net/packet/af_packet.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              PACKET - implements raw packet sockets.
   7  *
   8  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  11  *
  12  * Fixes:
  13  *              Alan Cox        :       verify_area() now used correctly
  14  *              Alan Cox        :       new skbuff lists, look ma no backlogs!
  15  *              Alan Cox        :       tidied skbuff lists.
  16  *              Alan Cox        :       Now uses generic datagram routines I
  17  *                                      added. Also fixed the peek/read crash
  18  *                                      from all old Linux datagram code.
  19  *              Alan Cox        :       Uses the improved datagram code.
  20  *              Alan Cox        :       Added NULL's for socket options.
  21  *              Alan Cox        :       Re-commented the code.
  22  *              Alan Cox        :       Use new kernel side addressing
  23  *              Rob Janssen     :       Correct MTU usage.
  24  *              Dave Platt      :       Counter leaks caused by incorrect
  25  *                                      interrupt locking and some slightly
  26  *                                      dubious gcc output. Can you read
  27  *                                      compiler: it said _VOLATILE_
  28  *      Richard Kooijman        :       Timestamp fixes.
  29  *              Alan Cox        :       New buffers. Use sk->mac.raw.
  30  *              Alan Cox        :       sendmsg/recvmsg support.
  31  *              Alan Cox        :       Protocol setting support
  32  *      Alexey Kuznetsov        :       Untied from IPv4 stack.
  33  *      Cyrus Durgin            :       Fixed kerneld for kmod.
  34  *      Michal Ostrowski        :       Module initialization cleanup.
  35  *         Ulises Alonso        :       Frame number limit removal and
  36  *                                      packet_set_ring memory leak.
  37  *              Eric Biederman  :       Allow for > 8 byte hardware addresses.
  38  *                                      The convention is that longer addresses
  39  *                                      will simply extend the hardware address
  40  *                                      byte arrays at the end of sockaddr_ll
  41  *                                      and packet_mreq.
  42  *              Johann Baudy    :       Added TX RING.
  43  *              Chetan Loke     :       Implemented TPACKET_V3 block abstraction
  44  *                                      layer.
  45  *                                      Copyright (C) 2011, <lokec@ccs.neu.edu>
  46  *
  47  *
  48  *              This program is free software; you can redistribute it and/or
  49  *              modify it under the terms of the GNU General Public License
  50  *              as published by the Free Software Foundation; either version
  51  *              2 of the License, or (at your option) any later version.
  52  *
  53  */
  54
  55 #include <linux/types.h>
  56 #include <linux/mm.h>
  57 #include <linux/capability.h>
  58 #include <linux/fcntl.h>
  59 #include <linux/socket.h>
  60 #include <linux/in.h>
  61 #include <linux/inet.h>
  62 #include <linux/netdevice.h>
  63 #include <linux/if_packet.h>
  64 #include <linux/wireless.h>
  65 #include <linux/kernel.h>
  66 #include <linux/kmod.h>
  67 #include <linux/slab.h>
  68 #include <linux/vmalloc.h>
  69 #include <net/net_namespace.h>
  70 #include <net/ip.h>
  71 #include <net/protocol.h>
  72 #include <linux/skbuff.h>
  73 #include <net/sock.h>
  74 #include <linux/errno.h>
  75 #include <linux/timer.h>
  76 #include <asm/uaccess.h>
  77 #include <asm/ioctls.h>
  78 #include <asm/page.h>
  79 #include <asm/cacheflush.h>
  80 #include <asm/io.h>
  81 #include <linux/proc_fs.h>
  82 #include <linux/seq_file.h>
  83 #include <linux/poll.h>
  84 #include <linux/module.h>
  85 #include <linux/init.h>
  86 #include <linux/mutex.h>
  87 #include <linux/if_vlan.h>
  88 #include <linux/virtio_net.h>
  89 #include <linux/errqueue.h>
  90 #include <linux/net_tstamp.h>
  91 #include <linux/percpu.h>
  92 #ifdef CONFIG_INET
  93 #include <net/inet_common.h>
  94 #endif
  95
  96 #include "internal.h"
  97
  98 /*
  99    Assumptions:
 100    - if device has no dev->hard_header routine, it adds and removes ll header
 101      inside itself. In this case ll header is invisible outside of device,
 102      but higher levels still should reserve dev->hard_header_len.
 103      Some devices are enough clever to reallocate skb, when header
 104      will not fit to reserved space (tunnel), another ones are silly
 105      (PPP).
 106    - packet socket receives packets with pulled ll header,
 107      so that SOCK_RAW should push it back.
 108
 109 On receive:
 110 -----------
 111
 112 Incoming, dev->hard_header!=NULL
 113    mac_header -> ll header
 114    data       -> data
 115
 116 Outgoing, dev->hard_header!=NULL
 117    mac_header -> ll header
 118    data       -> ll header
 119
 120 Incoming, dev->hard_header==NULL
 121    mac_header -> UNKNOWN position. It is very likely, that it points to ll
 122                  header.  PPP makes it, that is wrong, because introduce
 123                  assymetry between rx and tx paths.
 124    data       -> data
 125
 126 Outgoing, dev->hard_header==NULL
 127    mac_header -> data. ll header is still not built!
 128    data       -> data
 129
 130 Resume
 131   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 132
 133
 134 On transmit:
 135 ------------
 136
 137 dev->hard_header != NULL
 138    mac_header -> ll header
 139    data       -> ll header
 140
 141 dev->hard_header == NULL (ll header is added by device, we cannot control it)
 142    mac_header -> data
 143    data       -> data
 144
 145    We should set nh.raw on output to correct posistion,
 146    packet classifier depends on it.
 147  */
 148
 149 /* Private packet socket structures. */
 150
 151 /* identical to struct packet_mreq except it has
 152  * a longer address field.
 153  */
 154 struct packet_mreq_max {
 155         int             mr_ifindex;
 156         unsigned short  mr_type;
 157         unsigned short  mr_alen;
 158         unsigned char   mr_address[MAX_ADDR_LEN];
 159 };
 160
 161 union tpacket_uhdr {
 162         struct tpacket_hdr  *h1;
 163         struct tpacket2_hdr *h2;
 164         struct tpacket3_hdr *h3;
 165         void *raw;
 166 };
 167
 168 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 169                 int closing, int tx_ring);
 170
 171 #define V3_ALIGNMENT    (8)
 172
 173 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 174
 175 #define BLK_PLUS_PRIV(sz_of_priv) \
 176         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 177
 178 #define PGV_FROM_VMALLOC 1
 179
 180 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
 181 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
 182 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
 183 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
 184 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
 185 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
 186 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
 187
 188 struct packet_sock;
 189 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
 190 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 191                        struct packet_type *pt, struct net_device *orig_dev);
 192
 193 static void *packet_previous_frame(struct packet_sock *po,
 194                 struct packet_ring_buffer *rb,
 195                 int status);
 196 static void packet_increment_head(struct packet_ring_buffer *buff);
 197 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
 198                         struct tpacket_block_desc *);
 199 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 200                         struct packet_sock *);
 201 static void prb_retire_current_block(struct tpacket_kbdq_core *,
 202                 struct packet_sock *, unsigned int status);
 203 static int prb_queue_frozen(struct tpacket_kbdq_core *);
 204 static void prb_open_block(struct tpacket_kbdq_core *,
 205                 struct tpacket_block_desc *);
 206 static void prb_retire_rx_blk_timer_expired(unsigned long);
 207 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 208 static void prb_init_blk_timer(struct packet_sock *,
 209                 struct tpacket_kbdq_core *,
 210                 void (*func) (unsigned long));
 211 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 212 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 213                 struct tpacket3_hdr *);
 214 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 215                 struct tpacket3_hdr *);
 216 static void packet_flush_mclist(struct sock *sk);
 217
 218 struct packet_skb_cb {
 219         union {
 220                 struct sockaddr_pkt pkt;
 221                 union {
 222                         /* Trick: alias skb original length with
 223                          * ll.sll_family and ll.protocol in order
 224                          * to save room.
 225                          */
 226                         unsigned int origlen;
 227                         struct sockaddr_ll ll;
 228                 };
 229         } sa;
 230 };
 231
 232 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 233
 234 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 235 #define GET_PBLOCK_DESC(x, bid) \
 236         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 237 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 238         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 239 #define GET_NEXT_PRB_BLK_NUM(x) \
 240         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 241         ((x)->kactive_blk_num+1) : 0)
 242
 243 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 244 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 245
 246 static int packet_direct_xmit(struct sk_buff *skb)
 247 {
 248         struct net_device *dev = skb->dev;
 249         netdev_features_t features;
 250         struct netdev_queue *txq;
 251         int ret = NETDEV_TX_BUSY;
 252
 253         if (unlikely(!netif_running(dev) ||
 254                      !netif_carrier_ok(dev)))
 255                 goto drop;
 256
 257         features = netif_skb_features(skb);
 258         if (skb_needs_linearize(skb, features) &&
 259             __skb_linearize(skb))
 260                 goto drop;
 261
 262         txq = skb_get_tx_queue(dev, skb);
 263
 264         local_bh_disable();
 265
 266         HARD_TX_LOCK(dev, txq, smp_processor_id());
 267         if (!netif_xmit_frozen_or_drv_stopped(txq))
 268                 ret = netdev_start_xmit(skb, dev, txq, false);
 269         HARD_TX_UNLOCK(dev, txq);
 270
 271         local_bh_enable();
 272
 273         if (!dev_xmit_complete(ret))
 274                 kfree_skb(skb);
 275
 276         return ret;
 277 drop:
 278         atomic_long_inc(&dev->tx_dropped);
 279         kfree_skb(skb);
 280         return NET_XMIT_DROP;
 281 }
 282
 283 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
 284 {
 285         struct net_device *dev;
 286
 287         rcu_read_lock();
 288         dev = rcu_dereference(po->cached_dev);
 289         if (likely(dev))
 290                 dev_hold(dev);
 291         rcu_read_unlock();
 292
 293         return dev;
 294 }
 295
 296 static void packet_cached_dev_assign(struct packet_sock *po,
 297                                      struct net_device *dev)
 298 {
 299         rcu_assign_pointer(po->cached_dev, dev);
 300 }
 301
 302 static void packet_cached_dev_reset(struct packet_sock *po)
 303 {
 304         RCU_INIT_POINTER(po->cached_dev, NULL);
 305 }
 306
 307 static bool packet_use_direct_xmit(const struct packet_sock *po)
 308 {
 309         return po->xmit == packet_direct_xmit;
 310 }
 311
 312 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
 313 {
 314         return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
 315 }
 316
 317 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
 318 {
 319         const struct net_device_ops *ops = dev->netdev_ops;
 320         u16 queue_index;
 321
 322         if (ops->ndo_select_queue) {
 323                 queue_index = ops->ndo_select_queue(dev, skb, NULL,
 324                                                     __packet_pick_tx_queue);
 325                 queue_index = netdev_cap_txqueue(dev, queue_index);
 326         } else {
 327                 queue_index = __packet_pick_tx_queue(dev, skb);
 328         }
 329
 330         skb_set_queue_mapping(skb, queue_index);
 331 }
 332
 333 /* register_prot_hook must be invoked with the po->bind_lock held,
 334  * or from a context in which asynchronous accesses to the packet
 335  * socket is not possible (packet_create()).
 336  */
 337 static void register_prot_hook(struct sock *sk)
 338 {
 339         struct packet_sock *po = pkt_sk(sk);
 340
 341         if (!po->running) {
 342                 if (po->fanout)
 343                         __fanout_link(sk, po);
 344                 else
 345                         dev_add_pack(&po->prot_hook);
 346
 347                 sock_hold(sk);
 348                 po->running = 1;
 349         }
 350 }
 351
 352 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
 353  * held.   If the sync parameter is true, we will temporarily drop
 354  * the po->bind_lock and do a synchronize_net to make sure no
 355  * asynchronous packet processing paths still refer to the elements
 356  * of po->prot_hook.  If the sync parameter is false, it is the
 357  * callers responsibility to take care of this.
