net/packet/af_packet.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              PACKET - implements raw packet sockets.
   7  *
   8  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  11  *
  12  * Fixes:
  13  *              Alan Cox        :       verify_area() now used correctly
  14  *              Alan Cox        :       new skbuff lists, look ma no backlogs!
  15  *              Alan Cox        :       tidied skbuff lists.
  16  *              Alan Cox        :       Now uses generic datagram routines I
  17  *                                      added. Also fixed the peek/read crash
  18  *                                      from all old Linux datagram code.
  19  *              Alan Cox        :       Uses the improved datagram code.
  20  *              Alan Cox        :       Added NULL's for socket options.
  21  *              Alan Cox        :       Re-commented the code.
  22  *              Alan Cox        :       Use new kernel side addressing
  23  *              Rob Janssen     :       Correct MTU usage.
  24  *              Dave Platt      :       Counter leaks caused by incorrect
  25  *                                      interrupt locking and some slightly
  26  *                                      dubious gcc output. Can you read
  27  *                                      compiler: it said _VOLATILE_
  28  *      Richard Kooijman        :       Timestamp fixes.
  29  *              Alan Cox        :       New buffers. Use sk->mac.raw.
  30  *              Alan Cox        :       sendmsg/recvmsg support.
  31  *              Alan Cox        :       Protocol setting support
  32  *      Alexey Kuznetsov        :       Untied from IPv4 stack.
  33  *      Cyrus Durgin            :       Fixed kerneld for kmod.
  34  *      Michal Ostrowski        :       Module initialization cleanup.
  35  *         Ulises Alonso        :       Frame number limit removal and
  36  *                                      packet_set_ring memory leak.
  37  *              Eric Biederman  :       Allow for > 8 byte hardware addresses.
  38  *                                      The convention is that longer addresses
  39  *                                      will simply extend the hardware address
  40  *                                      byte arrays at the end of sockaddr_ll
  41  *                                      and packet_mreq.
  42  *              Johann Baudy    :       Added TX RING.
  43  *              Chetan Loke     :       Implemented TPACKET_V3 block abstraction
  44  *                                      layer.
  45  *                                      Copyright (C) 2011, <lokec@ccs.neu.edu>
  46  *
  47  *
  48  *              This program is free software; you can redistribute it and/or
  49  *              modify it under the terms of the GNU General Public License
  50  *              as published by the Free Software Foundation; either version
  51  *              2 of the License, or (at your option) any later version.
  52  *
  53  */
  54
  55 #include <linux/types.h>
  56 #include <linux/mm.h>
  57 #include <linux/capability.h>
  58 #include <linux/fcntl.h>
  59 #include <linux/socket.h>
  60 #include <linux/in.h>
  61 #include <linux/inet.h>
  62 #include <linux/netdevice.h>
  63 #include <linux/if_packet.h>
  64 #include <linux/wireless.h>
  65 #include <linux/kernel.h>
  66 #include <linux/kmod.h>
  67 #include <linux/slab.h>
  68 #include <linux/vmalloc.h>
  69 #include <net/net_namespace.h>
  70 #include <net/ip.h>
  71 #include <net/protocol.h>
  72 #include <linux/skbuff.h>
  73 #include <net/sock.h>
  74 #include <linux/errno.h>
  75 #include <linux/timer.h>
  76 #include <linux/uaccess.h>
  77 #include <asm/ioctls.h>
  78 #include <asm/page.h>
  79 #include <asm/cacheflush.h>
  80 #include <asm/io.h>
  81 #include <linux/proc_fs.h>
  82 #include <linux/seq_file.h>
  83 #include <linux/poll.h>
  84 #include <linux/module.h>
  85 #include <linux/init.h>
  86 #include <linux/mutex.h>
  87 #include <linux/if_vlan.h>
  88 #include <linux/virtio_net.h>
  89 #include <linux/errqueue.h>
  90 #include <linux/net_tstamp.h>
  91 #include <linux/percpu.h>
  92 #ifdef CONFIG_INET
  93 #include <net/inet_common.h>
  94 #endif
  95 #include <linux/bpf.h>
  96 #include <net/compat.h>
  97
  98 #include "internal.h"
  99
 100 /*
 101    Assumptions:
 102    - if device has no dev->hard_header routine, it adds and removes ll header
 103      inside itself. In this case ll header is invisible outside of device,
 104      but higher levels still should reserve dev->hard_header_len.
 105      Some devices are enough clever to reallocate skb, when header
 106      will not fit to reserved space (tunnel), another ones are silly
 107      (PPP).
 108    - packet socket receives packets with pulled ll header,
 109      so that SOCK_RAW should push it back.
 110
 111 On receive:
 112 -----------
 113
 114 Incoming, dev->hard_header!=NULL
 115    mac_header -> ll header
 116    data       -> data
 117
 118 Outgoing, dev->hard_header!=NULL
 119    mac_header -> ll header
 120    data       -> ll header
 121
 122 Incoming, dev->hard_header==NULL
 123    mac_header -> UNKNOWN position. It is very likely, that it points to ll
 124                  header.  PPP makes it, that is wrong, because introduce
 125                  assymetry between rx and tx paths.
 126    data       -> data
 127
 128 Outgoing, dev->hard_header==NULL
 129    mac_header -> data. ll header is still not built!
 130    data       -> data
 131
 132 Resume
 133   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 134
 135
 136 On transmit:
 137 ------------
 138
 139 dev->hard_header != NULL
 140    mac_header -> ll header
 141    data       -> ll header
 142
 143 dev->hard_header == NULL (ll header is added by device, we cannot control it)
 144    mac_header -> data
 145    data       -> data
 146
 147    We should set nh.raw on output to correct posistion,
 148    packet classifier depends on it.
 149  */
 150
 151 /* Private packet socket structures. */
 152
 153 /* identical to struct packet_mreq except it has
 154  * a longer address field.
 155  */
 156 struct packet_mreq_max {
 157         int             mr_ifindex;
 158         unsigned short  mr_type;
 159         unsigned short  mr_alen;
 160         unsigned char   mr_address[MAX_ADDR_LEN];
 161 };
 162
 163 union tpacket_uhdr {
 164         struct tpacket_hdr  *h1;
 165         struct tpacket2_hdr *h2;
 166         struct tpacket3_hdr *h3;
 167         void *raw;
 168 };
 169
 170 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 171                 int closing, int tx_ring);
 172
 173 #define V3_ALIGNMENT    (8)
 174
 175 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 176
 177 #define BLK_PLUS_PRIV(sz_of_priv) \
 178         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 179
 180 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
 181 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
 182 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
 183 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
 184 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
 185 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
 186 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
 187
 188 struct packet_sock;
 189 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 190                        struct packet_type *pt, struct net_device *orig_dev);
 191
 192 static void *packet_previous_frame(struct packet_sock *po,
 193                 struct packet_ring_buffer *rb,
 194                 int status);
 195 static void packet_increment_head(struct packet_ring_buffer *buff);
 196 static int prb_curr_blk_in_use(struct tpacket_block_desc *);
 197 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 198                         struct packet_sock *);
 199 static void prb_retire_current_block(struct tpacket_kbdq_core *,
 200                 struct packet_sock *, unsigned int status);
 201 static int prb_queue_frozen(struct tpacket_kbdq_core *);
 202 static void prb_open_block(struct tpacket_kbdq_core *,
 203                 struct tpacket_block_desc *);
 204 static void prb_retire_rx_blk_timer_expired(unsigned long);
 205 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 206 static void prb_init_blk_timer(struct packet_sock *,
 207                 struct tpacket_kbdq_core *,
 208                 void (*func) (unsigned long));
 209 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 210 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 211                 struct tpacket3_hdr *);
 212 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 213                 struct tpacket3_hdr *);
 214 static void packet_flush_mclist(struct sock *sk);
 215 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb);
 216
 217 struct packet_skb_cb {
 218         union {
 219                 struct sockaddr_pkt pkt;
 220                 union {
 221                         /* Trick: alias skb original length with
 222                          * ll.sll_family and ll.protocol in order
 223                          * to save room.
 224                          */
 225                         unsigned int origlen;
 226                         struct sockaddr_ll ll;
 227                 };
 228         } sa;
 229 };
 230
 231 #define vio_le() virtio_legacy_is_little_endian()
 232
 233 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 234
 235 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 236 #define GET_PBLOCK_DESC(x, bid) \
 237         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 238 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 239         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 240 #define GET_NEXT_PRB_BLK_NUM(x) \
 241         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 242         ((x)->kactive_blk_num+1) : 0)
 243
 244 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 245 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 246
 247 static int packet_direct_xmit(struct sk_buff *skb)
 248 {
 249         struct net_device *dev = skb->dev;
 250         struct sk_buff *orig_skb = skb;
 251         struct netdev_queue *txq;
 252         int ret = NETDEV_TX_BUSY;
 253
 254         if (unlikely(!netif_running(dev) ||
 255                      !netif_carrier_ok(dev)))
 256                 goto drop;
 257
 258         skb = validate_xmit_skb_list(skb, dev);
 259         if (skb != orig_skb)
 260                 goto drop;
 261
 262         packet_pick_tx_queue(dev, skb);
 263         txq = skb_get_tx_queue(dev, skb);
 264
 265         local_bh_disable();
 266
 267         HARD_TX_LOCK(dev, txq, smp_processor_id());
 268         if (!netif_xmit_frozen_or_drv_stopped(txq))
 269                 ret = netdev_start_xmit(skb, dev, txq, false);
 270         HARD_TX_UNLOCK(dev, txq);
 271
 272         local_bh_enable();
 273
 274         if (!dev_xmit_complete(ret))
 275                 kfree_skb(skb);
 276
 277         return ret;
 278 drop:
 279         atomic_long_inc(&dev->tx_dropped);
 280         kfree_skb_list(skb);
 281         return NET_XMIT_DROP;
 282 }
 283
 284 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
 285 {
 286         struct net_device *dev;
 287
 288         rcu_read_lock();
 289         dev = rcu_dereference(po->cached_dev);
 290         if (likely(dev))
 291                 dev_hold(dev);
 292         rcu_read_unlock();
 293
 294         return dev;
 295 }
 296
 297 static void packet_cached_dev_assign(struct packet_sock *po,
 298                                      struct net_device *dev)
 299 {
 300         rcu_assign_pointer(po->cached_dev, dev);
 301 }
 302
 303 static void packet_cached_dev_reset(struct packet_sock *po)
 304 {
 305         RCU_INIT_POINTER(po->cached_dev, NULL);
 306 }
 307
 308 static bool packet_use_direct_xmit(const struct packet_sock *po)
 309 {
 310         return po->xmit == packet_direct_xmit;
 311 }
 312
 313 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
 314 {
 315         return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
 316 }
 317
 318 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
 319 {
 320         const struct net_device_ops *ops = dev->netdev_ops;
 321         u16 queue_index;
 322
 323         if (ops->ndo_select_queue) {
 324                 queue_index = ops->ndo_select_queue(dev, skb, NULL,
 325                                                     __packet_pick_tx_queue);
 326                 queue_index = netdev_cap_txqueue(dev, queue_index);
 327         } else {
 328                 queue_index = __packet_pick_tx_queue(dev, skb);
 329         }
 330
 331         skb_set_queue_mapping(skb, queue_index);
 332 }
 333
 334 /* __register_prot_hook must be invoked through register_prot_hook
 335  * or from a context in which asynchronous accesses to the packet
 336  * socket is not possible (packet_create()).
 337  */
 338 static void __register_prot_hook(struct sock *sk)
 339 {
 340         struct packet_sock *po = pkt_sk(sk);
 341
 342         if (!po->running) {
 343                 if (po->fanout)
 344                         __fanout_link(sk, po);
 345                 else
 346                         dev_add_pack(&po->prot_hook);
 347
 348                 sock_hold(sk);
 349                 po->running = 1;
 350         }
 351 }
 352
 353 static void register_prot_hook(struct sock *sk)
 354 {
 355         lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
 356         __register_prot_hook(sk);
 357 }
 358
 359 /* If the sync parameter is true, we will temporarily drop
 360  * the po->bind_lock and do a synchronize_net to make sure no
 361  * asynchronous packet processing paths still refer to the elements
 362  * of po->prot_hook.  If the sync parameter is false, it is the
 363  * callers responsibility to take care of this.
 364  */
 365 static void __unregister_prot_hook(struct sock *sk, bool sync)
 366 {
 367         struct packet_sock *po = pkt_sk(sk);
 368
 369         lockdep_assert_held_once(&po->bind_lock);
 370
 371         po->running = 0;
 372
 373         if (po->fanout)
 374                 __fanout_unlink(sk, po);
 375         else
 376                 __dev_remove_pack(&po->prot_hook);
 377
 378         __sock_put(sk);
 379
 380         if (sync) {
 381                 spin_unlock(&po->bind_lock);
 382                 synchronize_net();
 383                 spin_lock(&po->bind_lock);
 384         }
 385 }
 386
 387 static void unregister_prot_hook(struct sock *sk, bool sync)
 388 {
 389         struct packet_sock *po = pkt_sk(sk);
 390
 391         if (po->running)
 392                 __unregister_prot_hook(sk, sync);
 393 }
 394
 395 static inline struct page * __pure pgv_to_page(void *addr)
 396 {
 397         if (is_vmalloc_addr(addr))
 398                 return vmalloc_to_page(addr);
 399         return virt_to_page(addr);
 400 }
 401
 402 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 403 {
 404         union tpacket_uhdr h;
 405
 406         /* WRITE_ONCE() are paired with READ_ONCE() in __packet_get_status */
 407
 408         h.raw = frame;
 409         switch (po->tp_version) {
 410         case TPACKET_V1:
 411                 WRITE_ONCE(h.h1->tp_status, status);
 412                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 413                 break;
 414         case TPACKET_V2:
 415                 WRITE_ONCE(h.h2->tp_status, status);
 416                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 417                 break;
 418         case TPACKET_V3:
 419                 WRITE_ONCE(h.h3->tp_status, status);
 420                 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
 421                 break;
 422         default:
 423                 WARN(1, "TPACKET version not supported.\n");
 424                 BUG();
 425         }
 426
 427         smp_wmb();
 428 }
 429
 430 static int __packet_get_status(struct packet_sock *po, void *frame)
 431 {
 432         union tpacket_uhdr h;
 433
 434         smp_rmb();
 435
 436         /* READ_ONCE() are paired with WRITE_ONCE() in __packet_set_status */
 437
 438         h.raw = frame;
 439         switch (po->tp_version) {
 440         case TPACKET_V1:
 441                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 442                 return READ_ONCE(h.h1->tp_status);
 443         case TPACKET_V2:
 444                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 445                 return READ_ONCE(h.h2->tp_status);
 446         case TPACKET_V3:
 447                 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
 448                 return READ_ONCE(h.h3->tp_status);
 449         default:
 450                 WARN(1, "TPACKET version not supported.\n");
 451                 BUG();
 452                 return 0;
 453         }
 454 }
 455
 456 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
 457                                    unsigned int flags)
 458 {
 459         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 460
 461         if (shhwtstamps &&
 462             (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
 463             ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
 464                 return TP_STATUS_TS_RAW_HARDWARE;
 465
 466         if (ktime_to_timespec_cond(skb->tstamp, ts))
 467                 return TP_STATUS_TS_SOFTWARE;
 468
 469         return 0;
 470 }
 471
 472 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
 473                                     struct sk_buff *skb)
 474 {
 475         union tpacket_uhdr h;
 476         struct timespec ts;
 477         __u32 ts_status;
 478
 479         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
 480                 return 0;
 481
 482         h.raw = frame;
 483         switch (po->tp_version) {
 484         case TPACKET_V1:
 485                 h.h1->tp_sec = ts.tv_sec;
 486                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
 487                 break;
 488         case TPACKET_V2:
 489                 h.h2->tp_sec = ts.tv_sec;
 490                 h.h2->tp_nsec = ts.tv_nsec;
 491                 break;
 492         case TPACKET_V3:
 493                 h.h3->tp_sec = ts.tv_sec;
 494                 h.h3->tp_nsec = ts.tv_nsec;
 495                 break;
 496         default:
 497                 WARN(1, "TPACKET version not supported.\n");
 498                 BUG();
 499         }
 500
 501         /* one flush is safe, as both fields always lie on the same cacheline */
 502         flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
 503         smp_wmb();
 504
 505         return ts_status;
 506 }
 507
 508 static void *packet_lookup_frame(struct packet_sock *po,
 509                 struct packet_ring_buffer *rb,
 510                 unsigned int position,
 511                 int status)
 512 {
 513         unsigned int pg_vec_pos, frame_offset;
 514         union tpacket_uhdr h;
 515
 516         pg_vec_pos = position / rb->frames_per_block;
 517         frame_offset = position % rb->frames_per_block;
 518
 519         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 520                 (frame_offset * rb->frame_size);
 521
 522         if (status != __packet_get_status(po, h.raw))
 523                 return NULL;
 524
 525         return h.raw;
 526 }
 527
 528 static void *packet_current_frame(struct packet_sock *po,
 529                 struct packet_ring_buffer *rb,
 530                 int status)
 531 {
 532         return packet_lookup_frame(po, rb, rb->head, status);
 533 }
 534
 535 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 536 {
 537         del_timer_sync(&pkc->retire_blk_timer);
 538 }
 539
 540 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 541                 struct sk_buff_head *rb_queue)
 542 {
 543         struct tpacket_kbdq_core *pkc;
 544
 545         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 546
 547         spin_lock_bh(&rb_queue->lock);
 548         pkc->delete_blk_timer = 1;
 549         spin_unlock_bh(&rb_queue->lock);
 550
 551         prb_del_retire_blk_timer(pkc);
 552 }
 553
 554 static void prb_init_blk_timer(struct packet_sock *po,
 555                 struct tpacket_kbdq_core *pkc,
 556                 void (*func) (unsigned long))
 557 {
 558         init_timer(&pkc->retire_blk_timer);
 559         pkc->retire_blk_timer.data = (long)po;
 560         pkc->retire_blk_timer.function = func;
 561         pkc->retire_blk_timer.expires = jiffies;
 562 }
 563
 564 static void prb_setup_retire_blk_timer(struct packet_sock *po)
 565 {
 566         struct tpacket_kbdq_core *pkc;
 567
 568         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 569         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
 570 }
 571
 572 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 573                                 int blk_size_in_bytes)
 574 {
 575         struct net_device *dev;
 576         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 577         struct ethtool_link_ksettings ecmd;
 578         int err;
 579
 580         rtnl_lock();
 581         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 582         if (unlikely(!dev)) {
 583                 rtnl_unlock();
 584                 return DEFAULT_PRB_RETIRE_TOV;
 585         }
 586         err = __ethtool_get_link_ksettings(dev, &ecmd);
 587         rtnl_unlock();
 588         if (!err) {
 589                 /*
 590                  * If the link speed is so slow you don't really
 591                  * need to worry about perf anyways
 592                  */
 593                 if (ecmd.base.speed < SPEED_1000 ||
 594                     ecmd.base.speed == SPEED_UNKNOWN) {
 595                         return DEFAULT_PRB_RETIRE_TOV;
 596                 } else {
 597                         msec = 1;
 598                         div = ecmd.base.speed / 1000;
 599                 }
 600         } else
 601                 return DEFAULT_PRB_RETIRE_TOV;
 602
 603         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 604
 605         if (div)
 606                 mbits /= div;
 607
 608         tmo = mbits * msec;
 609
 610         if (div)
 611                 return tmo+1;
 612         return tmo;
 613 }
 614
 615 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 616                         union tpacket_req_u *req_u)
 617 {
 618         p1->feature_req_word = req_u->req3.tp_feature_req_word;
 619 }
 620
 621 static void init_prb_bdqc(struct packet_sock *po,
 622                         struct packet_ring_buffer *rb,
 623                         struct pgv *pg_vec,
 624                         union tpacket_req_u *req_u)
 625 {
 626         struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
 627         struct tpacket_block_desc *pbd;
 628
 629         memset(p1, 0x0, sizeof(*p1));
 630
 631         p1->knxt_seq_num = 1;
 632         p1->pkbdq = pg_vec;
 633         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 634         p1->pkblk_start = pg_vec[0].buffer;
 635         p1->kblk_size = req_u->req3.tp_block_size;
 636         p1->knum_blocks = req_u->req3.tp_block_nr;
 637         p1->hdrlen = po->tp_hdrlen;
 638         p1->version = po->tp_version;
 639         p1->last_kactive_blk_num = 0;
 640         po->stats.stats3.tp_freeze_q_cnt = 0;
 641         if (req_u->req3.tp_retire_blk_tov)
 642                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 643         else
 644                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 645                                                 req_u->req3.tp_block_size);
 646         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 647         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 648
 649         p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
 650         prb_init_ft_ops(p1, req_u);
 651         prb_setup_retire_blk_timer(po);
 652         prb_open_block(p1, pbd);
 653 }
 654
 655 /*  Do NOT update the last_blk_num first.
