
1509 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1510 {
1511         struct bpf_prog *old;
1512
1513         spin_lock(&f->lock);
1514         old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1515         rcu_assign_pointer(f->bpf_prog, new);
1516         spin_unlock(&f->lock);
1517
1518         if (old) {
1519                 synchronize_net();
1520                 bpf_prog_destroy(old);
1521         }
1522 }
1523
1524 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1525                                 unsigned int len)
1526 {
1527         struct bpf_prog *new;
1528         struct sock_fprog fprog;
1529         int ret;
1530
1531         if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1532                 return -EPERM;
1533         if (len != sizeof(fprog))
1534                 return -EINVAL;
1535         if (copy_from_user(&fprog, data, len))
1536                 return -EFAULT;
1537
1538         ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1539         if (ret)
1540                 return ret;
1541
1542         __fanout_set_data_bpf(po->fanout, new);
1543         return 0;
1544 }
1545
1546 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1547                                 unsigned int len)
1548 {
1549         struct bpf_prog *new;
1550         u32 fd;
1551
1552         if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1553                 return -EPERM;
1554         if (len != sizeof(fd))
1555                 return -EINVAL;
1556         if (copy_from_user(&fd, data, len))
1557                 return -EFAULT;
1558
1559         new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1560         if (IS_ERR(new))
1561                 return PTR_ERR(new);
1562
1563         __fanout_set_data_bpf(po->fanout, new);
1564         return 0;
1565 }
1566
1567 static int fanout_set_data(struct packet_sock *po, char __user *data,
1568                            unsigned int len)
1569 {
1570         switch (po->fanout->type) {
1571         case PACKET_FANOUT_CBPF:
1572                 return fanout_set_data_cbpf(po, data, len);
1573         case PACKET_FANOUT_EBPF:
1574                 return fanout_set_data_ebpf(po, data, len);
1575         default:
1576                 return -EINVAL;
1577         }
1578 }
1579
1580 static void fanout_release_data(struct packet_fanout *f)
1581 {
1582         switch (f->type) {
1583         case PACKET_FANOUT_CBPF:
1584         case PACKET_FANOUT_EBPF:
1585                 __fanout_set_data_bpf(f, NULL);
1586         }
1587 }
1588
1589 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1590 {
1591         struct packet_fanout *f;
1592
1593         list_for_each_entry(f, &fanout_list, list) {
1594                 if (f->id == candidate_id &&
1595                     read_pnet(&f->net) == sock_net(sk)) {
1596                         return false;
1597                 }
1598         }
1599         return true;
1600 }
1601
1602 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1603 {
1604         u16 id = fanout_next_id;
1605
1606         do {
1607                 if (__fanout_id_is_free(sk, id)) {
1608                         *new_id = id;
1609                         fanout_next_id = id + 1;
1610                         return true;
1611                 }
1612
1613                 id++;
1614         } while (id != fanout_next_id);
1615
1616         return false;
1617 }
1618
1619 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1620 {
1621         struct packet_rollover *rollover = NULL;
1622         struct packet_sock *po = pkt_sk(sk);
1623         struct packet_fanout *f, *match;
1624         u8 type = type_flags & 0xff;
1625         u8 flags = type_flags >> 8;
1626         int err;
1627
1628         switch (type) {
1629         case PACKET_FANOUT_ROLLOVER:
1630                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1631                         return -EINVAL;
1632         case PACKET_FANOUT_HASH:
1633         case PACKET_FANOUT_LB:
1634         case PACKET_FANOUT_CPU:
1635         case PACKET_FANOUT_RND:
1636         case PACKET_FANOUT_QM:
1637         case PACKET_FANOUT_CBPF:
1638         case PACKET_FANOUT_EBPF:
1639                 break;
1640         default:
1641                 return -EINVAL;
1642         }
1643
1644         mutex_lock(&fanout_mutex);
1645
1646         err = -EALREADY;
1647         if (po->fanout)
1648                 goto out;
1649
1650         if (type == PACKET_FANOUT_ROLLOVER ||
1651             (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1652                 err = -ENOMEM;
1653                 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1654                 if (!rollover)
1655                         goto out;
1656                 atomic_long_set(&rollover->num, 0);
1657                 atomic_long_set(&rollover->num_huge, 0);
1658                 atomic_long_set(&rollover->num_failed, 0);
1659         }
1660
1661         if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1662                 if (id != 0) {
1663                         err = -EINVAL;
1664                         goto out;
1665                 }
1666                 if (!fanout_find_new_id(sk, &id)) {
1667                         err = -ENOMEM;
1668                         goto out;
1669                 }
1670                 /* ephemeral flag for the first socket in the group: drop it */
1671                 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1672         }
1673
1674         match = NULL;
1675         list_for_each_entry(f, &fanout_list, list) {
1676                 if (f->id == id &&
1677                     read_pnet(&f->net) == sock_net(sk)) {
1678                         match = f;
1679                         break;
1680                 }
1681         }
1682         err = -EINVAL;
1683         if (match && match->flags != flags)
1684                 goto out;
1685         if (!match) {
1686                 err = -ENOMEM;
1687                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1688                 if (!match)
1689                         goto out;
1690                 write_pnet(&match->net, sock_net(sk));
1691                 match->id = id;
1692                 match->type = type;
1693                 match->flags = flags;
1694                 INIT_LIST_HEAD(&match->list);
1695                 spin_lock_init(&match->lock);
1696                 refcount_set(&match->sk_ref, 0);
1697                 fanout_init_data(match);
1698                 match->prot_hook.type = po->prot_hook.type;
1699                 match->prot_hook.dev = po->prot_hook.dev;
1700                 match->prot_hook.func = packet_rcv_fanout;
1701                 match->prot_hook.af_packet_priv = match;
1702                 match->prot_hook.id_match = match_fanout_group;
1703                 list_add(&match->list, &fanout_list);
1704         }
1705         err = -EINVAL;
1706
1707         spin_lock(&po->bind_lock);
1708         if (po->running &&
1709             match->type == type &&
1710             match->prot_hook.type == po->prot_hook.type &&
1711             match->prot_hook.dev == po->prot_hook.dev) {
1712                 err = -ENOSPC;
1713                 if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1714                         __dev_remove_pack(&po->prot_hook);
1715                         po->fanout = match;
1716                         po->rollover = rollover;
1717                         rollover = NULL;
1718                         refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1719                         __fanout_link(sk, po);
1720                         err = 0;
1721                 }
1722         }
1723         spin_unlock(&po->bind_lock);
1724
1725         if (err && !refcount_read(&match->sk_ref)) {
1726                 list_del(&match->list);
1727                 kfree(match);
1728         }
1729
1730 out:
1731         kfree(rollover);
1732         mutex_unlock(&fanout_mutex);
1733         return err;
1734 }
1735
1736 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1737  * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1738  * It is the responsibility of the caller to call fanout_release_data() and
1739  * free the returned packet_fanout (after synchronize_net())
1740  */
1741 static struct packet_fanout *fanout_release(struct sock *sk)
1742 {
1743         struct packet_sock *po = pkt_sk(sk);
1744         struct packet_fanout *f;
1745
1746         mutex_lock(&fanout_mutex);
1747         f = po->fanout;
1748         if (f) {
1749                 po->fanout = NULL;
1750
1751                 if (refcount_dec_and_test(&f->sk_ref))
1752                         list_del(&f->list);
1753                 else
1754                         f = NULL;
1755         }
1756         mutex_unlock(&fanout_mutex);
1757
1758         return f;
1759 }
1760
1761 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1762                                           struct sk_buff *skb)
1763 {
1764         /* Earlier code assumed this would be a VLAN pkt, double-check
1765          * this now that we have the actual packet in hand. We can only
1766          * do this check on Ethernet devices.
1767          */
1768         if (unlikely(dev->type != ARPHRD_ETHER))
1769                 return false;
1770
1771         skb_reset_mac_header(skb);
1772         return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1773 }
1774
1775 static const struct proto_ops packet_ops;
1776
1777 static const struct proto_ops packet_ops_spkt;
1778
1779 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1780                            struct packet_type *pt, struct net_device *orig_dev)
1781 {
1782         struct sock *sk;
1783         struct sockaddr_pkt *spkt;
1784
1785         /*
1786          *      When we registered the protocol we saved the socket in the data
1787          *      field for just this event.
1788          */
1789
1790         sk = pt->af_packet_priv;
1791
1792         /*
1793          *      Yank back the headers [hope the device set this
1794          *      right or kerboom...]
1795          *
1796          *      Incoming packets have ll header pulled,
1797          *      push it back.
1798          *
1799          *      For outgoing ones skb->data == skb_mac_header(skb)
1800          *      so that this procedure is noop.
1801          */
1802
1803         if (skb->pkt_type == PACKET_LOOPBACK)
1804                 goto out;
1805
1806         if (!net_eq(dev_net(dev), sock_net(sk)))
1807                 goto out;
1808
1809         skb = skb_share_check(skb, GFP_ATOMIC);
1810         if (skb == NULL)
1811                 goto oom;
1812
1813         /* drop any routing info */
1814         skb_dst_drop(skb);
1815
1816         /* drop conntrack reference */
1817         nf_reset(skb);
1818
1819         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1820
1821         skb_push(skb, skb->data - skb_mac_header(skb));
1822
1823         /*
1824          *      The SOCK_PACKET socket receives _all_ frames.
1825          */
1826
1827         spkt->spkt_family = dev->type;
1828         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1829         spkt->spkt_protocol = skb->protocol;
1830
1831         /*
1832          *      Charge the memory to the socket. This is done specifically
1833          *      to prevent sockets using all the memory up.
1834          */
1835
1836         if (sock_queue_rcv_skb(sk, skb) == 0)
1837                 return 0;
1838
1839 out:
1840         kfree_skb(skb);
1841 oom:
1842         return 0;
1843 }
1844
1845 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1846 {
1847         if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1848             sock->type == SOCK_RAW) {
1849                 skb_reset_mac_header(skb);
1850                 skb->protocol = dev_parse_header_protocol(skb);
1851         }
1852
1853         skb_probe_transport_header(skb);
1854 }
1855
1856 /*
1857  *      Output a raw packet to a device layer. This bypasses all the other
1858  *      protocol layers and you must therefore supply it with a complete frame
1859  */
1860
1861 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1862                                size_t len)
1863 {
1864         struct sock *sk = sock->sk;
1865         DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1866         struct sk_buff *skb = NULL;
1867         struct net_device *dev;
1868         struct sockcm_cookie sockc;
1869         __be16 proto = 0;
1870         int err;
1871         int extra_len = 0;
1872
1873         /*
1874          *      Get and verify the address.
1875          */
1876
1877         if (saddr) {
1878                 if (msg->msg_namelen < sizeof(struct sockaddr))
1879                         return -EINVAL;
1880                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1881                         proto = saddr->spkt_protocol;
1882         } else
1883                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1884
1885         /*
1886          *      Find the device first to size check it
1887          */
1888
1889         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1890 retry:
1891         rcu_read_lock();
1892         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1893         err = -ENODEV;
1894         if (dev == NULL)
1895                 goto out_unlock;
1896
1897         err = -ENETDOWN;
1898         if (!(dev->flags & IFF_UP))
1899                 goto out_unlock;
1900
1901         /*
1902          * You may not queue a frame bigger than the mtu. This is the lowest level
1903          * raw protocol and you must do your own fragmentation at this level.
1904          */
1905
1906         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1907                 if (!netif_supports_nofcs(dev)) {
1908                         err = -EPROTONOSUPPORT;
1909                         goto out_unlock;
1910                 }
1911                 extra_len = 4; /* We're doing our own CRC */
1912         }
1913
1914         err = -EMSGSIZE;
1915         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1916                 goto out_unlock;
1917
1918         if (!skb) {
1919                 size_t reserved = LL_RESERVED_SPACE(dev);
1920                 int tlen = dev->needed_tailroom;
1921                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1922
1923                 rcu_read_unlock();
1924                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1925                 if (skb == NULL)
1926                         return -ENOBUFS;
1927                 /* FIXME: Save some space for broken drivers that write a hard
1928                  * header at transmission time by themselves. PPP is the notable
1929                  * one here. This should really be fixed at the driver level.