 358  */
 359 static void __unregister_prot_hook(struct sock *sk, bool sync)
 360 {
 361         struct packet_sock *po = pkt_sk(sk);
 362
 363         po->running = 0;
 364
 365         if (po->fanout)
 366                 __fanout_unlink(sk, po);
 367         else
 368                 __dev_remove_pack(&po->prot_hook);
 369
 370         __sock_put(sk);
 371
 372         if (sync) {
 373                 spin_unlock(&po->bind_lock);
 374                 synchronize_net();
 375                 spin_lock(&po->bind_lock);
 376         }
 377 }
 378
 379 static void unregister_prot_hook(struct sock *sk, bool sync)
 380 {
 381         struct packet_sock *po = pkt_sk(sk);
 382
 383         if (po->running)
 384                 __unregister_prot_hook(sk, sync);
 385 }
 386
 387 static inline struct page * __pure pgv_to_page(void *addr)
 388 {
 389         if (is_vmalloc_addr(addr))
 390                 return vmalloc_to_page(addr);
 391         return virt_to_page(addr);
 392 }
 393
 394 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 395 {
 396         union tpacket_uhdr h;
 397
 398         h.raw = frame;
 399         switch (po->tp_version) {
 400         case TPACKET_V1:
 401                 h.h1->tp_status = status;
 402                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 403                 break;
 404         case TPACKET_V2:
 405                 h.h2->tp_status = status;
 406                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 407                 break;
 408         case TPACKET_V3:
 409         default:
 410                 WARN(1, "TPACKET version not supported.\n");
 411                 BUG();
 412         }
 413
 414         smp_wmb();
 415 }
 416
 417 static int __packet_get_status(struct packet_sock *po, void *frame)
 418 {
 419         union tpacket_uhdr h;
 420
 421         smp_rmb();
 422
 423         h.raw = frame;
 424         switch (po->tp_version) {
 425         case TPACKET_V1:
 426                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 427                 return h.h1->tp_status;
 428         case TPACKET_V2:
 429                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 430                 return h.h2->tp_status;
 431         case TPACKET_V3:
 432         default:
 433                 WARN(1, "TPACKET version not supported.\n");
 434                 BUG();
 435                 return 0;
 436         }
 437 }
 438
 439 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
 440                                    unsigned int flags)
 441 {
 442         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 443
 444         if (shhwtstamps &&
 445             (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
 446             ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
 447                 return TP_STATUS_TS_RAW_HARDWARE;
 448
 449         if (ktime_to_timespec_cond(skb->tstamp, ts))
 450                 return TP_STATUS_TS_SOFTWARE;
 451
 452         return 0;
 453 }
 454
 455 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
 456                                     struct sk_buff *skb)
 457 {
 458         union tpacket_uhdr h;
 459         struct timespec ts;
 460         __u32 ts_status;
 461
 462         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
 463                 return 0;
 464
 465         h.raw = frame;
 466         switch (po->tp_version) {
 467         case TPACKET_V1:
 468                 h.h1->tp_sec = ts.tv_sec;
 469                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
 470                 break;
 471         case TPACKET_V2:
 472                 h.h2->tp_sec = ts.tv_sec;
 473                 h.h2->tp_nsec = ts.tv_nsec;
 474                 break;
 475         case TPACKET_V3:
 476         default:
 477                 WARN(1, "TPACKET version not supported.\n");
 478                 BUG();
 479         }
 480
 481         /* one flush is safe, as both fields always lie on the same cacheline */
 482         flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
 483         smp_wmb();
 484
 485         return ts_status;
 486 }
 487
 488 static void *packet_lookup_frame(struct packet_sock *po,
 489                 struct packet_ring_buffer *rb,
 490                 unsigned int position,
 491                 int status)
 492 {
 493         unsigned int pg_vec_pos, frame_offset;
 494         union tpacket_uhdr h;
 495
 496         pg_vec_pos = position / rb->frames_per_block;
 497         frame_offset = position % rb->frames_per_block;
 498
 499         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 500                 (frame_offset * rb->frame_size);
 501
 502         if (status != __packet_get_status(po, h.raw))
 503                 return NULL;
 504
 505         return h.raw;
 506 }
 507
 508 static void *packet_current_frame(struct packet_sock *po,
 509                 struct packet_ring_buffer *rb,
 510                 int status)
 511 {
 512         return packet_lookup_frame(po, rb, rb->head, status);
 513 }
 514
 515 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 516 {
 517         del_timer_sync(&pkc->retire_blk_timer);
 518 }
 519
 520 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 521                 int tx_ring,
 522                 struct sk_buff_head *rb_queue)
 523 {
 524         struct tpacket_kbdq_core *pkc;
 525
 526         pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
 527                         GET_PBDQC_FROM_RB(&po->rx_ring);
 528
 529         spin_lock_bh(&rb_queue->lock);
 530         pkc->delete_blk_timer = 1;
 531         spin_unlock_bh(&rb_queue->lock);
 532
 533         prb_del_retire_blk_timer(pkc);
 534 }
 535
 536 static void prb_init_blk_timer(struct packet_sock *po,
 537                 struct tpacket_kbdq_core *pkc,
 538                 void (*func) (unsigned long))
 539 {
 540         init_timer(&pkc->retire_blk_timer);
 541         pkc->retire_blk_timer.data = (long)po;
 542         pkc->retire_blk_timer.function = func;
 543         pkc->retire_blk_timer.expires = jiffies;
 544 }
 545
 546 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
 547 {
 548         struct tpacket_kbdq_core *pkc;
 549
 550         if (tx_ring)
 551                 BUG();
 552
 553         pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
 554                         GET_PBDQC_FROM_RB(&po->rx_ring);
 555         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
 556 }
 557
 558 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 559                                 int blk_size_in_bytes)
 560 {
 561         struct net_device *dev;
 562         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 563         struct ethtool_cmd ecmd;
 564         int err;
 565         u32 speed;
 566
 567         rtnl_lock();
 568         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 569         if (unlikely(!dev)) {
 570                 rtnl_unlock();
 571                 return DEFAULT_PRB_RETIRE_TOV;
 572         }
 573         err = __ethtool_get_settings(dev, &ecmd);
 574         speed = ethtool_cmd_speed(&ecmd);
 575         rtnl_unlock();
 576         if (!err) {
 577                 /*
 578                  * If the link speed is so slow you don't really
 579                  * need to worry about perf anyways
 580                  */
 581                 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
 582                         return DEFAULT_PRB_RETIRE_TOV;
 583                 } else {
 584                         msec = 1;
 585                         div = speed / 1000;
 586                 }
 587         }
 588
 589         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 590
 591         if (div)
 592                 mbits /= div;
 593
 594         tmo = mbits * msec;
 595
 596         if (div)
 597                 return tmo+1;
 598         return tmo;
 599 }
 600
 601 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 602                         union tpacket_req_u *req_u)
 603 {
 604         p1->feature_req_word = req_u->req3.tp_feature_req_word;
 605 }
 606
 607 static void init_prb_bdqc(struct packet_sock *po,
 608                         struct packet_ring_buffer *rb,
 609                         struct pgv *pg_vec,
 610                         union tpacket_req_u *req_u, int tx_ring)
 611 {
 612         struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
 613         struct tpacket_block_desc *pbd;
 614
 615         memset(p1, 0x0, sizeof(*p1));
 616
 617         p1->knxt_seq_num = 1;
 618         p1->pkbdq = pg_vec;
 619         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 620         p1->pkblk_start = pg_vec[0].buffer;
 621         p1->kblk_size = req_u->req3.tp_block_size;
 622         p1->knum_blocks = req_u->req3.tp_block_nr;
 623         p1->hdrlen = po->tp_hdrlen;
 624         p1->version = po->tp_version;
 625         p1->last_kactive_blk_num = 0;
 626         po->stats.stats3.tp_freeze_q_cnt = 0;
 627         if (req_u->req3.tp_retire_blk_tov)
 628                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 629         else
 630                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 631                                                 req_u->req3.tp_block_size);
 632         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 633         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 634
 635         p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
 636         prb_init_ft_ops(p1, req_u);
 637         prb_setup_retire_blk_timer(po, tx_ring);
 638         prb_open_block(p1, pbd);
 639 }
 640
 641 /*  Do NOT update the last_blk_num first.
 642  *  Assumes sk_buff_head lock is held.
 643  */
 644 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 645 {
 646         mod_timer(&pkc->retire_blk_timer,
 647                         jiffies + pkc->tov_in_jiffies);
 648         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 649 }
 650
 651 /*
 652  * Timer logic:
 653  * 1) We refresh the timer only when we open a block.
 654  *    By doing this we don't waste cycles refreshing the timer
 655  *        on packet-by-packet basis.
 656  *
 657  * With a 1MB block-size, on a 1Gbps line, it will take
 658  * i) ~8 ms to fill a block + ii) memcpy etc.
 659  * In this cut we are not accounting for the memcpy time.
 660  *
 661  * So, if the user sets the 'tmo' to 10ms then the timer
 662  * will never fire while the block is still getting filled
 663  * (which is what we want). However, the user could choose
 664  * to close a block early and that's fine.
 665  *
 666  * But when the timer does fire, we check whether or not to refresh it.
 667  * Since the tmo granularity is in msecs, it is not too expensive
 668  * to refresh the timer, lets say every '8' msecs.
 669  * Either the user can set the 'tmo' or we can derive it based on
 670  * a) line-speed and b) block-size.
 671  * prb_calc_retire_blk_tmo() calculates the tmo.
 672  *
 673  */
 674 static void prb_retire_rx_blk_timer_expired(unsigned long data)
 675 {
 676         struct packet_sock *po = (struct packet_sock *)data;
 677         struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 678         unsigned int frozen;
 679         struct tpacket_block_desc *pbd;
 680
 681         spin_lock(&po->sk.sk_receive_queue.lock);
 682
 683         frozen = prb_queue_frozen(pkc);
 684         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 685
 686         if (unlikely(pkc->delete_blk_timer))
 687                 goto out;
 688
 689         /* We only need to plug the race when the block is partially filled.
 690          * tpacket_rcv:
 691          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 692          *              copy_bits() is in progress ...
 693          *              timer fires on other cpu:
 694          *              we can't retire the current block because copy_bits
 695          *              is in progress.
 696          *
 697          */
 698         if (BLOCK_NUM_PKTS(pbd)) {
 699                 while (atomic_read(&pkc->blk_fill_in_prog)) {
 700                         /* Waiting for skb_copy_bits to finish... */
 701                         cpu_relax();
 702                 }
 703         }
 704
 705         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 706                 if (!frozen) {
 707                         if (!BLOCK_NUM_PKTS(pbd)) {
 708                                 /* An empty block. Just refresh the timer. */
 709                                 goto refresh_timer;
 710                         }
 711                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 712                         if (!prb_dispatch_next_block(pkc, po))
 713                                 goto refresh_timer;
 714                         else
 715                                 goto out;
 716                 } else {
 717                         /* Case 1. Queue was frozen because user-space was
 718                          *         lagging behind.
 719                          */
 720                         if (prb_curr_blk_in_use(pkc, pbd)) {
 721                                 /*
 722                                  * Ok, user-space is still behind.
 723                                  * So just refresh the timer.
 724                                  */
 725                                 goto refresh_timer;
 726                         } else {
 727                                /* Case 2. queue was frozen,user-space caught up,
 728                                 * now the link went idle && the timer fired.
 729                                 * We don't have a block to close.So we open this
 730                                 * block and restart the timer.
 731                                 * opening a block thaws the queue,restarts timer
 732                                 * Thawing/timer-refresh is a side effect.
 733                                 */
 734                                 prb_open_block(pkc, pbd);
 735                                 goto out;
 736                         }
 737                 }
 738         }
 739
 740 refresh_timer:
 741         _prb_refresh_rx_retire_blk_timer(pkc);
 742
 743 out:
 744         spin_unlock(&po->sk.sk_receive_queue.lock);
 745 }
 746
 747 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 748                 struct tpacket_block_desc *pbd1, __u32 status)
 749 {
 750         /* Flush everything minus the block header */
 751
 752 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 753         u8 *start, *end;
 754
 755         start = (u8 *)pbd1;
 756
 757         /* Skip the block header(we know header WILL fit in 4K) */
 758         start += PAGE_SIZE;
 759
 760         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 761         for (; start < end; start += PAGE_SIZE)
 762                 flush_dcache_page(pgv_to_page(start));
 763
 764         smp_wmb();
 765 #endif
 766
 767         /* Now update the block status. */
 768
 769         BLOCK_STATUS(pbd1) = status;
 770
 771         /* Flush the block header */
 772
 773 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 774         start = (u8 *)pbd1;
 775         flush_dcache_page(pgv_to_page(start));
 776
 777         smp_wmb();
 778 #endif
 779 }
 780
 781 /*
 782  * Side effect:
 783  *
 784  * 1) flush the block
 785  * 2) Increment active_blk_num
 786  *
 787  * Note:We DONT refresh the timer on purpose.
 788  *      Because almost always the next block will be opened.
 789  */
 790 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 791                 struct tpacket_block_desc *pbd1,
 792                 struct packet_sock *po, unsigned int stat)
 793 {
 794         __u32 status = TP_STATUS_USER | stat;
 795
 796         struct tpacket3_hdr *last_pkt;
 797         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 798         struct sock *sk = &po->sk;
 799
 800         if (po->stats.stats3.tp_drops)
 801                 status |= TP_STATUS_LOSING;
 802
 803         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 804         last_pkt->tp_next_offset = 0;
 805
 806         /* Get the ts of the last pkt */
 807         if (BLOCK_NUM_PKTS(pbd1)) {
 808                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 809                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 810         } else {
 811                 /* Ok, we tmo'd - so get the current time.
 812                  *
 813                  * It shouldn't really happen as we don't close empty
 814                  * blocks. See prb_retire_rx_blk_timer_expired().
 815                  */
 816                 struct timespec ts;
 817                 getnstimeofday(&ts);
 818                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 819                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 820         }
 821
 822         smp_wmb();
 823
 824         /* Flush the block */
 825         prb_flush_block(pkc1, pbd1, status);
 826
 827         sk->sk_data_ready(sk);
 828
 829         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 830 }
 831
 832 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 833 {
 834         pkc->reset_pending_on_curr_blk = 0;
 835 }
 836
 837 /*
 838  * Side effect of opening a block:
 839  *
 840  * 1) prb_queue is thawed.
 841  * 2) retire_blk_timer is refreshed.