 656  *  Assumes sk_buff_head lock is held.
 657  */
 658 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 659 {
 660         mod_timer(&pkc->retire_blk_timer,
 661                         jiffies + pkc->tov_in_jiffies);
 662         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 663 }
 664
 665 /*
 666  * Timer logic:
 667  * 1) We refresh the timer only when we open a block.
 668  *    By doing this we don't waste cycles refreshing the timer
 669  *        on packet-by-packet basis.
 670  *
 671  * With a 1MB block-size, on a 1Gbps line, it will take
 672  * i) ~8 ms to fill a block + ii) memcpy etc.
 673  * In this cut we are not accounting for the memcpy time.
 674  *
 675  * So, if the user sets the 'tmo' to 10ms then the timer
 676  * will never fire while the block is still getting filled
 677  * (which is what we want). However, the user could choose
 678  * to close a block early and that's fine.
 679  *
 680  * But when the timer does fire, we check whether or not to refresh it.
 681  * Since the tmo granularity is in msecs, it is not too expensive
 682  * to refresh the timer, lets say every '8' msecs.
 683  * Either the user can set the 'tmo' or we can derive it based on
 684  * a) line-speed and b) block-size.
 685  * prb_calc_retire_blk_tmo() calculates the tmo.
 686  *
 687  */
 688 static void prb_retire_rx_blk_timer_expired(unsigned long data)
 689 {
 690         struct packet_sock *po = (struct packet_sock *)data;
 691         struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 692         unsigned int frozen;
 693         struct tpacket_block_desc *pbd;
 694
 695         spin_lock(&po->sk.sk_receive_queue.lock);
 696
 697         frozen = prb_queue_frozen(pkc);
 698         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 699
 700         if (unlikely(pkc->delete_blk_timer))
 701                 goto out;
 702
 703         /* We only need to plug the race when the block is partially filled.
 704          * tpacket_rcv:
 705          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 706          *              copy_bits() is in progress ...
 707          *              timer fires on other cpu:
 708          *              we can't retire the current block because copy_bits
 709          *              is in progress.
 710          *
 711          */
 712         if (BLOCK_NUM_PKTS(pbd)) {
 713                 while (atomic_read(&pkc->blk_fill_in_prog)) {
 714                         /* Waiting for skb_copy_bits to finish... */
 715                         cpu_relax();
 716                 }
 717         }
 718
 719         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 720                 if (!frozen) {
 721                         if (!BLOCK_NUM_PKTS(pbd)) {
 722                                 /* An empty block. Just refresh the timer. */
 723                                 goto refresh_timer;
 724                         }
 725                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 726                         if (!prb_dispatch_next_block(pkc, po))
 727                                 goto refresh_timer;
 728                         else
 729                                 goto out;
 730                 } else {
 731                         /* Case 1. Queue was frozen because user-space was
 732                          *         lagging behind.
 733                          */
 734                         if (prb_curr_blk_in_use(pbd)) {
 735                                 /*
 736                                  * Ok, user-space is still behind.
 737                                  * So just refresh the timer.
 738                                  */
 739                                 goto refresh_timer;
 740                         } else {
 741                                /* Case 2. queue was frozen,user-space caught up,
 742                                 * now the link went idle && the timer fired.
 743                                 * We don't have a block to close.So we open this
 744                                 * block and restart the timer.
 745                                 * opening a block thaws the queue,restarts timer
 746                                 * Thawing/timer-refresh is a side effect.
 747                                 */
 748                                 prb_open_block(pkc, pbd);
 749                                 goto out;
 750                         }
 751                 }
 752         }
 753
 754 refresh_timer:
 755         _prb_refresh_rx_retire_blk_timer(pkc);
 756
 757 out:
 758         spin_unlock(&po->sk.sk_receive_queue.lock);
 759 }
 760
 761 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 762                 struct tpacket_block_desc *pbd1, __u32 status)
 763 {
 764         /* Flush everything minus the block header */
 765
 766 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 767         u8 *start, *end;
 768
 769         start = (u8 *)pbd1;
 770
 771         /* Skip the block header(we know header WILL fit in 4K) */
 772         start += PAGE_SIZE;
 773
 774         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 775         for (; start < end; start += PAGE_SIZE)
 776                 flush_dcache_page(pgv_to_page(start));
 777
 778         smp_wmb();
 779 #endif
 780
 781         /* Now update the block status. */
 782
 783         BLOCK_STATUS(pbd1) = status;
 784
 785         /* Flush the block header */
 786
 787 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 788         start = (u8 *)pbd1;
 789         flush_dcache_page(pgv_to_page(start));
 790
 791         smp_wmb();
 792 #endif
 793 }
 794
 795 /*
 796  * Side effect:
 797  *
 798  * 1) flush the block
 799  * 2) Increment active_blk_num
 800  *
 801  * Note:We DONT refresh the timer on purpose.
 802  *      Because almost always the next block will be opened.
 803  */
 804 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 805                 struct tpacket_block_desc *pbd1,
 806                 struct packet_sock *po, unsigned int stat)
 807 {
 808         __u32 status = TP_STATUS_USER | stat;
 809
 810         struct tpacket3_hdr *last_pkt;
 811         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 812         struct sock *sk = &po->sk;
 813
 814         if (po->stats.stats3.tp_drops)
 815                 status |= TP_STATUS_LOSING;
 816
 817         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 818         last_pkt->tp_next_offset = 0;
 819
 820         /* Get the ts of the last pkt */
 821         if (BLOCK_NUM_PKTS(pbd1)) {
 822                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 823                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 824         } else {
 825                 /* Ok, we tmo'd - so get the current time.
 826                  *
 827                  * It shouldn't really happen as we don't close empty
 828                  * blocks. See prb_retire_rx_blk_timer_expired().
 829                  */
 830                 struct timespec ts;
 831                 getnstimeofday(&ts);
 832                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 833                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 834         }
 835
 836         smp_wmb();
 837
 838         /* Flush the block */
 839         prb_flush_block(pkc1, pbd1, status);
 840
 841         sk->sk_data_ready(sk);
 842
 843         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 844 }
 845
 846 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 847 {
 848         pkc->reset_pending_on_curr_blk = 0;
 849 }
 850
 851 /*
 852  * Side effect of opening a block:
 853  *
 854  * 1) prb_queue is thawed.
 855  * 2) retire_blk_timer is refreshed.
 856  *
 857  */
 858 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 859         struct tpacket_block_desc *pbd1)
 860 {
 861         struct timespec ts;
 862         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 863
 864         smp_rmb();
 865
 866         /* We could have just memset this but we will lose the
 867          * flexibility of making the priv area sticky
 868          */
 869
 870         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 871         BLOCK_NUM_PKTS(pbd1) = 0;
 872         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 873
 874         getnstimeofday(&ts);
 875
 876         h1->ts_first_pkt.ts_sec = ts.tv_sec;
 877         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 878
 879         pkc1->pkblk_start = (char *)pbd1;
 880         pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 881
 882         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 883         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 884
 885         pbd1->version = pkc1->version;
 886         pkc1->prev = pkc1->nxt_offset;
 887         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 888
 889         prb_thaw_queue(pkc1);
 890         _prb_refresh_rx_retire_blk_timer(pkc1);
 891
 892         smp_wmb();
 893 }
 894
 895 /*
 896  * Queue freeze logic:
 897  * 1) Assume tp_block_nr = 8 blocks.
 898  * 2) At time 't0', user opens Rx ring.
 899  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 900  * 4) user-space is either sleeping or processing block '0'.
 901  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 902  *    it will close block-7,loop around and try to fill block '0'.
 903  *    call-flow:
 904  *    __packet_lookup_frame_in_block
 905  *      prb_retire_current_block()
 906  *      prb_dispatch_next_block()
 907  *        |->(BLOCK_STATUS == USER) evaluates to true
 908  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 909  * 6) Now there are two cases:
 910  *    6.1) Link goes idle right after the queue is frozen.
 911  *         But remember, the last open_block() refreshed the timer.
 912  *         When this timer expires,it will refresh itself so that we can
 913  *         re-open block-0 in near future.
 914  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 915  *         case and __packet_lookup_frame_in_block will check if block-0
 916  *         is free and can now be re-used.
 917  */
 918 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 919                                   struct packet_sock *po)
 920 {
 921         pkc->reset_pending_on_curr_blk = 1;
 922         po->stats.stats3.tp_freeze_q_cnt++;
 923 }
 924
 925 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 926
 927 /*
 928  * If the next block is free then we will dispatch it
 929  * and return a good offset.
 930  * Else, we will freeze the queue.
 931  * So, caller must check the return value.
 932  */
 933 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 934                 struct packet_sock *po)
 935 {
 936         struct tpacket_block_desc *pbd;
 937
 938         smp_rmb();
 939
 940         /* 1. Get current block num */
 941         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 942
 943         /* 2. If this block is currently in_use then freeze the queue */
 944         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 945                 prb_freeze_queue(pkc, po);
 946                 return NULL;
 947         }
 948
 949         /*
 950          * 3.
 951          * open this block and return the offset where the first packet
 952          * needs to get stored.
 953          */
 954         prb_open_block(pkc, pbd);
 955         return (void *)pkc->nxt_offset;
 956 }
 957
 958 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 959                 struct packet_sock *po, unsigned int status)
 960 {
 961         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 962
 963         /* retire/close the current block */
 964         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 965                 /*
 966                  * Plug the case where copy_bits() is in progress on
 967                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 968                  * have space to copy the pkt in the current block and
 969                  * called prb_retire_current_block()
 970                  *
 971                  * We don't need to worry about the TMO case because
 972                  * the timer-handler already handled this case.
 973                  */
 974                 if (!(status & TP_STATUS_BLK_TMO)) {
 975                         while (atomic_read(&pkc->blk_fill_in_prog)) {
 976                                 /* Waiting for skb_copy_bits to finish... */
 977                                 cpu_relax();
 978                         }
 979                 }
 980                 prb_close_block(pkc, pbd, po, status);
 981                 return;
 982         }
 983 }
 984
 985 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
 986 {
 987         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 988 }
 989
 990 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 991 {
 992         return pkc->reset_pending_on_curr_blk;
 993 }
 994
 995 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 996         __releases(&pkc->blk_fill_in_prog_lock)
 997 {
 998         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 999         atomic_dec(&pkc->blk_fill_in_prog);
1000 }
1001
1002 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
1003                         struct tpacket3_hdr *ppd)
1004 {
1005         ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
1006 }
1007
1008 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
1009                         struct tpacket3_hdr *ppd)
1010 {
1011         ppd->hv1.tp_rxhash = 0;
1012 }
1013
1014 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1015                         struct tpacket3_hdr *ppd)
1016 {
1017         if (skb_vlan_tag_present(pkc->skb)) {
1018                 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1019                 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1020                 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1021         } else {
1022                 ppd->hv1.tp_vlan_tci = 0;
1023                 ppd->hv1.tp_vlan_tpid = 0;
1024                 ppd->tp_status = TP_STATUS_AVAILABLE;
1025         }
1026 }
1027
1028 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1029                         struct tpacket3_hdr *ppd)
1030 {
1031         ppd->hv1.tp_padding = 0;
1032         prb_fill_vlan_info(pkc, ppd);
1033
1034         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1035                 prb_fill_rxhash(pkc, ppd);
1036         else
1037                 prb_clear_rxhash(pkc, ppd);
1038 }
1039
1040 static void prb_fill_curr_block(char *curr,
1041                                 struct tpacket_kbdq_core *pkc,
1042                                 struct tpacket_block_desc *pbd,
1043                                 unsigned int len)
1044         __acquires(&pkc->blk_fill_in_prog_lock)
1045 {
1046         struct tpacket3_hdr *ppd;
1047
1048         ppd  = (struct tpacket3_hdr *)curr;
1049         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1050         pkc->prev = curr;
1051         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1052         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1053         BLOCK_NUM_PKTS(pbd) += 1;
1054         atomic_inc(&pkc->blk_fill_in_prog);
1055         prb_run_all_ft_ops(pkc, ppd);
1056 }
1057
1058 /* Assumes caller has the sk->rx_queue.lock */
1059 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1060                                             struct sk_buff *skb,
1061                                                 int status,
1062                                             unsigned int len
1063                                             )
1064 {
1065         struct tpacket_kbdq_core *pkc;
1066         struct tpacket_block_desc *pbd;
1067         char *curr, *end;
1068
1069         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1070         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1071
1072         /* Queue is frozen when user space is lagging behind */
1073         if (prb_queue_frozen(pkc)) {
1074                 /*
1075                  * Check if that last block which caused the queue to freeze,
1076                  * is still in_use by user-space.
1077                  */
1078                 if (prb_curr_blk_in_use(pbd)) {
1079                         /* Can't record this packet */
1080                         return NULL;
1081                 } else {
1082                         /*
1083                          * Ok, the block was released by user-space.
1084                          * Now let's open that block.
1085                          * opening a block also thaws the queue.
1086                          * Thawing is a side effect.
1087                          */
1088                         prb_open_block(pkc, pbd);
1089                 }
1090         }
1091
1092         smp_mb();
1093         curr = pkc->nxt_offset;
1094         pkc->skb = skb;
1095         end = (char *)pbd + pkc->kblk_size;
1096
1097         /* first try the current block */
1098         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1099                 prb_fill_curr_block(curr, pkc, pbd, len);
1100                 return (void *)curr;
1101         }
1102
1103         /* Ok, close the current block */
1104         prb_retire_current_block(pkc, po, 0);
1105
1106         /* Now, try to dispatch the next block */
1107         curr = (char *)prb_dispatch_next_block(pkc, po);
1108         if (curr) {
1109                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1110                 prb_fill_curr_block(curr, pkc, pbd, len);
1111                 return (void *)curr;
1112         }
1113
1114         /*
1115          * No free blocks are available.user_space hasn't caught up yet.
1116          * Queue was just frozen and now this packet will get dropped.