1930                  */
1931                 skb_reserve(skb, reserved);
1932                 skb_reset_network_header(skb);
1933
1934                 /* Try to align data part correctly */
1935                 if (hhlen) {
1936                         skb->data -= hhlen;
1937                         skb->tail -= hhlen;
1938                         if (len < hhlen)
1939                                 skb_reset_network_header(skb);
1940                 }
1941                 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1942                 if (err)
1943                         goto out_free;
1944                 goto retry;
1945         }
1946
1947         if (!dev_validate_header(dev, skb->data, len)) {
1948                 err = -EINVAL;
1949                 goto out_unlock;
1950         }
1951         if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1952             !packet_extra_vlan_len_allowed(dev, skb)) {
1953                 err = -EMSGSIZE;
1954                 goto out_unlock;
1955         }
1956
1957         sockcm_init(&sockc, sk);
1958         if (msg->msg_controllen) {
1959                 err = sock_cmsg_send(sk, msg, &sockc);
1960                 if (unlikely(err))
1961                         goto out_unlock;
1962         }
1963
1964         skb->protocol = proto;
1965         skb->dev = dev;
1966         skb->priority = sk->sk_priority;
1967         skb->mark = sk->sk_mark;
1968         skb->tstamp = sockc.transmit_time;
1969
1970         skb_setup_tx_timestamp(skb, sockc.tsflags);
1971
1972         if (unlikely(extra_len == 4))
1973                 skb->no_fcs = 1;
1974
1975         packet_parse_headers(skb, sock);
1976
1977         dev_queue_xmit(skb);
1978         rcu_read_unlock();
1979         return len;
1980
1981 out_unlock:
1982         rcu_read_unlock();
1983 out_free:
1984         kfree_skb(skb);
1985         return err;
1986 }
1987
1988 static unsigned int run_filter(struct sk_buff *skb,
1989                                const struct sock *sk,
1990                                unsigned int res)
1991 {
1992         struct sk_filter *filter;
1993
1994         rcu_read_lock();
1995         filter = rcu_dereference(sk->sk_filter);
1996         if (filter != NULL)
1997                 res = bpf_prog_run_clear_cb(filter->prog, skb);
1998         rcu_read_unlock();
1999
2000         return res;
2001 }
2002
2003 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2004                            size_t *len)
2005 {
2006         struct virtio_net_hdr vnet_hdr;
2007
2008         if (*len < sizeof(vnet_hdr))
2009                 return -EINVAL;
2010         *len -= sizeof(vnet_hdr);
2011
2012         if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2013                 return -EINVAL;
2014
2015         return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2016 }
2017
2018 /*
2019  * This function makes lazy skb cloning in hope that most of packets
2020  * are discarded by BPF.
2021  *
2022  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2023  * and skb->cb are mangled. It works because (and until) packets
2024  * falling here are owned by current CPU. Output packets are cloned
2025  * by dev_queue_xmit_nit(), input packets are processed by net_bh
2026  * sequencially, so that if we return skb to original state on exit,
2027  * we will not harm anyone.
2028  */
2029
2030 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2031                       struct packet_type *pt, struct net_device *orig_dev)
2032 {
2033         struct sock *sk;
2034         struct sockaddr_ll *sll;
2035         struct packet_sock *po;
2036         u8 *skb_head = skb->data;
2037         int skb_len = skb->len;
2038         unsigned int snaplen, res;
2039         bool is_drop_n_account = false;
2040
2041         if (skb->pkt_type == PACKET_LOOPBACK)
2042                 goto drop;
2043
2044         sk = pt->af_packet_priv;
2045         po = pkt_sk(sk);
2046
2047         if (!net_eq(dev_net(dev), sock_net(sk)))
2048                 goto drop;
2049
2050         skb->dev = dev;
2051
2052         if (dev->header_ops) {
2053                 /* The device has an explicit notion of ll header,
2054                  * exported to higher levels.
2055                  *
2056                  * Otherwise, the device hides details of its frame
2057                  * structure, so that corresponding packet head is
2058                  * never delivered to user.
2059                  */
2060                 if (sk->sk_type != SOCK_DGRAM)
2061                         skb_push(skb, skb->data - skb_mac_header(skb));
2062                 else if (skb->pkt_type == PACKET_OUTGOING) {
2063                         /* Special case: outgoing packets have ll header at head */
2064                         skb_pull(skb, skb_network_offset(skb));
2065                 }
2066         }
2067
2068         snaplen = skb->len;
2069
2070         res = run_filter(skb, sk, snaplen);
2071         if (!res)
2072                 goto drop_n_restore;
2073         if (snaplen > res)
2074                 snaplen = res;
2075
2076         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2077                 goto drop_n_acct;
2078
2079         if (skb_shared(skb)) {
2080                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2081                 if (nskb == NULL)
2082                         goto drop_n_acct;
2083
2084                 if (skb_head != skb->data) {
2085                         skb->data = skb_head;
2086                         skb->len = skb_len;
2087                 }
2088                 consume_skb(skb);
2089                 skb = nskb;
2090         }
2091
2092         sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2093
2094         sll = &PACKET_SKB_CB(skb)->sa.ll;
2095         sll->sll_hatype = dev->type;
2096         sll->sll_pkttype = skb->pkt_type;
2097         if (unlikely(po->origdev))
2098                 sll->sll_ifindex = orig_dev->ifindex;
2099         else
2100                 sll->sll_ifindex = dev->ifindex;
2101
2102         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2103
2104         /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2105          * Use their space for storing the original skb length.
2106          */
2107         PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2108
2109         if (pskb_trim(skb, snaplen))
2110                 goto drop_n_acct;
2111
2112         skb_set_owner_r(skb, sk);
2113         skb->dev = NULL;
2114         skb_dst_drop(skb);
2115
2116         /* drop conntrack reference */
2117         nf_reset(skb);
2118
2119         spin_lock(&sk->sk_receive_queue.lock);
2120         po->stats.stats1.tp_packets++;
2121         sock_skb_set_dropcount(sk, skb);
2122         __skb_queue_tail(&sk->sk_receive_queue, skb);
2123         spin_unlock(&sk->sk_receive_queue.lock);
2124         sk->sk_data_ready(sk);
2125         return 0;
2126
2127 drop_n_acct:
2128         is_drop_n_account = true;
2129         atomic_inc(&po->tp_drops);
2130         atomic_inc(&sk->sk_drops);
2131
2132 drop_n_restore:
2133         if (skb_head != skb->data && skb_shared(skb)) {
2134                 skb->data = skb_head;
2135                 skb->len = skb_len;
2136         }
2137 drop:
2138         if (!is_drop_n_account)
2139                 consume_skb(skb);
2140         else
2141                 kfree_skb(skb);
2142         return 0;
2143 }
2144
2145 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2146                        struct packet_type *pt, struct net_device *orig_dev)
2147 {
2148         struct sock *sk;
2149         struct packet_sock *po;
2150         struct sockaddr_ll *sll;
2151         union tpacket_uhdr h;
2152         u8 *skb_head = skb->data;
2153         int skb_len = skb->len;
2154         unsigned int snaplen, res;
2155         unsigned long status = TP_STATUS_USER;
2156         unsigned short macoff, netoff, hdrlen;
2157         struct sk_buff *copy_skb = NULL;
2158         struct timespec ts;
2159         __u32 ts_status;
2160         bool is_drop_n_account = false;
2161         bool do_vnet = false;
2162
2163         /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2164          * We may add members to them until current aligned size without forcing
2165          * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2166          */
2167         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2168         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2169
2170         if (skb->pkt_type == PACKET_LOOPBACK)
2171                 goto drop;
2172
2173         sk = pt->af_packet_priv;
2174         po = pkt_sk(sk);
2175
2176         if (!net_eq(dev_net(dev), sock_net(sk)))
2177                 goto drop;
2178
2179         if (dev->header_ops) {
2180                 if (sk->sk_type != SOCK_DGRAM)
2181                         skb_push(skb, skb->data - skb_mac_header(skb));
2182                 else if (skb->pkt_type == PACKET_OUTGOING) {
2183                         /* Special case: outgoing packets have ll header at head */
2184                         skb_pull(skb, skb_network_offset(skb));
2185                 }
2186         }
2187
2188         snaplen = skb->len;
2189
2190         res = run_filter(skb, sk, snaplen);
2191         if (!res)
2192                 goto drop_n_restore;
2193
2194         /* If we are flooded, just give up */
2195         if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2196                 atomic_inc(&po->tp_drops);
2197                 goto drop_n_restore;
2198         }
2199
2200         if (skb->ip_summed == CHECKSUM_PARTIAL)
2201                 status |= TP_STATUS_CSUMNOTREADY;
2202         else if (skb->pkt_type != PACKET_OUTGOING &&
2203                  (skb->ip_summed == CHECKSUM_COMPLETE ||
2204                   skb_csum_unnecessary(skb)))
2205                 status |= TP_STATUS_CSUM_VALID;
2206
2207         if (snaplen > res)
2208                 snaplen = res;
2209
2210         if (sk->sk_type == SOCK_DGRAM) {
2211                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2212                                   po->tp_reserve;
2213         } else {
2214                 unsigned int maclen = skb_network_offset(skb);
2215                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2216                                        (maclen < 16 ? 16 : maclen)) +
2217                                        po->tp_reserve;
2218                 if (po->has_vnet_hdr) {
2219                         netoff += sizeof(struct virtio_net_hdr);
2220                         do_vnet = true;
2221                 }
2222                 macoff = netoff - maclen;
2223         }
2224         if (po->tp_version <= TPACKET_V2) {
2225                 if (macoff + snaplen > po->rx_ring.frame_size) {
2226                         if (po->copy_thresh &&
2227                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2228                                 if (skb_shared(skb)) {
2229                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
2230                                 } else {
2231                                         copy_skb = skb_get(skb);
2232                                         skb_head = skb->data;
2233                                 }
2234                                 if (copy_skb)
2235                                         skb_set_owner_r(copy_skb, sk);
2236                         }
2237                         snaplen = po->rx_ring.frame_size - macoff;
2238                         if ((int)snaplen < 0) {
2239                                 snaplen = 0;
2240                                 do_vnet = false;
2241                         }
2242                 }
2243         } else if (unlikely(macoff + snaplen >
2244                             GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2245                 u32 nval;
2246
2247                 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2248                 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2249                             snaplen, nval, macoff);
2250                 snaplen = nval;
2251                 if (unlikely((int)snaplen < 0)) {
2252                         snaplen = 0;
2253                         macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2254                         do_vnet = false;
2255                 }
2256         }
2257         spin_lock(&sk->sk_receive_queue.lock);
2258         h.raw = packet_current_rx_frame(po, skb,
2259                                         TP_STATUS_KERNEL, (macoff+snaplen));
2260         if (!h.raw)
2261                 goto drop_n_account;
2262         if (po->tp_version <= TPACKET_V2) {
2263                 packet_increment_rx_head(po, &po->rx_ring);
2264         /*
2265          * LOSING will be reported till you read the stats,
2266          * because it's COR - Clear On Read.
2267          * Anyways, moving it for V1/V2 only as V3 doesn't need this
2268          * at packet level.
2269          */
2270                 if (atomic_read(&po->tp_drops))
2271                         status |= TP_STATUS_LOSING;
2272         }
2273
2274         if (do_vnet &&
2275             virtio_net_hdr_from_skb(skb, h.raw + macoff -
2276                                     sizeof(struct virtio_net_hdr),
2277                                     vio_le(), true, 0))
2278                 goto drop_n_account;
2279
2280         po->stats.stats1.tp_packets++;
2281         if (copy_skb) {
2282                 status |= TP_STATUS_COPY;
2283                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2284         }
2285         spin_unlock(&sk->sk_receive_queue.lock);
2286
2287         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2288
2289         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2290                 getnstimeofday(&ts);
2291
2292         status |= ts_status;
2293
2294         switch (po->tp_version) {
2295         case TPACKET_V1:
2296                 h.h1->tp_len = skb->len;
2297                 h.h1->tp_snaplen = snaplen;
2298                 h.h1->tp_mac = macoff;
2299                 h.h1->tp_net = netoff;
2300                 h.h1->tp_sec = ts.tv_sec;
2301                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2302                 hdrlen = sizeof(*h.h1);
2303                 break;
2304         case TPACKET_V2:
2305                 h.h2->tp_len = skb->len;
2306                 h.h2->tp_snaplen = snaplen;
2307                 h.h2->tp_mac = macoff;
2308                 h.h2->tp_net = netoff;
2309                 h.h2->tp_sec = ts.tv_sec;
2310                 h.h2->tp_nsec = ts.tv_nsec;
2311                 if (skb_vlan_tag_present(skb)) {
2312                         h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2313                         h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2314                         status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2315                 } else {
2316                         h.h2->tp_vlan_tci = 0;
2317                         h.h2->tp_vlan_tpid = 0;
2318                 }
2319                 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2320                 hdrlen = sizeof(*h.h2);
2321                 break;
2322         case TPACKET_V3:
2323                 /* tp_nxt_offset,vlan are already populated above.