 842  *
 843  */
 844 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 845         struct tpacket_block_desc *pbd1)
 846 {
 847         struct timespec ts;
 848         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 849
 850         smp_rmb();
 851
 852         /* We could have just memset this but we will lose the
 853          * flexibility of making the priv area sticky
 854          */
 855
 856         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 857         BLOCK_NUM_PKTS(pbd1) = 0;
 858         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 859
 860         getnstimeofday(&ts);
 861
 862         h1->ts_first_pkt.ts_sec = ts.tv_sec;
 863         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 864
 865         pkc1->pkblk_start = (char *)pbd1;
 866         pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 867
 868         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 869         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 870
 871         pbd1->version = pkc1->version;
 872         pkc1->prev = pkc1->nxt_offset;
 873         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 874
 875         prb_thaw_queue(pkc1);
 876         _prb_refresh_rx_retire_blk_timer(pkc1);
 877
 878         smp_wmb();
 879 }
 880
 881 /*
 882  * Queue freeze logic:
 883  * 1) Assume tp_block_nr = 8 blocks.
 884  * 2) At time 't0', user opens Rx ring.
 885  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 886  * 4) user-space is either sleeping or processing block '0'.
 887  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 888  *    it will close block-7,loop around and try to fill block '0'.
 889  *    call-flow:
 890  *    __packet_lookup_frame_in_block
 891  *      prb_retire_current_block()
 892  *      prb_dispatch_next_block()
 893  *        |->(BLOCK_STATUS == USER) evaluates to true
 894  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 895  * 6) Now there are two cases:
 896  *    6.1) Link goes idle right after the queue is frozen.
 897  *         But remember, the last open_block() refreshed the timer.
 898  *         When this timer expires,it will refresh itself so that we can
 899  *         re-open block-0 in near future.
 900  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 901  *         case and __packet_lookup_frame_in_block will check if block-0
 902  *         is free and can now be re-used.
 903  */
 904 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 905                                   struct packet_sock *po)
 906 {
 907         pkc->reset_pending_on_curr_blk = 1;
 908         po->stats.stats3.tp_freeze_q_cnt++;
 909 }
 910
 911 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 912
 913 /*
 914  * If the next block is free then we will dispatch it
 915  * and return a good offset.
 916  * Else, we will freeze the queue.
 917  * So, caller must check the return value.
 918  */
 919 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 920                 struct packet_sock *po)
 921 {
 922         struct tpacket_block_desc *pbd;
 923
 924         smp_rmb();
 925
 926         /* 1. Get current block num */
 927         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 928
 929         /* 2. If this block is currently in_use then freeze the queue */
 930         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 931                 prb_freeze_queue(pkc, po);
 932                 return NULL;
 933         }
 934
 935         /*
 936          * 3.
 937          * open this block and return the offset where the first packet
 938          * needs to get stored.
 939          */
 940         prb_open_block(pkc, pbd);
 941         return (void *)pkc->nxt_offset;
 942 }
 943
 944 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 945                 struct packet_sock *po, unsigned int status)
 946 {
 947         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 948
 949         /* retire/close the current block */
 950         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 951                 /*
 952                  * Plug the case where copy_bits() is in progress on
 953                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 954                  * have space to copy the pkt in the current block and
 955                  * called prb_retire_current_block()
 956                  *
 957                  * We don't need to worry about the TMO case because
 958                  * the timer-handler already handled this case.
 959                  */
 960                 if (!(status & TP_STATUS_BLK_TMO)) {
 961                         while (atomic_read(&pkc->blk_fill_in_prog)) {
 962                                 /* Waiting for skb_copy_bits to finish... */
 963                                 cpu_relax();
 964                         }
 965                 }
 966                 prb_close_block(pkc, pbd, po, status);
 967                 return;
 968         }
 969 }
 970
 971 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
 972                                       struct tpacket_block_desc *pbd)
 973 {
 974         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 975 }
 976
 977 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 978 {
 979         return pkc->reset_pending_on_curr_blk;
 980 }
 981
 982 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 983 {
 984         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 985         atomic_dec(&pkc->blk_fill_in_prog);
 986 }
 987
 988 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 989                         struct tpacket3_hdr *ppd)
 990 {
 991         ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
 992 }
 993
 994 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 995                         struct tpacket3_hdr *ppd)
 996 {
 997         ppd->hv1.tp_rxhash = 0;
 998 }
 999
1000 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1001                         struct tpacket3_hdr *ppd)
1002 {
1003         if (skb_vlan_tag_present(pkc->skb)) {
1004                 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1005                 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1006                 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1007         } else {
1008                 ppd->hv1.tp_vlan_tci = 0;
1009                 ppd->hv1.tp_vlan_tpid = 0;
1010                 ppd->tp_status = TP_STATUS_AVAILABLE;
1011         }
1012 }
1013
1014 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1015                         struct tpacket3_hdr *ppd)
1016 {
1017         ppd->hv1.tp_padding = 0;
1018         prb_fill_vlan_info(pkc, ppd);
1019
1020         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1021                 prb_fill_rxhash(pkc, ppd);
1022         else
1023                 prb_clear_rxhash(pkc, ppd);
1024 }
1025
1026 static void prb_fill_curr_block(char *curr,
1027                                 struct tpacket_kbdq_core *pkc,
1028                                 struct tpacket_block_desc *pbd,
1029                                 unsigned int len)
1030 {
1031         struct tpacket3_hdr *ppd;
1032
1033         ppd  = (struct tpacket3_hdr *)curr;
1034         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1035         pkc->prev = curr;
1036         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1037         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1038         BLOCK_NUM_PKTS(pbd) += 1;
1039         atomic_inc(&pkc->blk_fill_in_prog);
1040         prb_run_all_ft_ops(pkc, ppd);
1041 }
1042
1043 /* Assumes caller has the sk->rx_queue.lock */
1044 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1045                                             struct sk_buff *skb,
1046                                                 int status,
1047                                             unsigned int len
1048                                             )
1049 {
1050         struct tpacket_kbdq_core *pkc;
1051         struct tpacket_block_desc *pbd;
1052         char *curr, *end;
1053
1054         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1055         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1056
1057         /* Queue is frozen when user space is lagging behind */
1058         if (prb_queue_frozen(pkc)) {
1059                 /*
1060                  * Check if that last block which caused the queue to freeze,
1061                  * is still in_use by user-space.
1062                  */
1063                 if (prb_curr_blk_in_use(pkc, pbd)) {
1064                         /* Can't record this packet */
1065                         return NULL;
1066                 } else {
1067                         /*
1068                          * Ok, the block was released by user-space.
1069                          * Now let's open that block.
1070                          * opening a block also thaws the queue.
1071                          * Thawing is a side effect.
1072                          */
1073                         prb_open_block(pkc, pbd);
1074                 }
1075         }
1076
1077         smp_mb();
1078         curr = pkc->nxt_offset;
1079         pkc->skb = skb;
1080         end = (char *)pbd + pkc->kblk_size;
1081
1082         /* first try the current block */
1083         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1084                 prb_fill_curr_block(curr, pkc, pbd, len);
1085                 return (void *)curr;
1086         }
1087
1088         /* Ok, close the current block */
1089         prb_retire_current_block(pkc, po, 0);
1090
1091         /* Now, try to dispatch the next block */
1092         curr = (char *)prb_dispatch_next_block(pkc, po);
1093         if (curr) {
1094                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1095                 prb_fill_curr_block(curr, pkc, pbd, len);
1096                 return (void *)curr;
1097         }
1098
1099         /*
1100          * No free blocks are available.user_space hasn't caught up yet.
1101          * Queue was just frozen and now this packet will get dropped.
1102          */
1103         return NULL;
1104 }
1105
1106 static void *packet_current_rx_frame(struct packet_sock *po,
1107                                             struct sk_buff *skb,
1108                                             int status, unsigned int len)
1109 {
1110         char *curr = NULL;
1111         switch (po->tp_version) {
1112         case TPACKET_V1:
1113         case TPACKET_V2:
1114                 curr = packet_lookup_frame(po, &po->rx_ring,
1115                                         po->rx_ring.head, status);
1116                 return curr;
1117         case TPACKET_V3:
1118                 return __packet_lookup_frame_in_block(po, skb, status, len);
1119         default:
1120                 WARN(1, "TPACKET version not supported\n");
1121                 BUG();
1122                 return NULL;
1123         }
1124 }
1125
1126 static void *prb_lookup_block(struct packet_sock *po,
1127                                      struct packet_ring_buffer *rb,
1128                                      unsigned int idx,
1129                                      int status)
1130 {
1131         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1132         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1133
1134         if (status != BLOCK_STATUS(pbd))
1135                 return NULL;
1136         return pbd;
1137 }
1138
1139 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1140 {
1141         unsigned int prev;
1142         if (rb->prb_bdqc.kactive_blk_num)
1143                 prev = rb->prb_bdqc.kactive_blk_num-1;
1144         else
1145                 prev = rb->prb_bdqc.knum_blocks-1;
1146         return prev;
1147 }
1148
1149 /* Assumes caller has held the rx_queue.lock */
1150 static void *__prb_previous_block(struct packet_sock *po,
1151                                          struct packet_ring_buffer *rb,
1152                                          int status)
1153 {
1154         unsigned int previous = prb_previous_blk_num(rb);
1155         return prb_lookup_block(po, rb, previous, status);
1156 }
1157
1158 static void *packet_previous_rx_frame(struct packet_sock *po,
1159                                              struct packet_ring_buffer *rb,
1160                                              int status)
1161 {
1162         if (po->tp_version <= TPACKET_V2)
1163                 return packet_previous_frame(po, rb, status);
1164
1165         return __prb_previous_block(po, rb, status);
1166 }
1167
1168 static void packet_increment_rx_head(struct packet_sock *po,
1169                                             struct packet_ring_buffer *rb)
1170 {
1171         switch (po->tp_version) {
1172         case TPACKET_V1:
1173         case TPACKET_V2:
1174                 return packet_increment_head(rb);
1175         case TPACKET_V3:
1176         default:
1177                 WARN(1, "TPACKET version not supported.\n");
1178                 BUG();
1179                 return;
1180         }
1181 }
1182
1183 static void *packet_previous_frame(struct packet_sock *po,
1184                 struct packet_ring_buffer *rb,
1185                 int status)
1186 {
1187         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1188         return packet_lookup_frame(po, rb, previous, status);
1189 }
1190
1191 static void packet_increment_head(struct packet_ring_buffer *buff)
1192 {
1193         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1194 }
1195
1196 static void packet_inc_pending(struct packet_ring_buffer *rb)
1197 {
1198         this_cpu_inc(*rb->pending_refcnt);
1199 }
1200
1201 static void packet_dec_pending(struct packet_ring_buffer *rb)
1202 {
1203         this_cpu_dec(*rb->pending_refcnt);
1204 }
1205
1206 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1207 {
1208         unsigned int refcnt = 0;
1209         int cpu;
1210
1211         /* We don't use pending refcount in rx_ring. */
1212         if (rb->pending_refcnt == NULL)
1213                 return 0;
1214
1215         for_each_possible_cpu(cpu)
1216                 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1217
1218         return refcnt;
1219 }
1220
1221 static int packet_alloc_pending(struct packet_sock *po)
1222 {
1223         po->rx_ring.pending_refcnt = NULL;
1224
1225         po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1226         if (unlikely(po->tx_ring.pending_refcnt == NULL))
1227                 return -ENOBUFS;
1228
1229         return 0;
1230 }
1231
1232 static void packet_free_pending(struct packet_sock *po)
1233 {
1234         free_percpu(po->tx_ring.pending_refcnt);
1235 }
1236
1237 #define ROOM_POW_OFF    2
1238 #define ROOM_NONE       0x0
1239 #define ROOM_LOW        0x1
1240 #define ROOM_NORMAL     0x2
1241
1242 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1243 {
1244         int idx, len;
1245
1246         len = po->rx_ring.frame_max + 1;
1247         idx = po->rx_ring.head;
1248         if (pow_off)
1249                 idx += len >> pow_off;
1250         if (idx >= len)
1251                 idx -= len;
1252         return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1253 }
1254
1255 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1256 {
1257         int idx, len;
1258
1259         len = po->rx_ring.prb_bdqc.knum_blocks;
1260         idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1261         if (pow_off)
1262                 idx += len >> pow_off;
1263         if (idx >= len)
1264                 idx -= len;
1265         return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1266 }
1267
1268 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1269 {
1270         struct sock *sk = &po->sk;
1271         int ret = ROOM_NONE;
1272
1273         if (po->prot_hook.func != tpacket_rcv) {
1274                 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1275                                           - (skb ? skb->truesize : 0);
1276                 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1277                         return ROOM_NORMAL;
1278                 else if (avail > 0)
1279                         return ROOM_LOW;
1280                 else
1281                         return ROOM_NONE;
1282         }
1283
1284         if (po->tp_version == TPACKET_V3) {