1117          */
1118         return NULL;
1119 }
1120
1121 static void *packet_current_rx_frame(struct packet_sock *po,
1122                                             struct sk_buff *skb,
1123                                             int status, unsigned int len)
1124 {
1125         char *curr = NULL;
1126         switch (po->tp_version) {
1127         case TPACKET_V1:
1128         case TPACKET_V2:
1129                 curr = packet_lookup_frame(po, &po->rx_ring,
1130                                         po->rx_ring.head, status);
1131                 return curr;
1132         case TPACKET_V3:
1133                 return __packet_lookup_frame_in_block(po, skb, status, len);
1134         default:
1135                 WARN(1, "TPACKET version not supported\n");
1136                 BUG();
1137                 return NULL;
1138         }
1139 }
1140
1141 static void *prb_lookup_block(struct packet_sock *po,
1142                                      struct packet_ring_buffer *rb,
1143                                      unsigned int idx,
1144                                      int status)
1145 {
1146         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1147         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1148
1149         if (status != BLOCK_STATUS(pbd))
1150                 return NULL;
1151         return pbd;
1152 }
1153
1154 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1155 {
1156         unsigned int prev;
1157         if (rb->prb_bdqc.kactive_blk_num)
1158                 prev = rb->prb_bdqc.kactive_blk_num-1;
1159         else
1160                 prev = rb->prb_bdqc.knum_blocks-1;
1161         return prev;
1162 }
1163
1164 /* Assumes caller has held the rx_queue.lock */
1165 static void *__prb_previous_block(struct packet_sock *po,
1166                                          struct packet_ring_buffer *rb,
1167                                          int status)
1168 {
1169         unsigned int previous = prb_previous_blk_num(rb);
1170         return prb_lookup_block(po, rb, previous, status);
1171 }
1172
1173 static void *packet_previous_rx_frame(struct packet_sock *po,
1174                                              struct packet_ring_buffer *rb,
1175                                              int status)
1176 {
1177         if (po->tp_version <= TPACKET_V2)
1178                 return packet_previous_frame(po, rb, status);
1179
1180         return __prb_previous_block(po, rb, status);
1181 }
1182
1183 static void packet_increment_rx_head(struct packet_sock *po,
1184                                             struct packet_ring_buffer *rb)
1185 {
1186         switch (po->tp_version) {
1187         case TPACKET_V1:
1188         case TPACKET_V2:
1189                 return packet_increment_head(rb);
1190         case TPACKET_V3:
1191         default:
1192                 WARN(1, "TPACKET version not supported.\n");
1193                 BUG();
1194                 return;
1195         }
1196 }
1197
1198 static void *packet_previous_frame(struct packet_sock *po,
1199                 struct packet_ring_buffer *rb,
1200                 int status)
1201 {
1202         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1203         return packet_lookup_frame(po, rb, previous, status);
1204 }
1205
1206 static void packet_increment_head(struct packet_ring_buffer *buff)
1207 {
1208         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1209 }
1210
1211 static void packet_inc_pending(struct packet_ring_buffer *rb)
1212 {
1213         this_cpu_inc(*rb->pending_refcnt);
1214 }
1215
1216 static void packet_dec_pending(struct packet_ring_buffer *rb)
1217 {
1218         this_cpu_dec(*rb->pending_refcnt);
1219 }
1220
1221 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1222 {
1223         unsigned int refcnt = 0;
1224         int cpu;
1225
1226         /* We don't use pending refcount in rx_ring. */
1227         if (rb->pending_refcnt == NULL)
1228                 return 0;
1229
1230         for_each_possible_cpu(cpu)
1231                 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1232
1233         return refcnt;
1234 }
1235
1236 static int packet_alloc_pending(struct packet_sock *po)
1237 {
1238         po->rx_ring.pending_refcnt = NULL;
1239
1240         po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1241         if (unlikely(po->tx_ring.pending_refcnt == NULL))
1242                 return -ENOBUFS;
1243
1244         return 0;
1245 }
1246
1247 static void packet_free_pending(struct packet_sock *po)
1248 {
1249         free_percpu(po->tx_ring.pending_refcnt);
1250 }
1251
1252 #define ROOM_POW_OFF    2
1253 #define ROOM_NONE       0x0
1254 #define ROOM_LOW        0x1
1255 #define ROOM_NORMAL     0x2
1256
1257 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1258 {
1259         int idx, len;
1260
1261         len = po->rx_ring.frame_max + 1;
1262         idx = po->rx_ring.head;
1263         if (pow_off)
1264                 idx += len >> pow_off;
1265         if (idx >= len)
1266                 idx -= len;
1267         return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1268 }
1269
1270 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1271 {
1272         int idx, len;
1273
1274         len = po->rx_ring.prb_bdqc.knum_blocks;
1275         idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1276         if (pow_off)
1277                 idx += len >> pow_off;
1278         if (idx >= len)
1279                 idx -= len;
1280         return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1281 }
1282
1283 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1284 {
1285         struct sock *sk = &po->sk;
1286         int ret = ROOM_NONE;
1287
1288         if (po->prot_hook.func != tpacket_rcv) {
1289                 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1290                                           - (skb ? skb->truesize : 0);
1291                 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1292                         return ROOM_NORMAL;
1293                 else if (avail > 0)
1294                         return ROOM_LOW;
1295                 else
1296                         return ROOM_NONE;
1297         }
1298
1299         if (po->tp_version == TPACKET_V3) {
1300                 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1301                         ret = ROOM_NORMAL;
1302                 else if (__tpacket_v3_has_room(po, 0))
1303                         ret = ROOM_LOW;
1304         } else {
1305                 if (__tpacket_has_room(po, ROOM_POW_OFF))
1306                         ret = ROOM_NORMAL;
1307                 else if (__tpacket_has_room(po, 0))
1308                         ret = ROOM_LOW;
1309         }
1310
1311         return ret;
1312 }
1313
1314 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1315 {
1316         int ret;
1317         bool has_room;
1318
1319         spin_lock_bh(&po->sk.sk_receive_queue.lock);
1320         ret = __packet_rcv_has_room(po, skb);
1321         has_room = ret == ROOM_NORMAL;
1322         if (po->pressure == has_room)
1323                 po->pressure = !has_room;
1324         spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1325
1326         return ret;
1327 }
1328
1329 static void packet_sock_destruct(struct sock *sk)
1330 {
1331         skb_queue_purge(&sk->sk_error_queue);
1332
1333         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1334         WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1335
1336         if (!sock_flag(sk, SOCK_DEAD)) {
1337                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1338                 return;
1339         }
1340
1341         sk_refcnt_debug_dec(sk);
1342 }
1343
1344 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1345 {
1346         u32 *history = po->rollover->history;
1347         u32 victim, rxhash;
1348         int i, count = 0;
1349
1350         rxhash = skb_get_hash(skb);
1351         for (i = 0; i < ROLLOVER_HLEN; i++)
1352                 if (READ_ONCE(history[i]) == rxhash)
1353                         count++;
1354
1355         victim = prandom_u32() % ROLLOVER_HLEN;
1356
1357         /* Avoid dirtying the cache line if possible */
1358         if (READ_ONCE(history[victim]) != rxhash)
1359                 WRITE_ONCE(history[victim], rxhash);
1360
1361         return count > (ROLLOVER_HLEN >> 1);
1362 }
1363
1364 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1365                                       struct sk_buff *skb,
1366                                       unsigned int num)
1367 {
1368         return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1369 }
1370
1371 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1372                                     struct sk_buff *skb,
1373                                     unsigned int num)
1374 {
1375         unsigned int val = atomic_inc_return(&f->rr_cur);
1376
1377         return val % num;
1378 }
1379
1380 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1381                                      struct sk_buff *skb,
1382                                      unsigned int num)
1383 {
1384         return smp_processor_id() % num;
1385 }
1386
1387 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1388                                      struct sk_buff *skb,
1389                                      unsigned int num)
1390 {
1391         return prandom_u32_max(num);
1392 }
1393
1394 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1395                                           struct sk_buff *skb,
1396                                           unsigned int idx, bool try_self,
1397                                           unsigned int num)
1398 {
1399         struct packet_sock *po, *po_next, *po_skip = NULL;
1400         unsigned int i, j, room = ROOM_NONE;
1401
1402         po = pkt_sk(f->arr[idx]);
1403
1404         if (try_self) {
1405                 room = packet_rcv_has_room(po, skb);
1406                 if (room == ROOM_NORMAL ||
1407                     (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1408                         return idx;
1409                 po_skip = po;
1410         }
1411
1412         i = j = min_t(int, po->rollover->sock, num - 1);
1413         do {
1414                 po_next = pkt_sk(f->arr[i]);
1415                 if (po_next != po_skip && !po_next->pressure &&
1416                     packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1417                         if (i != j)
1418                                 po->rollover->sock = i;
1419                         atomic_long_inc(&po->rollover->num);
1420                         if (room == ROOM_LOW)
1421                                 atomic_long_inc(&po->rollover->num_huge);
1422                         return i;
1423                 }
1424
1425                 if (++i == num)
1426                         i = 0;
1427         } while (i != j);
1428
1429         atomic_long_inc(&po->rollover->num_failed);
1430         return idx;
1431 }
1432
1433 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1434                                     struct sk_buff *skb,
1435                                     unsigned int num)
1436 {
1437         return skb_get_queue_mapping(skb) % num;
1438 }
1439
1440 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1441                                      struct sk_buff *skb,
1442                                      unsigned int num)
1443 {
1444         struct bpf_prog *prog;
1445         unsigned int ret = 0;
1446
1447         rcu_read_lock();
1448         prog = rcu_dereference(f->bpf_prog);
1449         if (prog)
1450                 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1451         rcu_read_unlock();
1452
1453         return ret;
1454 }
1455
1456 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1457 {
1458         return f->flags & (flag >> 8);
1459 }
1460
1461 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1462                              struct packet_type *pt, struct net_device *orig_dev)
1463 {
1464         struct packet_fanout *f = pt->af_packet_priv;
1465         unsigned int num = READ_ONCE(f->num_members);
1466         struct net *net = read_pnet(&f->net);
1467         struct packet_sock *po;
1468         unsigned int idx;
1469
1470         if (!net_eq(dev_net(dev), net) || !num) {
1471                 kfree_skb(skb);
1472                 return 0;
1473         }
1474
1475         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1476                 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1477                 if (!skb)
1478                         return 0;
1479         }
1480         switch (f->type) {
1481         case PACKET_FANOUT_HASH:
1482         default:
1483                 idx = fanout_demux_hash(f, skb, num);
1484                 break;
1485         case PACKET_FANOUT_LB:
1486                 idx = fanout_demux_lb(f, skb, num);
1487                 break;
1488         case PACKET_FANOUT_CPU:
1489                 idx = fanout_demux_cpu(f, skb, num);
1490                 break;
1491         case PACKET_FANOUT_RND:
1492                 idx = fanout_demux_rnd(f, skb, num);
1493                 break;
1494         case PACKET_FANOUT_QM:
1495                 idx = fanout_demux_qm(f, skb, num);
1496                 break;
1497         case PACKET_FANOUT_ROLLOVER:
1498                 idx = fanout_demux_rollover(f, skb, 0, false, num);
1499                 break;
1500         case PACKET_FANOUT_CBPF:
1501         case PACKET_FANOUT_EBPF:
1502                 idx = fanout_demux_bpf(f, skb, num);
1503                 break;
1504         }
1505
1506         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1507                 idx = fanout_demux_rollover(f, skb, idx, true, num);
1508
1509         po = pkt_sk(f->arr[idx]);
1510         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1511 }
1512
1513 DEFINE_MUTEX(fanout_mutex);
1514 EXPORT_SYMBOL_GPL(fanout_mutex);
1515 static LIST_HEAD(fanout_list);
1516 static u16 fanout_next_id;
1517
1518 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1519 {
1520         struct packet_fanout *f = po->fanout;
1521
1522         spin_lock(&f->lock);
1523         f->arr[f->num_members] = sk;
1524         smp_wmb();
1525         f->num_members++;
1526         if (f->num_members == 1)
1527                 dev_add_pack(&f->prot_hook);
1528         spin_unlock(&f->lock);
1529 }
1530
1531 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1532 {
1533         struct packet_fanout *f = po->fanout;
1534         int i;
1535
1536         spin_lock(&f->lock);
1537         for (i = 0; i < f->num_members; i++) {
1538                 if (f->arr[i] == sk)
1539                         break;
1540         }
1541         BUG_ON(i >= f->num_members);
1542         f->arr[i] = f->arr[f->num_members - 1];
1543         f->num_members--;
1544         if (f->num_members == 0)
1545                 __dev_remove_pack(&f->prot_hook);
1546         spin_unlock(&f->lock);
1547 }
1548
1549 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1550 {
1551         if (sk->sk_family != PF_PACKET)
1552                 return false;
1553
1554         return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1555 }
1556
1557 static void fanout_init_data(struct packet_fanout *f)
1558 {
1559         switch (f->type) {
1560         case PACKET_FANOUT_LB:
1561                 atomic_set(&f->rr_cur, 0);
1562                 break;
1563         case PACKET_FANOUT_CBPF:
1564         case PACKET_FANOUT_EBPF:
1565                 RCU_INIT_POINTER(f->bpf_prog, NULL);
1566                 break;
1567         }
1568 }
1569
1570 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1571 {
1572         struct bpf_prog *old;
1573
1574         spin_lock(&f->lock);
1575         old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1576         rcu_assign_pointer(f->bpf_prog, new);
1577         spin_unlock(&f->lock);
1578
1579         if (old) {
1580                 synchronize_net();
1581                 bpf_prog_destroy(old);
1582         }
1583 }
1584
1585 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1586                                 unsigned int len)
1587 {
1588         struct bpf_prog *new;
1589         struct sock_fprog fprog;
1590         int ret;
1591
1592         if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1593                 return -EPERM;
1594         if (len != sizeof(fprog))
1595                 return -EINVAL;
1596         if (copy_from_user(&fprog, data, len))
1597                 return -EFAULT;
1598
1599         ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1600         if (ret)
1601                 return ret;
1602
1603         __fanout_set_data_bpf(po->fanout, new);
1604         return 0;
1605 }
1606
1607 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1608                                 unsigned int len)
1609 {
1610         struct bpf_prog *new;
1611         u32 fd;
1612
1613         if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1614                 return -EPERM;
1615         if (len != sizeof(fd))
1616                 return -EINVAL;
1617         if (copy_from_user(&fd, data, len))
1618                 return -EFAULT;
1619
1620         new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1621         if (IS_ERR(new))
1622                 return PTR_ERR(new);
1623
1624         __fanout_set_data_bpf(po->fanout, new);
1625         return 0;
1626 }
1627
1628 static int fanout_set_data(struct packet_sock *po, char __user *data,
1629                            unsigned int len)
1630 {
1631         switch (po->fanout->type) {
1632         case PACKET_FANOUT_CBPF:
1633                 return fanout_set_data_cbpf(po, data, len);
1634         case PACKET_FANOUT_EBPF:
1635                 return fanout_set_data_ebpf(po, data, len);
1636         default:
1637                 return -EINVAL;
1638         };
1639 }
1640
1641 static void fanout_release_data(struct packet_fanout *f)
1642 {
1643         switch (f->type) {
1644         case PACKET_FANOUT_CBPF:
1645         case PACKET_FANOUT_EBPF:
1646                 __fanout_set_data_bpf(f, NULL);
1647         };
1648 }
1649
1650 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1651 {
1652         struct packet_fanout *f;
1653
1654         list_for_each_entry(f, &fanout_list, list) {
1655                 if (f->id == candidate_id &&
1656                     read_pnet(&f->net) == sock_net(sk)) {
1657                         return false;
1658                 }
1659         }
1660         return true;
1661 }
1662
1663 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1664 {
1665         u16 id = fanout_next_id;
1666
1667         do {
1668                 if (__fanout_id_is_free(sk, id)) {
1669                         *new_id = id;
1670                         fanout_next_id = id + 1;
1671                         return true;
1672                 }
1673
1674                 id++;
1675         } while (id != fanout_next_id);
1676
1677         return false;
1678 }
1679
1680 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1681 {
1682         struct packet_rollover *rollover = NULL;
1683         struct packet_sock *po = pkt_sk(sk);
1684         struct packet_fanout *f, *match;
1685         u8 type = type_flags & 0xff;
1686         u8 flags = type_flags >> 8;
1687         int err;
1688
1689         switch (type) {
1690         case PACKET_FANOUT_ROLLOVER:
1691                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1692                         return -EINVAL;
1693         case PACKET_FANOUT_HASH:
1694         case PACKET_FANOUT_LB:
1695         case PACKET_FANOUT_CPU:
1696         case PACKET_FANOUT_RND:
1697         case PACKET_FANOUT_QM:
1698         case PACKET_FANOUT_CBPF:
1699         case PACKET_FANOUT_EBPF:
1700                 break;
1701         default:
1702                 return -EINVAL;
1703         }
1704
1705         mutex_lock(&fanout_mutex);
1706
1707         err = -EALREADY;
1708         if (po->fanout)
1709                 goto out;
1710
1711         if (type == PACKET_FANOUT_ROLLOVER ||
1712             (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1713                 err = -ENOMEM;
1714                 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1715                 if (!rollover)
1716                         goto out;
1717                 atomic_long_set(&rollover->num, 0);
1718                 atomic_long_set(&rollover->num_huge, 0);
1719                 atomic_long_set(&rollover->num_failed, 0);
1720         }
1721
1722         if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1723                 if (id != 0) {
1724                         err = -EINVAL;
1725                         goto out;
1726                 }
1727                 if (!fanout_find_new_id(sk, &id)) {
1728                         err = -ENOMEM;
1729                         goto out;
1730                 }
1731                 /* ephemeral flag for the first socket in the group: drop it */
1732                 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1733         }
1734
1735         match = NULL;
1736         list_for_each_entry(f, &fanout_list, list) {
1737                 if (f->id == id &&
1738                     read_pnet(&f->net) == sock_net(sk)) {
1739                         match = f;
1740                         break;
1741                 }
1742         }
1743         err = -EINVAL;
1744         if (match && match->flags != flags)
1745                 goto out;
1746         if (!match) {
1747                 err = -ENOMEM;
1748                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1749                 if (!match)
1750                         goto out;
1751                 write_pnet(&match->net, sock_net(sk));
1752                 match->id = id;
1753                 match->type = type;
1754                 match->flags = flags;
1755                 INIT_LIST_HEAD(&match->list);
1756                 spin_lock_init(&match->lock);
1757                 refcount_set(&match->sk_ref, 0);
1758                 fanout_init_data(match);
1759                 match->prot_hook.type = po->prot_hook.type;
1760                 match->prot_hook.dev = po->prot_hook.dev;
1761                 match->prot_hook.func = packet_rcv_fanout;
1762                 match->prot_hook.af_packet_priv = match;
1763                 match->prot_hook.af_packet_net = read_pnet(&match->net);
1764                 match->prot_hook.id_match = match_fanout_group;
1765                 list_add(&match->list, &fanout_list);
1766         }
1767         err = -EINVAL;
1768
1769         spin_lock(&po->bind_lock);
1770         if (po->running &&
1771             match->type == type &&
1772             match->prot_hook.type == po->prot_hook.type &&
1773             match->prot_hook.dev == po->prot_hook.dev) {
1774                 err = -ENOSPC;
1775                 if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1776                         __dev_remove_pack(&po->prot_hook);
1777
1778                         /* Paired with packet_setsockopt(PACKET_FANOUT_DATA) */
1779                         WRITE_ONCE(po->fanout, match);
1780
1781                         po->rollover = rollover;
1782                         rollover = NULL;
1783                         refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1784                         __fanout_link(sk, po);
1785                         err = 0;
1786                 }
1787         }
1788         spin_unlock(&po->bind_lock);
1789
1790         if (err && !refcount_read(&match->sk_ref)) {
1791                 list_del(&match->list);
1792                 kfree(match);
1793         }
1794
1795 out:
1796         kfree(rollover);
1797         mutex_unlock(&fanout_mutex);
1798         return err;
1799 }
1800
1801 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1802  * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1803  * It is the responsibility of the caller to call fanout_release_data() and
1804  * free the returned packet_fanout (after synchronize_net())
1805  */
1806 static struct packet_fanout *fanout_release(struct sock *sk)
1807 {
1808         struct packet_sock *po = pkt_sk(sk);
1809         struct packet_fanout *f;
1810
1811         mutex_lock(&fanout_mutex);
1812         f = po->fanout;
1813         if (f) {
1814                 po->fanout = NULL;
1815
1816                 if (refcount_dec_and_test(&f->sk_ref))
1817                         list_del(&f->list);
1818                 else
1819                         f = NULL;
1820         }
1821         mutex_unlock(&fanout_mutex);
1822
1823         return f;
1824 }
1825
1826 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1827                                           struct sk_buff *skb)
1828 {
1829         /* Earlier code assumed this would be a VLAN pkt, double-check
1830          * this now that we have the actual packet in hand. We can only
1831          * do this check on Ethernet devices.
