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