1285                 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1286                         ret = ROOM_NORMAL;
1287                 else if (__tpacket_v3_has_room(po, 0))
1288                         ret = ROOM_LOW;
1289         } else {
1290                 if (__tpacket_has_room(po, ROOM_POW_OFF))
1291                         ret = ROOM_NORMAL;
1292                 else if (__tpacket_has_room(po, 0))
1293                         ret = ROOM_LOW;
1294         }
1295
1296         return ret;
1297 }
1298
1299 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1300 {
1301         int ret;
1302         bool has_room;
1303
1304         spin_lock_bh(&po->sk.sk_receive_queue.lock);
1305         ret = __packet_rcv_has_room(po, skb);
1306         has_room = ret == ROOM_NORMAL;
1307         if (po->pressure == has_room)
1308                 po->pressure = !has_room;
1309         spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1310
1311         return ret;
1312 }
1313
1314 static void packet_sock_destruct(struct sock *sk)
1315 {
1316         skb_queue_purge(&sk->sk_error_queue);
1317
1318         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1319         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1320
1321         if (!sock_flag(sk, SOCK_DEAD)) {
1322                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1323                 return;
1324         }
1325
1326         sk_refcnt_debug_dec(sk);
1327 }
1328
1329 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1330 {
1331         int x = atomic_read(&f->rr_cur) + 1;
1332
1333         if (x >= num)
1334                 x = 0;
1335
1336         return x;
1337 }
1338
1339 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1340 {
1341         u32 rxhash;
1342         int i, count = 0;
1343
1344         rxhash = skb_get_hash(skb);
1345         for (i = 0; i < ROLLOVER_HLEN; i++)
1346                 if (po->rollover->history[i] == rxhash)
1347                         count++;
1348
1349         po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1350         return count > (ROLLOVER_HLEN >> 1);
1351 }
1352
1353 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1354                                       struct sk_buff *skb,
1355                                       unsigned int num)
1356 {
1357         return reciprocal_scale(skb_get_hash(skb), num);
1358 }
1359
1360 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1361                                     struct sk_buff *skb,
1362                                     unsigned int num)
1363 {
1364         int cur, old;
1365
1366         cur = atomic_read(&f->rr_cur);
1367         while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1368                                      fanout_rr_next(f, num))) != cur)
1369                 cur = old;
1370         return cur;
1371 }
1372
1373 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1374                                      struct sk_buff *skb,
1375                                      unsigned int num)
1376 {
1377         return smp_processor_id() % num;
1378 }
1379
1380 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1381                                      struct sk_buff *skb,
1382                                      unsigned int num)
1383 {
1384         return prandom_u32_max(num);
1385 }
1386
1387 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1388                                           struct sk_buff *skb,
1389                                           unsigned int idx, bool try_self,
1390                                           unsigned int num)
1391 {
1392         struct packet_sock *po, *po_next, *po_skip = NULL;
1393         unsigned int i, j, room = ROOM_NONE;
1394
1395         po = pkt_sk(f->arr[idx]);
1396
1397         if (try_self) {
1398                 room = packet_rcv_has_room(po, skb);
1399                 if (room == ROOM_NORMAL ||
1400                     (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1401                         return idx;
1402                 po_skip = po;
1403         }
1404
1405         i = j = min_t(int, po->rollover->sock, num - 1);
1406         do {
1407                 po_next = pkt_sk(f->arr[i]);
1408                 if (po_next != po_skip && !po_next->pressure &&
1409                     packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1410                         if (i != j)
1411                                 po->rollover->sock = i;
1412                         atomic_long_inc(&po->rollover->num);
1413                         if (room == ROOM_LOW)
1414                                 atomic_long_inc(&po->rollover->num_huge);
1415                         return i;
1416                 }
1417
1418                 if (++i == num)
1419                         i = 0;
1420         } while (i != j);
1421
1422         atomic_long_inc(&po->rollover->num_failed);
1423         return idx;
1424 }
1425
1426 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1427                                     struct sk_buff *skb,
1428                                     unsigned int num)
1429 {
1430         return skb_get_queue_mapping(skb) % num;
1431 }
1432
1433 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1434 {
1435         return f->flags & (flag >> 8);
1436 }
1437
1438 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1439                              struct packet_type *pt, struct net_device *orig_dev)
1440 {
1441         struct packet_fanout *f = pt->af_packet_priv;
1442         unsigned int num = f->num_members;
1443         struct packet_sock *po;
1444         unsigned int idx;
1445
1446         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1447             !num) {
1448                 kfree_skb(skb);
1449                 return 0;
1450         }
1451
1452         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1453                 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1454                 if (!skb)
1455                         return 0;
1456         }
1457         switch (f->type) {
1458         case PACKET_FANOUT_HASH:
1459         default:
1460                 idx = fanout_demux_hash(f, skb, num);
1461                 break;
1462         case PACKET_FANOUT_LB:
1463                 idx = fanout_demux_lb(f, skb, num);
1464                 break;
1465         case PACKET_FANOUT_CPU:
1466                 idx = fanout_demux_cpu(f, skb, num);
1467                 break;
1468         case PACKET_FANOUT_RND:
1469                 idx = fanout_demux_rnd(f, skb, num);
1470                 break;
1471         case PACKET_FANOUT_QM:
1472                 idx = fanout_demux_qm(f, skb, num);
1473                 break;
1474         case PACKET_FANOUT_ROLLOVER:
1475                 idx = fanout_demux_rollover(f, skb, 0, false, num);
1476                 break;
1477         }
1478
1479         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1480                 idx = fanout_demux_rollover(f, skb, idx, true, num);
1481
1482         po = pkt_sk(f->arr[idx]);
1483         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1484 }
1485
1486 DEFINE_MUTEX(fanout_mutex);
1487 EXPORT_SYMBOL_GPL(fanout_mutex);
1488 static LIST_HEAD(fanout_list);
1489
1490 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1491 {
1492         struct packet_fanout *f = po->fanout;
1493
1494         spin_lock(&f->lock);
1495         f->arr[f->num_members] = sk;
1496         smp_wmb();
1497         f->num_members++;
1498         spin_unlock(&f->lock);
1499 }
1500
1501 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1502 {
1503         struct packet_fanout *f = po->fanout;
1504         int i;
1505
1506         spin_lock(&f->lock);
1507         for (i = 0; i < f->num_members; i++) {
1508                 if (f->arr[i] == sk)
1509                         break;
1510         }
1511         BUG_ON(i >= f->num_members);
1512         f->arr[i] = f->arr[f->num_members - 1];
1513         f->num_members--;
1514         spin_unlock(&f->lock);
1515 }
1516
1517 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1518 {
1519         if (ptype->af_packet_priv == (void *)((struct packet_sock *)sk)->fanout)
1520                 return true;
1521
1522         return false;
1523 }
1524
1525 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1526 {
1527         struct packet_sock *po = pkt_sk(sk);
1528         struct packet_fanout *f, *match;
1529         u8 type = type_flags & 0xff;
1530         u8 flags = type_flags >> 8;
1531         int err;
1532
1533         switch (type) {
1534         case PACKET_FANOUT_ROLLOVER:
1535                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1536                         return -EINVAL;
1537         case PACKET_FANOUT_HASH:
1538         case PACKET_FANOUT_LB:
1539         case PACKET_FANOUT_CPU:
1540         case PACKET_FANOUT_RND:
1541         case PACKET_FANOUT_QM:
1542                 break;
1543         default:
1544                 return -EINVAL;
1545         }
1546
1547         if (!po->running)
1548                 return -EINVAL;
1549
1550         if (po->fanout)
1551                 return -EALREADY;
1552
1553         if (type == PACKET_FANOUT_ROLLOVER ||
1554             (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1555                 po->rollover = kzalloc(sizeof(*po->rollover), GFP_KERNEL);
1556                 if (!po->rollover)
1557                         return -ENOMEM;
1558                 atomic_long_set(&po->rollover->num, 0);
1559                 atomic_long_set(&po->rollover->num_huge, 0);
1560                 atomic_long_set(&po->rollover->num_failed, 0);
1561         }
1562
1563         mutex_lock(&fanout_mutex);
1564         match = NULL;
1565         list_for_each_entry(f, &fanout_list, list) {
1566                 if (f->id == id &&
1567                     read_pnet(&f->net) == sock_net(sk)) {
1568                         match = f;
1569                         break;
1570                 }
1571         }
1572         err = -EINVAL;
1573         if (match && match->flags != flags)
1574                 goto out;
1575         if (!match) {
1576                 err = -ENOMEM;
1577                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1578                 if (!match)
1579                         goto out;
1580                 write_pnet(&match->net, sock_net(sk));
1581                 match->id = id;
1582                 match->type = type;
1583                 match->flags = flags;
1584                 atomic_set(&match->rr_cur, 0);
1585                 INIT_LIST_HEAD(&match->list);
1586                 spin_lock_init(&match->lock);
1587                 atomic_set(&match->sk_ref, 0);
1588                 match->prot_hook.type = po->prot_hook.type;
1589                 match->prot_hook.dev = po->prot_hook.dev;
1590                 match->prot_hook.func = packet_rcv_fanout;
1591                 match->prot_hook.af_packet_priv = match;
1592                 match->prot_hook.id_match = match_fanout_group;
1593                 dev_add_pack(&match->prot_hook);
1594                 list_add(&match->list, &fanout_list);
1595         }
1596         err = -EINVAL;
1597         if (match->type == type &&
1598             match->prot_hook.type == po->prot_hook.type &&
1599             match->prot_hook.dev == po->prot_hook.dev) {
1600                 err = -ENOSPC;
1601                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1602                         __dev_remove_pack(&po->prot_hook);
1603                         po->fanout = match;
1604                         atomic_inc(&match->sk_ref);
1605                         __fanout_link(sk, po);
1606                         err = 0;
1607                 }
1608         }
1609 out:
1610         mutex_unlock(&fanout_mutex);
1611         if (err) {
1612                 kfree(po->rollover);
1613                 po->rollover = NULL;
1614         }
1615         return err;
1616 }
1617
1618 static void fanout_release(struct sock *sk)
1619 {
1620         struct packet_sock *po = pkt_sk(sk);
1621         struct packet_fanout *f;
1622
1623         f = po->fanout;
1624         if (!f)
1625                 return;
1626
1627         mutex_lock(&fanout_mutex);
1628         po->fanout = NULL;
1629
1630         if (atomic_dec_and_test(&f->sk_ref)) {
1631                 list_del(&f->list);
1632                 dev_remove_pack(&f->prot_hook);
1633                 kfree(f);
1634         }
1635         mutex_unlock(&fanout_mutex);
1636
1637         if (po->rollover)
1638                 kfree_rcu(po->rollover, rcu);
1639 }
1640
1641 static const struct proto_ops packet_ops;
1642
1643 static const struct proto_ops packet_ops_spkt;
1644
1645 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1646                            struct packet_type *pt, struct net_device *orig_dev)
1647 {
1648         struct sock *sk;
1649         struct sockaddr_pkt *spkt;
1650
1651         /*
1652          *      When we registered the protocol we saved the socket in the data
1653          *      field for just this event.
1654          */
1655
1656         sk = pt->af_packet_priv;
1657
1658         /*
1659          *      Yank back the headers [hope the device set this
1660          *      right or kerboom...]
1661          *
1662          *      Incoming packets have ll header pulled,
1663          *      push it back.
1664          *
1665          *      For outgoing ones skb->data == skb_mac_header(skb)
1666          *      so that this procedure is noop.
1667          */
1668
1669         if (skb->pkt_type == PACKET_LOOPBACK)
1670                 goto out;
1671
1672         if (!net_eq(dev_net(dev), sock_net(sk)))
1673                 goto out;
1674
1675         skb = skb_share_check(skb, GFP_ATOMIC);
1676         if (skb == NULL)
1677                 goto oom;
1678
1679         /* drop any routing info */
1680         skb_dst_drop(skb);
1681
1682         /* drop conntrack reference */
1683         nf_reset(skb);
1684
1685         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1686
1687         skb_push(skb, skb->data - skb_mac_header(skb));
1688
1689         /*
1690          *      The SOCK_PACKET socket receives _all_ frames.
1691          */
1692
1693         spkt->spkt_family = dev->type;
1694         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1695         spkt->spkt_protocol = skb->protocol;
1696
1697         /*
1698          *      Charge the memory to the socket. This is done specifically
1699          *      to prevent sockets using all the memory up.
1700          */
1701
1702         if (sock_queue_rcv_skb(sk, skb) == 0)
1703                 return 0;
1704
1705 out:
1706         kfree_skb(skb);
1707 oom:
1708         return 0;
1709 }
1710
1711
1712 /*
1713  *      Output a raw packet to a device layer. This bypasses all the other
1714  *      protocol layers and you must therefore supply it with a complete frame
1715  */
1716
1717 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1718                                size_t len)
1719 {
1720         struct sock *sk = sock->sk;
1721         DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1722         struct sk_buff *skb = NULL;
1723         struct net_device *dev;
1724         __be16 proto = 0;
1725         int err;
1726         int extra_len = 0;
1727
1728         /*
1729          *      Get and verify the address.
1730          */
1731
1732         if (saddr) {
1733                 if (msg->msg_namelen < sizeof(struct sockaddr))
1734                         return -EINVAL;
1735                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1736                         proto = saddr->spkt_protocol;
1737         } else
1738                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1739
1740         /*
1741          *      Find the device first to size check it
1742          */
1743
1744         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1745 retry:
1746         rcu_read_lock();
1747         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1748         err = -ENODEV;
1749         if (dev == NULL)
1750                 goto out_unlock;
1751
1752         err = -ENETDOWN;
1753         if (!(dev->flags & IFF_UP))
1754                 goto out_unlock;
1755
1756         /*
1757          * You may not queue a frame bigger than the mtu. This is the lowest level
1758          * raw protocol and you must do your own fragmentation at this level.