1832          */
1833         if (unlikely(dev->type != ARPHRD_ETHER))
1834                 return false;
1835
1836         skb_reset_mac_header(skb);
1837         return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1838 }
1839
1840 static const struct proto_ops packet_ops;
1841
1842 static const struct proto_ops packet_ops_spkt;
1843
1844 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1845                            struct packet_type *pt, struct net_device *orig_dev)
1846 {
1847         struct sock *sk;
1848         struct sockaddr_pkt *spkt;
1849
1850         /*
1851          *      When we registered the protocol we saved the socket in the data
1852          *      field for just this event.
1853          */
1854
1855         sk = pt->af_packet_priv;
1856
1857         /*
1858          *      Yank back the headers [hope the device set this
1859          *      right or kerboom...]
1860          *
1861          *      Incoming packets have ll header pulled,
1862          *      push it back.
1863          *
1864          *      For outgoing ones skb->data == skb_mac_header(skb)
1865          *      so that this procedure is noop.
1866          */
1867
1868         if (skb->pkt_type == PACKET_LOOPBACK)
1869                 goto out;
1870
1871         if (!net_eq(dev_net(dev), sock_net(sk)))
1872                 goto out;
1873
1874         skb = skb_share_check(skb, GFP_ATOMIC);
1875         if (skb == NULL)
1876                 goto oom;
1877
1878         /* drop any routing info */
1879         skb_dst_drop(skb);
1880
1881         /* drop conntrack reference */
1882         nf_reset(skb);
1883
1884         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1885
1886         skb_push(skb, skb->data - skb_mac_header(skb));
1887
1888         /*
1889          *      The SOCK_PACKET socket receives _all_ frames.
1890          */
1891
1892         spkt->spkt_family = dev->type;
1893         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1894         spkt->spkt_protocol = skb->protocol;
1895
1896         /*
1897          *      Charge the memory to the socket. This is done specifically
1898          *      to prevent sockets using all the memory up.
1899          */
1900
1901         if (sock_queue_rcv_skb(sk, skb) == 0)
1902                 return 0;
1903
1904 out:
1905         kfree_skb(skb);
1906 oom:
1907         return 0;
1908 }
1909
1910
1911 /*
1912  *      Output a raw packet to a device layer. This bypasses all the other
1913  *      protocol layers and you must therefore supply it with a complete frame
1914  */
1915
1916 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1917                                size_t len)
1918 {
1919         struct sock *sk = sock->sk;
1920         DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1921         struct sk_buff *skb = NULL;
1922         struct net_device *dev;
1923         struct sockcm_cookie sockc;
1924         __be16 proto = 0;
1925         int err;
1926         int extra_len = 0;
1927
1928         /*
1929          *      Get and verify the address.
1930          */
1931
1932         if (saddr) {
1933                 if (msg->msg_namelen < sizeof(struct sockaddr))
1934                         return -EINVAL;
1935                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1936                         proto = saddr->spkt_protocol;
1937         } else
1938                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1939
1940         /*
1941          *      Find the device first to size check it
1942          */
1943
1944         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1945 retry:
1946         rcu_read_lock();
1947         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1948         err = -ENODEV;
1949         if (dev == NULL)
1950                 goto out_unlock;
1951
1952         err = -ENETDOWN;
1953         if (!(dev->flags & IFF_UP))
1954                 goto out_unlock;
1955
1956         /*
1957          * You may not queue a frame bigger than the mtu. This is the lowest level
1958          * raw protocol and you must do your own fragmentation at this level.
1959          */
1960
1961         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1962                 if (!netif_supports_nofcs(dev)) {
1963                         err = -EPROTONOSUPPORT;
1964                         goto out_unlock;
1965                 }
1966                 extra_len = 4; /* We're doing our own CRC */
1967         }
1968
1969         err = -EMSGSIZE;
1970         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1971                 goto out_unlock;
1972
1973         if (!skb) {
1974                 size_t reserved = LL_RESERVED_SPACE(dev);
1975                 int tlen = dev->needed_tailroom;
1976                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1977
1978                 rcu_read_unlock();
1979                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1980                 if (skb == NULL)
1981                         return -ENOBUFS;
1982                 /* FIXME: Save some space for broken drivers that write a hard
1983                  * header at transmission time by themselves. PPP is the notable
1984                  * one here. This should really be fixed at the driver level.
1985                  */
1986                 skb_reserve(skb, reserved);
1987                 skb_reset_network_header(skb);
1988
1989                 /* Try to align data part correctly */
1990                 if (hhlen) {
1991                         skb->data -= hhlen;
1992                         skb->tail -= hhlen;
1993                         if (len < hhlen)
1994                                 skb_reset_network_header(skb);
1995                 }
1996                 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1997                 if (err)
1998                         goto out_free;
1999                 goto retry;
2000         }
2001
2002         if (!dev_validate_header(dev, skb->data, len) || !skb->len) {
2003                 err = -EINVAL;
2004                 goto out_unlock;
2005         }
2006         if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
2007             !packet_extra_vlan_len_allowed(dev, skb)) {
2008                 err = -EMSGSIZE;
2009                 goto out_unlock;
2010         }
2011
2012         sockc.tsflags = sk->sk_tsflags;
2013         if (msg->msg_controllen) {
2014                 err = sock_cmsg_send(sk, msg, &sockc);
2015                 if (unlikely(err))
2016                         goto out_unlock;
2017         }
2018
2019         skb->protocol = proto;
2020         skb->dev = dev;
2021         skb->priority = sk->sk_priority;
2022         skb->mark = sk->sk_mark;
2023
2024         skb_setup_tx_timestamp(skb, sockc.tsflags);
2025
2026         if (unlikely(extra_len == 4))
2027                 skb->no_fcs = 1;
2028
2029         skb_probe_transport_header(skb, 0);
2030
2031         dev_queue_xmit(skb);
2032         rcu_read_unlock();
2033         return len;
2034
2035 out_unlock:
2036         rcu_read_unlock();
2037 out_free:
2038         kfree_skb(skb);
2039         return err;
2040 }
2041
2042 static unsigned int run_filter(struct sk_buff *skb,
2043                                const struct sock *sk,
2044                                unsigned int res)
2045 {
2046         struct sk_filter *filter;
2047
2048         rcu_read_lock();
2049         filter = rcu_dereference(sk->sk_filter);
2050         if (filter != NULL)
2051                 res = bpf_prog_run_clear_cb(filter->prog, skb);
2052         rcu_read_unlock();
2053
2054         return res;
2055 }
2056
2057 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2058                            size_t *len)
2059 {
2060         struct virtio_net_hdr vnet_hdr;
2061
2062         if (*len < sizeof(vnet_hdr))
2063                 return -EINVAL;
2064         *len -= sizeof(vnet_hdr);
2065
2066         if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2067                 return -EINVAL;
2068
2069         return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2070 }
2071
2072 /*
2073  * This function makes lazy skb cloning in hope that most of packets
2074  * are discarded by BPF.
2075  *
2076  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2077  * and skb->cb are mangled. It works because (and until) packets
2078  * falling here are owned by current CPU. Output packets are cloned
2079  * by dev_queue_xmit_nit(), input packets are processed by net_bh
2080  * sequencially, so that if we return skb to original state on exit,
2081  * we will not harm anyone.
2082  */
2083
2084 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2085                       struct packet_type *pt, struct net_device *orig_dev)
2086 {
2087         struct sock *sk;
2088         struct sockaddr_ll *sll;
2089         struct packet_sock *po;
2090         u8 *skb_head = skb->data;
2091         int skb_len = skb->len;
2092         unsigned int snaplen, res;
2093         bool is_drop_n_account = false;
2094
2095         if (skb->pkt_type == PACKET_LOOPBACK)
2096                 goto drop;
2097
2098         sk = pt->af_packet_priv;
2099         po = pkt_sk(sk);
2100
2101         if (!net_eq(dev_net(dev), sock_net(sk)))
2102                 goto drop;
2103
2104         skb->dev = dev;
2105
2106         if (dev->header_ops) {
2107                 /* The device has an explicit notion of ll header,
2108                  * exported to higher levels.
2109                  *
2110                  * Otherwise, the device hides details of its frame
2111                  * structure, so that corresponding packet head is
2112                  * never delivered to user.
2113                  */
2114                 if (sk->sk_type != SOCK_DGRAM)
2115                         skb_push(skb, skb->data - skb_mac_header(skb));
2116                 else if (skb->pkt_type == PACKET_OUTGOING) {
2117                         /* Special case: outgoing packets have ll header at head */
2118                         skb_pull(skb, skb_network_offset(skb));
2119                 }
2120         }
2121
2122         snaplen = skb->len;
2123
2124         res = run_filter(skb, sk, snaplen);
2125         if (!res)
2126                 goto drop_n_restore;
2127         if (snaplen > res)
2128                 snaplen = res;
2129
2130         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2131                 goto drop_n_acct;
2132
2133         if (skb_shared(skb)) {
2134                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2135                 if (nskb == NULL)
2136                         goto drop_n_acct;
2137
2138                 if (skb_head != skb->data) {
2139                         skb->data = skb_head;
2140                         skb->len = skb_len;
2141                 }
2142                 consume_skb(skb);
2143                 skb = nskb;
2144         }
2145
2146         sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2147
2148         sll = &PACKET_SKB_CB(skb)->sa.ll;
2149         sll->sll_hatype = dev->type;
2150         sll->sll_pkttype = skb->pkt_type;
2151         if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV)))
2152                 sll->sll_ifindex = orig_dev->ifindex;
2153         else
2154                 sll->sll_ifindex = dev->ifindex;
2155
2156         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2157
2158         /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2159          * Use their space for storing the original skb length.
2160          */
2161         PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2162
2163         if (pskb_trim(skb, snaplen))
2164                 goto drop_n_acct;
2165
2166         skb_set_owner_r(skb, sk);
2167         skb->dev = NULL;
2168         skb_dst_drop(skb);
2169
2170         /* drop conntrack reference */
2171         nf_reset(skb);
2172
2173         spin_lock(&sk->sk_receive_queue.lock);
2174         po->stats.stats1.tp_packets++;
2175         sock_skb_set_dropcount(sk, skb);
2176         __skb_queue_tail(&sk->sk_receive_queue, skb);
2177         spin_unlock(&sk->sk_receive_queue.lock);
2178         sk->sk_data_ready(sk);
2179         return 0;
2180
2181 drop_n_acct:
2182         is_drop_n_account = true;
2183         spin_lock(&sk->sk_receive_queue.lock);
2184         po->stats.stats1.tp_drops++;
2185         atomic_inc(&sk->sk_drops);
2186         spin_unlock(&sk->sk_receive_queue.lock);
2187
2188 drop_n_restore:
2189         if (skb_head != skb->data && skb_shared(skb)) {
2190                 skb->data = skb_head;
2191                 skb->len = skb_len;
2192         }
2193 drop:
2194         if (!is_drop_n_account)
2195                 consume_skb(skb);
2196         else
2197                 kfree_skb(skb);
2198         return 0;
2199 }
2200
2201 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2202                        struct packet_type *pt, struct net_device *orig_dev)
2203 {
2204         struct sock *sk;
2205         struct packet_sock *po;
2206         struct sockaddr_ll *sll;
2207         union tpacket_uhdr h;
2208         u8 *skb_head = skb->data;
2209         int skb_len = skb->len;
2210         unsigned int snaplen, res;
2211         unsigned long status = TP_STATUS_USER;
2212         unsigned short macoff, hdrlen;
2213         unsigned int netoff;
2214         struct sk_buff *copy_skb = NULL;
2215         struct timespec ts;
2216         __u32 ts_status;
2217         bool is_drop_n_account = false;
2218         unsigned int slot_id = 0;
2219         bool do_vnet = false;
2220
2221         /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2222          * We may add members to them until current aligned size without forcing
2223          * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2224          */
2225         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2226         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2227
2228         if (skb->pkt_type == PACKET_LOOPBACK)
2229                 goto drop;
2230
2231         sk = pt->af_packet_priv;
2232         po = pkt_sk(sk);
2233
2234         if (!net_eq(dev_net(dev), sock_net(sk)))
2235                 goto drop;
2236
2237         if (dev->header_ops) {
2238                 if (sk->sk_type != SOCK_DGRAM)
2239                         skb_push(skb, skb->data - skb_mac_header(skb));
2240                 else if (skb->pkt_type == PACKET_OUTGOING) {
2241                         /* Special case: outgoing packets have ll header at head */
2242                         skb_pull(skb, skb_network_offset(skb));
2243                 }
2244         }
2245
2246         snaplen = skb->len;
2247
2248         res = run_filter(skb, sk, snaplen);
2249         if (!res)
2250                 goto drop_n_restore;
2251
2252         if (skb->ip_summed == CHECKSUM_PARTIAL)
2253                 status |= TP_STATUS_CSUMNOTREADY;
2254         else if (skb->pkt_type != PACKET_OUTGOING &&
2255                  skb_csum_unnecessary(skb))
2256                 status |= TP_STATUS_CSUM_VALID;
2257
2258         if (snaplen > res)
2259                 snaplen = res;
2260
2261         if (sk->sk_type == SOCK_DGRAM) {
2262                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2263                                   po->tp_reserve;
2264         } else {
2265                 unsigned int maclen = skb_network_offset(skb);
2266                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2267                                        (maclen < 16 ? 16 : maclen)) +
2268                                        po->tp_reserve;
2269                 if (po->has_vnet_hdr) {
2270                         netoff += sizeof(struct virtio_net_hdr);
2271                         do_vnet = true;
2272                 }
2273                 macoff = netoff - maclen;
2274         }
2275         if (netoff > USHRT_MAX) {
2276                 spin_lock(&sk->sk_receive_queue.lock);
2277                 po->stats.stats1.tp_drops++;
2278                 spin_unlock(&sk->sk_receive_queue.lock);
2279                 goto drop_n_restore;
2280         }
2281         if (po->tp_version <= TPACKET_V2) {
2282                 if (macoff + snaplen > po->rx_ring.frame_size) {
2283                         if (po->copy_thresh &&
2284                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2285                                 if (skb_shared(skb)) {
2286                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
2287                                 } else {
2288                                         copy_skb = skb_get(skb);
2289                                         skb_head = skb->data;
2290                                 }
2291                                 if (copy_skb) {
2292                                         memset(&PACKET_SKB_CB(copy_skb)->sa.ll, 0,
2293                                                sizeof(PACKET_SKB_CB(copy_skb)->sa.ll));
2294                                         skb_set_owner_r(copy_skb, sk);
2295                                 }
2296                         }
2297                         snaplen = po->rx_ring.frame_size - macoff;
2298                         if ((int)snaplen < 0) {
2299                                 snaplen = 0;
2300                                 do_vnet = false;
2301                         }
2302                 }
2303         } else if (unlikely(macoff + snaplen >
2304                             GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2305                 u32 nval;
2306
2307                 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2308                 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2309                             snaplen, nval, macoff);
2310                 snaplen = nval;
2311                 if (unlikely((int)snaplen < 0)) {
2312                         snaplen = 0;
2313                         macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2314                         do_vnet = false;
2315                 }
2316         }
2317         spin_lock(&sk->sk_receive_queue.lock);
2318         h.raw = packet_current_rx_frame(po, skb,
2319                                         TP_STATUS_KERNEL, (macoff+snaplen));
2320         if (!h.raw)
2321                 goto drop_n_account;
2322
2323         if (po->tp_version <= TPACKET_V2) {
2324                 slot_id = po->rx_ring.head;
2325                 if (test_bit(slot_id, po->rx_ring.rx_owner_map))
2326                         goto drop_n_account;
2327                 __set_bit(slot_id, po->rx_ring.rx_owner_map);
2328         }
2329
2330         if (do_vnet &&
2331             virtio_net_hdr_from_skb(skb, h.raw + macoff -
2332                                     sizeof(struct virtio_net_hdr),
2333                                     vio_le(), true, 0)) {
2334                 if (po->tp_version == TPACKET_V3)
2335                         prb_clear_blk_fill_status(&po->rx_ring);
2336                 goto drop_n_account;
2337         }
2338
2339         if (po->tp_version <= TPACKET_V2) {
2340                 packet_increment_rx_head(po, &po->rx_ring);
2341         /*
2342          * LOSING will be reported till you read the stats,
2343          * because it's COR - Clear On Read.
2344          * Anyways, moving it for V1/V2 only as V3 doesn't need this
2345          * at packet level.
2346          */
2347                 if (po->stats.stats1.tp_drops)
2348                         status |= TP_STATUS_LOSING;
2349         }
2350
2351         po->stats.stats1.tp_packets++;
2352         if (copy_skb) {
2353                 status |= TP_STATUS_COPY;
2354                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2355         }
2356         spin_unlock(&sk->sk_receive_queue.lock);
2357
2358         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2359
2360         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2361                 getnstimeofday(&ts);
2362
2363         status |= ts_status;
2364
2365         switch (po->tp_version) {
2366         case TPACKET_V1:
2367                 h.h1->tp_len = skb->len;
2368                 h.h1->tp_snaplen = snaplen;
2369                 h.h1->tp_mac = macoff;
2370                 h.h1->tp_net = netoff;
2371                 h.h1->tp_sec = ts.tv_sec;
2372                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2373                 hdrlen = sizeof(*h.h1);
2374                 break;
2375         case TPACKET_V2:
2376                 h.h2->tp_len = skb->len;
2377                 h.h2->tp_snaplen = snaplen;
2378                 h.h2->tp_mac = macoff;
2379                 h.h2->tp_net = netoff;
2380                 h.h2->tp_sec = ts.tv_sec;
2381                 h.h2->tp_nsec = ts.tv_nsec;
2382                 if (skb_vlan_tag_present(skb)) {
2383                         h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2384                         h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2385                         status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2386                 } else {
2387                         h.h2->tp_vlan_tci = 0;
2388                         h.h2->tp_vlan_tpid = 0;
2389                 }
2390                 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2391                 hdrlen = sizeof(*h.h2);
2392                 break;
2393         case TPACKET_V3:
2394                 /* tp_nxt_offset,vlan are already populated above.