2324                  * So DONT clear those fields here
2325                  */
2326                 h.h3->tp_status |= status;
2327                 h.h3->tp_len = skb->len;
2328                 h.h3->tp_snaplen = snaplen;
2329                 h.h3->tp_mac = macoff;
2330                 h.h3->tp_net = netoff;
2331                 h.h3->tp_sec  = ts.tv_sec;
2332                 h.h3->tp_nsec = ts.tv_nsec;
2333                 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2334                 hdrlen = sizeof(*h.h3);
2335                 break;
2336         default:
2337                 BUG();
2338         }
2339
2340         sll = h.raw + TPACKET_ALIGN(hdrlen);
2341         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2342         sll->sll_family = AF_PACKET;
2343         sll->sll_hatype = dev->type;
2344         sll->sll_protocol = skb->protocol;
2345         sll->sll_pkttype = skb->pkt_type;
2346         if (unlikely(po->origdev))
2347                 sll->sll_ifindex = orig_dev->ifindex;
2348         else
2349                 sll->sll_ifindex = dev->ifindex;
2350
2351         smp_mb();
2352
2353 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2354         if (po->tp_version <= TPACKET_V2) {
2355                 u8 *start, *end;
2356
2357                 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2358                                         macoff + snaplen);
2359
2360                 for (start = h.raw; start < end; start += PAGE_SIZE)
2361                         flush_dcache_page(pgv_to_page(start));
2362         }
2363         smp_wmb();
2364 #endif
2365
2366         if (po->tp_version <= TPACKET_V2) {
2367                 __packet_set_status(po, h.raw, status);
2368                 sk->sk_data_ready(sk);
2369         } else {
2370                 prb_clear_blk_fill_status(&po->rx_ring);
2371         }
2372
2373 drop_n_restore:
2374         if (skb_head != skb->data && skb_shared(skb)) {
2375                 skb->data = skb_head;
2376                 skb->len = skb_len;
2377         }
2378 drop:
2379         if (!is_drop_n_account)
2380                 consume_skb(skb);
2381         else
2382                 kfree_skb(skb);
2383         return 0;
2384
2385 drop_n_account:
2386         spin_unlock(&sk->sk_receive_queue.lock);
2387         atomic_inc(&po->tp_drops);
2388         is_drop_n_account = true;
2389
2390         sk->sk_data_ready(sk);
2391         kfree_skb(copy_skb);
2392         goto drop_n_restore;
2393 }
2394
2395 static void tpacket_destruct_skb(struct sk_buff *skb)
2396 {
2397         struct packet_sock *po = pkt_sk(skb->sk);
2398
2399         if (likely(po->tx_ring.pg_vec)) {
2400                 void *ph;
2401                 __u32 ts;
2402
2403                 ph = skb_zcopy_get_nouarg(skb);
2404                 packet_dec_pending(&po->tx_ring);
2405
2406                 ts = __packet_set_timestamp(po, ph, skb);
2407                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2408         }
2409
2410         sock_wfree(skb);
2411 }
2412
2413 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2414 {
2415         if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2416             (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2417              __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2418               __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2419                 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2420                          __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2421                         __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2422
2423         if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2424                 return -EINVAL;
2425
2426         return 0;
2427 }
2428
2429 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2430                                  struct virtio_net_hdr *vnet_hdr)
2431 {
2432         if (*len < sizeof(*vnet_hdr))
2433                 return -EINVAL;
2434         *len -= sizeof(*vnet_hdr);
2435
2436         if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2437                 return -EFAULT;
2438
2439         return __packet_snd_vnet_parse(vnet_hdr, *len);
2440 }
2441
2442 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2443                 void *frame, struct net_device *dev, void *data, int tp_len,
2444                 __be16 proto, unsigned char *addr, int hlen, int copylen,
2445                 const struct sockcm_cookie *sockc)
2446 {
2447         union tpacket_uhdr ph;
2448         int to_write, offset, len, nr_frags, len_max;
2449         struct socket *sock = po->sk.sk_socket;
2450         struct page *page;
2451         int err;
2452
2453         ph.raw = frame;
2454
2455         skb->protocol = proto;
2456         skb->dev = dev;
2457         skb->priority = po->sk.sk_priority;
2458         skb->mark = po->sk.sk_mark;
2459         skb->tstamp = sockc->transmit_time;
2460         skb_setup_tx_timestamp(skb, sockc->tsflags);
2461         skb_zcopy_set_nouarg(skb, ph.raw);
2462
2463         skb_reserve(skb, hlen);
2464         skb_reset_network_header(skb);
2465
2466         to_write = tp_len;
2467
2468         if (sock->type == SOCK_DGRAM) {
2469                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2470                                 NULL, tp_len);
2471                 if (unlikely(err < 0))
2472                         return -EINVAL;
2473         } else if (copylen) {
2474                 int hdrlen = min_t(int, copylen, tp_len);
2475
2476                 skb_push(skb, dev->hard_header_len);
2477                 skb_put(skb, copylen - dev->hard_header_len);
2478                 err = skb_store_bits(skb, 0, data, hdrlen);
2479                 if (unlikely(err))
2480                         return err;
2481                 if (!dev_validate_header(dev, skb->data, hdrlen))
2482                         return -EINVAL;
2483
2484                 data += hdrlen;
2485                 to_write -= hdrlen;
2486         }
2487
2488         offset = offset_in_page(data);
2489         len_max = PAGE_SIZE - offset;
2490         len = ((to_write > len_max) ? len_max : to_write);
2491
2492         skb->data_len = to_write;
2493         skb->len += to_write;
2494         skb->truesize += to_write;
2495         refcount_add(to_write, &po->sk.sk_wmem_alloc);
2496
2497         while (likely(to_write)) {
2498                 nr_frags = skb_shinfo(skb)->nr_frags;
2499
2500                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2501                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2502                                MAX_SKB_FRAGS);
2503                         return -EFAULT;
2504                 }
2505
2506                 page = pgv_to_page(data);
2507                 data += len;
2508                 flush_dcache_page(page);
2509                 get_page(page);
2510                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2511                 to_write -= len;
2512                 offset = 0;
2513                 len_max = PAGE_SIZE;
2514                 len = ((to_write > len_max) ? len_max : to_write);
2515         }
2516
2517         packet_parse_headers(skb, sock);
2518
2519         return tp_len;
2520 }
2521
2522 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2523                                 int size_max, void **data)
2524 {
2525         union tpacket_uhdr ph;
2526         int tp_len, off;
2527
2528         ph.raw = frame;
2529
2530         switch (po->tp_version) {
2531         case TPACKET_V3:
2532                 if (ph.h3->tp_next_offset != 0) {
2533                         pr_warn_once("variable sized slot not supported");
2534                         return -EINVAL;
2535                 }
2536                 tp_len = ph.h3->tp_len;
2537                 break;
2538         case TPACKET_V2:
2539                 tp_len = ph.h2->tp_len;
2540                 break;
2541         default:
2542                 tp_len = ph.h1->tp_len;
2543                 break;
2544         }
2545         if (unlikely(tp_len > size_max)) {
2546                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2547                 return -EMSGSIZE;
2548         }
2549
2550         if (unlikely(po->tp_tx_has_off)) {
2551                 int off_min, off_max;
2552
2553                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2554                 off_max = po->tx_ring.frame_size - tp_len;
2555                 if (po->sk.sk_type == SOCK_DGRAM) {
2556                         switch (po->tp_version) {
2557                         case TPACKET_V3:
2558                                 off = ph.h3->tp_net;
2559                                 break;
2560                         case TPACKET_V2:
2561                                 off = ph.h2->tp_net;
2562                                 break;
2563                         default:
2564                                 off = ph.h1->tp_net;
2565                                 break;
2566                         }
2567                 } else {
2568                         switch (po->tp_version) {
2569                         case TPACKET_V3:
2570                                 off = ph.h3->tp_mac;
2571                                 break;
2572                         case TPACKET_V2:
2573                                 off = ph.h2->tp_mac;
2574                                 break;
2575                         default:
2576                                 off = ph.h1->tp_mac;
2577                                 break;
2578                         }
2579                 }
2580                 if (unlikely((off < off_min) || (off_max < off)))
2581                         return -EINVAL;
2582         } else {
2583                 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2584         }
2585
2586         *data = frame + off;
2587         return tp_len;
2588 }
2589
2590 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2591 {
2592         struct sk_buff *skb;
2593         struct net_device *dev;
2594         struct virtio_net_hdr *vnet_hdr = NULL;
2595         struct sockcm_cookie sockc;
2596         __be16 proto;
2597         int err, reserve = 0;
2598         void *ph;
2599         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2600         bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2601         unsigned char *addr = NULL;
2602         int tp_len, size_max;
2603         void *data;
2604         int len_sum = 0;
2605         int status = TP_STATUS_AVAILABLE;
2606         int hlen, tlen, copylen = 0;
2607
2608         mutex_lock(&po->pg_vec_lock);
2609
2610         if (likely(saddr == NULL)) {
2611                 dev     = packet_cached_dev_get(po);
2612                 proto   = po->num;
2613         } else {
2614                 err = -EINVAL;
2615                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2616                         goto out;
2617                 if (msg->msg_namelen < (saddr->sll_halen
2618                                         + offsetof(struct sockaddr_ll,
2619                                                 sll_addr)))
2620                         goto out;
2621                 proto   = saddr->sll_protocol;
2622                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2623                 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2624                         if (dev && msg->msg_namelen < dev->addr_len +
2625                                    offsetof(struct sockaddr_ll, sll_addr))
2626                                 goto out_put;
2627                         addr = saddr->sll_addr;
2628                 }
2629         }
2630
2631         err = -ENXIO;
2632         if (unlikely(dev == NULL))
2633                 goto out;
2634         err = -ENETDOWN;
2635         if (unlikely(!(dev->flags & IFF_UP)))
2636                 goto out_put;
2637
2638         sockcm_init(&sockc, &po->sk);
2639         if (msg->msg_controllen) {
2640                 err = sock_cmsg_send(&po->sk, msg, &sockc);
2641                 if (unlikely(err))
2642                         goto out_put;
2643         }
2644
2645         if (po->sk.sk_socket->type == SOCK_RAW)
2646                 reserve = dev->hard_header_len;
2647         size_max = po->tx_ring.frame_size
2648                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2649
2650         if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2651                 size_max = dev->mtu + reserve + VLAN_HLEN;
2652
2653         do {
2654                 ph = packet_current_frame(po, &po->tx_ring,
2655                                           TP_STATUS_SEND_REQUEST);
2656                 if (unlikely(ph == NULL)) {
2657                         if (need_wait && need_resched())
2658                                 schedule();
2659                         continue;
2660                 }
2661
2662                 skb = NULL;
2663                 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2664                 if (tp_len < 0)
2665                         goto tpacket_error;
2666
2667                 status = TP_STATUS_SEND_REQUEST;
2668                 hlen = LL_RESERVED_SPACE(dev);
2669                 tlen = dev->needed_tailroom;
2670                 if (po->has_vnet_hdr) {
2671                         vnet_hdr = data;
2672                         data += sizeof(*vnet_hdr);
2673                         tp_len -= sizeof(*vnet_hdr);
2674                         if (tp_len < 0 ||
2675                             __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2676                                 tp_len = -EINVAL;
2677                                 goto tpacket_error;
2678                         }
2679                         copylen = __virtio16_to_cpu(vio_le(),
2680                                                     vnet_hdr->hdr_len);
2681                 }
2682                 copylen = max_t(int, copylen, dev->hard_header_len);
2683                 skb = sock_alloc_send_skb(&po->sk,
2684                                 hlen + tlen + sizeof(struct sockaddr_ll) +
2685                                 (copylen - dev->hard_header_len),
2686                                 !need_wait, &err);
2687
2688                 if (unlikely(skb == NULL)) {
2689                         /* we assume the socket was initially writeable ... */
2690                         if (likely(len_sum > 0))
2691                                 err = len_sum;
2692                         goto out_status;
2693                 }
2694                 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2695                                           addr, hlen, copylen, &sockc);
2696                 if (likely(tp_len >= 0) &&
2697                     tp_len > dev->mtu + reserve &&
2698                     !po->has_vnet_hdr &&
2699                     !packet_extra_vlan_len_allowed(dev, skb))
2700                         tp_len = -EMSGSIZE;
2701
2702                 if (unlikely(tp_len < 0)) {
2703 tpacket_error:
2704                         if (po->tp_loss) {
2705                                 __packet_set_status(po, ph,
2706                                                 TP_STATUS_AVAILABLE);
2707                                 packet_increment_head(&po->tx_ring);
2708                                 kfree_skb(skb);
2709                                 continue;
2710                         } else {
2711                                 status = TP_STATUS_WRONG_FORMAT;
2712                                 err = tp_len;
2713                                 goto out_status;
2714                         }
2715                 }
2716
2717                 if (po->has_vnet_hdr) {
2718                         if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2719                                 tp_len = -EINVAL;
2720                                 goto tpacket_error;
2721                         }
2722                         virtio_net_hdr_set_proto(skb, vnet_hdr);
2723                 }
2724
2725                 skb->destructor = tpacket_destruct_skb;
2726                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2727                 packet_inc_pending(&po->tx_ring);
2728
2729                 status = TP_STATUS_SEND_REQUEST;
2730                 err = po->xmit(skb);
2731                 if (unlikely(err > 0)) {
2732                         err = net_xmit_errno(err);
2733                         if (err && __packet_get_status(po, ph) ==
2734                                    TP_STATUS_AVAILABLE) {
2735                                 /* skb was destructed already */
2736                                 skb = NULL;
2737                                 goto out_status;
2738                         }
2739                         /*
2740                          * skb was dropped but not destructed yet;
2741                          * let's treat it like congestion or err < 0
2742                          */
2743                         err = 0;
2744                 }
2745                 packet_increment_head(&po->tx_ring);
2746                 len_sum += tp_len;
2747         } while (likely((ph != NULL) ||
2748                 /* Note: packet_read_pending() might be slow if we have
2749                  * to call it as it's per_cpu variable, but in fast-path
2750                  * we already short-circuit the loop with the first
2751                  * condition, and luckily don't have to go that path
2752                  * anyway.