1759          */
1760
1761         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1762                 if (!netif_supports_nofcs(dev)) {
1763                         err = -EPROTONOSUPPORT;
1764                         goto out_unlock;
1765                 }
1766                 extra_len = 4; /* We're doing our own CRC */
1767         }
1768
1769         err = -EMSGSIZE;
1770         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1771                 goto out_unlock;
1772
1773         if (!skb) {
1774                 size_t reserved = LL_RESERVED_SPACE(dev);
1775                 int tlen = dev->needed_tailroom;
1776                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1777
1778                 rcu_read_unlock();
1779                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1780                 if (skb == NULL)
1781                         return -ENOBUFS;
1782                 /* FIXME: Save some space for broken drivers that write a hard
1783                  * header at transmission time by themselves. PPP is the notable
1784                  * one here. This should really be fixed at the driver level.
1785                  */
1786                 skb_reserve(skb, reserved);
1787                 skb_reset_network_header(skb);
1788
1789                 /* Try to align data part correctly */
1790                 if (hhlen) {
1791                         skb->data -= hhlen;
1792                         skb->tail -= hhlen;
1793                         if (len < hhlen)
1794                                 skb_reset_network_header(skb);
1795                 }
1796                 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1797                 if (err)
1798                         goto out_free;
1799                 goto retry;
1800         }
1801
1802         if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1803                 /* Earlier code assumed this would be a VLAN pkt,
1804                  * double-check this now that we have the actual
1805                  * packet in hand.
1806                  */
1807                 struct ethhdr *ehdr;
1808                 skb_reset_mac_header(skb);
1809                 ehdr = eth_hdr(skb);
1810                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1811                         err = -EMSGSIZE;
1812                         goto out_unlock;
1813                 }
1814         }
1815
1816         skb->protocol = proto;
1817         skb->dev = dev;
1818         skb->priority = sk->sk_priority;
1819         skb->mark = sk->sk_mark;
1820
1821         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1822
1823         if (unlikely(extra_len == 4))
1824                 skb->no_fcs = 1;
1825
1826         skb_probe_transport_header(skb, 0);
1827
1828         dev_queue_xmit(skb);
1829         rcu_read_unlock();
1830         return len;
1831
1832 out_unlock:
1833         rcu_read_unlock();
1834 out_free:
1835         kfree_skb(skb);
1836         return err;
1837 }
1838
1839 static unsigned int run_filter(const struct sk_buff *skb,
1840                                       const struct sock *sk,
1841                                       unsigned int res)
1842 {
1843         struct sk_filter *filter;
1844
1845         rcu_read_lock();
1846         filter = rcu_dereference(sk->sk_filter);
1847         if (filter != NULL)
1848                 res = SK_RUN_FILTER(filter, skb);
1849         rcu_read_unlock();
1850
1851         return res;
1852 }
1853
1854 /*
1855  * This function makes lazy skb cloning in hope that most of packets
1856  * are discarded by BPF.
1857  *
1858  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1859  * and skb->cb are mangled. It works because (and until) packets
1860  * falling here are owned by current CPU. Output packets are cloned
1861  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1862  * sequencially, so that if we return skb to original state on exit,
1863  * we will not harm anyone.
1864  */
1865
1866 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1867                       struct packet_type *pt, struct net_device *orig_dev)
1868 {
1869         struct sock *sk;
1870         struct sockaddr_ll *sll;
1871         struct packet_sock *po;
1872         u8 *skb_head = skb->data;
1873         int skb_len = skb->len;
1874         unsigned int snaplen, res;
1875
1876         if (skb->pkt_type == PACKET_LOOPBACK)
1877                 goto drop;
1878
1879         sk = pt->af_packet_priv;
1880         po = pkt_sk(sk);
1881
1882         if (!net_eq(dev_net(dev), sock_net(sk)))
1883                 goto drop;
1884
1885         skb->dev = dev;
1886
1887         if (dev->header_ops) {
1888                 /* The device has an explicit notion of ll header,
1889                  * exported to higher levels.
1890                  *
1891                  * Otherwise, the device hides details of its frame
1892                  * structure, so that corresponding packet head is
1893                  * never delivered to user.
1894                  */
1895                 if (sk->sk_type != SOCK_DGRAM)
1896                         skb_push(skb, skb->data - skb_mac_header(skb));
1897                 else if (skb->pkt_type == PACKET_OUTGOING) {
1898                         /* Special case: outgoing packets have ll header at head */
1899                         skb_pull(skb, skb_network_offset(skb));
1900                 }
1901         }
1902
1903         snaplen = skb->len;
1904
1905         res = run_filter(skb, sk, snaplen);
1906         if (!res)
1907                 goto drop_n_restore;
1908         if (snaplen > res)
1909                 snaplen = res;
1910
1911         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1912                 goto drop_n_acct;
1913
1914         if (skb_shared(skb)) {
1915                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1916                 if (nskb == NULL)
1917                         goto drop_n_acct;
1918
1919                 if (skb_head != skb->data) {
1920                         skb->data = skb_head;
1921                         skb->len = skb_len;
1922                 }
1923                 consume_skb(skb);
1924                 skb = nskb;
1925         }
1926
1927         sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
1928
1929         sll = &PACKET_SKB_CB(skb)->sa.ll;
1930         sll->sll_hatype = dev->type;
1931         sll->sll_pkttype = skb->pkt_type;
1932         if (unlikely(po->origdev))
1933                 sll->sll_ifindex = orig_dev->ifindex;
1934         else
1935                 sll->sll_ifindex = dev->ifindex;
1936
1937         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1938
1939         /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
1940          * Use their space for storing the original skb length.
1941          */
1942         PACKET_SKB_CB(skb)->sa.origlen = skb->len;
1943
1944         if (pskb_trim(skb, snaplen))
1945                 goto drop_n_acct;
1946
1947         skb_set_owner_r(skb, sk);
1948         skb->dev = NULL;
1949         skb_dst_drop(skb);
1950
1951         /* drop conntrack reference */
1952         nf_reset(skb);
1953
1954         spin_lock(&sk->sk_receive_queue.lock);
1955         po->stats.stats1.tp_packets++;
1956         sock_skb_set_dropcount(sk, skb);
1957         __skb_queue_tail(&sk->sk_receive_queue, skb);
1958         spin_unlock(&sk->sk_receive_queue.lock);
1959         sk->sk_data_ready(sk);
1960         return 0;
1961
1962 drop_n_acct:
1963         spin_lock(&sk->sk_receive_queue.lock);
1964         po->stats.stats1.tp_drops++;
1965         atomic_inc(&sk->sk_drops);
1966         spin_unlock(&sk->sk_receive_queue.lock);
1967
1968 drop_n_restore:
1969         if (skb_head != skb->data && skb_shared(skb)) {
1970                 skb->data = skb_head;
1971                 skb->len = skb_len;
1972         }
1973 drop:
1974         consume_skb(skb);
1975         return 0;
1976 }
1977
1978 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1979                        struct packet_type *pt, struct net_device *orig_dev)
1980 {
1981         struct sock *sk;
1982         struct packet_sock *po;
1983         struct sockaddr_ll *sll;
1984         union tpacket_uhdr h;
1985         u8 *skb_head = skb->data;
1986         int skb_len = skb->len;
1987         unsigned int snaplen, res;
1988         unsigned long status = TP_STATUS_USER;
1989         unsigned short macoff, netoff, hdrlen;
1990         struct sk_buff *copy_skb = NULL;
1991         struct timespec ts;
1992         __u32 ts_status;
1993
1994         /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
1995          * We may add members to them until current aligned size without forcing
1996          * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
1997          */
1998         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
1999         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2000
2001         if (skb->pkt_type == PACKET_LOOPBACK)
2002                 goto drop;
2003
2004         sk = pt->af_packet_priv;
2005         po = pkt_sk(sk);
2006
2007         if (!net_eq(dev_net(dev), sock_net(sk)))
2008                 goto drop;
2009
2010         if (dev->header_ops) {
2011                 if (sk->sk_type != SOCK_DGRAM)
2012                         skb_push(skb, skb->data - skb_mac_header(skb));
2013                 else if (skb->pkt_type == PACKET_OUTGOING) {
2014                         /* Special case: outgoing packets have ll header at head */
2015                         skb_pull(skb, skb_network_offset(skb));
2016                 }
2017         }
2018
2019         snaplen = skb->len;
2020
2021         res = run_filter(skb, sk, snaplen);
2022         if (!res)
2023                 goto drop_n_restore;
2024
2025         if (skb->ip_summed == CHECKSUM_PARTIAL)
2026                 status |= TP_STATUS_CSUMNOTREADY;
2027         else if (skb->pkt_type != PACKET_OUTGOING &&
2028                  (skb->ip_summed == CHECKSUM_COMPLETE ||
2029                   skb_csum_unnecessary(skb)))
2030                 status |= TP_STATUS_CSUM_VALID;
2031
2032         if (snaplen > res)
2033                 snaplen = res;
2034
2035         if (sk->sk_type == SOCK_DGRAM) {
2036                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2037                                   po->tp_reserve;
2038         } else {
2039                 unsigned int maclen = skb_network_offset(skb);
2040                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2041                                        (maclen < 16 ? 16 : maclen)) +
2042                         po->tp_reserve;
2043                 macoff = netoff - maclen;
2044         }
2045         if (po->tp_version <= TPACKET_V2) {
2046                 if (macoff + snaplen > po->rx_ring.frame_size) {
2047                         if (po->copy_thresh &&
2048                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2049                                 if (skb_shared(skb)) {
2050                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
2051                                 } else {
2052                                         copy_skb = skb_get(skb);
2053                                         skb_head = skb->data;
2054                                 }
2055                                 if (copy_skb)
2056                                         skb_set_owner_r(copy_skb, sk);
2057                         }
2058                         snaplen = po->rx_ring.frame_size - macoff;
2059                         if ((int)snaplen < 0)
2060                                 snaplen = 0;
2061                 }
2062         } else if (unlikely(macoff + snaplen >
2063                             GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2064                 u32 nval;
2065
2066                 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2067                 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2068                             snaplen, nval, macoff);
2069                 snaplen = nval;
2070                 if (unlikely((int)snaplen < 0)) {
2071                         snaplen = 0;
2072                         macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2073                 }
2074         }
2075         spin_lock(&sk->sk_receive_queue.lock);
2076         h.raw = packet_current_rx_frame(po, skb,
2077                                         TP_STATUS_KERNEL, (macoff+snaplen));
2078         if (!h.raw)
2079                 goto ring_is_full;
2080         if (po->tp_version <= TPACKET_V2) {
2081                 packet_increment_rx_head(po, &po->rx_ring);
2082         /*
2083          * LOSING will be reported till you read the stats,
2084          * because it's COR - Clear On Read.
2085          * Anyways, moving it for V1/V2 only as V3 doesn't need this
2086          * at packet level.
2087          */
2088                 if (po->stats.stats1.tp_drops)
2089                         status |= TP_STATUS_LOSING;
2090         }
2091         po->stats.stats1.tp_packets++;
2092         if (copy_skb) {
2093                 status |= TP_STATUS_COPY;
2094                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2095         }
2096         spin_unlock(&sk->sk_receive_queue.lock);
2097
2098         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2099
2100         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2101                 getnstimeofday(&ts);
2102
2103         status |= ts_status;
2104
2105         switch (po->tp_version) {
2106         case TPACKET_V1:
2107                 h.h1->tp_len = skb->len;
2108                 h.h1->tp_snaplen = snaplen;
2109                 h.h1->tp_mac = macoff;
2110                 h.h1->tp_net = netoff;
2111                 h.h1->tp_sec = ts.tv_sec;
2112                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2113                 hdrlen = sizeof(*h.h1);
2114                 break;
2115         case TPACKET_V2:
2116                 h.h2->tp_len = skb->len;
2117                 h.h2->tp_snaplen = snaplen;
2118                 h.h2->tp_mac = macoff;
2119                 h.h2->tp_net = netoff;
2120                 h.h2->tp_sec = ts.tv_sec;
2121                 h.h2->tp_nsec = ts.tv_nsec;
2122                 if (skb_vlan_tag_present(skb)) {
2123                         h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2124                         h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2125                         status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2126                 } else {
2127                         h.h2->tp_vlan_tci = 0;
2128                         h.h2->tp_vlan_tpid = 0;
2129                 }
2130                 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2131                 hdrlen = sizeof(*h.h2);
2132                 break;
2133         case TPACKET_V3:
2134                 /* tp_nxt_offset,vlan are already populated above.