2395                  * So DONT clear those fields here
2396                  */
2397                 h.h3->tp_status |= status;
2398                 h.h3->tp_len = skb->len;
2399                 h.h3->tp_snaplen = snaplen;
2400                 h.h3->tp_mac = macoff;
2401                 h.h3->tp_net = netoff;
2402                 h.h3->tp_sec  = ts.tv_sec;
2403                 h.h3->tp_nsec = ts.tv_nsec;
2404                 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2405                 hdrlen = sizeof(*h.h3);
2406                 break;
2407         default:
2408                 BUG();
2409         }
2410
2411         sll = h.raw + TPACKET_ALIGN(hdrlen);
2412         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2413         sll->sll_family = AF_PACKET;
2414         sll->sll_hatype = dev->type;
2415         sll->sll_protocol = skb->protocol;
2416         sll->sll_pkttype = skb->pkt_type;
2417         if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV)))
2418                 sll->sll_ifindex = orig_dev->ifindex;
2419         else
2420                 sll->sll_ifindex = dev->ifindex;
2421
2422         smp_mb();
2423
2424 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2425         if (po->tp_version <= TPACKET_V2) {
2426                 u8 *start, *end;
2427
2428                 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2429                                         macoff + snaplen);
2430
2431                 for (start = h.raw; start < end; start += PAGE_SIZE)
2432                         flush_dcache_page(pgv_to_page(start));
2433         }
2434         smp_wmb();
2435 #endif
2436
2437         if (po->tp_version <= TPACKET_V2) {
2438                 spin_lock(&sk->sk_receive_queue.lock);
2439                 __packet_set_status(po, h.raw, status);
2440                 __clear_bit(slot_id, po->rx_ring.rx_owner_map);
2441                 spin_unlock(&sk->sk_receive_queue.lock);
2442                 sk->sk_data_ready(sk);
2443         } else if (po->tp_version == TPACKET_V3) {
2444                 prb_clear_blk_fill_status(&po->rx_ring);
2445         }
2446
2447 drop_n_restore:
2448         if (skb_head != skb->data && skb_shared(skb)) {
2449                 skb->data = skb_head;
2450                 skb->len = skb_len;
2451         }
2452 drop:
2453         if (!is_drop_n_account)
2454                 consume_skb(skb);
2455         else
2456                 kfree_skb(skb);
2457         return 0;
2458
2459 drop_n_account:
2460         is_drop_n_account = true;
2461         po->stats.stats1.tp_drops++;
2462         spin_unlock(&sk->sk_receive_queue.lock);
2463
2464         sk->sk_data_ready(sk);
2465         kfree_skb(copy_skb);
2466         goto drop_n_restore;
2467 }
2468
2469 static void tpacket_destruct_skb(struct sk_buff *skb)
2470 {
2471         struct packet_sock *po = pkt_sk(skb->sk);
2472
2473         if (likely(po->tx_ring.pg_vec)) {
2474                 void *ph;
2475                 __u32 ts;
2476
2477                 ph = skb_zcopy_get_nouarg(skb);
2478                 packet_dec_pending(&po->tx_ring);
2479
2480                 ts = __packet_set_timestamp(po, ph, skb);
2481                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2482
2483                 if (!packet_read_pending(&po->tx_ring))
2484                         complete(&po->skb_completion);
2485         }
2486
2487         sock_wfree(skb);
2488 }
2489
2490 static void tpacket_set_protocol(const struct net_device *dev,
2491                                  struct sk_buff *skb)
2492 {
2493         if (dev->type == ARPHRD_ETHER) {
2494                 skb_reset_mac_header(skb);
2495                 skb->protocol = eth_hdr(skb)->h_proto;
2496         }
2497 }
2498
2499 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2500 {
2501         if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2502             (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2503              __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2504               __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2505                 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2506                          __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2507                         __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2508
2509         if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2510                 return -EINVAL;
2511
2512         return 0;
2513 }
2514
2515 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2516                                  struct virtio_net_hdr *vnet_hdr)
2517 {
2518         if (*len < sizeof(*vnet_hdr))
2519                 return -EINVAL;
2520         *len -= sizeof(*vnet_hdr);
2521
2522         if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2523                 return -EFAULT;
2524
2525         return __packet_snd_vnet_parse(vnet_hdr, *len);
2526 }
2527
2528 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2529                 void *frame, struct net_device *dev, void *data, int tp_len,
2530                 __be16 proto, unsigned char *addr, int hlen, int copylen,
2531                 const struct sockcm_cookie *sockc)
2532 {
2533         union tpacket_uhdr ph;
2534         int to_write, offset, len, nr_frags, len_max;
2535         struct socket *sock = po->sk.sk_socket;
2536         struct page *page;
2537         int err;
2538
2539         ph.raw = frame;
2540
2541         skb->protocol = proto;
2542         skb->dev = dev;
2543         skb->priority = po->sk.sk_priority;
2544         skb->mark = po->sk.sk_mark;
2545         skb_setup_tx_timestamp(skb, sockc->tsflags);
2546         skb_zcopy_set_nouarg(skb, ph.raw);
2547
2548         skb_reserve(skb, hlen);
2549         skb_reset_network_header(skb);
2550
2551         to_write = tp_len;
2552
2553         if (sock->type == SOCK_DGRAM) {
2554                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2555                                 NULL, tp_len);
2556                 if (unlikely(err < 0))
2557                         return -EINVAL;
2558         } else if (copylen) {
2559                 int hdrlen = min_t(int, copylen, tp_len);
2560
2561                 skb_push(skb, dev->hard_header_len);
2562                 skb_put(skb, copylen - dev->hard_header_len);
2563                 err = skb_store_bits(skb, 0, data, hdrlen);
2564                 if (unlikely(err))
2565                         return err;
2566                 if (!dev_validate_header(dev, skb->data, hdrlen))
2567                         return -EINVAL;
2568                 if (!skb->protocol)
2569                         tpacket_set_protocol(dev, skb);
2570
2571                 data += hdrlen;
2572                 to_write -= hdrlen;
2573         }
2574
2575         offset = offset_in_page(data);
2576         len_max = PAGE_SIZE - offset;
2577         len = ((to_write > len_max) ? len_max : to_write);
2578
2579         skb->data_len = to_write;
2580         skb->len += to_write;
2581         skb->truesize += to_write;
2582         refcount_add(to_write, &po->sk.sk_wmem_alloc);
2583
2584         while (likely(to_write)) {
2585                 nr_frags = skb_shinfo(skb)->nr_frags;
2586
2587                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2588                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2589                                MAX_SKB_FRAGS);
2590                         return -EFAULT;
2591                 }
2592
2593                 page = pgv_to_page(data);
2594                 data += len;
2595                 flush_dcache_page(page);
2596                 get_page(page);
2597                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2598                 to_write -= len;
2599                 offset = 0;
2600                 len_max = PAGE_SIZE;
2601                 len = ((to_write > len_max) ? len_max : to_write);
2602         }
2603
2604         skb_probe_transport_header(skb, 0);
2605
2606         return tp_len;
2607 }
2608
2609 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2610                                 int size_max, void **data)
2611 {
2612         union tpacket_uhdr ph;
2613         int tp_len, off;
2614
2615         ph.raw = frame;
2616
2617         switch (po->tp_version) {
2618         case TPACKET_V3:
2619                 if (ph.h3->tp_next_offset != 0) {
2620                         pr_warn_once("variable sized slot not supported");
2621                         return -EINVAL;
2622                 }
2623                 tp_len = ph.h3->tp_len;
2624                 break;
2625         case TPACKET_V2:
2626                 tp_len = ph.h2->tp_len;
2627                 break;
2628         default:
2629                 tp_len = ph.h1->tp_len;
2630                 break;
2631         }
2632         if (unlikely(tp_len > size_max)) {
2633                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2634                 return -EMSGSIZE;
2635         }
2636
2637         if (unlikely(po->tp_tx_has_off)) {
2638                 int off_min, off_max;
2639
2640                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2641                 off_max = po->tx_ring.frame_size - tp_len;
2642                 if (po->sk.sk_type == SOCK_DGRAM) {
2643                         switch (po->tp_version) {
2644                         case TPACKET_V3:
2645                                 off = ph.h3->tp_net;
2646                                 break;
2647                         case TPACKET_V2:
2648                                 off = ph.h2->tp_net;
2649                                 break;
2650                         default:
2651                                 off = ph.h1->tp_net;
2652                                 break;
2653                         }
2654                 } else {
2655                         switch (po->tp_version) {
2656                         case TPACKET_V3:
2657                                 off = ph.h3->tp_mac;
2658                                 break;
2659                         case TPACKET_V2:
2660                                 off = ph.h2->tp_mac;
2661                                 break;
2662                         default:
2663                                 off = ph.h1->tp_mac;
2664                                 break;
2665                         }
2666                 }
2667                 if (unlikely((off < off_min) || (off_max < off)))
2668                         return -EINVAL;
2669         } else {
2670                 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2671         }
2672
2673         *data = frame + off;
2674         return tp_len;
2675 }
2676
2677 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2678 {
2679         struct sk_buff *skb = NULL;
2680         struct net_device *dev;
2681         struct virtio_net_hdr *vnet_hdr = NULL;
2682         struct sockcm_cookie sockc;
2683         __be16 proto;
2684         int err, reserve = 0;
2685         void *ph;
2686         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2687         bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2688         unsigned char *addr = NULL;
2689         int tp_len, size_max;
2690         void *data;
2691         int len_sum = 0;
2692         int status = TP_STATUS_AVAILABLE;
2693         int hlen, tlen, copylen = 0;
2694         long timeo = 0;
2695
2696         mutex_lock(&po->pg_vec_lock);
2697
2698         /* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2699          * we need to confirm it under protection of pg_vec_lock.
2700          */
2701         if (unlikely(!po->tx_ring.pg_vec)) {
2702                 err = -EBUSY;
2703                 goto out;
2704         }
2705         if (likely(saddr == NULL)) {
2706                 dev     = packet_cached_dev_get(po);
2707                 proto   = READ_ONCE(po->num);
2708         } else {
2709                 err = -EINVAL;
2710                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2711                         goto out;
2712                 if (msg->msg_namelen < (saddr->sll_halen
2713                                         + offsetof(struct sockaddr_ll,
2714                                                 sll_addr)))
2715                         goto out;
2716                 proto   = saddr->sll_protocol;
2717                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2718                 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2719                         if (dev && msg->msg_namelen < dev->addr_len +
2720                                    offsetof(struct sockaddr_ll, sll_addr))
2721                                 goto out_put;
2722                         addr = saddr->sll_addr;
2723                 }
2724         }
2725
2726         err = -ENXIO;
2727         if (unlikely(dev == NULL))
2728                 goto out;
2729         err = -ENETDOWN;
2730         if (unlikely(!(dev->flags & IFF_UP)))
2731                 goto out_put;
2732
2733         sockc.tsflags = po->sk.sk_tsflags;
2734         if (msg->msg_controllen) {
2735                 err = sock_cmsg_send(&po->sk, msg, &sockc);
2736                 if (unlikely(err))
2737                         goto out_put;
2738         }
2739
2740         if (po->sk.sk_socket->type == SOCK_RAW)
2741                 reserve = dev->hard_header_len;
2742         size_max = po->tx_ring.frame_size
2743                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2744
2745         if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2746                 size_max = dev->mtu + reserve + VLAN_HLEN;
2747
2748         reinit_completion(&po->skb_completion);
2749
2750         do {
2751                 ph = packet_current_frame(po, &po->tx_ring,
2752                                           TP_STATUS_SEND_REQUEST);
2753                 if (unlikely(ph == NULL)) {
2754                         if (need_wait && skb) {
2755                                 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2756                                 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2757                                 if (timeo <= 0) {
2758                                         err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2759                                         goto out_put;
2760                                 }
2761                         }
2762                         /* check for additional frames */
2763                         continue;
2764                 }
2765
2766                 skb = NULL;
2767                 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2768                 if (tp_len < 0)
2769                         goto tpacket_error;
2770
2771                 status = TP_STATUS_SEND_REQUEST;
2772                 hlen = LL_RESERVED_SPACE(dev);
2773                 tlen = dev->needed_tailroom;
2774                 if (po->has_vnet_hdr) {
2775                         vnet_hdr = data;
2776                         data += sizeof(*vnet_hdr);
2777                         tp_len -= sizeof(*vnet_hdr);
2778                         if (tp_len < 0 ||
2779                             __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2780                                 tp_len = -EINVAL;
2781                                 goto tpacket_error;
2782                         }
2783                         copylen = __virtio16_to_cpu(vio_le(),
2784                                                     vnet_hdr->hdr_len);
2785                 }
2786                 copylen = max_t(int, copylen, dev->hard_header_len);
2787                 skb = sock_alloc_send_skb(&po->sk,
2788                                 hlen + tlen + sizeof(struct sockaddr_ll) +
2789                                 (copylen - dev->hard_header_len),
2790                                 !need_wait, &err);
2791
2792                 if (unlikely(skb == NULL)) {
2793                         /* we assume the socket was initially writeable ... */
2794                         if (likely(len_sum > 0))
2795                                 err = len_sum;
2796                         goto out_status;
2797                 }
2798                 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2799                                           addr, hlen, copylen, &sockc);
2800                 if (likely(tp_len >= 0) &&
2801                     tp_len > dev->mtu + reserve &&
2802                     !po->has_vnet_hdr &&
2803                     !packet_extra_vlan_len_allowed(dev, skb))
2804                         tp_len = -EMSGSIZE;
2805
2806                 if (unlikely(tp_len < 0)) {
2807 tpacket_error:
2808                         if (po->tp_loss) {
2809                                 __packet_set_status(po, ph,
2810                                                 TP_STATUS_AVAILABLE);
2811                                 packet_increment_head(&po->tx_ring);
2812                                 kfree_skb(skb);
2813                                 continue;
2814                         } else {
2815                                 status = TP_STATUS_WRONG_FORMAT;
2816                                 err = tp_len;
2817                                 goto out_status;
2818                         }
2819                 }
2820
2821                 if (po->has_vnet_hdr) {
2822                         if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2823                                 tp_len = -EINVAL;
2824                                 goto tpacket_error;
2825                         }
2826                         virtio_net_hdr_set_proto(skb, vnet_hdr);
2827                 }
2828
2829                 skb->destructor = tpacket_destruct_skb;
2830                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2831                 packet_inc_pending(&po->tx_ring);
2832
2833                 status = TP_STATUS_SEND_REQUEST;
2834                 err = po->xmit(skb);
2835                 if (unlikely(err != 0)) {
2836                         if (err > 0)
2837                                 err = net_xmit_errno(err);
2838                         if (err && __packet_get_status(po, ph) ==
2839                                    TP_STATUS_AVAILABLE) {
2840                                 /* skb was destructed already */
2841                                 skb = NULL;
2842                                 goto out_status;
2843                         }
2844                         /*
2845                          * skb was dropped but not destructed yet;
2846                          * let's treat it like congestion or err < 0
2847                          */
2848                         err = 0;
2849                 }
2850                 packet_increment_head(&po->tx_ring);
2851                 len_sum += tp_len;
2852         } while (likely((ph != NULL) ||
2853                 /* Note: packet_read_pending() might be slow if we have
2854                  * to call it as it's per_cpu variable, but in fast-path
2855                  * we already short-circuit the loop with the first
2856                  * condition, and luckily don't have to go that path
2857                  * anyway.
2858                  */
2859                  (need_wait && packet_read_pending(&po->tx_ring))));
2860
2861         err = len_sum;
2862         goto out_put;
2863
2864 out_status:
2865         __packet_set_status(po, ph, status);
2866         kfree_skb(skb);
2867 out_put:
2868         dev_put(dev);
2869 out:
2870         mutex_unlock(&po->pg_vec_lock);
2871         return err;
2872 }
2873
2874 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2875                                         size_t reserve, size_t len,
2876                                         size_t linear, int noblock,
2877                                         int *err)
2878 {
2879         struct sk_buff *skb;
2880
2881         /* Under a page?  Don't bother with paged skb. */
2882         if (prepad + len < PAGE_SIZE || !linear)
2883                 linear = len;
2884
2885         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2886                                    err, 0);
2887         if (!skb)
2888                 return NULL;
2889
2890         skb_reserve(skb, reserve);
2891         skb_put(skb, linear);
2892         skb->data_len = len - linear;
2893         skb->len += len - linear;
2894
2895         return skb;
2896 }
2897
2898 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2899 {
2900         struct sock *sk = sock->sk;
2901         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2902         struct sk_buff *skb;
2903         struct net_device *dev;
2904         __be16 proto;
2905         unsigned char *addr = NULL;
2906         int err, reserve = 0;
2907         struct sockcm_cookie sockc;
2908         struct virtio_net_hdr vnet_hdr = { 0 };
2909         int offset = 0;
2910         struct packet_sock *po = pkt_sk(sk);
2911         bool has_vnet_hdr = false;
2912         int hlen, tlen, linear;
2913         int extra_len = 0;
2914
2915         /*
2916          *      Get and verify the address.