2753                  */
2754                  (need_wait && packet_read_pending(&po->tx_ring))));
2755
2756         err = len_sum;
2757         goto out_put;
2758
2759 out_status:
2760         __packet_set_status(po, ph, status);
2761         kfree_skb(skb);
2762 out_put:
2763         dev_put(dev);
2764 out:
2765         mutex_unlock(&po->pg_vec_lock);
2766         return err;
2767 }
2768
2769 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2770                                         size_t reserve, size_t len,
2771                                         size_t linear, int noblock,
2772                                         int *err)
2773 {
2774         struct sk_buff *skb;
2775
2776         /* Under a page?  Don't bother with paged skb. */
2777         if (prepad + len < PAGE_SIZE || !linear)
2778                 linear = len;
2779
2780         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2781                                    err, 0);
2782         if (!skb)
2783                 return NULL;
2784
2785         skb_reserve(skb, reserve);
2786         skb_put(skb, linear);
2787         skb->data_len = len - linear;
2788         skb->len += len - linear;
2789
2790         return skb;
2791 }
2792
2793 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2794 {
2795         struct sock *sk = sock->sk;
2796         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2797         struct sk_buff *skb;
2798         struct net_device *dev;
2799         __be16 proto;
2800         unsigned char *addr = NULL;
2801         int err, reserve = 0;
2802         struct sockcm_cookie sockc;
2803         struct virtio_net_hdr vnet_hdr = { 0 };
2804         int offset = 0;
2805         struct packet_sock *po = pkt_sk(sk);
2806         bool has_vnet_hdr = false;
2807         int hlen, tlen, linear;
2808         int extra_len = 0;
2809
2810         /*
2811          *      Get and verify the address.
2812          */
2813
2814         if (likely(saddr == NULL)) {
2815                 dev     = packet_cached_dev_get(po);
2816                 proto   = po->num;
2817         } else {
2818                 err = -EINVAL;
2819                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2820                         goto out;
2821                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2822                         goto out;
2823                 proto   = saddr->sll_protocol;
2824                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2825                 if (sock->type == SOCK_DGRAM) {
2826                         if (dev && msg->msg_namelen < dev->addr_len +
2827                                    offsetof(struct sockaddr_ll, sll_addr))
2828                                 goto out_unlock;
2829                         addr = saddr->sll_addr;
2830                 }
2831         }
2832
2833         err = -ENXIO;
2834         if (unlikely(dev == NULL))
2835                 goto out_unlock;
2836         err = -ENETDOWN;
2837         if (unlikely(!(dev->flags & IFF_UP)))
2838                 goto out_unlock;
2839
2840         sockcm_init(&sockc, sk);
2841         sockc.mark = sk->sk_mark;
2842         if (msg->msg_controllen) {
2843                 err = sock_cmsg_send(sk, msg, &sockc);
2844                 if (unlikely(err))
2845                         goto out_unlock;
2846         }
2847
2848         if (sock->type == SOCK_RAW)
2849                 reserve = dev->hard_header_len;
2850         if (po->has_vnet_hdr) {
2851                 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2852                 if (err)
2853                         goto out_unlock;
2854                 has_vnet_hdr = true;
2855         }
2856
2857         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2858                 if (!netif_supports_nofcs(dev)) {
2859                         err = -EPROTONOSUPPORT;
2860                         goto out_unlock;
2861                 }
2862                 extra_len = 4; /* We're doing our own CRC */
2863         }
2864
2865         err = -EMSGSIZE;
2866         if (!vnet_hdr.gso_type &&
2867             (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2868                 goto out_unlock;
2869
2870         err = -ENOBUFS;
2871         hlen = LL_RESERVED_SPACE(dev);
2872         tlen = dev->needed_tailroom;
2873         linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2874         linear = max(linear, min_t(int, len, dev->hard_header_len));
2875         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2876                                msg->msg_flags & MSG_DONTWAIT, &err);
2877         if (skb == NULL)
2878                 goto out_unlock;
2879
2880         skb_reset_network_header(skb);
2881
2882         err = -EINVAL;
2883         if (sock->type == SOCK_DGRAM) {
2884                 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2885                 if (unlikely(offset < 0))
2886                         goto out_free;
2887         } else if (reserve) {
2888                 skb_reserve(skb, -reserve);
2889                 if (len < reserve + sizeof(struct ipv6hdr) &&
2890                     dev->min_header_len != dev->hard_header_len)
2891                         skb_reset_network_header(skb);
2892         }
2893
2894         /* Returns -EFAULT on error */
2895         err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2896         if (err)
2897                 goto out_free;
2898
2899         if (sock->type == SOCK_RAW &&
2900             !dev_validate_header(dev, skb->data, len)) {
2901                 err = -EINVAL;
2902                 goto out_free;
2903         }
2904
2905         skb_setup_tx_timestamp(skb, sockc.tsflags);
2906
2907         if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2908             !packet_extra_vlan_len_allowed(dev, skb)) {
2909                 err = -EMSGSIZE;
2910                 goto out_free;
2911         }
2912
2913         skb->protocol = proto;
2914         skb->dev = dev;
2915         skb->priority = sk->sk_priority;
2916         skb->mark = sockc.mark;
2917         skb->tstamp = sockc.transmit_time;
2918
2919         if (has_vnet_hdr) {
2920                 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2921                 if (err)
2922                         goto out_free;
2923                 len += sizeof(vnet_hdr);
2924                 virtio_net_hdr_set_proto(skb, &vnet_hdr);
2925         }
2926
2927         packet_parse_headers(skb, sock);
2928
2929         if (unlikely(extra_len == 4))
2930                 skb->no_fcs = 1;
2931
2932         err = po->xmit(skb);
2933         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2934                 goto out_unlock;
2935
2936         dev_put(dev);
2937
2938         return len;
2939
2940 out_free:
2941         kfree_skb(skb);
2942 out_unlock:
2943         if (dev)
2944                 dev_put(dev);
2945 out:
2946         return err;
2947 }
2948
2949 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2950 {
2951         struct sock *sk = sock->sk;
2952         struct packet_sock *po = pkt_sk(sk);
2953
2954         if (po->tx_ring.pg_vec)
2955                 return tpacket_snd(po, msg);
2956         else
2957                 return packet_snd(sock, msg, len);
2958 }
2959
2960 /*
2961  *      Close a PACKET socket. This is fairly simple. We immediately go
2962  *      to 'closed' state and remove our protocol entry in the device list.
2963  */
2964
2965 static int packet_release(struct socket *sock)
2966 {
2967         struct sock *sk = sock->sk;
2968         struct packet_sock *po;
2969         struct packet_fanout *f;
2970         struct net *net;
2971         union tpacket_req_u req_u;
2972
2973         if (!sk)
2974                 return 0;
2975
2976         net = sock_net(sk);
2977         po = pkt_sk(sk);
2978
2979         mutex_lock(&net->packet.sklist_lock);
2980         sk_del_node_init_rcu(sk);
2981         mutex_unlock(&net->packet.sklist_lock);
2982
2983         preempt_disable();
2984         sock_prot_inuse_add(net, sk->sk_prot, -1);
2985         preempt_enable();
2986
2987         spin_lock(&po->bind_lock);
2988         unregister_prot_hook(sk, false);
2989         packet_cached_dev_reset(po);
2990
2991         if (po->prot_hook.dev) {
2992                 dev_put(po->prot_hook.dev);
2993                 po->prot_hook.dev = NULL;
2994         }
2995         spin_unlock(&po->bind_lock);
2996
2997         packet_flush_mclist(sk);
2998
2999         lock_sock(sk);
3000         if (po->rx_ring.pg_vec) {
3001                 memset(&req_u, 0, sizeof(req_u));
3002                 packet_set_ring(sk, &req_u, 1, 0);
3003         }
3004
3005         if (po->tx_ring.pg_vec) {
3006                 memset(&req_u, 0, sizeof(req_u));
3007                 packet_set_ring(sk, &req_u, 1, 1);
3008         }
3009         release_sock(sk);
3010
3011         f = fanout_release(sk);
3012
3013         synchronize_net();
3014
3015         kfree(po->rollover);
3016         if (f) {
3017                 fanout_release_data(f);
3018                 kfree(f);
3019         }
3020         /*
3021          *      Now the socket is dead. No more input will appear.
3022          */
3023         sock_orphan(sk);
3024         sock->sk = NULL;
3025
3026         /* Purge queues */
3027
3028         skb_queue_purge(&sk->sk_receive_queue);
3029         packet_free_pending(po);
3030         sk_refcnt_debug_release(sk);
3031
3032         sock_put(sk);
3033         return 0;
3034 }
3035
3036 /*
3037  *      Attach a packet hook.