2135                  * So DONT clear those fields here
2136                  */
2137                 h.h3->tp_status |= status;
2138                 h.h3->tp_len = skb->len;
2139                 h.h3->tp_snaplen = snaplen;
2140                 h.h3->tp_mac = macoff;
2141                 h.h3->tp_net = netoff;
2142                 h.h3->tp_sec  = ts.tv_sec;
2143                 h.h3->tp_nsec = ts.tv_nsec;
2144                 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2145                 hdrlen = sizeof(*h.h3);
2146                 break;
2147         default:
2148                 BUG();
2149         }
2150
2151         sll = h.raw + TPACKET_ALIGN(hdrlen);
2152         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2153         sll->sll_family = AF_PACKET;
2154         sll->sll_hatype = dev->type;
2155         sll->sll_protocol = skb->protocol;
2156         sll->sll_pkttype = skb->pkt_type;
2157         if (unlikely(po->origdev))
2158                 sll->sll_ifindex = orig_dev->ifindex;
2159         else
2160                 sll->sll_ifindex = dev->ifindex;
2161
2162         smp_mb();
2163
2164 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2165         if (po->tp_version <= TPACKET_V2) {
2166                 u8 *start, *end;
2167
2168                 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2169                                         macoff + snaplen);
2170
2171                 for (start = h.raw; start < end; start += PAGE_SIZE)
2172                         flush_dcache_page(pgv_to_page(start));
2173         }
2174         smp_wmb();
2175 #endif
2176
2177         if (po->tp_version <= TPACKET_V2) {
2178                 __packet_set_status(po, h.raw, status);
2179                 sk->sk_data_ready(sk);
2180         } else {
2181                 prb_clear_blk_fill_status(&po->rx_ring);
2182         }
2183
2184 drop_n_restore:
2185         if (skb_head != skb->data && skb_shared(skb)) {
2186                 skb->data = skb_head;
2187                 skb->len = skb_len;
2188         }
2189 drop:
2190         kfree_skb(skb);
2191         return 0;
2192
2193 ring_is_full:
2194         po->stats.stats1.tp_drops++;
2195         spin_unlock(&sk->sk_receive_queue.lock);
2196
2197         sk->sk_data_ready(sk);
2198         kfree_skb(copy_skb);
2199         goto drop_n_restore;
2200 }
2201
2202 static void tpacket_destruct_skb(struct sk_buff *skb)
2203 {
2204         struct packet_sock *po = pkt_sk(skb->sk);
2205
2206         if (likely(po->tx_ring.pg_vec)) {
2207                 void *ph;
2208                 __u32 ts;
2209
2210                 ph = skb_shinfo(skb)->destructor_arg;
2211                 packet_dec_pending(&po->tx_ring);
2212
2213                 ts = __packet_set_timestamp(po, ph, skb);
2214                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2215         }
2216
2217         sock_wfree(skb);
2218 }
2219
2220 static bool ll_header_truncated(const struct net_device *dev, int len)
2221 {
2222         /* net device doesn't like empty head */
2223         if (unlikely(len <= dev->hard_header_len)) {
2224                 net_warn_ratelimited("%s: packet size is too short (%d <= %d)\n",
2225                                      current->comm, len, dev->hard_header_len);
2226                 return true;
2227         }
2228
2229         return false;
2230 }
2231
2232 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2233                 void *frame, struct net_device *dev, int size_max,
2234                 __be16 proto, unsigned char *addr, int hlen)
2235 {
2236         union tpacket_uhdr ph;
2237         int to_write, offset, len, tp_len, nr_frags, len_max;
2238         struct socket *sock = po->sk.sk_socket;
2239         struct page *page;
2240         void *data;
2241         int err;
2242
2243         ph.raw = frame;
2244
2245         skb->protocol = proto;
2246         skb->dev = dev;
2247         skb->priority = po->sk.sk_priority;
2248         skb->mark = po->sk.sk_mark;
2249         sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2250         skb_shinfo(skb)->destructor_arg = ph.raw;
2251
2252         switch (po->tp_version) {
2253         case TPACKET_V2:
2254                 tp_len = ph.h2->tp_len;
2255                 break;
2256         default:
2257                 tp_len = ph.h1->tp_len;
2258                 break;
2259         }
2260         if (unlikely(tp_len > size_max)) {
2261                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2262                 return -EMSGSIZE;
2263         }
2264
2265         skb_reserve(skb, hlen);
2266         skb_reset_network_header(skb);
2267
2268         if (!packet_use_direct_xmit(po))
2269                 skb_probe_transport_header(skb, 0);
2270         if (unlikely(po->tp_tx_has_off)) {
2271                 int off_min, off_max, off;
2272                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2273                 off_max = po->tx_ring.frame_size - tp_len;
2274                 if (sock->type == SOCK_DGRAM) {
2275                         switch (po->tp_version) {
2276                         case TPACKET_V2:
2277                                 off = ph.h2->tp_net;
2278                                 break;
2279                         default:
2280                                 off = ph.h1->tp_net;
2281                                 break;
2282                         }
2283                 } else {
2284                         switch (po->tp_version) {
2285                         case TPACKET_V2:
2286                                 off = ph.h2->tp_mac;
2287                                 break;
2288                         default:
2289                                 off = ph.h1->tp_mac;
2290                                 break;
2291                         }
2292                 }
2293                 if (unlikely((off < off_min) || (off_max < off)))
2294                         return -EINVAL;
2295                 data = ph.raw + off;
2296         } else {
2297                 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2298         }
2299         to_write = tp_len;
2300
2301         if (sock->type == SOCK_DGRAM) {
2302                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2303                                 NULL, tp_len);
2304                 if (unlikely(err < 0))
2305                         return -EINVAL;
2306         } else if (dev->hard_header_len) {
2307                 if (ll_header_truncated(dev, tp_len))
2308                         return -EINVAL;
2309
2310                 skb_push(skb, dev->hard_header_len);
2311                 err = skb_store_bits(skb, 0, data,
2312                                 dev->hard_header_len);
2313                 if (unlikely(err))
2314                         return err;
2315
2316                 data += dev->hard_header_len;
2317                 to_write -= dev->hard_header_len;
2318         }
2319
2320         offset = offset_in_page(data);
2321         len_max = PAGE_SIZE - offset;
2322         len = ((to_write > len_max) ? len_max : to_write);
2323
2324         skb->data_len = to_write;
2325         skb->len += to_write;
2326         skb->truesize += to_write;
2327         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2328
2329         while (likely(to_write)) {
2330                 nr_frags = skb_shinfo(skb)->nr_frags;
2331
2332                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2333                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2334                                MAX_SKB_FRAGS);
2335                         return -EFAULT;
2336                 }
2337
2338                 page = pgv_to_page(data);
2339                 data += len;
2340                 flush_dcache_page(page);
2341                 get_page(page);
2342                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2343                 to_write -= len;
2344                 offset = 0;
2345                 len_max = PAGE_SIZE;
2346                 len = ((to_write > len_max) ? len_max : to_write);
2347         }
2348
2349         return tp_len;
2350 }
2351
2352 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2353 {
2354         struct sk_buff *skb;
2355         struct net_device *dev;
2356         __be16 proto;
2357         int err, reserve = 0;
2358         void *ph;
2359         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2360         bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2361         int tp_len, size_max;
2362         unsigned char *addr;
2363         int len_sum = 0;
2364         int status = TP_STATUS_AVAILABLE;
2365         int hlen, tlen;
2366
2367         mutex_lock(&po->pg_vec_lock);
2368
2369         if (likely(saddr == NULL)) {
2370                 dev     = packet_cached_dev_get(po);
2371                 proto   = po->num;
2372                 addr    = NULL;
2373         } else {
2374                 err = -EINVAL;
2375                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2376                         goto out;
2377                 if (msg->msg_namelen < (saddr->sll_halen
2378                                         + offsetof(struct sockaddr_ll,
2379                                                 sll_addr)))
2380                         goto out;
2381                 proto   = saddr->sll_protocol;
2382                 addr    = saddr->sll_addr;
2383                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2384         }
2385
2386         err = -ENXIO;
2387         if (unlikely(dev == NULL))
2388                 goto out;
2389         err = -ENETDOWN;
2390         if (unlikely(!(dev->flags & IFF_UP)))
2391                 goto out_put;
2392
2393         reserve = dev->hard_header_len + VLAN_HLEN;
2394         size_max = po->tx_ring.frame_size
2395                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2396
2397         if (size_max > dev->mtu + reserve)
2398                 size_max = dev->mtu + reserve;
2399
2400         do {
2401                 ph = packet_current_frame(po, &po->tx_ring,
2402                                           TP_STATUS_SEND_REQUEST);
2403                 if (unlikely(ph == NULL)) {
2404                         if (need_wait && need_resched())
2405                                 schedule();
2406                         continue;
2407                 }
2408
2409                 status = TP_STATUS_SEND_REQUEST;
2410                 hlen = LL_RESERVED_SPACE(dev);
2411                 tlen = dev->needed_tailroom;
2412                 skb = sock_alloc_send_skb(&po->sk,
2413                                 hlen + tlen + sizeof(struct sockaddr_ll),
2414                                 !need_wait, &err);
2415
2416                 if (unlikely(skb == NULL)) {
2417                         /* we assume the socket was initially writeable ... */
2418                         if (likely(len_sum > 0))
2419                                 err = len_sum;
2420                         goto out_status;
2421                 }
2422                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2423                                           addr, hlen);
2424                 if (tp_len > dev->mtu + dev->hard_header_len) {
2425                         struct ethhdr *ehdr;
2426                         /* Earlier code assumed this would be a VLAN pkt,
2427                          * double-check this now that we have the actual
2428                          * packet in hand.
2429                          */
2430
2431                         skb_reset_mac_header(skb);
2432                         ehdr = eth_hdr(skb);
2433                         if (ehdr->h_proto != htons(ETH_P_8021Q))
2434                                 tp_len = -EMSGSIZE;
2435                 }
2436                 if (unlikely(tp_len < 0)) {
2437                         if (po->tp_loss) {
2438                                 __packet_set_status(po, ph,
2439                                                 TP_STATUS_AVAILABLE);
2440                                 packet_increment_head(&po->tx_ring);
2441                                 kfree_skb(skb);
2442                                 continue;
2443                         } else {
2444                                 status = TP_STATUS_WRONG_FORMAT;
2445                                 err = tp_len;
2446                                 goto out_status;
2447                         }
2448                 }
2449
2450                 packet_pick_tx_queue(dev, skb);
2451
2452                 skb->destructor = tpacket_destruct_skb;
2453                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2454                 packet_inc_pending(&po->tx_ring);
2455
2456                 status = TP_STATUS_SEND_REQUEST;
2457                 err = po->xmit(skb);
2458                 if (unlikely(err > 0)) {
2459                         err = net_xmit_errno(err);
2460                         if (err && __packet_get_status(po, ph) ==
2461                                    TP_STATUS_AVAILABLE) {
2462                                 /* skb was destructed already */
2463                                 skb = NULL;
2464                                 goto out_status;
2465                         }
2466                         /*
2467                          * skb was dropped but not destructed yet;
2468                          * let's treat it like congestion or err < 0
2469                          */
2470                         err = 0;
2471                 }
2472                 packet_increment_head(&po->tx_ring);
2473                 len_sum += tp_len;
2474         } while (likely((ph != NULL) ||
2475                 /* Note: packet_read_pending() might be slow if we have
2476                  * to call it as it's per_cpu variable, but in fast-path
2477                  * we already short-circuit the loop with the first
2478                  * condition, and luckily don't have to go that path
2479                  * anyway.
2480                  */
2481                  (need_wait && packet_read_pending(&po->tx_ring))));
2482
2483         err = len_sum;
2484         goto out_put;
2485
2486 out_status:
2487         __packet_set_status(po, ph, status);
2488         kfree_skb(skb);
2489 out_put:
2490         dev_put(dev);
2491 out:
2492         mutex_unlock(&po->pg_vec_lock);
2493         return err;
2494 }
2495
2496 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2497                                         size_t reserve, size_t len,
2498                                         size_t linear, int noblock,
2499                                         int *err)
2500 {
2501         struct sk_buff *skb;
2502
2503         /* Under a page?  Don't bother with paged skb. */
2504         if (prepad + len < PAGE_SIZE || !linear)
2505                 linear = len;
2506
2507         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2508                                    err, 0);
2509         if (!skb)
2510                 return NULL;
2511
2512         skb_reserve(skb, reserve);
2513         skb_put(skb, linear);
2514         skb->data_len = len - linear;
2515         skb->len += len - linear;
2516
2517         return skb;
2518 }
2519
2520 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2521 {
2522         struct sock *sk = sock->sk;
2523         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2524         struct sk_buff *skb;
2525         struct net_device *dev;
2526         __be16 proto;
2527         unsigned char *addr;
2528         int err, reserve = 0;
2529         struct virtio_net_hdr vnet_hdr = { 0 };
2530         int offset = 0;
2531         int vnet_hdr_len;
2532         struct packet_sock *po = pkt_sk(sk);
2533         unsigned short gso_type = 0;
2534         int hlen, tlen;
2535         int extra_len = 0;
2536         ssize_t n;
2537
2538         /*
2539          *      Get and verify the address.