2917          */
2918
2919         if (likely(saddr == NULL)) {
2920                 dev     = packet_cached_dev_get(po);
2921                 proto   = READ_ONCE(po->num);
2922         } else {
2923                 err = -EINVAL;
2924                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2925                         goto out;
2926                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2927                         goto out;
2928                 proto   = saddr->sll_protocol;
2929                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2930                 if (sock->type == SOCK_DGRAM) {
2931                         if (dev && msg->msg_namelen < dev->addr_len +
2932                                    offsetof(struct sockaddr_ll, sll_addr))
2933                                 goto out_unlock;
2934                         addr = saddr->sll_addr;
2935                 }
2936         }
2937
2938         err = -ENXIO;
2939         if (unlikely(dev == NULL))
2940                 goto out_unlock;
2941         err = -ENETDOWN;
2942         if (unlikely(!(dev->flags & IFF_UP)))
2943                 goto out_unlock;
2944
2945         sockc.tsflags = sk->sk_tsflags;
2946         sockc.mark = sk->sk_mark;
2947         if (msg->msg_controllen) {
2948                 err = sock_cmsg_send(sk, msg, &sockc);
2949                 if (unlikely(err))
2950                         goto out_unlock;
2951         }
2952
2953         if (sock->type == SOCK_RAW)
2954                 reserve = dev->hard_header_len;
2955         if (po->has_vnet_hdr) {
2956                 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2957                 if (err)
2958                         goto out_unlock;
2959                 has_vnet_hdr = true;
2960         }
2961
2962         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2963                 if (!netif_supports_nofcs(dev)) {
2964                         err = -EPROTONOSUPPORT;
2965                         goto out_unlock;
2966                 }
2967                 extra_len = 4; /* We're doing our own CRC */
2968         }
2969
2970         err = -EMSGSIZE;
2971         if (!vnet_hdr.gso_type &&
2972             (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2973                 goto out_unlock;
2974
2975         err = -ENOBUFS;
2976         hlen = LL_RESERVED_SPACE(dev);
2977         tlen = dev->needed_tailroom;
2978         linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2979         linear = max(linear, min_t(int, len, dev->hard_header_len));
2980         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2981                                msg->msg_flags & MSG_DONTWAIT, &err);
2982         if (skb == NULL)
2983                 goto out_unlock;
2984
2985         skb_reset_network_header(skb);
2986
2987         err = -EINVAL;
2988         if (sock->type == SOCK_DGRAM) {
2989                 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2990                 if (unlikely(offset < 0))
2991                         goto out_free;
2992         } else if (reserve) {
2993                 skb_reserve(skb, -reserve);
2994                 if (len < reserve)
2995                         skb_reset_network_header(skb);
2996         }
2997
2998         /* Returns -EFAULT on error */
2999         err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
3000         if (err)
3001                 goto out_free;
3002
3003         if (sock->type == SOCK_RAW &&
3004             !dev_validate_header(dev, skb->data, len)) {
3005                 err = -EINVAL;
3006                 goto out_free;
3007         }
3008
3009         skb_setup_tx_timestamp(skb, sockc.tsflags);
3010
3011         if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
3012             !packet_extra_vlan_len_allowed(dev, skb)) {
3013                 err = -EMSGSIZE;
3014                 goto out_free;
3015         }
3016
3017         skb->protocol = proto;
3018         skb->dev = dev;
3019         skb->priority = sk->sk_priority;
3020         skb->mark = sockc.mark;
3021
3022         if (has_vnet_hdr) {
3023                 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
3024                 if (err)
3025                         goto out_free;
3026                 len += sizeof(vnet_hdr);
3027                 virtio_net_hdr_set_proto(skb, &vnet_hdr);
3028         }
3029
3030         skb_probe_transport_header(skb, reserve);
3031
3032         if (unlikely(extra_len == 4))
3033                 skb->no_fcs = 1;
3034
3035         err = po->xmit(skb);
3036         if (unlikely(err != 0)) {
3037                 if (err > 0)
3038                         err = net_xmit_errno(err);
3039                 if (err)
3040                         goto out_unlock;
3041         }
3042
3043         dev_put(dev);
3044
3045         return len;
3046
3047 out_free:
3048         kfree_skb(skb);
3049 out_unlock:
3050         if (dev)
3051                 dev_put(dev);
3052 out:
3053         return err;
3054 }
3055
3056 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
3057 {
3058         struct sock *sk = sock->sk;
3059         struct packet_sock *po = pkt_sk(sk);
3060
3061         if (po->tx_ring.pg_vec)
3062                 return tpacket_snd(po, msg);
3063         else
3064                 return packet_snd(sock, msg, len);
3065 }
3066
3067 /*
3068  *      Close a PACKET socket. This is fairly simple. We immediately go
3069  *      to 'closed' state and remove our protocol entry in the device list.
3070  */
3071
3072 static int packet_release(struct socket *sock)
3073 {
3074         struct sock *sk = sock->sk;
3075         struct packet_sock *po;
3076         struct packet_fanout *f;
3077         struct net *net;
3078         union tpacket_req_u req_u;
3079
3080         if (!sk)
3081                 return 0;
3082
3083         net = sock_net(sk);
3084         po = pkt_sk(sk);
3085
3086         mutex_lock(&net->packet.sklist_lock);
3087         sk_del_node_init_rcu(sk);
3088         mutex_unlock(&net->packet.sklist_lock);
3089
3090         preempt_disable();
3091         sock_prot_inuse_add(net, sk->sk_prot, -1);
3092         preempt_enable();
3093
3094         spin_lock(&po->bind_lock);
3095         unregister_prot_hook(sk, false);
3096         packet_cached_dev_reset(po);
3097
3098         if (po->prot_hook.dev) {
3099                 dev_put(po->prot_hook.dev);
3100                 po->prot_hook.dev = NULL;
3101         }
3102         spin_unlock(&po->bind_lock);
3103
3104         packet_flush_mclist(sk);
3105
3106         lock_sock(sk);
3107         if (po->rx_ring.pg_vec) {
3108                 memset(&req_u, 0, sizeof(req_u));
3109                 packet_set_ring(sk, &req_u, 1, 0);
3110         }
3111
3112         if (po->tx_ring.pg_vec) {
3113                 memset(&req_u, 0, sizeof(req_u));
3114                 packet_set_ring(sk, &req_u, 1, 1);
3115         }
3116         release_sock(sk);
3117
3118         f = fanout_release(sk);
3119
3120         synchronize_net();
3121
3122         if (f) {
3123                 kfree(po->rollover);
3124                 fanout_release_data(f);
3125                 kfree(f);
3126         }
3127         /*
3128          *      Now the socket is dead. No more input will appear.
3129          */
3130         sock_orphan(sk);
3131         sock->sk = NULL;
3132
3133         /* Purge queues */
3134
3135         skb_queue_purge(&sk->sk_receive_queue);
3136         packet_free_pending(po);
3137         sk_refcnt_debug_release(sk);
3138
3139         sock_put(sk);
3140         return 0;
3141 }
3142
3143 /*
3144  *      Attach a packet hook.
3145  */
3146
3147 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3148                           __be16 proto)
3149 {
3150         struct packet_sock *po = pkt_sk(sk);
3151         struct net_device *dev_curr;
3152         __be16 proto_curr;
3153         bool need_rehook;
3154         struct net_device *dev = NULL;
3155         int ret = 0;
3156         bool unlisted = false;
3157
3158         lock_sock(sk);
3159         spin_lock(&po->bind_lock);
3160         if (!proto)
3161                 proto = po->num;
3162
3163         rcu_read_lock();
3164
3165         if (po->fanout) {
3166                 ret = -EINVAL;
3167                 goto out_unlock;
3168         }
3169
3170         if (name) {
3171                 dev = dev_get_by_name_rcu(sock_net(sk), name);
3172                 if (!dev) {
3173                         ret = -ENODEV;
3174                         goto out_unlock;
3175                 }
3176         } else if (ifindex) {
3177                 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3178                 if (!dev) {
3179                         ret = -ENODEV;
3180                         goto out_unlock;
3181                 }
3182         }
3183
3184         if (dev)
3185                 dev_hold(dev);
3186
3187         proto_curr = po->prot_hook.type;
3188         dev_curr = po->prot_hook.dev;
3189
3190         need_rehook = proto_curr != proto || dev_curr != dev;
3191
3192         if (need_rehook) {
3193                 if (po->running) {
3194                         rcu_read_unlock();
3195                         /* prevents packet_notifier() from calling
3196                          * register_prot_hook()
3197                          */
3198                         WRITE_ONCE(po->num, 0);
3199                         __unregister_prot_hook(sk, true);
3200                         rcu_read_lock();
3201                         dev_curr = po->prot_hook.dev;
3202                         if (dev)
3203                                 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3204                                                                  dev->ifindex);
3205                 }
3206
3207                 BUG_ON(po->running);
3208                 WRITE_ONCE(po->num, proto);
3209                 po->prot_hook.type = proto;
3210
3211                 if (unlikely(unlisted)) {
3212                         dev_put(dev);
3213                         po->prot_hook.dev = NULL;
3214                         WRITE_ONCE(po->ifindex, -1);
3215                         packet_cached_dev_reset(po);
3216                 } else {
3217                         po->prot_hook.dev = dev;
3218                         WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0);
3219                         packet_cached_dev_assign(po, dev);
3220                 }
3221         }
3222         if (dev_curr)
3223                 dev_put(dev_curr);
3224
3225         if (proto == 0 || !need_rehook)
3226                 goto out_unlock;
3227
3228         if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3229                 register_prot_hook(sk);
3230         } else {
3231                 sk->sk_err = ENETDOWN;
3232                 if (!sock_flag(sk, SOCK_DEAD))
3233                         sk->sk_error_report(sk);
3234         }
3235
3236 out_unlock:
3237         rcu_read_unlock();
3238         spin_unlock(&po->bind_lock);
3239         release_sock(sk);
3240         return ret;
3241 }
3242
3243 /*
3244  *      Bind a packet socket to a device
3245  */
3246
3247 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3248                             int addr_len)
3249 {
3250         struct sock *sk = sock->sk;
3251         char name[sizeof(uaddr->sa_data) + 1];
3252
3253         /*
3254          *      Check legality
3255          */
3256
3257         if (addr_len != sizeof(struct sockaddr))
3258                 return -EINVAL;
3259         /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3260          * zero-terminated.
3261          */
3262         memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3263         name[sizeof(uaddr->sa_data)] = 0;
3264
3265         return packet_do_bind(sk, name, 0, 0);
3266 }
3267
3268 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3269 {
3270         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3271         struct sock *sk = sock->sk;
3272
3273         /*
3274          *      Check legality
3275          */
3276
3277         if (addr_len < sizeof(struct sockaddr_ll))
3278                 return -EINVAL;
3279         if (sll->sll_family != AF_PACKET)
3280                 return -EINVAL;
3281
3282         return packet_do_bind(sk, NULL, sll->sll_ifindex, sll->sll_protocol);
3283 }
3284
3285 static struct proto packet_proto = {
3286         .name     = "PACKET",
3287         .owner    = THIS_MODULE,
3288         .obj_size = sizeof(struct packet_sock),
3289 };
3290
3291 /*
3292  *      Create a packet of type SOCK_PACKET.
3293  */
3294
3295 static int packet_create(struct net *net, struct socket *sock, int protocol,
3296                          int kern)
3297 {
3298         struct sock *sk;
3299         struct packet_sock *po;
3300         __be16 proto = (__force __be16)protocol; /* weird, but documented */
3301         int err;
3302
3303         if (!ns_capable(net->user_ns, CAP_NET_RAW))
3304                 return -EPERM;
3305         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3306             sock->type != SOCK_PACKET)
3307                 return -ESOCKTNOSUPPORT;
3308
3309         sock->state = SS_UNCONNECTED;
3310
3311         err = -ENOBUFS;
3312         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3313         if (sk == NULL)
3314                 goto out;
3315
3316         sock->ops = &packet_ops;
3317         if (sock->type == SOCK_PACKET)
3318                 sock->ops = &packet_ops_spkt;
3319
3320         sock_init_data(sock, sk);
3321
3322         po = pkt_sk(sk);
3323         init_completion(&po->skb_completion);
3324         sk->sk_family = PF_PACKET;
3325         po->num = proto;
3326         po->xmit = dev_queue_xmit;
3327
3328         err = packet_alloc_pending(po);
3329         if (err)
3330                 goto out2;
3331
3332         packet_cached_dev_reset(po);
3333
3334         sk->sk_destruct = packet_sock_destruct;
3335         sk_refcnt_debug_inc(sk);
3336
3337         /*
3338          *      Attach a protocol block
3339          */
3340
3341         spin_lock_init(&po->bind_lock);
3342         mutex_init(&po->pg_vec_lock);
3343         po->rollover = NULL;
3344         po->prot_hook.func = packet_rcv;
3345
3346         if (sock->type == SOCK_PACKET)
3347                 po->prot_hook.func = packet_rcv_spkt;
3348
3349         po->prot_hook.af_packet_priv = sk;
3350         po->prot_hook.af_packet_net = sock_net(sk);
3351
3352         if (proto) {
3353                 po->prot_hook.type = proto;
3354                 __register_prot_hook(sk);
3355         }
3356
3357         mutex_lock(&net->packet.sklist_lock);
3358         sk_add_node_tail_rcu(sk, &net->packet.sklist);
3359         mutex_unlock(&net->packet.sklist_lock);
3360
3361         preempt_disable();
3362         sock_prot_inuse_add(net, &packet_proto, 1);
3363         preempt_enable();
3364
3365         return 0;
3366 out2:
3367         sk_free(sk);
3368 out:
3369         return err;
3370 }
3371
3372 /*
3373  *      Pull a packet from our receive queue and hand it to the user.
3374  *      If necessary we block.
3375  */
3376
3377 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3378                           int flags)
3379 {
3380         struct sock *sk = sock->sk;
3381         struct sk_buff *skb;
3382         int copied, err;
3383         int vnet_hdr_len = 0;
3384         unsigned int origlen = 0;
3385
3386         err = -EINVAL;
3387         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3388                 goto out;
3389
3390 #if 0
3391         /* What error should we return now? EUNATTACH? */
3392         if (pkt_sk(sk)->ifindex < 0)
3393                 return -ENODEV;
3394 #endif
3395
3396         if (flags & MSG_ERRQUEUE) {
3397                 err = sock_recv_errqueue(sk, msg, len,
3398                                          SOL_PACKET, PACKET_TX_TIMESTAMP);
3399                 goto out;
3400         }
3401
3402         /*
3403          *      Call the generic datagram receiver. This handles all sorts
3404          *      of horrible races and re-entrancy so we can forget about it
3405          *      in the protocol layers.
3406          *
3407          *      Now it will return ENETDOWN, if device have just gone down,
3408          *      but then it will block.
3409          */
3410
3411         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3412
3413         /*
3414          *      An error occurred so return it. Because skb_recv_datagram()
3415          *      handles the blocking we don't see and worry about blocking
3416          *      retries.
3417          */
3418
3419         if (skb == NULL)
3420                 goto out;
3421
3422         if (pkt_sk(sk)->pressure)
3423                 packet_rcv_has_room(pkt_sk(sk), NULL);
3424
3425         if (pkt_sk(sk)->has_vnet_hdr) {
3426                 err = packet_rcv_vnet(msg, skb, &len);
3427                 if (err)
3428                         goto out_free;
3429                 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3430         }
3431
3432         /* You lose any data beyond the buffer you gave. If it worries
3433          * a user program they can ask the device for its MTU
3434          * anyway.
3435          */
3436         copied = skb->len;
3437         if (copied > len) {
3438                 copied = len;
3439                 msg->msg_flags |= MSG_TRUNC;
3440         }
3441
3442         err = skb_copy_datagram_msg(skb, 0, msg, copied);
3443         if (err)
3444                 goto out_free;
3445
3446         if (sock->type != SOCK_PACKET) {
3447                 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3448
3449                 /* Original length was stored in sockaddr_ll fields */
3450                 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3451                 sll->sll_family = AF_PACKET;
3452                 sll->sll_protocol = skb->protocol;
3453         }
3454
3455         sock_recv_ts_and_drops(msg, sk, skb);
3456
3457         if (msg->msg_name) {
3458                 const size_t max_len = min(sizeof(skb->cb),
3459                                            sizeof(struct sockaddr_storage));
3460                 int copy_len;
3461
3462                 /* If the address length field is there to be filled
3463                  * in, we fill it in now.
3464                  */
3465                 if (sock->type == SOCK_PACKET) {
3466                         __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3467                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
3468                         copy_len = msg->msg_namelen;
3469                 } else {
3470                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3471
3472                         msg->msg_namelen = sll->sll_halen +
3473                                 offsetof(struct sockaddr_ll, sll_addr);
3474                         copy_len = msg->msg_namelen;
3475                         if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3476                                 memset(msg->msg_name +
3477                                        offsetof(struct sockaddr_ll, sll_addr),
3478                                        0, sizeof(sll->sll_addr));
3479                                 msg->msg_namelen = sizeof(struct sockaddr_ll);
3480                         }
3481                 }
3482                 if (WARN_ON_ONCE(copy_len > max_len)) {
3483                         copy_len = max_len;
3484                         msg->msg_namelen = copy_len;
3485                 }
3486                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3487         }
3488
3489         if (packet_sock_flag(pkt_sk(sk), PACKET_SOCK_AUXDATA)) {
3490                 struct tpacket_auxdata aux;
3491
3492                 aux.tp_status = TP_STATUS_USER;
3493                 if (skb->ip_summed == CHECKSUM_PARTIAL)
3494                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3495                 else if (skb->pkt_type != PACKET_OUTGOING &&
3496                          skb_csum_unnecessary(skb))
3497                         aux.tp_status |= TP_STATUS_CSUM_VALID;
3498
3499                 aux.tp_len = origlen;
3500                 aux.tp_snaplen = skb->len;
3501                 aux.tp_mac = 0;
3502                 aux.tp_net = skb_network_offset(skb);
3503                 if (skb_vlan_tag_present(skb)) {
3504                         aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3505                         aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3506                         aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3507                 } else {
3508                         aux.tp_vlan_tci = 0;
3509                         aux.tp_vlan_tpid = 0;
3510                 }
3511                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3512         }
3513
3514         /*
3515          *      Free or return the buffer as appropriate. Again this
3516          *      hides all the races and re-entrancy issues from us.