3038  */
3039
3040 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3041                           __be16 proto)
3042 {
3043         struct packet_sock *po = pkt_sk(sk);
3044         struct net_device *dev_curr;
3045         __be16 proto_curr;
3046         bool need_rehook;
3047         struct net_device *dev = NULL;
3048         int ret = 0;
3049         bool unlisted = false;
3050
3051         lock_sock(sk);
3052         spin_lock(&po->bind_lock);
3053         rcu_read_lock();
3054
3055         if (po->fanout) {
3056                 ret = -EINVAL;
3057                 goto out_unlock;
3058         }
3059
3060         if (name) {
3061                 dev = dev_get_by_name_rcu(sock_net(sk), name);
3062                 if (!dev) {
3063                         ret = -ENODEV;
3064                         goto out_unlock;
3065                 }
3066         } else if (ifindex) {
3067                 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3068                 if (!dev) {
3069                         ret = -ENODEV;
3070                         goto out_unlock;
3071                 }
3072         }
3073
3074         if (dev)
3075                 dev_hold(dev);
3076
3077         proto_curr = po->prot_hook.type;
3078         dev_curr = po->prot_hook.dev;
3079
3080         need_rehook = proto_curr != proto || dev_curr != dev;
3081
3082         if (need_rehook) {
3083                 if (po->running) {
3084                         rcu_read_unlock();
3085                         /* prevents packet_notifier() from calling
3086                          * register_prot_hook()
3087                          */
3088                         po->num = 0;
3089                         __unregister_prot_hook(sk, true);
3090                         rcu_read_lock();
3091                         dev_curr = po->prot_hook.dev;
3092                         if (dev)
3093                                 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3094                                                                  dev->ifindex);
3095                 }
3096
3097                 BUG_ON(po->running);
3098                 po->num = proto;
3099                 po->prot_hook.type = proto;
3100
3101                 if (unlikely(unlisted)) {
3102                         dev_put(dev);
3103                         po->prot_hook.dev = NULL;
3104                         po->ifindex = -1;
3105                         packet_cached_dev_reset(po);
3106                 } else {
3107                         po->prot_hook.dev = dev;
3108                         po->ifindex = dev ? dev->ifindex : 0;
3109                         packet_cached_dev_assign(po, dev);
3110                 }
3111         }
3112         if (dev_curr)
3113                 dev_put(dev_curr);
3114
3115         if (proto == 0 || !need_rehook)
3116                 goto out_unlock;
3117
3118         if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3119                 register_prot_hook(sk);
3120         } else {
3121                 sk->sk_err = ENETDOWN;
3122                 if (!sock_flag(sk, SOCK_DEAD))
3123                         sk->sk_error_report(sk);
3124         }
3125
3126 out_unlock:
3127         rcu_read_unlock();
3128         spin_unlock(&po->bind_lock);
3129         release_sock(sk);
3130         return ret;
3131 }
3132
3133 /*
3134  *      Bind a packet socket to a device
3135  */
3136
3137 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3138                             int addr_len)
3139 {
3140         struct sock *sk = sock->sk;
3141         char name[sizeof(uaddr->sa_data) + 1];
3142
3143         /*
3144          *      Check legality
3145          */
3146
3147         if (addr_len != sizeof(struct sockaddr))
3148                 return -EINVAL;
3149         /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3150          * zero-terminated.
3151          */
3152         memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3153         name[sizeof(uaddr->sa_data)] = 0;
3154
3155         return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3156 }
3157
3158 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3159 {
3160         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3161         struct sock *sk = sock->sk;
3162
3163         /*
3164          *      Check legality
3165          */
3166
3167         if (addr_len < sizeof(struct sockaddr_ll))
3168                 return -EINVAL;
3169         if (sll->sll_family != AF_PACKET)
3170                 return -EINVAL;
3171
3172         return packet_do_bind(sk, NULL, sll->sll_ifindex,
3173                               sll->sll_protocol ? : pkt_sk(sk)->num);
3174 }
3175
3176 static struct proto packet_proto = {
3177         .name     = "PACKET",
3178         .owner    = THIS_MODULE,
3179         .obj_size = sizeof(struct packet_sock),
3180 };
3181
3182 /*
3183  *      Create a packet of type SOCK_PACKET.
3184  */
3185
3186 static int packet_create(struct net *net, struct socket *sock, int protocol,
3187                          int kern)
3188 {
3189         struct sock *sk;
3190         struct packet_sock *po;
3191         __be16 proto = (__force __be16)protocol; /* weird, but documented */
3192         int err;
3193
3194         if (!ns_capable(net->user_ns, CAP_NET_RAW))
3195                 return -EPERM;
3196         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3197             sock->type != SOCK_PACKET)
3198                 return -ESOCKTNOSUPPORT;
3199
3200         sock->state = SS_UNCONNECTED;
3201
3202         err = -ENOBUFS;
3203         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3204         if (sk == NULL)
3205                 goto out;
3206
3207         sock->ops = &packet_ops;
3208         if (sock->type == SOCK_PACKET)
3209                 sock->ops = &packet_ops_spkt;
3210
3211         sock_init_data(sock, sk);
3212
3213         po = pkt_sk(sk);
3214         sk->sk_family = PF_PACKET;
3215         po->num = proto;
3216         po->xmit = dev_queue_xmit;
3217
3218         err = packet_alloc_pending(po);
3219         if (err)
3220                 goto out2;
3221
3222         packet_cached_dev_reset(po);
3223
3224         sk->sk_destruct = packet_sock_destruct;
3225         sk_refcnt_debug_inc(sk);
3226
3227         /*
3228          *      Attach a protocol block
3229          */
3230
3231         spin_lock_init(&po->bind_lock);
3232         mutex_init(&po->pg_vec_lock);
3233         po->rollover = NULL;
3234         po->prot_hook.func = packet_rcv;
3235
3236         if (sock->type == SOCK_PACKET)
3237                 po->prot_hook.func = packet_rcv_spkt;
3238
3239         po->prot_hook.af_packet_priv = sk;
3240
3241         if (proto) {
3242                 po->prot_hook.type = proto;
3243                 __register_prot_hook(sk);
3244         }
3245
3246         mutex_lock(&net->packet.sklist_lock);
3247         sk_add_node_tail_rcu(sk, &net->packet.sklist);
3248         mutex_unlock(&net->packet.sklist_lock);
3249
3250         preempt_disable();
3251         sock_prot_inuse_add(net, &packet_proto, 1);
3252         preempt_enable();
3253
3254         return 0;
3255 out2:
3256         sk_free(sk);
3257 out:
3258         return err;
3259 }
3260
3261 /*
3262  *      Pull a packet from our receive queue and hand it to the user.
3263  *      If necessary we block.
3264  */
3265
3266 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3267                           int flags)
3268 {
3269         struct sock *sk = sock->sk;
3270         struct sk_buff *skb;
3271         int copied, err;
3272         int vnet_hdr_len = 0;
3273         unsigned int origlen = 0;
3274
3275         err = -EINVAL;
3276         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3277                 goto out;
3278
3279 #if 0
3280         /* What error should we return now? EUNATTACH? */
3281         if (pkt_sk(sk)->ifindex < 0)
3282                 return -ENODEV;
3283 #endif
3284
3285         if (flags & MSG_ERRQUEUE) {
3286                 err = sock_recv_errqueue(sk, msg, len,
3287                                          SOL_PACKET, PACKET_TX_TIMESTAMP);
3288                 goto out;
3289         }
3290
3291         /*
3292          *      Call the generic datagram receiver. This handles all sorts
3293          *      of horrible races and re-entrancy so we can forget about it
3294          *      in the protocol layers.
3295          *
3296          *      Now it will return ENETDOWN, if device have just gone down,
3297          *      but then it will block.
3298          */
3299
3300         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3301
3302         /*
3303          *      An error occurred so return it. Because skb_recv_datagram()
3304          *      handles the blocking we don't see and worry about blocking
3305          *      retries.
3306          */
3307
3308         if (skb == NULL)
3309                 goto out;
3310
3311         if (READ_ONCE(pkt_sk(sk)->pressure))
3312                 packet_rcv_has_room(pkt_sk(sk), NULL);
3313
3314         if (pkt_sk(sk)->has_vnet_hdr) {
3315                 err = packet_rcv_vnet(msg, skb, &len);
3316                 if (err)
3317                         goto out_free;
3318                 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3319         }
3320
3321         /* You lose any data beyond the buffer you gave. If it worries
3322          * a user program they can ask the device for its MTU
3323          * anyway.
3324          */
3325         copied = skb->len;
3326         if (copied > len) {
3327                 copied = len;
3328                 msg->msg_flags |= MSG_TRUNC;
3329         }
3330
3331         err = skb_copy_datagram_msg(skb, 0, msg, copied);
3332         if (err)
3333                 goto out_free;
3334
3335         if (sock->type != SOCK_PACKET) {
3336                 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3337
3338                 /* Original length was stored in sockaddr_ll fields */
3339                 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3340                 sll->sll_family = AF_PACKET;
3341                 sll->sll_protocol = skb->protocol;
3342         }
3343
3344         sock_recv_ts_and_drops(msg, sk, skb);
3345
3346         if (msg->msg_name) {
3347                 int copy_len;
3348
3349                 /* If the address length field is there to be filled
3350                  * in, we fill it in now.
3351                  */
3352                 if (sock->type == SOCK_PACKET) {
3353                         __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3354                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
3355                         copy_len = msg->msg_namelen;
3356                 } else {
3357                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3358
3359                         msg->msg_namelen = sll->sll_halen +
3360                                 offsetof(struct sockaddr_ll, sll_addr);
3361                         copy_len = msg->msg_namelen;
3362                         if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3363                                 memset(msg->msg_name +
3364                                        offsetof(struct sockaddr_ll, sll_addr),
3365                                        0, sizeof(sll->sll_addr));
3366                                 msg->msg_namelen = sizeof(struct sockaddr_ll);
3367                         }
3368                 }
3369                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3370         }
3371
3372         if (pkt_sk(sk)->auxdata) {
3373                 struct tpacket_auxdata aux;
3374
3375                 aux.tp_status = TP_STATUS_USER;
3376                 if (skb->ip_summed == CHECKSUM_PARTIAL)
3377                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3378                 else if (skb->pkt_type != PACKET_OUTGOING &&
3379                          (skb->ip_summed == CHECKSUM_COMPLETE ||
3380                           skb_csum_unnecessary(skb)))
3381                         aux.tp_status |= TP_STATUS_CSUM_VALID;
3382
3383                 aux.tp_len = origlen;
3384                 aux.tp_snaplen = skb->len;
3385                 aux.tp_mac = 0;
3386                 aux.tp_net = skb_network_offset(skb);
3387                 if (skb_vlan_tag_present(skb)) {
3388                         aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3389                         aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3390                         aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3391                 } else {
3392                         aux.tp_vlan_tci = 0;
3393                         aux.tp_vlan_tpid = 0;
3394                 }
3395                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3396         }
3397
3398         /*
3399          *      Free or return the buffer as appropriate. Again this
3400          *      hides all the races and re-entrancy issues from us.