2540          */
2541
2542         if (likely(saddr == NULL)) {
2543                 dev     = packet_cached_dev_get(po);
2544                 proto   = po->num;
2545                 addr    = NULL;
2546         } else {
2547                 err = -EINVAL;
2548                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2549                         goto out;
2550                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2551                         goto out;
2552                 proto   = saddr->sll_protocol;
2553                 addr    = saddr->sll_addr;
2554                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2555         }
2556
2557         err = -ENXIO;
2558         if (unlikely(dev == NULL))
2559                 goto out_unlock;
2560         err = -ENETDOWN;
2561         if (unlikely(!(dev->flags & IFF_UP)))
2562                 goto out_unlock;
2563
2564         if (sock->type == SOCK_RAW)
2565                 reserve = dev->hard_header_len;
2566         if (po->has_vnet_hdr) {
2567                 vnet_hdr_len = sizeof(vnet_hdr);
2568
2569                 err = -EINVAL;
2570                 if (len < vnet_hdr_len)
2571                         goto out_unlock;
2572
2573                 len -= vnet_hdr_len;
2574
2575                 err = -EFAULT;
2576                 n = copy_from_iter(&vnet_hdr, vnet_hdr_len, &msg->msg_iter);
2577                 if (n != vnet_hdr_len)
2578                         goto out_unlock;
2579
2580                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2581                     (__virtio16_to_cpu(false, vnet_hdr.csum_start) +
2582                      __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2 >
2583                       __virtio16_to_cpu(false, vnet_hdr.hdr_len)))
2584                         vnet_hdr.hdr_len = __cpu_to_virtio16(false,
2585                                  __virtio16_to_cpu(false, vnet_hdr.csum_start) +
2586                                 __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2);
2587
2588                 err = -EINVAL;
2589                 if (__virtio16_to_cpu(false, vnet_hdr.hdr_len) > len)
2590                         goto out_unlock;
2591
2592                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2593                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2594                         case VIRTIO_NET_HDR_GSO_TCPV4:
2595                                 gso_type = SKB_GSO_TCPV4;
2596                                 break;
2597                         case VIRTIO_NET_HDR_GSO_TCPV6:
2598                                 gso_type = SKB_GSO_TCPV6;
2599                                 break;
2600                         case VIRTIO_NET_HDR_GSO_UDP:
2601                                 gso_type = SKB_GSO_UDP;
2602                                 break;
2603                         default:
2604                                 goto out_unlock;
2605                         }
2606
2607                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2608                                 gso_type |= SKB_GSO_TCP_ECN;
2609
2610                         if (vnet_hdr.gso_size == 0)
2611                                 goto out_unlock;
2612
2613                 }
2614         }
2615
2616         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2617                 if (!netif_supports_nofcs(dev)) {
2618                         err = -EPROTONOSUPPORT;
2619                         goto out_unlock;
2620                 }
2621                 extra_len = 4; /* We're doing our own CRC */
2622         }
2623
2624         err = -EMSGSIZE;
2625         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2626                 goto out_unlock;
2627
2628         err = -ENOBUFS;
2629         hlen = LL_RESERVED_SPACE(dev);
2630         tlen = dev->needed_tailroom;
2631         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
2632                                __virtio16_to_cpu(false, vnet_hdr.hdr_len),
2633                                msg->msg_flags & MSG_DONTWAIT, &err);
2634         if (skb == NULL)
2635                 goto out_unlock;
2636
2637         skb_set_network_header(skb, reserve);
2638
2639         err = -EINVAL;
2640         if (sock->type == SOCK_DGRAM) {
2641                 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2642                 if (unlikely(offset < 0))
2643                         goto out_free;
2644         } else {
2645                 if (ll_header_truncated(dev, len))
2646                         goto out_free;
2647         }
2648
2649         /* Returns -EFAULT on error */
2650         err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2651         if (err)
2652                 goto out_free;
2653
2654         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2655
2656         if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2657                 /* Earlier code assumed this would be a VLAN pkt,
2658                  * double-check this now that we have the actual
2659                  * packet in hand.
2660                  */
2661                 struct ethhdr *ehdr;
2662                 skb_reset_mac_header(skb);
2663                 ehdr = eth_hdr(skb);
2664                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2665                         err = -EMSGSIZE;
2666                         goto out_free;
2667                 }
2668         }
2669
2670         skb->protocol = proto;
2671         skb->dev = dev;
2672         skb->priority = sk->sk_priority;
2673         skb->mark = sk->sk_mark;
2674
2675         packet_pick_tx_queue(dev, skb);
2676
2677         if (po->has_vnet_hdr) {
2678                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2679                         u16 s = __virtio16_to_cpu(false, vnet_hdr.csum_start);
2680                         u16 o = __virtio16_to_cpu(false, vnet_hdr.csum_offset);
2681                         if (!skb_partial_csum_set(skb, s, o)) {
2682                                 err = -EINVAL;
2683                                 goto out_free;
2684                         }
2685                 }
2686
2687                 skb_shinfo(skb)->gso_size =
2688                         __virtio16_to_cpu(false, vnet_hdr.gso_size);
2689                 skb_shinfo(skb)->gso_type = gso_type;
2690
2691                 /* Header must be checked, and gso_segs computed. */
2692                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2693                 skb_shinfo(skb)->gso_segs = 0;
2694
2695                 len += vnet_hdr_len;
2696         }
2697
2698         if (!packet_use_direct_xmit(po))
2699                 skb_probe_transport_header(skb, reserve);
2700         if (unlikely(extra_len == 4))
2701                 skb->no_fcs = 1;
2702
2703         err = po->xmit(skb);
2704         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2705                 goto out_unlock;
2706
2707         dev_put(dev);
2708
2709         return len;
2710
2711 out_free:
2712         kfree_skb(skb);
2713 out_unlock:
2714         if (dev)
2715                 dev_put(dev);
2716 out:
2717         return err;
2718 }
2719
2720 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2721 {
2722         struct sock *sk = sock->sk;
2723         struct packet_sock *po = pkt_sk(sk);
2724
2725         if (po->tx_ring.pg_vec)
2726                 return tpacket_snd(po, msg);
2727         else
2728                 return packet_snd(sock, msg, len);
2729 }
2730
2731 /*
2732  *      Close a PACKET socket. This is fairly simple. We immediately go
2733  *      to 'closed' state and remove our protocol entry in the device list.
2734  */
2735
2736 static int packet_release(struct socket *sock)
2737 {
2738         struct sock *sk = sock->sk;
2739         struct packet_sock *po;
2740         struct net *net;
2741         union tpacket_req_u req_u;
2742
2743         if (!sk)
2744                 return 0;
2745
2746         net = sock_net(sk);
2747         po = pkt_sk(sk);
2748
2749         mutex_lock(&net->packet.sklist_lock);
2750         sk_del_node_init_rcu(sk);
2751         mutex_unlock(&net->packet.sklist_lock);
2752
2753         preempt_disable();
2754         sock_prot_inuse_add(net, sk->sk_prot, -1);
2755         preempt_enable();
2756
2757         spin_lock(&po->bind_lock);
2758         unregister_prot_hook(sk, false);
2759         packet_cached_dev_reset(po);
2760
2761         if (po->prot_hook.dev) {
2762                 dev_put(po->prot_hook.dev);
2763                 po->prot_hook.dev = NULL;
2764         }
2765         spin_unlock(&po->bind_lock);
2766
2767         packet_flush_mclist(sk);
2768
2769         if (po->rx_ring.pg_vec) {
2770                 memset(&req_u, 0, sizeof(req_u));
2771                 packet_set_ring(sk, &req_u, 1, 0);
2772         }
2773
2774         if (po->tx_ring.pg_vec) {
2775                 memset(&req_u, 0, sizeof(req_u));
2776                 packet_set_ring(sk, &req_u, 1, 1);
2777         }
2778
2779         fanout_release(sk);
2780
2781         synchronize_net();
2782         /*
2783          *      Now the socket is dead. No more input will appear.
2784          */
2785         sock_orphan(sk);
2786         sock->sk = NULL;
2787
2788         /* Purge queues */
2789
2790         skb_queue_purge(&sk->sk_receive_queue);
2791         packet_free_pending(po);
2792         sk_refcnt_debug_release(sk);
2793
2794         sock_put(sk);
2795         return 0;
2796 }
2797
2798 /*
2799  *      Attach a packet hook.
2800  */
2801
2802 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 proto)
2803 {
2804         struct packet_sock *po = pkt_sk(sk);
2805         const struct net_device *dev_curr;
2806         __be16 proto_curr;
2807         bool need_rehook;
2808
2809         if (po->fanout) {
2810                 if (dev)
2811                         dev_put(dev);
2812
2813                 return -EINVAL;
2814         }
2815
2816         lock_sock(sk);
2817         spin_lock(&po->bind_lock);
2818
2819         proto_curr = po->prot_hook.type;
2820         dev_curr = po->prot_hook.dev;
2821
2822         need_rehook = proto_curr != proto || dev_curr != dev;
2823
2824         if (need_rehook) {
2825                 unregister_prot_hook(sk, true);
2826
2827                 po->num = proto;
2828                 po->prot_hook.type = proto;
2829
2830                 if (po->prot_hook.dev)
2831                         dev_put(po->prot_hook.dev);
2832
2833                 po->prot_hook.dev = dev;
2834
2835                 po->ifindex = dev ? dev->ifindex : 0;
2836                 packet_cached_dev_assign(po, dev);
2837         }
2838
2839         if (proto == 0 || !need_rehook)
2840                 goto out_unlock;
2841
2842         if (!dev || (dev->flags & IFF_UP)) {
2843                 register_prot_hook(sk);
2844         } else {
2845                 sk->sk_err = ENETDOWN;
2846                 if (!sock_flag(sk, SOCK_DEAD))
2847                         sk->sk_error_report(sk);
2848         }
2849
2850 out_unlock:
2851         spin_unlock(&po->bind_lock);
2852         release_sock(sk);
2853         return 0;
2854 }
2855
2856 /*
2857  *      Bind a packet socket to a device
2858  */
2859
2860 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2861                             int addr_len)
2862 {
2863         struct sock *sk = sock->sk;
2864         char name[15];
2865         struct net_device *dev;
2866         int err = -ENODEV;
2867
2868         /*
2869          *      Check legality
2870          */
2871
2872         if (addr_len != sizeof(struct sockaddr))
2873                 return -EINVAL;
2874         strlcpy(name, uaddr->sa_data, sizeof(name));
2875
2876         dev = dev_get_by_name(sock_net(sk), name);
2877         if (dev)
2878                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2879         return err;
2880 }
2881
2882 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2883 {
2884         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2885         struct sock *sk = sock->sk;
2886         struct net_device *dev = NULL;
2887         int err;
2888
2889
2890         /*
2891          *      Check legality
2892          */
2893
2894         if (addr_len < sizeof(struct sockaddr_ll))
2895                 return -EINVAL;
2896         if (sll->sll_family != AF_PACKET)
2897                 return -EINVAL;
2898
2899         if (sll->sll_ifindex) {
2900                 err = -ENODEV;
2901                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2902                 if (dev == NULL)
2903                         goto out;
2904         }
2905         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2906
2907 out:
2908         return err;
2909 }
2910
2911 static struct proto packet_proto = {
2912         .name     = "PACKET",
2913         .owner    = THIS_MODULE,
2914         .obj_size = sizeof(struct packet_sock),
2915 };
2916
2917 /*
2918  *      Create a packet of type SOCK_PACKET.
2919  */
2920
2921 static int packet_create(struct net *net, struct socket *sock, int protocol,
2922                          int kern)
2923 {
2924         struct sock *sk;
2925         struct packet_sock *po;
2926         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2927         int err;
2928
2929         if (!ns_capable(net->user_ns, CAP_NET_RAW))
2930                 return -EPERM;
2931         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2932             sock->type != SOCK_PACKET)
2933                 return -ESOCKTNOSUPPORT;
2934
2935         sock->state = SS_UNCONNECTED;
2936
2937         err = -ENOBUFS;
2938         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
2939         if (sk == NULL)
2940                 goto out;
2941
2942         sock->ops = &packet_ops;
2943         if (sock->type == SOCK_PACKET)
2944                 sock->ops = &packet_ops_spkt;
2945
2946         sock_init_data(sock, sk);
2947
2948         po = pkt_sk(sk);
2949         sk->sk_family = PF_PACKET;
2950         po->num = proto;
2951         po->xmit = dev_queue_xmit;
2952
2953         err = packet_alloc_pending(po);
2954         if (err)
2955                 goto out2;
2956
2957         packet_cached_dev_reset(po);
2958
2959         sk->sk_destruct = packet_sock_destruct;
2960         sk_refcnt_debug_inc(sk);
2961
2962         /*
2963          *      Attach a protocol block
2964          */
2965
2966         spin_lock_init(&po->bind_lock);
2967         mutex_init(&po->pg_vec_lock);
2968         po->rollover = NULL;
2969         po->prot_hook.func = packet_rcv;
2970
2971         if (sock->type == SOCK_PACKET)
2972                 po->prot_hook.func = packet_rcv_spkt;
2973
2974         po->prot_hook.af_packet_priv = sk;
2975
2976         if (proto) {
2977                 po->prot_hook.type = proto;
2978                 register_prot_hook(sk);
2979         }
2980
2981         mutex_lock(&net->packet.sklist_lock);
2982         sk_add_node_rcu(sk, &net->packet.sklist);
2983         mutex_unlock(&net->packet.sklist_lock);
2984
2985         preempt_disable();
2986         sock_prot_inuse_add(net, &packet_proto, 1);
2987         preempt_enable();
2988
2989         return 0;
2990 out2:
2991         sk_free(sk);
2992 out:
2993         return err;
2994 }
2995
2996 /*
2997  *      Pull a packet from our receive queue and hand it to the user.