3517          */
3518         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3519
3520 out_free:
3521         skb_free_datagram(sk, skb);
3522 out:
3523         return err;
3524 }
3525
3526 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3527                                int *uaddr_len, int peer)
3528 {
3529         struct net_device *dev;
3530         struct sock *sk = sock->sk;
3531
3532         if (peer)
3533                 return -EOPNOTSUPP;
3534
3535         uaddr->sa_family = AF_PACKET;
3536         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3537         rcu_read_lock();
3538         dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex));
3539         if (dev)
3540                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3541         rcu_read_unlock();
3542         *uaddr_len = sizeof(*uaddr);
3543
3544         return 0;
3545 }
3546
3547 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3548                           int *uaddr_len, int peer)
3549 {
3550         struct net_device *dev;
3551         struct sock *sk = sock->sk;
3552         struct packet_sock *po = pkt_sk(sk);
3553         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3554         int ifindex;
3555
3556         if (peer)
3557                 return -EOPNOTSUPP;
3558
3559         ifindex = READ_ONCE(po->ifindex);
3560         sll->sll_family = AF_PACKET;
3561         sll->sll_ifindex = ifindex;
3562         sll->sll_protocol = READ_ONCE(po->num);
3563         sll->sll_pkttype = 0;
3564         rcu_read_lock();
3565         dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3566         if (dev) {
3567                 sll->sll_hatype = dev->type;
3568                 sll->sll_halen = dev->addr_len;
3569                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3570         } else {
3571                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
3572                 sll->sll_halen = 0;
3573         }
3574         rcu_read_unlock();
3575         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3576
3577         return 0;
3578 }
3579
3580 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3581                          int what)
3582 {
3583         switch (i->type) {
3584         case PACKET_MR_MULTICAST:
3585                 if (i->alen != dev->addr_len)
3586                         return -EINVAL;
3587                 if (what > 0)
3588                         return dev_mc_add(dev, i->addr);
3589                 else
3590                         return dev_mc_del(dev, i->addr);
3591                 break;
3592         case PACKET_MR_PROMISC:
3593                 return dev_set_promiscuity(dev, what);
3594         case PACKET_MR_ALLMULTI:
3595                 return dev_set_allmulti(dev, what);
3596         case PACKET_MR_UNICAST:
3597                 if (i->alen != dev->addr_len)
3598                         return -EINVAL;
3599                 if (what > 0)
3600                         return dev_uc_add(dev, i->addr);
3601                 else
3602                         return dev_uc_del(dev, i->addr);
3603                 break;
3604         default:
3605                 break;
3606         }
3607         return 0;
3608 }
3609
3610 static void packet_dev_mclist_delete(struct net_device *dev,
3611                                      struct packet_mclist **mlp)
3612 {
3613         struct packet_mclist *ml;
3614
3615         while ((ml = *mlp) != NULL) {
3616                 if (ml->ifindex == dev->ifindex) {
3617                         packet_dev_mc(dev, ml, -1);
3618                         *mlp = ml->next;
3619                         kfree(ml);
3620                 } else
3621                         mlp = &ml->next;
3622         }
3623 }
3624
3625 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3626 {
3627         struct packet_sock *po = pkt_sk(sk);
3628         struct packet_mclist *ml, *i;
3629         struct net_device *dev;
3630         int err;
3631
3632         rtnl_lock();
3633
3634         err = -ENODEV;
3635         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3636         if (!dev)
3637                 goto done;
3638
3639         err = -EINVAL;
3640         if (mreq->mr_alen > dev->addr_len)
3641                 goto done;
3642
3643         err = -ENOBUFS;
3644         i = kmalloc(sizeof(*i), GFP_KERNEL);
3645         if (i == NULL)
3646                 goto done;
3647
3648         err = 0;
3649         for (ml = po->mclist; ml; ml = ml->next) {
3650                 if (ml->ifindex == mreq->mr_ifindex &&
3651                     ml->type == mreq->mr_type &&
3652                     ml->alen == mreq->mr_alen &&
3653                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3654                         ml->count++;
3655                         /* Free the new element ... */
3656                         kfree(i);
3657                         goto done;
3658                 }
3659         }
3660
3661         i->type = mreq->mr_type;
3662         i->ifindex = mreq->mr_ifindex;
3663         i->alen = mreq->mr_alen;
3664         memcpy(i->addr, mreq->mr_address, i->alen);
3665         memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3666         i->count = 1;
3667         i->next = po->mclist;
3668         po->mclist = i;
3669         err = packet_dev_mc(dev, i, 1);
3670         if (err) {
3671                 po->mclist = i->next;
3672                 kfree(i);
3673         }
3674
3675 done:
3676         rtnl_unlock();
3677         return err;
3678 }
3679
3680 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3681 {
3682         struct packet_mclist *ml, **mlp;
3683
3684         rtnl_lock();
3685
3686         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3687                 if (ml->ifindex == mreq->mr_ifindex &&
3688                     ml->type == mreq->mr_type &&
3689                     ml->alen == mreq->mr_alen &&
3690                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3691                         if (--ml->count == 0) {
3692                                 struct net_device *dev;
3693                                 *mlp = ml->next;
3694                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3695                                 if (dev)
3696                                         packet_dev_mc(dev, ml, -1);
3697                                 kfree(ml);
3698                         }
3699                         break;
3700                 }
3701         }
3702         rtnl_unlock();
3703         return 0;
3704 }
3705
3706 static void packet_flush_mclist(struct sock *sk)
3707 {
3708         struct packet_sock *po = pkt_sk(sk);
3709         struct packet_mclist *ml;
3710
3711         if (!po->mclist)
3712                 return;
3713
3714         rtnl_lock();
3715         while ((ml = po->mclist) != NULL) {
3716                 struct net_device *dev;
3717
3718                 po->mclist = ml->next;
3719                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3720                 if (dev != NULL)
3721                         packet_dev_mc(dev, ml, -1);
3722                 kfree(ml);
3723         }
3724         rtnl_unlock();
3725 }
3726
3727 static int
3728 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3729 {
3730         struct sock *sk = sock->sk;
3731         struct packet_sock *po = pkt_sk(sk);
3732         int ret;
3733
3734         if (level != SOL_PACKET)
3735                 return -ENOPROTOOPT;
3736
3737         switch (optname) {
3738         case PACKET_ADD_MEMBERSHIP:
3739         case PACKET_DROP_MEMBERSHIP:
3740         {
3741                 struct packet_mreq_max mreq;
3742                 int len = optlen;
3743                 memset(&mreq, 0, sizeof(mreq));
3744                 if (len < sizeof(struct packet_mreq))
3745                         return -EINVAL;
3746                 if (len > sizeof(mreq))
3747                         len = sizeof(mreq);
3748                 if (copy_from_user(&mreq, optval, len))
3749                         return -EFAULT;
3750                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3751                         return -EINVAL;
3752                 if (optname == PACKET_ADD_MEMBERSHIP)
3753                         ret = packet_mc_add(sk, &mreq);
3754                 else
3755                         ret = packet_mc_drop(sk, &mreq);
3756                 return ret;
3757         }
3758
3759         case PACKET_RX_RING:
3760         case PACKET_TX_RING:
3761         {
3762                 union tpacket_req_u req_u;
3763                 int len;
3764
3765                 lock_sock(sk);
3766                 switch (po->tp_version) {
3767                 case TPACKET_V1:
3768                 case TPACKET_V2:
3769                         len = sizeof(req_u.req);
3770                         break;
3771                 case TPACKET_V3:
3772                 default:
3773                         len = sizeof(req_u.req3);
3774                         break;
3775                 }
3776                 if (optlen < len) {
3777                         ret = -EINVAL;
3778                 } else {
3779                         if (copy_from_user(&req_u.req, optval, len))
3780                                 ret = -EFAULT;
3781                         else
3782                                 ret = packet_set_ring(sk, &req_u, 0,
3783                                                     optname == PACKET_TX_RING);
3784                 }
3785                 release_sock(sk);
3786                 return ret;
3787         }
3788         case PACKET_COPY_THRESH:
3789         {
3790                 int val;
3791
3792                 if (optlen != sizeof(val))
3793                         return -EINVAL;
3794                 if (copy_from_user(&val, optval, sizeof(val)))
3795                         return -EFAULT;
3796
3797                 pkt_sk(sk)->copy_thresh = val;
3798                 return 0;
3799         }
3800         case PACKET_VERSION:
3801         {
3802                 int val;
3803
3804                 if (optlen != sizeof(val))
3805                         return -EINVAL;
3806                 if (copy_from_user(&val, optval, sizeof(val)))
3807                         return -EFAULT;
3808                 switch (val) {
3809                 case TPACKET_V1:
3810                 case TPACKET_V2:
3811                 case TPACKET_V3:
3812                         break;
3813                 default:
3814                         return -EINVAL;
3815                 }
3816                 lock_sock(sk);
3817                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3818                         ret = -EBUSY;
3819                 } else {
3820                         po->tp_version = val;
3821                         ret = 0;
3822                 }
3823                 release_sock(sk);
3824                 return ret;
3825         }
3826         case PACKET_RESERVE:
3827         {
3828                 unsigned int val;
3829
3830                 if (optlen != sizeof(val))
3831                         return -EINVAL;
3832                 if (copy_from_user(&val, optval, sizeof(val)))
3833                         return -EFAULT;
3834                 if (val > INT_MAX)
3835                         return -EINVAL;
3836                 lock_sock(sk);
3837                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3838                         ret = -EBUSY;
3839                 } else {
3840                         po->tp_reserve = val;
3841                         ret = 0;
3842                 }
3843                 release_sock(sk);
3844                 return ret;
3845         }
3846         case PACKET_LOSS:
3847         {
3848                 unsigned int val;
3849
3850                 if (optlen != sizeof(val))
3851                         return -EINVAL;
3852                 if (copy_from_user(&val, optval, sizeof(val)))
3853                         return -EFAULT;
3854
3855                 lock_sock(sk);
3856                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3857                         ret = -EBUSY;
3858                 } else {
3859                         po->tp_loss = !!val;
3860                         ret = 0;
3861                 }
3862                 release_sock(sk);
3863                 return ret;
3864         }
3865         case PACKET_AUXDATA:
3866         {
3867                 int val;
3868
3869                 if (optlen < sizeof(val))
3870                         return -EINVAL;
3871                 if (copy_from_user(&val, optval, sizeof(val)))
3872                         return -EFAULT;
3873
3874                 packet_sock_flag_set(po, PACKET_SOCK_AUXDATA, val);
3875                 return 0;
3876         }
3877         case PACKET_ORIGDEV:
3878         {
3879                 int val;
3880
3881                 if (optlen < sizeof(val))
3882                         return -EINVAL;
3883                 if (copy_from_user(&val, optval, sizeof(val)))
3884                         return -EFAULT;
3885
3886                 packet_sock_flag_set(po, PACKET_SOCK_ORIGDEV, val);
3887                 return 0;
3888         }
3889         case PACKET_VNET_HDR:
3890         {
3891                 int val;
3892
3893                 if (sock->type != SOCK_RAW)
3894                         return -EINVAL;
3895                 if (optlen < sizeof(val))
3896                         return -EINVAL;
3897                 if (copy_from_user(&val, optval, sizeof(val)))
3898                         return -EFAULT;
3899
3900                 lock_sock(sk);
3901                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3902                         ret = -EBUSY;
3903                 } else {
3904                         po->has_vnet_hdr = !!val;
3905                         ret = 0;
3906                 }
3907                 release_sock(sk);
3908                 return ret;
3909         }
3910         case PACKET_TIMESTAMP:
3911         {
3912                 int val;
3913
3914                 if (optlen != sizeof(val))
3915                         return -EINVAL;
3916                 if (copy_from_user(&val, optval, sizeof(val)))
3917                         return -EFAULT;
3918
3919                 po->tp_tstamp = val;
3920                 return 0;
3921         }
3922         case PACKET_FANOUT:
3923         {
3924                 int val;
3925
3926                 if (optlen != sizeof(val))
3927                         return -EINVAL;
3928                 if (copy_from_user(&val, optval, sizeof(val)))
3929                         return -EFAULT;
3930
3931                 return fanout_add(sk, val & 0xffff, val >> 16);
3932         }
3933         case PACKET_FANOUT_DATA:
3934         {
3935                 /* Paired with the WRITE_ONCE() in fanout_add() */
3936                 if (!READ_ONCE(po->fanout))
3937                         return -EINVAL;
3938
3939                 return fanout_set_data(po, optval, optlen);
3940         }
3941         case PACKET_TX_HAS_OFF:
3942         {
3943                 unsigned int val;
3944
3945                 if (optlen != sizeof(val))
3946                         return -EINVAL;
3947                 if (copy_from_user(&val, optval, sizeof(val)))
3948                         return -EFAULT;
3949
3950                 lock_sock(sk);
3951                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3952                         ret = -EBUSY;
3953                 } else {
3954                         po->tp_tx_has_off = !!val;
3955                         ret = 0;
3956                 }
3957                 release_sock(sk);
3958                 return 0;
3959         }
3960         case PACKET_QDISC_BYPASS:
3961         {
3962                 int val;
3963
3964                 if (optlen != sizeof(val))
3965                         return -EINVAL;
3966                 if (copy_from_user(&val, optval, sizeof(val)))
3967                         return -EFAULT;
3968
3969                 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3970                 return 0;
3971         }
3972         default:
3973                 return -ENOPROTOOPT;
3974         }
3975 }
3976
3977 static int packet_getsockopt(struct socket *sock, int level, int optname,
3978                              char __user *optval, int __user *optlen)
3979 {
3980         int len;
3981         int val, lv = sizeof(val);
3982         struct sock *sk = sock->sk;
3983         struct packet_sock *po = pkt_sk(sk);
3984         void *data = &val;
3985         union tpacket_stats_u st;
3986         struct tpacket_rollover_stats rstats;
3987
3988         if (level != SOL_PACKET)
3989                 return -ENOPROTOOPT;
3990
3991         if (get_user(len, optlen))
3992                 return -EFAULT;
3993
3994         if (len < 0)
3995                 return -EINVAL;
3996
3997         switch (optname) {
3998         case PACKET_STATISTICS:
3999                 spin_lock_bh(&sk->sk_receive_queue.lock);
4000                 memcpy(&st, &po->stats, sizeof(st));
4001                 memset(&po->stats, 0, sizeof(po->stats));
4002                 spin_unlock_bh(&sk->sk_receive_queue.lock);
4003
4004                 if (po->tp_version == TPACKET_V3) {
4005                         lv = sizeof(struct tpacket_stats_v3);
4006                         st.stats3.tp_packets += st.stats3.tp_drops;
4007                         data = &st.stats3;
4008                 } else {
4009                         lv = sizeof(struct tpacket_stats);
4010                         st.stats1.tp_packets += st.stats1.tp_drops;
4011                         data = &st.stats1;
4012                 }
4013
4014                 break;
4015         case PACKET_AUXDATA:
4016                 val = packet_sock_flag(po, PACKET_SOCK_AUXDATA);
4017                 break;
4018         case PACKET_ORIGDEV:
4019                 val = packet_sock_flag(po, PACKET_SOCK_ORIGDEV);
4020                 break;
4021         case PACKET_VNET_HDR:
4022                 val = po->has_vnet_hdr;
4023                 break;
4024         case PACKET_VERSION:
4025                 val = po->tp_version;
4026                 break;
4027         case PACKET_HDRLEN:
4028                 if (len > sizeof(int))
4029                         len = sizeof(int);
4030                 if (len < sizeof(int))
4031                         return -EINVAL;
4032                 if (copy_from_user(&val, optval, len))
4033                         return -EFAULT;
4034                 switch (val) {
4035                 case TPACKET_V1:
4036                         val = sizeof(struct tpacket_hdr);
4037                         break;
4038                 case TPACKET_V2:
4039                         val = sizeof(struct tpacket2_hdr);
4040                         break;
4041                 case TPACKET_V3:
4042                         val = sizeof(struct tpacket3_hdr);
4043                         break;
4044                 default:
4045                         return -EINVAL;
4046                 }
4047                 break;
4048         case PACKET_RESERVE:
4049                 val = po->tp_reserve;
4050                 break;
4051         case PACKET_LOSS:
4052                 val = po->tp_loss;
4053                 break;
4054         case PACKET_TIMESTAMP:
4055                 val = po->tp_tstamp;
4056                 break;
4057         case PACKET_FANOUT:
4058                 val = (po->fanout ?
4059                        ((u32)po->fanout->id |
4060                         ((u32)po->fanout->type << 16) |
4061                         ((u32)po->fanout->flags << 24)) :
4062                        0);
4063                 break;
4064         case PACKET_ROLLOVER_STATS:
4065                 if (!po->rollover)
4066                         return -EINVAL;
4067                 rstats.tp_all = atomic_long_read(&po->rollover->num);
4068                 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4069                 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4070                 data = &rstats;
4071                 lv = sizeof(rstats);
4072                 break;
4073         case PACKET_TX_HAS_OFF:
4074                 val = po->tp_tx_has_off;
4075                 break;
4076         case PACKET_QDISC_BYPASS:
4077                 val = packet_use_direct_xmit(po);
4078                 break;
4079         default:
4080                 return -ENOPROTOOPT;
4081         }
4082
4083         if (len > lv)
4084                 len = lv;
4085         if (put_user(len, optlen))
4086                 return -EFAULT;
4087         if (copy_to_user(optval, data, len))
4088                 return -EFAULT;
4089         return 0;
4090 }
4091
4092
4093 #ifdef CONFIG_COMPAT
4094 static int compat_packet_setsockopt(struct socket *sock, int level, int optname,
4095                                     char __user *optval, unsigned int optlen)
4096 {
4097         struct packet_sock *po = pkt_sk(sock->sk);
4098
4099         if (level != SOL_PACKET)
4100                 return -ENOPROTOOPT;
4101
4102         if (optname == PACKET_FANOUT_DATA &&
4103             po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) {
4104                 optval = (char __user *)get_compat_bpf_fprog(optval);
4105                 if (!optval)
4106                         return -EFAULT;
4107                 optlen = sizeof(struct sock_fprog);
4108         }
4109
4110         return packet_setsockopt(sock, level, optname, optval, optlen);
4111 }
4112 #endif
4113
4114 static int packet_notifier(struct notifier_block *this,
4115                            unsigned long msg, void *ptr)
4116 {
4117         struct sock *sk;
4118         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4119         struct net *net = dev_net(dev);
4120
4121         rcu_read_lock();
4122         sk_for_each_rcu(sk, &net->packet.sklist) {
4123                 struct packet_sock *po = pkt_sk(sk);
4124
4125                 switch (msg) {
4126                 case NETDEV_UNREGISTER:
4127                         if (po->mclist)
4128                                 packet_dev_mclist_delete(dev, &po->mclist);
4129                         /* fallthrough */
4130
4131                 case NETDEV_DOWN:
4132                         if (dev->ifindex == po->ifindex) {
4133                                 spin_lock(&po->bind_lock);
4134                                 if (po->running) {
4135                                         __unregister_prot_hook(sk, false);
4136                                         sk->sk_err = ENETDOWN;
4137                                         if (!sock_flag(sk, SOCK_DEAD))
4138                                                 sk->sk_error_report(sk);
4139                                 }
4140                                 if (msg == NETDEV_UNREGISTER) {
4141                                         packet_cached_dev_reset(po);
4142                                         WRITE_ONCE(po->ifindex, -1);
4143                                         if (po->prot_hook.dev)
4144                                                 dev_put(po->prot_hook.dev);
4145                                         po->prot_hook.dev = NULL;
4146                                 }
4147                                 spin_unlock(&po->bind_lock);
4148                         }
4149                         break;
4150                 case NETDEV_UP:
4151                         if (dev->ifindex == po->ifindex) {
4152                                 spin_lock(&po->bind_lock);
4153                                 if (po->num)
4154                                         register_prot_hook(sk);
4155                                 spin_unlock(&po->bind_lock);
4156                         }
4157                         break;
4158                 }
4159         }
4160         rcu_read_unlock();
4161         return NOTIFY_DONE;
4162 }
4163
4164
4165 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4166                         unsigned long arg)
4167 {
4168         struct sock *sk = sock->sk;
4169
4170         switch (cmd) {
4171         case SIOCOUTQ:
4172         {
4173                 int amount = sk_wmem_alloc_get(sk);
4174
4175                 return put_user(amount, (int __user *)arg);
4176         }
4177         case SIOCINQ:
4178         {
4179                 struct sk_buff *skb;
4180                 int amount = 0;
4181
4182                 spin_lock_bh(&sk->sk_receive_queue.lock);
4183                 skb = skb_peek(&sk->sk_receive_queue);
4184                 if (skb)
4185                         amount = skb->len;
4186                 spin_unlock_bh(&sk->sk_receive_queue.lock);
4187                 return put_user(amount, (int __user *)arg);
4188         }
4189         case SIOCGSTAMP:
4190                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
4191         case SIOCGSTAMPNS:
4192                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
4193
4194 #ifdef CONFIG_INET
4195         case SIOCADDRT:
4196         case SIOCDELRT:
4197         case SIOCDARP:
4198         case SIOCGARP:
4199         case SIOCSARP:
4200         case SIOCGIFADDR:
4201         case SIOCSIFADDR:
4202         case SIOCGIFBRDADDR:
4203         case SIOCSIFBRDADDR:
4204         case SIOCGIFNETMASK:
4205         case SIOCSIFNETMASK:
4206         case SIOCGIFDSTADDR:
4207         case SIOCSIFDSTADDR:
4208         case SIOCSIFFLAGS:
4209                 return inet_dgram_ops.ioctl(sock, cmd, arg);
4210 #endif
4211
4212         default:
4213                 return -ENOIOCTLCMD;
4214         }
4215         return 0;
4216 }
4217
4218 static unsigned int packet_poll(struct file *file, struct socket *sock,
4219                                 poll_table *wait)
4220 {
4221         struct sock *sk = sock->sk;
4222         struct packet_sock *po = pkt_sk(sk);
4223         unsigned int mask = datagram_poll(file, sock, wait);
4224
4225         spin_lock_bh(&sk->sk_receive_queue.lock);
4226         if (po->rx_ring.pg_vec) {
4227                 if (!packet_previous_rx_frame(po, &po->rx_ring,
4228                         TP_STATUS_KERNEL))
4229                         mask |= POLLIN | POLLRDNORM;
4230         }
4231         if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
4232                 po->pressure = 0;
4233         spin_unlock_bh(&sk->sk_receive_queue.lock);
4234         spin_lock_bh(&sk->sk_write_queue.lock);
4235         if (po->tx_ring.pg_vec) {
4236                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4237                         mask |= POLLOUT | POLLWRNORM;
4238         }
4239         spin_unlock_bh(&sk->sk_write_queue.lock);
4240         return mask;
4241 }
4242
4243
4244 /* Dirty? Well, I still did not learn better way to account
4245  * for user mmaps.