3401          */
3402         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3403
3404 out_free:
3405         skb_free_datagram(sk, skb);
3406 out:
3407         return err;
3408 }
3409
3410 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3411                                int peer)
3412 {
3413         struct net_device *dev;
3414         struct sock *sk = sock->sk;
3415
3416         if (peer)
3417                 return -EOPNOTSUPP;
3418
3419         uaddr->sa_family = AF_PACKET;
3420         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3421         rcu_read_lock();
3422         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3423         if (dev)
3424                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3425         rcu_read_unlock();
3426
3427         return sizeof(*uaddr);
3428 }
3429
3430 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3431                           int peer)
3432 {
3433         struct net_device *dev;
3434         struct sock *sk = sock->sk;
3435         struct packet_sock *po = pkt_sk(sk);
3436         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3437
3438         if (peer)
3439                 return -EOPNOTSUPP;
3440
3441         sll->sll_family = AF_PACKET;
3442         sll->sll_ifindex = po->ifindex;
3443         sll->sll_protocol = po->num;
3444         sll->sll_pkttype = 0;
3445         rcu_read_lock();
3446         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3447         if (dev) {
3448                 sll->sll_hatype = dev->type;
3449                 sll->sll_halen = dev->addr_len;
3450                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3451         } else {
3452                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
3453                 sll->sll_halen = 0;
3454         }
3455         rcu_read_unlock();
3456
3457         return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3458 }
3459
3460 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3461                          int what)
3462 {
3463         switch (i->type) {
3464         case PACKET_MR_MULTICAST:
3465                 if (i->alen != dev->addr_len)
3466                         return -EINVAL;
3467                 if (what > 0)
3468                         return dev_mc_add(dev, i->addr);
3469                 else
3470                         return dev_mc_del(dev, i->addr);
3471                 break;
3472         case PACKET_MR_PROMISC:
3473                 return dev_set_promiscuity(dev, what);
3474         case PACKET_MR_ALLMULTI:
3475                 return dev_set_allmulti(dev, what);
3476         case PACKET_MR_UNICAST:
3477                 if (i->alen != dev->addr_len)
3478                         return -EINVAL;
3479                 if (what > 0)
3480                         return dev_uc_add(dev, i->addr);
3481                 else
3482                         return dev_uc_del(dev, i->addr);
3483                 break;
3484         default:
3485                 break;
3486         }
3487         return 0;
3488 }
3489
3490 static void packet_dev_mclist_delete(struct net_device *dev,
3491                                      struct packet_mclist **mlp)
3492 {
3493         struct packet_mclist *ml;
3494
3495         while ((ml = *mlp) != NULL) {
3496                 if (ml->ifindex == dev->ifindex) {
3497                         packet_dev_mc(dev, ml, -1);
3498                         *mlp = ml->next;
3499                         kfree(ml);
3500                 } else
3501                         mlp = &ml->next;
3502         }
3503 }
3504
3505 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3506 {
3507         struct packet_sock *po = pkt_sk(sk);
3508         struct packet_mclist *ml, *i;
3509         struct net_device *dev;
3510         int err;
3511
3512         rtnl_lock();
3513
3514         err = -ENODEV;
3515         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3516         if (!dev)
3517                 goto done;
3518
3519         err = -EINVAL;
3520         if (mreq->mr_alen > dev->addr_len)
3521                 goto done;
3522
3523         err = -ENOBUFS;
3524         i = kmalloc(sizeof(*i), GFP_KERNEL);
3525         if (i == NULL)
3526                 goto done;
3527
3528         err = 0;
3529         for (ml = po->mclist; ml; ml = ml->next) {
3530                 if (ml->ifindex == mreq->mr_ifindex &&
3531                     ml->type == mreq->mr_type &&
3532                     ml->alen == mreq->mr_alen &&
3533                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3534                         ml->count++;
3535                         /* Free the new element ... */
3536                         kfree(i);
3537                         goto done;
3538                 }
3539         }
3540
3541         i->type = mreq->mr_type;
3542         i->ifindex = mreq->mr_ifindex;
3543         i->alen = mreq->mr_alen;
3544         memcpy(i->addr, mreq->mr_address, i->alen);
3545         memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3546         i->count = 1;
3547         i->next = po->mclist;
3548         po->mclist = i;
3549         err = packet_dev_mc(dev, i, 1);
3550         if (err) {
3551                 po->mclist = i->next;
3552                 kfree(i);
3553         }
3554
3555 done:
3556         rtnl_unlock();
3557         return err;
3558 }
3559
3560 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3561 {
3562         struct packet_mclist *ml, **mlp;
3563
3564         rtnl_lock();
3565
3566         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3567                 if (ml->ifindex == mreq->mr_ifindex &&
3568                     ml->type == mreq->mr_type &&
3569                     ml->alen == mreq->mr_alen &&
3570                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3571                         if (--ml->count == 0) {
3572                                 struct net_device *dev;
3573                                 *mlp = ml->next;
3574                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3575                                 if (dev)
3576                                         packet_dev_mc(dev, ml, -1);
3577                                 kfree(ml);
3578                         }
3579                         break;
3580                 }
3581         }
3582         rtnl_unlock();
3583         return 0;
3584 }
3585
3586 static void packet_flush_mclist(struct sock *sk)
3587 {
3588         struct packet_sock *po = pkt_sk(sk);
3589         struct packet_mclist *ml;
3590
3591         if (!po->mclist)
3592                 return;
3593
3594         rtnl_lock();
3595         while ((ml = po->mclist) != NULL) {
3596                 struct net_device *dev;
3597
3598                 po->mclist = ml->next;
3599                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3600                 if (dev != NULL)
3601                         packet_dev_mc(dev, ml, -1);
3602                 kfree(ml);
3603         }
3604         rtnl_unlock();
3605 }
3606
3607 static int
3608 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3609 {
3610         struct sock *sk = sock->sk;
3611         struct packet_sock *po = pkt_sk(sk);
3612         int ret;
3613
3614         if (level != SOL_PACKET)
3615                 return -ENOPROTOOPT;
3616
3617         switch (optname) {
3618         case PACKET_ADD_MEMBERSHIP:
3619         case PACKET_DROP_MEMBERSHIP:
3620         {
3621                 struct packet_mreq_max mreq;
3622                 int len = optlen;
3623                 memset(&mreq, 0, sizeof(mreq));
3624                 if (len < sizeof(struct packet_mreq))
3625                         return -EINVAL;
3626                 if (len > sizeof(mreq))
3627                         len = sizeof(mreq);
3628                 if (copy_from_user(&mreq, optval, len))
3629                         return -EFAULT;
3630                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3631                         return -EINVAL;
3632                 if (optname == PACKET_ADD_MEMBERSHIP)
3633                         ret = packet_mc_add(sk, &mreq);
3634                 else
3635                         ret = packet_mc_drop(sk, &mreq);
3636                 return ret;
3637         }
3638
3639         case PACKET_RX_RING:
3640         case PACKET_TX_RING:
3641         {
3642                 union tpacket_req_u req_u;
3643                 int len;
3644
3645                 lock_sock(sk);
3646                 switch (po->tp_version) {
3647                 case TPACKET_V1:
3648                 case TPACKET_V2:
3649                         len = sizeof(req_u.req);
3650                         break;
3651                 case TPACKET_V3:
3652                 default:
3653                         len = sizeof(req_u.req3);
3654                         break;
3655                 }
3656                 if (optlen < len) {
3657                         ret = -EINVAL;
3658                 } else {
3659                         if (copy_from_user(&req_u.req, optval, len))
3660                                 ret = -EFAULT;
3661                         else
3662                                 ret = packet_set_ring(sk, &req_u, 0,
3663                                                     optname == PACKET_TX_RING);
3664                 }
3665                 release_sock(sk);
3666                 return ret;
3667         }
3668         case PACKET_COPY_THRESH:
3669         {
3670                 int val;
3671
3672                 if (optlen != sizeof(val))
3673                         return -EINVAL;
3674                 if (copy_from_user(&val, optval, sizeof(val)))
3675                         return -EFAULT;
3676
3677                 pkt_sk(sk)->copy_thresh = val;
3678                 return 0;
3679         }
3680         case PACKET_VERSION:
3681         {
3682                 int val;
3683
3684                 if (optlen != sizeof(val))
3685                         return -EINVAL;
3686                 if (copy_from_user(&val, optval, sizeof(val)))
3687                         return -EFAULT;
3688                 switch (val) {
3689                 case TPACKET_V1:
3690                 case TPACKET_V2:
3691                 case TPACKET_V3:
3692                         break;
3693                 default:
3694                         return -EINVAL;
3695                 }
3696                 lock_sock(sk);
3697                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3698                         ret = -EBUSY;
3699                 } else {
3700                         po->tp_version = val;
3701                         ret = 0;
3702                 }
3703                 release_sock(sk);
3704                 return ret;
3705         }
3706         case PACKET_RESERVE:
3707         {
3708                 unsigned int val;
3709
3710                 if (optlen != sizeof(val))
3711                         return -EINVAL;
3712                 if (copy_from_user(&val, optval, sizeof(val)))
3713                         return -EFAULT;
3714                 if (val > INT_MAX)
3715                         return -EINVAL;
3716                 lock_sock(sk);
3717                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3718                         ret = -EBUSY;
3719                 } else {
3720                         po->tp_reserve = val;
3721                         ret = 0;
3722                 }
3723                 release_sock(sk);
3724                 return ret;
3725         }
3726         case PACKET_LOSS:
3727         {
3728                 unsigned int val;
3729
3730                 if (optlen != sizeof(val))
3731                         return -EINVAL;
3732                 if (copy_from_user(&val, optval, sizeof(val)))
3733                         return -EFAULT;
3734
3735                 lock_sock(sk);
3736                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3737                         ret = -EBUSY;
3738                 } else {
3739                         po->tp_loss = !!val;
3740                         ret = 0;
3741                 }
3742                 release_sock(sk);
3743                 return ret;
3744         }
3745         case PACKET_AUXDATA:
3746         {
3747                 int val;
3748
3749                 if (optlen < sizeof(val))
3750                         return -EINVAL;
3751                 if (copy_from_user(&val, optval, sizeof(val)))
3752                         return -EFAULT;
3753
3754                 lock_sock(sk);
3755                 po->auxdata = !!val;
3756                 release_sock(sk);
3757                 return 0;
3758         }
3759         case PACKET_ORIGDEV:
3760         {
3761                 int val;
3762
3763                 if (optlen < sizeof(val))
3764                         return -EINVAL;
3765                 if (copy_from_user(&val, optval, sizeof(val)))
3766                         return -EFAULT;
3767
3768                 lock_sock(sk);
3769                 po->origdev = !!val;
3770                 release_sock(sk);
3771                 return 0;
3772         }
3773         case PACKET_VNET_HDR:
3774         {
3775                 int val;
3776
3777                 if (sock->type != SOCK_RAW)
3778                         return -EINVAL;
3779                 if (optlen < sizeof(val))
3780                         return -EINVAL;
3781                 if (copy_from_user(&val, optval, sizeof(val)))
3782                         return -EFAULT;
3783
3784                 lock_sock(sk);
3785                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3786                         ret = -EBUSY;
3787                 } else {
3788                         po->has_vnet_hdr = !!val;
3789                         ret = 0;
3790                 }
3791                 release_sock(sk);
3792                 return ret;
3793         }
3794         case PACKET_TIMESTAMP:
3795         {
3796                 int val;
3797
3798                 if (optlen != sizeof(val))
3799                         return -EINVAL;
3800                 if (copy_from_user(&val, optval, sizeof(val)))
3801                         return -EFAULT;
3802
3803                 po->tp_tstamp = val;
3804                 return 0;
3805         }
3806         case PACKET_FANOUT:
3807         {
3808                 int val;
3809
3810                 if (optlen != sizeof(val))
3811                         return -EINVAL;
3812                 if (copy_from_user(&val, optval, sizeof(val)))
3813                         return -EFAULT;
3814
3815                 return fanout_add(sk, val & 0xffff, val >> 16);
3816         }
3817         case PACKET_FANOUT_DATA:
3818         {
3819                 if (!po->fanout)
3820                         return -EINVAL;
3821
3822                 return fanout_set_data(po, optval, optlen);
3823         }
3824         case PACKET_IGNORE_OUTGOING:
3825         {
3826                 int val;
3827
3828                 if (optlen != sizeof(val))
3829                         return -EINVAL;
3830                 if (copy_from_user(&val, optval, sizeof(val)))
3831                         return -EFAULT;
3832                 if (val < 0 || val > 1)
3833                         return -EINVAL;
3834
3835                 po->prot_hook.ignore_outgoing = !!val;
3836                 return 0;
3837         }
3838         case PACKET_TX_HAS_OFF:
3839         {
3840                 unsigned int val;
3841
3842                 if (optlen != sizeof(val))
3843                         return -EINVAL;
3844                 if (copy_from_user(&val, optval, sizeof(val)))
3845                         return -EFAULT;
3846
3847                 lock_sock(sk);
3848                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3849                         ret = -EBUSY;
3850                 } else {
3851                         po->tp_tx_has_off = !!val;
3852                         ret = 0;
3853                 }
3854                 release_sock(sk);
3855                 return 0;
3856         }
3857         case PACKET_QDISC_BYPASS:
3858         {
3859                 int val;
3860
3861                 if (optlen != sizeof(val))
3862                         return -EINVAL;
3863                 if (copy_from_user(&val, optval, sizeof(val)))
3864                         return -EFAULT;
3865
3866                 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3867                 return 0;
3868         }
3869         default:
3870                 return -ENOPROTOOPT;
3871         }
3872 }
3873
3874 static int packet_getsockopt(struct socket *sock, int level, int optname,
3875                              char __user *optval, int __user *optlen)
3876 {
3877         int len;
3878         int val, lv = sizeof(val);
3879         struct sock *sk = sock->sk;
3880         struct packet_sock *po = pkt_sk(sk);
3881         void *data = &val;
3882         union tpacket_stats_u st;
3883         struct tpacket_rollover_stats rstats;
3884         int drops;
3885
3886         if (level != SOL_PACKET)
3887                 return -ENOPROTOOPT;
3888
3889         if (get_user(len, optlen))
3890                 return -EFAULT;
3891
3892         if (len < 0)
3893                 return -EINVAL;
3894
3895         switch (optname) {
3896         case PACKET_STATISTICS:
3897                 spin_lock_bh(&sk->sk_receive_queue.lock);
3898                 memcpy(&st, &po->stats, sizeof(st));
3899                 memset(&po->stats, 0, sizeof(po->stats));
3900                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3901                 drops = atomic_xchg(&po->tp_drops, 0);
3902
3903                 if (po->tp_version == TPACKET_V3) {
3904                         lv = sizeof(struct tpacket_stats_v3);
3905                         st.stats3.tp_drops = drops;
3906                         st.stats3.tp_packets += drops;
3907                         data = &st.stats3;
3908                 } else {
3909                         lv = sizeof(struct tpacket_stats);
3910                         st.stats1.tp_drops = drops;
3911                         st.stats1.tp_packets += drops;
3912                         data = &st.stats1;
3913                 }
3914
3915                 break;
3916         case PACKET_AUXDATA:
3917                 val = po->auxdata;
3918                 break;
3919         case PACKET_ORIGDEV:
3920                 val = po->origdev;
3921                 break;
3922         case PACKET_VNET_HDR:
3923                 val = po->has_vnet_hdr;
3924                 break;
3925         case PACKET_VERSION:
3926                 val = po->tp_version;
3927                 break;
3928         case PACKET_HDRLEN:
3929                 if (len > sizeof(int))
3930                         len = sizeof(int);
3931                 if (len < sizeof(int))
3932                         return -EINVAL;
3933                 if (copy_from_user(&val, optval, len))
3934                         return -EFAULT;
3935                 switch (val) {
3936                 case TPACKET_V1:
3937                         val = sizeof(struct tpacket_hdr);
3938                         break;
3939                 case TPACKET_V2:
3940                         val = sizeof(struct tpacket2_hdr);
3941                         break;
3942                 case TPACKET_V3:
3943                         val = sizeof(struct tpacket3_hdr);
3944                         break;
3945                 default:
3946                         return -EINVAL;
3947                 }
3948                 break;
3949         case PACKET_RESERVE:
3950                 val = po->tp_reserve;
3951                 break;
3952         case PACKET_LOSS:
3953                 val = po->tp_loss;
3954                 break;
3955         case PACKET_TIMESTAMP:
3956                 val = po->tp_tstamp;
3957                 break;
3958         case PACKET_FANOUT:
3959                 val = (po->fanout ?
3960                        ((u32)po->fanout->id |
3961                         ((u32)po->fanout->type << 16) |
3962                         ((u32)po->fanout->flags << 24)) :
3963                        0);
3964                 break;
3965         case PACKET_IGNORE_OUTGOING:
3966                 val = po->prot_hook.ignore_outgoing;
3967                 break;
3968         case PACKET_ROLLOVER_STATS:
3969                 if (!po->rollover)
3970                         return -EINVAL;
3971                 rstats.tp_all = atomic_long_read(&po->rollover->num);
3972                 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3973                 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3974                 data = &rstats;
3975                 lv = sizeof(rstats);
3976                 break;
3977         case PACKET_TX_HAS_OFF:
3978                 val = po->tp_tx_has_off;
3979                 break;
3980         case PACKET_QDISC_BYPASS:
3981                 val = packet_use_direct_xmit(po);
3982                 break;
3983         default:
3984                 return -ENOPROTOOPT;
3985         }
3986
3987         if (len > lv)
3988                 len = lv;
3989         if (put_user(len, optlen))
3990                 return -EFAULT;
3991         if (copy_to_user(optval, data, len))
3992                 return -EFAULT;
3993         return 0;
3994 }
3995
3996
3997 #ifdef CONFIG_COMPAT
3998 static int compat_packet_setsockopt(struct socket *sock, int level, int optname,
3999                                     char __user *optval, unsigned int optlen)
4000 {
4001         struct packet_sock *po = pkt_sk(sock->sk);
4002
4003         if (level != SOL_PACKET)
4004                 return -ENOPROTOOPT;
4005
4006         if (optname == PACKET_FANOUT_DATA &&
4007             po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) {
4008                 optval = (char __user *)get_compat_bpf_fprog(optval);
4009                 if (!optval)
4010                         return -EFAULT;
4011                 optlen = sizeof(struct sock_fprog);
4012         }
4013
4014         return packet_setsockopt(sock, level, optname, optval, optlen);
4015 }
4016 #endif
4017
4018 static int packet_notifier(struct notifier_block *this,
4019                            unsigned long msg, void *ptr)
4020 {
4021         struct sock *sk;
4022         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4023         struct net *net = dev_net(dev);
4024
4025         rcu_read_lock();
4026         sk_for_each_rcu(sk, &net->packet.sklist) {
4027                 struct packet_sock *po = pkt_sk(sk);
4028
4029                 switch (msg) {
4030                 case NETDEV_UNREGISTER:
4031                         if (po->mclist)
4032                                 packet_dev_mclist_delete(dev, &po->mclist);
4033                         /* fallthrough */
4034
4035                 case NETDEV_DOWN:
4036                         if (dev->ifindex == po->ifindex) {
4037                                 spin_lock(&po->bind_lock);
4038                                 if (po->running) {
4039                                         __unregister_prot_hook(sk, false);
4040                                         sk->sk_err = ENETDOWN;
4041                                         if (!sock_flag(sk, SOCK_DEAD))
4042                                                 sk->sk_error_report(sk);
4043                                 }
4044                                 if (msg == NETDEV_UNREGISTER) {
4045                                         packet_cached_dev_reset(po);
4046                                         po->ifindex = -1;
4047                                         if (po->prot_hook.dev)
4048                                                 dev_put(po->prot_hook.dev);
4049                                         po->prot_hook.dev = NULL;
4050                                 }
4051                                 spin_unlock(&po->bind_lock);
4052                         }
4053                         break;
4054                 case NETDEV_UP:
4055                         if (dev->ifindex == po->ifindex) {
4056                                 spin_lock(&po->bind_lock);
4057                                 if (po->num)
4058                                         register_prot_hook(sk);
4059                                 spin_unlock(&po->bind_lock);
4060                         }
4061                         break;
4062                 }
4063         }
4064         rcu_read_unlock();
4065         return NOTIFY_DONE;
4066 }
4067
4068
4069 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4070                         unsigned long arg)
4071 {
4072         struct sock *sk = sock->sk;
4073
4074         switch (cmd) {
4075         case SIOCOUTQ:
4076         {
4077                 int amount = sk_wmem_alloc_get(sk);
4078
4079                 return put_user(amount, (int __user *)arg);
4080         }
4081         case SIOCINQ:
4082         {
4083                 struct sk_buff *skb;
4084                 int amount = 0;
4085
4086                 spin_lock_bh(&sk->sk_receive_queue.lock);
4087                 skb = skb_peek(&sk->sk_receive_queue);
4088                 if (skb)
4089                         amount = skb->len;
4090                 spin_unlock_bh(&sk->sk_receive_queue.lock);
4091                 return put_user(amount, (int __user *)arg);
4092         }
4093 #ifdef CONFIG_INET
4094         case SIOCADDRT:
4095         case SIOCDELRT:
4096         case SIOCDARP:
4097         case SIOCGARP:
4098         case SIOCSARP:
4099         case SIOCGIFADDR:
4100         case SIOCSIFADDR:
4101         case SIOCGIFBRDADDR:
4102         case SIOCSIFBRDADDR:
4103         case SIOCGIFNETMASK:
4104         case SIOCSIFNETMASK:
4105         case SIOCGIFDSTADDR:
4106         case SIOCSIFDSTADDR:
4107         case SIOCSIFFLAGS:
4108                 return inet_dgram_ops.ioctl(sock, cmd, arg);
4109 #endif
4110
4111         default:
4112                 return -ENOIOCTLCMD;
4113         }
4114         return 0;
4115 }
4116
4117 static __poll_t packet_poll(struct file *file, struct socket *sock,
4118                                 poll_table *wait)
4119 {
4120         struct sock *sk = sock->sk;
4121         struct packet_sock *po = pkt_sk(sk);
4122         __poll_t mask = datagram_poll(file, sock, wait);
4123
4124         spin_lock_bh(&sk->sk_receive_queue.lock);
4125         if (po->rx_ring.pg_vec) {
4126                 if (!packet_previous_rx_frame(po, &po->rx_ring,
4127                         TP_STATUS_KERNEL))
4128                         mask |= EPOLLIN | EPOLLRDNORM;
4129         }
4130         if (READ_ONCE(po->pressure) && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
4131                 WRITE_ONCE(po->pressure, 0);
4132         spin_unlock_bh(&sk->sk_receive_queue.lock);
4133         spin_lock_bh(&sk->sk_write_queue.lock);
4134         if (po->tx_ring.pg_vec) {
4135                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4136                         mask |= EPOLLOUT | EPOLLWRNORM;
4137         }
4138         spin_unlock_bh(&sk->sk_write_queue.lock);
4139         return mask;
4140 }
4141
4142
4143 /* Dirty? Well, I still did not learn better way to account
4144  * for user mmaps.