4246  */
4247
4248 static void packet_mm_open(struct vm_area_struct *vma)
4249 {
4250         struct file *file = vma->vm_file;
4251         struct socket *sock = file->private_data;
4252         struct sock *sk = sock->sk;
4253
4254         if (sk)
4255                 atomic_inc(&pkt_sk(sk)->mapped);
4256 }
4257
4258 static void packet_mm_close(struct vm_area_struct *vma)
4259 {
4260         struct file *file = vma->vm_file;
4261         struct socket *sock = file->private_data;
4262         struct sock *sk = sock->sk;
4263
4264         if (sk)
4265                 atomic_dec(&pkt_sk(sk)->mapped);
4266 }
4267
4268 static const struct vm_operations_struct packet_mmap_ops = {
4269         .open   =       packet_mm_open,
4270         .close  =       packet_mm_close,
4271 };
4272
4273 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4274                         unsigned int len)
4275 {
4276         int i;
4277
4278         for (i = 0; i < len; i++) {
4279                 if (likely(pg_vec[i].buffer)) {
4280                         if (is_vmalloc_addr(pg_vec[i].buffer))
4281                                 vfree(pg_vec[i].buffer);
4282                         else
4283                                 free_pages((unsigned long)pg_vec[i].buffer,
4284                                            order);
4285                         pg_vec[i].buffer = NULL;
4286                 }
4287         }
4288         kfree(pg_vec);
4289 }
4290
4291 static char *alloc_one_pg_vec_page(unsigned long order)
4292 {
4293         char *buffer;
4294         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4295                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4296
4297         buffer = (char *) __get_free_pages(gfp_flags, order);
4298         if (buffer)
4299                 return buffer;
4300
4301         /* __get_free_pages failed, fall back to vmalloc */
4302         buffer = vzalloc((1 << order) * PAGE_SIZE);
4303         if (buffer)
4304                 return buffer;
4305
4306         /* vmalloc failed, lets dig into swap here */
4307         gfp_flags &= ~__GFP_NORETRY;
4308         buffer = (char *) __get_free_pages(gfp_flags, order);
4309         if (buffer)
4310                 return buffer;
4311
4312         /* complete and utter failure */
4313         return NULL;
4314 }
4315
4316 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4317 {
4318         unsigned int block_nr = req->tp_block_nr;
4319         struct pgv *pg_vec;
4320         int i;
4321
4322         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4323         if (unlikely(!pg_vec))
4324                 goto out;
4325
4326         for (i = 0; i < block_nr; i++) {
4327                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4328                 if (unlikely(!pg_vec[i].buffer))
4329                         goto out_free_pgvec;
4330         }
4331
4332 out:
4333         return pg_vec;
4334
4335 out_free_pgvec:
4336         free_pg_vec(pg_vec, order, block_nr);
4337         pg_vec = NULL;
4338         goto out;
4339 }
4340
4341 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4342                 int closing, int tx_ring)
4343 {
4344         struct pgv *pg_vec = NULL;
4345         struct packet_sock *po = pkt_sk(sk);
4346         unsigned long *rx_owner_map = NULL;
4347         int was_running, order = 0;
4348         struct packet_ring_buffer *rb;
4349         struct sk_buff_head *rb_queue;
4350         __be16 num;
4351         int err = -EINVAL;
4352         /* Added to avoid minimal code churn */
4353         struct tpacket_req *req = &req_u->req;
4354
4355         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4356         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4357
4358         err = -EBUSY;
4359         if (!closing) {
4360                 if (atomic_read(&po->mapped))
4361                         goto out;
4362                 if (packet_read_pending(rb))
4363                         goto out;
4364         }
4365
4366         if (req->tp_block_nr) {
4367                 unsigned int min_frame_size;
4368
4369                 /* Sanity tests and some calculations */
4370                 err = -EBUSY;
4371                 if (unlikely(rb->pg_vec))
4372                         goto out;
4373
4374                 switch (po->tp_version) {
4375                 case TPACKET_V1:
4376                         po->tp_hdrlen = TPACKET_HDRLEN;
4377                         break;
4378                 case TPACKET_V2:
4379                         po->tp_hdrlen = TPACKET2_HDRLEN;
4380                         break;
4381                 case TPACKET_V3:
4382                         po->tp_hdrlen = TPACKET3_HDRLEN;
4383                         break;
4384                 }
4385
4386                 err = -EINVAL;
4387                 if (unlikely((int)req->tp_block_size <= 0))
4388                         goto out;
4389                 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4390                         goto out;
4391                 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4392                 if (po->tp_version >= TPACKET_V3 &&
4393                     req->tp_block_size <
4394                     BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4395                         goto out;
4396                 if (unlikely(req->tp_frame_size < min_frame_size))
4397                         goto out;
4398                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4399                         goto out;
4400
4401                 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4402                 if (unlikely(rb->frames_per_block == 0))
4403                         goto out;
4404                 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4405                         goto out;
4406                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4407                                         req->tp_frame_nr))
4408                         goto out;
4409
4410                 err = -ENOMEM;
4411                 order = get_order(req->tp_block_size);
4412                 pg_vec = alloc_pg_vec(req, order);
4413                 if (unlikely(!pg_vec))
4414                         goto out;
4415                 switch (po->tp_version) {
4416                 case TPACKET_V3:
4417                         /* Block transmit is not supported yet */
4418                         if (!tx_ring) {
4419                                 init_prb_bdqc(po, rb, pg_vec, req_u);
4420                         } else {
4421                                 struct tpacket_req3 *req3 = &req_u->req3;
4422
4423                                 if (req3->tp_retire_blk_tov ||
4424                                     req3->tp_sizeof_priv ||
4425                                     req3->tp_feature_req_word) {
4426                                         err = -EINVAL;
4427                                         goto out_free_pg_vec;
4428                                 }
4429                         }
4430                         break;
4431                 default:
4432                         if (!tx_ring) {
4433                                 rx_owner_map = bitmap_alloc(req->tp_frame_nr,
4434                                         GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
4435                                 if (!rx_owner_map)
4436                                         goto out_free_pg_vec;
4437                         }
4438                         break;
4439                 }
4440         }
4441         /* Done */
4442         else {
4443                 err = -EINVAL;
4444                 if (unlikely(req->tp_frame_nr))
4445                         goto out;
4446         }
4447
4448
4449         /* Detach socket from network */
4450         spin_lock(&po->bind_lock);
4451         was_running = po->running;
4452         num = po->num;
4453         if (was_running) {
4454                 WRITE_ONCE(po->num, 0);
4455                 __unregister_prot_hook(sk, false);
4456         }
4457         spin_unlock(&po->bind_lock);
4458
4459         synchronize_net();
4460
4461         err = -EBUSY;
4462         mutex_lock(&po->pg_vec_lock);
4463         if (closing || atomic_read(&po->mapped) == 0) {
4464                 err = 0;
4465                 spin_lock_bh(&rb_queue->lock);
4466                 swap(rb->pg_vec, pg_vec);
4467                 if (po->tp_version <= TPACKET_V2)
4468                         swap(rb->rx_owner_map, rx_owner_map);
4469                 rb->frame_max = (req->tp_frame_nr - 1);
4470                 rb->head = 0;
4471                 rb->frame_size = req->tp_frame_size;
4472                 spin_unlock_bh(&rb_queue->lock);
4473
4474                 swap(rb->pg_vec_order, order);
4475                 swap(rb->pg_vec_len, req->tp_block_nr);
4476
4477                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4478                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4479                                                 tpacket_rcv : packet_rcv;
4480                 skb_queue_purge(rb_queue);
4481                 if (atomic_read(&po->mapped))
4482                         pr_err("packet_mmap: vma is busy: %d\n",
4483                                atomic_read(&po->mapped));
4484         }
4485         mutex_unlock(&po->pg_vec_lock);
4486
4487         spin_lock(&po->bind_lock);
4488         if (was_running) {
4489                 WRITE_ONCE(po->num, num);
4490                 register_prot_hook(sk);
4491         }
4492         spin_unlock(&po->bind_lock);
4493         if (pg_vec && (po->tp_version > TPACKET_V2)) {
4494                 /* Because we don't support block-based V3 on tx-ring */
4495                 if (!tx_ring)
4496                         prb_shutdown_retire_blk_timer(po, rb_queue);
4497         }
4498
4499 out_free_pg_vec:
4500         if (pg_vec) {
4501                 bitmap_free(rx_owner_map);
4502                 free_pg_vec(pg_vec, order, req->tp_block_nr);
4503         }
4504 out:
4505         return err;
4506 }
4507
4508 static int packet_mmap(struct file *file, struct socket *sock,
4509                 struct vm_area_struct *vma)
4510 {
4511         struct sock *sk = sock->sk;
4512         struct packet_sock *po = pkt_sk(sk);
4513         unsigned long size, expected_size;
4514         struct packet_ring_buffer *rb;
4515         unsigned long start;
4516         int err = -EINVAL;
4517         int i;
4518
4519         if (vma->vm_pgoff)
4520                 return -EINVAL;
4521
4522         mutex_lock(&po->pg_vec_lock);
4523
4524         expected_size = 0;
4525         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4526                 if (rb->pg_vec) {
4527                         expected_size += rb->pg_vec_len
4528                                                 * rb->pg_vec_pages
4529                                                 * PAGE_SIZE;
4530                 }
4531         }
4532
4533         if (expected_size == 0)
4534                 goto out;
4535
4536         size = vma->vm_end - vma->vm_start;
4537         if (size != expected_size)
4538                 goto out;
4539
4540         start = vma->vm_start;
4541         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4542                 if (rb->pg_vec == NULL)
4543                         continue;
4544
4545                 for (i = 0; i < rb->pg_vec_len; i++) {
4546                         struct page *page;
4547                         void *kaddr = rb->pg_vec[i].buffer;
4548                         int pg_num;
4549
4550                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4551                                 page = pgv_to_page(kaddr);
4552                                 err = vm_insert_page(vma, start, page);
4553                                 if (unlikely(err))
4554                                         goto out;
4555                                 start += PAGE_SIZE;
4556                                 kaddr += PAGE_SIZE;
4557                         }
4558                 }
4559         }
4560
4561         atomic_inc(&po->mapped);
4562         vma->vm_ops = &packet_mmap_ops;
4563         err = 0;
4564
4565 out:
4566         mutex_unlock(&po->pg_vec_lock);
4567         return err;
4568 }
4569
4570 static const struct proto_ops packet_ops_spkt = {
4571         .family =       PF_PACKET,
4572         .owner =        THIS_MODULE,
4573         .release =      packet_release,
4574         .bind =         packet_bind_spkt,
4575         .connect =      sock_no_connect,
4576         .socketpair =   sock_no_socketpair,
4577         .accept =       sock_no_accept,
4578         .getname =      packet_getname_spkt,
4579         .poll =         datagram_poll,
4580         .ioctl =        packet_ioctl,
4581         .listen =       sock_no_listen,
4582         .shutdown =     sock_no_shutdown,
4583         .setsockopt =   sock_no_setsockopt,
4584         .getsockopt =   sock_no_getsockopt,
4585         .sendmsg =      packet_sendmsg_spkt,
4586         .recvmsg =      packet_recvmsg,
4587         .mmap =         sock_no_mmap,
4588         .sendpage =     sock_no_sendpage,
4589 };
4590
4591 static const struct proto_ops packet_ops = {
4592         .family =       PF_PACKET,
4593         .owner =        THIS_MODULE,
4594         .release =      packet_release,
4595         .bind =         packet_bind,
4596         .connect =      sock_no_connect,
4597         .socketpair =   sock_no_socketpair,
4598         .accept =       sock_no_accept,
4599         .getname =      packet_getname,
4600         .poll =         packet_poll,
4601         .ioctl =        packet_ioctl,
4602         .listen =       sock_no_listen,
4603         .shutdown =     sock_no_shutdown,
4604         .setsockopt =   packet_setsockopt,
4605         .getsockopt =   packet_getsockopt,
4606 #ifdef CONFIG_COMPAT
4607         .compat_setsockopt = compat_packet_setsockopt,
4608 #endif
4609         .sendmsg =      packet_sendmsg,
4610         .recvmsg =      packet_recvmsg,
4611         .mmap =         packet_mmap,
4612         .sendpage =     sock_no_sendpage,
4613 };
4614
4615 static const struct net_proto_family packet_family_ops = {
4616         .family =       PF_PACKET,
4617         .create =       packet_create,
4618         .owner  =       THIS_MODULE,
4619 };
4620
4621 static struct notifier_block packet_netdev_notifier = {
4622         .notifier_call =        packet_notifier,
4623 };
4624
4625 #ifdef CONFIG_PROC_FS
4626
4627 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4628         __acquires(RCU)
4629 {
4630         struct net *net = seq_file_net(seq);
4631
4632         rcu_read_lock();
4633         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4634 }
4635
4636 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4637 {
4638         struct net *net = seq_file_net(seq);
4639         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4640 }
4641
4642 static void packet_seq_stop(struct seq_file *seq, void *v)
4643         __releases(RCU)
4644 {
4645         rcu_read_unlock();
4646 }
4647
4648 static int packet_seq_show(struct seq_file *seq, void *v)
4649 {
4650         if (v == SEQ_START_TOKEN)
4651                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
4652         else {
4653                 struct sock *s = sk_entry(v);
4654                 const struct packet_sock *po = pkt_sk(s);
4655
4656                 seq_printf(seq,
4657                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4658                            s,
4659                            refcount_read(&s->sk_refcnt),
4660                            s->sk_type,
4661                            ntohs(READ_ONCE(po->num)),
4662                            READ_ONCE(po->ifindex),
4663                            po->running,
4664                            atomic_read(&s->sk_rmem_alloc),
4665                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4666                            sock_i_ino(s));
4667         }
4668
4669         return 0;
4670 }
4671
4672 static const struct seq_operations packet_seq_ops = {
4673         .start  = packet_seq_start,
4674         .next   = packet_seq_next,
4675         .stop   = packet_seq_stop,
4676         .show   = packet_seq_show,
4677 };
4678
4679 static int packet_seq_open(struct inode *inode, struct file *file)
4680 {
4681         return seq_open_net(inode, file, &packet_seq_ops,
4682                             sizeof(struct seq_net_private));
4683 }
4684
4685 static const struct file_operations packet_seq_fops = {
4686         .owner          = THIS_MODULE,
4687         .open           = packet_seq_open,
4688         .read           = seq_read,
4689         .llseek         = seq_lseek,
4690         .release        = seq_release_net,
4691 };
4692
4693 #endif
4694
4695 static int __net_init packet_net_init(struct net *net)
4696 {
4697         mutex_init(&net->packet.sklist_lock);
4698         INIT_HLIST_HEAD(&net->packet.sklist);
4699
4700         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4701                 return -ENOMEM;
4702
4703         return 0;
4704 }
4705
4706 static void __net_exit packet_net_exit(struct net *net)
4707 {
4708         remove_proc_entry("packet", net->proc_net);
4709 }
4710
4711 static struct pernet_operations packet_net_ops = {
4712         .init = packet_net_init,
4713         .exit = packet_net_exit,
4714 };
4715
4716
4717 static void __exit packet_exit(void)
4718 {
4719         unregister_netdevice_notifier(&packet_netdev_notifier);
4720         unregister_pernet_subsys(&packet_net_ops);
4721         sock_unregister(PF_PACKET);
4722         proto_unregister(&packet_proto);
4723 }
4724
4725 static int __init packet_init(void)
4726 {
4727         int rc;
4728
4729         rc = proto_register(&packet_proto, 0);
4730         if (rc)
4731                 goto out;
4732         rc = sock_register(&packet_family_ops);
4733         if (rc)
4734                 goto out_proto;
4735         rc = register_pernet_subsys(&packet_net_ops);
4736         if (rc)
4737                 goto out_sock;
4738         rc = register_netdevice_notifier(&packet_netdev_notifier);
4739         if (rc)
4740                 goto out_pernet;
4741
4742         return 0;
4743
4744 out_pernet:
4745         unregister_pernet_subsys(&packet_net_ops);
4746 out_sock:
4747         sock_unregister(PF_PACKET);
4748 out_proto:
4749         proto_unregister(&packet_proto);
4750 out:
4751         return rc;
4752 }
4753
4754 module_init(packet_init);
4755 module_exit(packet_exit);
4756 MODULE_LICENSE("GPL");
4757 MODULE_ALIAS_NETPROTO(PF_PACKET);