4145  */
4146
4147 static void packet_mm_open(struct vm_area_struct *vma)
4148 {
4149         struct file *file = vma->vm_file;
4150         struct socket *sock = file->private_data;
4151         struct sock *sk = sock->sk;
4152
4153         if (sk)
4154                 atomic_inc(&pkt_sk(sk)->mapped);
4155 }
4156
4157 static void packet_mm_close(struct vm_area_struct *vma)
4158 {
4159         struct file *file = vma->vm_file;
4160         struct socket *sock = file->private_data;
4161         struct sock *sk = sock->sk;
4162
4163         if (sk)
4164                 atomic_dec(&pkt_sk(sk)->mapped);
4165 }
4166
4167 static const struct vm_operations_struct packet_mmap_ops = {
4168         .open   =       packet_mm_open,
4169         .close  =       packet_mm_close,
4170 };
4171
4172 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4173                         unsigned int len)
4174 {
4175         int i;
4176
4177         for (i = 0; i < len; i++) {
4178                 if (likely(pg_vec[i].buffer)) {
4179                         if (is_vmalloc_addr(pg_vec[i].buffer))
4180                                 vfree(pg_vec[i].buffer);
4181                         else
4182                                 free_pages((unsigned long)pg_vec[i].buffer,
4183                                            order);
4184                         pg_vec[i].buffer = NULL;
4185                 }
4186         }
4187         kfree(pg_vec);
4188 }
4189
4190 static char *alloc_one_pg_vec_page(unsigned long order)
4191 {
4192         char *buffer;
4193         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4194                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4195
4196         buffer = (char *) __get_free_pages(gfp_flags, order);
4197         if (buffer)
4198                 return buffer;
4199
4200         /* __get_free_pages failed, fall back to vmalloc */
4201         buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4202         if (buffer)
4203                 return buffer;
4204
4205         /* vmalloc failed, lets dig into swap here */
4206         gfp_flags &= ~__GFP_NORETRY;
4207         buffer = (char *) __get_free_pages(gfp_flags, order);
4208         if (buffer)
4209                 return buffer;
4210
4211         /* complete and utter failure */
4212         return NULL;
4213 }
4214
4215 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4216 {
4217         unsigned int block_nr = req->tp_block_nr;
4218         struct pgv *pg_vec;
4219         int i;
4220
4221         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4222         if (unlikely(!pg_vec))
4223                 goto out;
4224
4225         for (i = 0; i < block_nr; i++) {
4226                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4227                 if (unlikely(!pg_vec[i].buffer))
4228                         goto out_free_pgvec;
4229         }
4230
4231 out:
4232         return pg_vec;
4233
4234 out_free_pgvec:
4235         free_pg_vec(pg_vec, order, block_nr);
4236         pg_vec = NULL;
4237         goto out;
4238 }
4239
4240 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4241                 int closing, int tx_ring)
4242 {
4243         struct pgv *pg_vec = NULL;
4244         struct packet_sock *po = pkt_sk(sk);
4245         int was_running, order = 0;
4246         struct packet_ring_buffer *rb;
4247         struct sk_buff_head *rb_queue;
4248         __be16 num;
4249         int err = -EINVAL;
4250         /* Added to avoid minimal code churn */
4251         struct tpacket_req *req = &req_u->req;
4252
4253         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4254         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4255
4256         err = -EBUSY;
4257         if (!closing) {
4258                 if (atomic_read(&po->mapped))
4259                         goto out;
4260                 if (packet_read_pending(rb))
4261                         goto out;
4262         }
4263
4264         if (req->tp_block_nr) {
4265                 unsigned int min_frame_size;
4266
4267                 /* Sanity tests and some calculations */
4268                 err = -EBUSY;
4269                 if (unlikely(rb->pg_vec))
4270                         goto out;
4271
4272                 switch (po->tp_version) {
4273                 case TPACKET_V1:
4274                         po->tp_hdrlen = TPACKET_HDRLEN;
4275                         break;
4276                 case TPACKET_V2:
4277                         po->tp_hdrlen = TPACKET2_HDRLEN;
4278                         break;
4279                 case TPACKET_V3:
4280                         po->tp_hdrlen = TPACKET3_HDRLEN;
4281                         break;
4282                 }
4283
4284                 err = -EINVAL;
4285                 if (unlikely((int)req->tp_block_size <= 0))
4286                         goto out;
4287                 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4288                         goto out;
4289                 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4290                 if (po->tp_version >= TPACKET_V3 &&
4291                     req->tp_block_size <
4292                     BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4293                         goto out;
4294                 if (unlikely(req->tp_frame_size < min_frame_size))
4295                         goto out;
4296                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4297                         goto out;
4298
4299                 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4300                 if (unlikely(rb->frames_per_block == 0))
4301                         goto out;
4302                 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4303                         goto out;
4304                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4305                                         req->tp_frame_nr))
4306                         goto out;
4307
4308                 err = -ENOMEM;
4309                 order = get_order(req->tp_block_size);
4310                 pg_vec = alloc_pg_vec(req, order);
4311                 if (unlikely(!pg_vec))
4312                         goto out;
4313                 switch (po->tp_version) {
4314                 case TPACKET_V3:
4315                         /* Block transmit is not supported yet */
4316                         if (!tx_ring) {
4317                                 init_prb_bdqc(po, rb, pg_vec, req_u);
4318                         } else {
4319                                 struct tpacket_req3 *req3 = &req_u->req3;
4320
4321                                 if (req3->tp_retire_blk_tov ||
4322                                     req3->tp_sizeof_priv ||
4323                                     req3->tp_feature_req_word) {
4324                                         err = -EINVAL;
4325                                         goto out;
4326                                 }
4327                         }
4328                         break;
4329                 default:
4330                         break;
4331                 }
4332         }
4333         /* Done */
4334         else {
4335                 err = -EINVAL;
4336                 if (unlikely(req->tp_frame_nr))
4337                         goto out;
4338         }
4339
4340
4341         /* Detach socket from network */
4342         spin_lock(&po->bind_lock);
4343         was_running = po->running;
4344         num = po->num;
4345         if (was_running) {
4346                 po->num = 0;
4347                 __unregister_prot_hook(sk, false);
4348         }
4349         spin_unlock(&po->bind_lock);
4350
4351         synchronize_net();
4352
4353         err = -EBUSY;
4354         mutex_lock(&po->pg_vec_lock);
4355         if (closing || atomic_read(&po->mapped) == 0) {
4356                 err = 0;
4357                 spin_lock_bh(&rb_queue->lock);
4358                 swap(rb->pg_vec, pg_vec);
4359                 rb->frame_max = (req->tp_frame_nr - 1);
4360                 rb->head = 0;
4361                 rb->frame_size = req->tp_frame_size;
4362                 spin_unlock_bh(&rb_queue->lock);
4363
4364                 swap(rb->pg_vec_order, order);
4365                 swap(rb->pg_vec_len, req->tp_block_nr);
4366
4367                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4368                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4369                                                 tpacket_rcv : packet_rcv;
4370                 skb_queue_purge(rb_queue);
4371                 if (atomic_read(&po->mapped))
4372                         pr_err("packet_mmap: vma is busy: %d\n",
4373                                atomic_read(&po->mapped));
4374         }
4375         mutex_unlock(&po->pg_vec_lock);
4376
4377         spin_lock(&po->bind_lock);
4378         if (was_running) {
4379                 po->num = num;
4380                 register_prot_hook(sk);
4381         }
4382         spin_unlock(&po->bind_lock);
4383         if (pg_vec && (po->tp_version > TPACKET_V2)) {
4384                 /* Because we don't support block-based V3 on tx-ring */
4385                 if (!tx_ring)
4386                         prb_shutdown_retire_blk_timer(po, rb_queue);
4387         }
4388
4389         if (pg_vec)
4390                 free_pg_vec(pg_vec, order, req->tp_block_nr);
4391 out:
4392         return err;
4393 }
4394
4395 static int packet_mmap(struct file *file, struct socket *sock,
4396                 struct vm_area_struct *vma)
4397 {
4398         struct sock *sk = sock->sk;
4399         struct packet_sock *po = pkt_sk(sk);
4400         unsigned long size, expected_size;
4401         struct packet_ring_buffer *rb;
4402         unsigned long start;
4403         int err = -EINVAL;
4404         int i;
4405
4406         if (vma->vm_pgoff)
4407                 return -EINVAL;
4408
4409         mutex_lock(&po->pg_vec_lock);
4410
4411         expected_size = 0;
4412         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4413                 if (rb->pg_vec) {
4414                         expected_size += rb->pg_vec_len
4415                                                 * rb->pg_vec_pages
4416                                                 * PAGE_SIZE;
4417                 }
4418         }
4419
4420         if (expected_size == 0)
4421                 goto out;
4422
4423         size = vma->vm_end - vma->vm_start;
4424         if (size != expected_size)
4425                 goto out;
4426
4427         start = vma->vm_start;
4428         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4429                 if (rb->pg_vec == NULL)
4430                         continue;
4431
4432                 for (i = 0; i < rb->pg_vec_len; i++) {
4433                         struct page *page;
4434                         void *kaddr = rb->pg_vec[i].buffer;
4435                         int pg_num;
4436
4437                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4438                                 page = pgv_to_page(kaddr);
4439                                 err = vm_insert_page(vma, start, page);
4440                                 if (unlikely(err))
4441                                         goto out;
4442                                 start += PAGE_SIZE;
4443                                 kaddr += PAGE_SIZE;
4444                         }
4445                 }
4446         }
4447
4448         atomic_inc(&po->mapped);
4449         vma->vm_ops = &packet_mmap_ops;
4450         err = 0;
4451
4452 out:
4453         mutex_unlock(&po->pg_vec_lock);
4454         return err;
4455 }
4456
4457 static const struct proto_ops packet_ops_spkt = {
4458         .family =       PF_PACKET,
4459         .owner =        THIS_MODULE,
4460         .release =      packet_release,
4461         .bind =         packet_bind_spkt,
4462         .connect =      sock_no_connect,
4463         .socketpair =   sock_no_socketpair,
4464         .accept =       sock_no_accept,
4465         .getname =      packet_getname_spkt,
4466         .poll =         datagram_poll,
4467         .ioctl =        packet_ioctl,
4468         .gettstamp =    sock_gettstamp,
4469         .listen =       sock_no_listen,
4470         .shutdown =     sock_no_shutdown,
4471         .setsockopt =   sock_no_setsockopt,
4472         .getsockopt =   sock_no_getsockopt,
4473         .sendmsg =      packet_sendmsg_spkt,
4474         .recvmsg =      packet_recvmsg,
4475         .mmap =         sock_no_mmap,
4476         .sendpage =     sock_no_sendpage,
4477 };
4478
4479 static const struct proto_ops packet_ops = {
4480         .family =       PF_PACKET,
4481         .owner =        THIS_MODULE,
4482         .release =      packet_release,
4483         .bind =         packet_bind,
4484         .connect =      sock_no_connect,
4485         .socketpair =   sock_no_socketpair,
4486         .accept =       sock_no_accept,
4487         .getname =      packet_getname,
4488         .poll =         packet_poll,
4489         .ioctl =        packet_ioctl,
4490         .gettstamp =    sock_gettstamp,
4491         .listen =       sock_no_listen,
4492         .shutdown =     sock_no_shutdown,
4493         .setsockopt =   packet_setsockopt,
4494         .getsockopt =   packet_getsockopt,
4495 #ifdef CONFIG_COMPAT
4496         .compat_setsockopt = compat_packet_setsockopt,
4497 #endif
4498         .sendmsg =      packet_sendmsg,
4499         .recvmsg =      packet_recvmsg,
4500         .mmap =         packet_mmap,
4501         .sendpage =     sock_no_sendpage,
4502 };
4503
4504 static const struct net_proto_family packet_family_ops = {
4505         .family =       PF_PACKET,
4506         .create =       packet_create,
4507         .owner  =       THIS_MODULE,
4508 };
4509
4510 static struct notifier_block packet_netdev_notifier = {
4511         .notifier_call =        packet_notifier,
4512 };
4513
4514 #ifdef CONFIG_PROC_FS
4515
4516 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4517         __acquires(RCU)
4518 {
4519         struct net *net = seq_file_net(seq);
4520
4521         rcu_read_lock();
4522         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4523 }
4524
4525 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4526 {
4527         struct net *net = seq_file_net(seq);
4528         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4529 }
4530
4531 static void packet_seq_stop(struct seq_file *seq, void *v)
4532         __releases(RCU)
4533 {
4534         rcu_read_unlock();
4535 }
4536
4537 static int packet_seq_show(struct seq_file *seq, void *v)
4538 {
4539         if (v == SEQ_START_TOKEN)
4540                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
4541         else {
4542                 struct sock *s = sk_entry(v);
4543                 const struct packet_sock *po = pkt_sk(s);
4544
4545                 seq_printf(seq,
4546                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4547                            s,
4548                            refcount_read(&s->sk_refcnt),
4549                            s->sk_type,
4550                            ntohs(po->num),
4551                            po->ifindex,
4552                            po->running,
4553                            atomic_read(&s->sk_rmem_alloc),
4554                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4555                            sock_i_ino(s));
4556         }
4557
4558         return 0;
4559 }
4560
4561 static const struct seq_operations packet_seq_ops = {
4562         .start  = packet_seq_start,
4563         .next   = packet_seq_next,
4564         .stop   = packet_seq_stop,
4565         .show   = packet_seq_show,
4566 };
4567 #endif
4568
4569 static int __net_init packet_net_init(struct net *net)
4570 {
4571         mutex_init(&net->packet.sklist_lock);
4572         INIT_HLIST_HEAD(&net->packet.sklist);
4573
4574         if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4575                         sizeof(struct seq_net_private)))
4576                 return -ENOMEM;
4577
4578         return 0;
4579 }
4580
4581 static void __net_exit packet_net_exit(struct net *net)
4582 {
4583         remove_proc_entry("packet", net->proc_net);
4584         WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4585 }
4586
4587 static struct pernet_operations packet_net_ops = {
4588         .init = packet_net_init,
4589         .exit = packet_net_exit,
4590 };
4591
4592
4593 static void __exit packet_exit(void)
4594 {
4595         unregister_netdevice_notifier(&packet_netdev_notifier);
4596         unregister_pernet_subsys(&packet_net_ops);
4597         sock_unregister(PF_PACKET);
4598         proto_unregister(&packet_proto);
4599 }
4600
4601 static int __init packet_init(void)
4602 {
4603         int rc;
4604
4605         rc = proto_register(&packet_proto, 0);
4606         if (rc)
4607                 goto out;
4608         rc = sock_register(&packet_family_ops);
4609         if (rc)
4610                 goto out_proto;
4611         rc = register_pernet_subsys(&packet_net_ops);
4612         if (rc)
4613                 goto out_sock;
4614         rc = register_netdevice_notifier(&packet_netdev_notifier);
4615         if (rc)
4616                 goto out_pernet;
4617
4618         return 0;
4619
4620 out_pernet:
4621         unregister_pernet_subsys(&packet_net_ops);
4622 out_sock:
4623         sock_unregister(PF_PACKET);
4624 out_proto:
4625         proto_unregister(&packet_proto);
4626 out:
4627         return rc;
4628 }
4629
4630 module_init(packet_init);
4631 module_exit(packet_exit);
4632 MODULE_LICENSE("GPL");
4633 MODULE_ALIAS_NETPROTO(PF_PACKET);