2998  *      If necessary we block.
2999  */
3000
3001 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3002                           int flags)
3003 {
3004         struct sock *sk = sock->sk;
3005         struct sk_buff *skb;
3006         int copied, err;
3007         int vnet_hdr_len = 0;
3008         unsigned int origlen = 0;
3009
3010         err = -EINVAL;
3011         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3012                 goto out;
3013
3014 #if 0
3015         /* What error should we return now? EUNATTACH? */
3016         if (pkt_sk(sk)->ifindex < 0)
3017                 return -ENODEV;
3018 #endif
3019
3020         if (flags & MSG_ERRQUEUE) {
3021                 err = sock_recv_errqueue(sk, msg, len,
3022                                          SOL_PACKET, PACKET_TX_TIMESTAMP);
3023                 goto out;
3024         }
3025
3026         /*
3027          *      Call the generic datagram receiver. This handles all sorts
3028          *      of horrible races and re-entrancy so we can forget about it
3029          *      in the protocol layers.
3030          *
3031          *      Now it will return ENETDOWN, if device have just gone down,
3032          *      but then it will block.
3033          */
3034
3035         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3036
3037         /*
3038          *      An error occurred so return it. Because skb_recv_datagram()
3039          *      handles the blocking we don't see and worry about blocking
3040          *      retries.
3041          */
3042
3043         if (skb == NULL)
3044                 goto out;
3045
3046         if (pkt_sk(sk)->pressure)
3047                 packet_rcv_has_room(pkt_sk(sk), NULL);
3048
3049         if (pkt_sk(sk)->has_vnet_hdr) {
3050                 struct virtio_net_hdr vnet_hdr = { 0 };
3051
3052                 err = -EINVAL;
3053                 vnet_hdr_len = sizeof(vnet_hdr);
3054                 if (len < vnet_hdr_len)
3055                         goto out_free;
3056
3057                 len -= vnet_hdr_len;
3058
3059                 if (skb_is_gso(skb)) {
3060                         struct skb_shared_info *sinfo = skb_shinfo(skb);
3061
3062                         /* This is a hint as to how much should be linear. */
3063                         vnet_hdr.hdr_len =
3064                                 __cpu_to_virtio16(false, skb_headlen(skb));
3065                         vnet_hdr.gso_size =
3066                                 __cpu_to_virtio16(false, sinfo->gso_size);
3067                         if (sinfo->gso_type & SKB_GSO_TCPV4)
3068                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
3069                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
3070                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
3071                         else if (sinfo->gso_type & SKB_GSO_UDP)
3072                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
3073                         else if (sinfo->gso_type & SKB_GSO_FCOE)
3074                                 goto out_free;
3075                         else
3076                                 BUG();
3077                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
3078                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
3079                 } else
3080                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
3081
3082                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3083                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
3084                         vnet_hdr.csum_start = __cpu_to_virtio16(false,
3085                                           skb_checksum_start_offset(skb));
3086                         vnet_hdr.csum_offset = __cpu_to_virtio16(false,
3087                                                          skb->csum_offset);
3088                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3089                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
3090                 } /* else everything is zero */
3091
3092                 err = memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_len);
3093                 if (err < 0)
3094                         goto out_free;
3095         }
3096
3097         /* You lose any data beyond the buffer you gave. If it worries
3098          * a user program they can ask the device for its MTU
3099          * anyway.
3100          */
3101         copied = skb->len;
3102         if (copied > len) {
3103                 copied = len;
3104                 msg->msg_flags |= MSG_TRUNC;
3105         }
3106
3107         err = skb_copy_datagram_msg(skb, 0, msg, copied);
3108         if (err)
3109                 goto out_free;
3110
3111         if (sock->type != SOCK_PACKET) {
3112                 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3113
3114                 /* Original length was stored in sockaddr_ll fields */
3115                 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3116                 sll->sll_family = AF_PACKET;
3117                 sll->sll_protocol = skb->protocol;
3118         }
3119
3120         sock_recv_ts_and_drops(msg, sk, skb);
3121
3122         if (msg->msg_name) {
3123                 /* If the address length field is there to be filled
3124                  * in, we fill it in now.
3125                  */
3126                 if (sock->type == SOCK_PACKET) {
3127                         __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3128                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
3129                 } else {
3130                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3131
3132                         msg->msg_namelen = sll->sll_halen +
3133                                 offsetof(struct sockaddr_ll, sll_addr);
3134                 }
3135                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3136                        msg->msg_namelen);
3137         }
3138
3139         if (pkt_sk(sk)->auxdata) {
3140                 struct tpacket_auxdata aux;
3141
3142                 aux.tp_status = TP_STATUS_USER;
3143                 if (skb->ip_summed == CHECKSUM_PARTIAL)
3144                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3145                 else if (skb->pkt_type != PACKET_OUTGOING &&
3146                          (skb->ip_summed == CHECKSUM_COMPLETE ||
3147                           skb_csum_unnecessary(skb)))
3148                         aux.tp_status |= TP_STATUS_CSUM_VALID;
3149
3150                 aux.tp_len = origlen;
3151                 aux.tp_snaplen = skb->len;
3152                 aux.tp_mac = 0;
3153                 aux.tp_net = skb_network_offset(skb);
3154                 if (skb_vlan_tag_present(skb)) {
3155                         aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3156                         aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3157                         aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3158                 } else {
3159                         aux.tp_vlan_tci = 0;
3160                         aux.tp_vlan_tpid = 0;
3161                 }
3162                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3163         }
3164
3165         /*
3166          *      Free or return the buffer as appropriate. Again this
3167          *      hides all the races and re-entrancy issues from us.
3168          */
3169         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3170
3171 out_free:
3172         skb_free_datagram(sk, skb);
3173 out:
3174         return err;
3175 }
3176
3177 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3178                                int *uaddr_len, int peer)
3179 {
3180         struct net_device *dev;
3181         struct sock *sk = sock->sk;
3182
3183         if (peer)
3184                 return -EOPNOTSUPP;
3185
3186         uaddr->sa_family = AF_PACKET;
3187         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3188         rcu_read_lock();
3189         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3190         if (dev)
3191                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3192         rcu_read_unlock();
3193         *uaddr_len = sizeof(*uaddr);
3194
3195         return 0;
3196 }
3197
3198 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3199                           int *uaddr_len, int peer)
3200 {
3201         struct net_device *dev;
3202         struct sock *sk = sock->sk;
3203         struct packet_sock *po = pkt_sk(sk);
3204         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3205
3206         if (peer)
3207                 return -EOPNOTSUPP;
3208
3209         sll->sll_family = AF_PACKET;
3210         sll->sll_ifindex = po->ifindex;
3211         sll->sll_protocol = po->num;
3212         sll->sll_pkttype = 0;
3213         rcu_read_lock();
3214         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3215         if (dev) {
3216                 sll->sll_hatype = dev->type;
3217                 sll->sll_halen = dev->addr_len;
3218                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3219         } else {
3220                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
3221                 sll->sll_halen = 0;
3222         }
3223         rcu_read_unlock();
3224         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3225
3226         return 0;
3227 }
3228
3229 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3230                          int what)
3231 {
3232         switch (i->type) {
3233         case PACKET_MR_MULTICAST:
3234                 if (i->alen != dev->addr_len)
3235                         return -EINVAL;
3236                 if (what > 0)
3237                         return dev_mc_add(dev, i->addr);
3238                 else
3239                         return dev_mc_del(dev, i->addr);
3240                 break;
3241         case PACKET_MR_PROMISC:
3242                 return dev_set_promiscuity(dev, what);
3243         case PACKET_MR_ALLMULTI:
3244                 return dev_set_allmulti(dev, what);
3245         case PACKET_MR_UNICAST:
3246                 if (i->alen != dev->addr_len)
3247                         return -EINVAL;
3248                 if (what > 0)
3249                         return dev_uc_add(dev, i->addr);
3250                 else
3251                         return dev_uc_del(dev, i->addr);
3252                 break;
3253         default:
3254                 break;
3255         }
3256         return 0;
3257 }
3258
3259 static void packet_dev_mclist_delete(struct net_device *dev,
3260                                      struct packet_mclist **mlp)
3261 {
3262         struct packet_mclist *ml;
3263
3264         while ((ml = *mlp) != NULL) {
3265                 if (ml->ifindex == dev->ifindex) {
3266                         packet_dev_mc(dev, ml, -1);
3267                         *mlp = ml->next;
3268                         kfree(ml);
3269                 } else
3270                         mlp = &ml->next;
3271         }
3272 }
3273
3274 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3275 {
3276         struct packet_sock *po = pkt_sk(sk);
3277         struct packet_mclist *ml, *i;
3278         struct net_device *dev;
3279         int err;
3280
3281         rtnl_lock();
3282
3283         err = -ENODEV;
3284         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3285         if (!dev)
3286                 goto done;
3287
3288         err = -EINVAL;
3289         if (mreq->mr_alen > dev->addr_len)
3290                 goto done;
3291
3292         err = -ENOBUFS;
3293         i = kmalloc(sizeof(*i), GFP_KERNEL);
3294         if (i == NULL)
3295                 goto done;
3296
3297         err = 0;
3298         for (ml = po->mclist; ml; ml = ml->next) {
3299                 if (ml->ifindex == mreq->mr_ifindex &&
3300                     ml->type == mreq->mr_type &&
3301                     ml->alen == mreq->mr_alen &&
3302                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3303                         ml->count++;
3304                         /* Free the new element ... */
3305                         kfree(i);
3306                         goto done;
3307                 }
3308         }
3309
3310         i->type = mreq->mr_type;
3311         i->ifindex = mreq->mr_ifindex;
3312         i->alen = mreq->mr_alen;
3313         memcpy(i->addr, mreq->mr_address, i->alen);
3314         i->count = 1;
3315         i->next = po->mclist;
3316         po->mclist = i;
3317         err = packet_dev_mc(dev, i, 1);
3318         if (err) {
3319                 po->mclist = i->next;
3320                 kfree(i);
3321         }
3322
3323 done:
3324         rtnl_unlock();
3325         return err;
3326 }
3327
3328 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3329 {
3330         struct packet_mclist *ml, **mlp;
3331
3332         rtnl_lock();
3333
3334         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3335                 if (ml->ifindex == mreq->mr_ifindex &&
3336                     ml->type == mreq->mr_type &&
3337                     ml->alen == mreq->mr_alen &&
3338                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3339                         if (--ml->count == 0) {
3340                                 struct net_device *dev;
3341                                 *mlp = ml->next;
3342                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3343                                 if (dev)
3344                                         packet_dev_mc(dev, ml, -1);
3345                                 kfree(ml);
3346                         }
3347                         break;
3348                 }
3349         }
3350         rtnl_unlock();
3351         return 0;
3352 }
3353
3354 static void packet_flush_mclist(struct sock *sk)
3355 {
3356         struct packet_sock *po = pkt_sk(sk);
3357         struct packet_mclist *ml;
3358
3359         if (!po->mclist)
3360                 return;
3361
3362         rtnl_lock();
3363         while ((ml = po->mclist) != NULL) {
3364                 struct net_device *dev;
3365
3366                 po->mclist = ml->next;
3367                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3368                 if (dev != NULL)
3369                         packet_dev_mc(dev, ml, -1);
3370                 kfree(ml);
3371         }
3372         rtnl_unlock();
3373 }
3374
3375 static int
3376 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3377 {
3378         struct sock *sk = sock->sk;
3379         struct packet_sock *po = pkt_sk(sk);
3380         int ret;
3381
3382         if (level != SOL_PACKET)
3383                 return -ENOPROTOOPT;
3384
3385         switch (optname) {
3386         case PACKET_ADD_MEMBERSHIP:
3387         case PACKET_DROP_MEMBERSHIP:
3388         {
3389                 struct packet_mreq_max mreq;
3390                 int len = optlen;
3391                 memset(&mreq, 0, sizeof(mreq));
3392                 if (len < sizeof(struct packet_mreq))
3393                         return -EINVAL;
3394                 if (len > sizeof(mreq))
3395                         len = sizeof(mreq);
3396                 if (copy_from_user(&mreq, optval, len))
3397                         return -EFAULT;
3398                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3399                         return -EINVAL;
3400                 if (optname == PACKET_ADD_MEMBERSHIP)
3401                         ret = packet_mc_add(sk, &mreq);
3402                 else
3403                         ret = packet_mc_drop(sk, &mreq);
3404                 return ret;
3405         }
3406
3407         case PACKET_RX_RING:
3408         case PACKET_TX_RING:
3409         {
3410                 union tpacket_req_u req_u;
3411                 int len;
3412
3413                 switch (po->tp_version) {
3414                 case TPACKET_V1:
3415                 case TPACKET_V2:
3416                         len = sizeof(req_u.req);
3417                         break;
3418                 case TPACKET_V3:
3419                 default:
3420                         len = sizeof(req_u.req3);
3421                         break;
3422                 }
3423                 if (optlen < len)
3424                         return -EINVAL;
3425                 if (pkt_sk(sk)->has_vnet_hdr)
3426                         return -EINVAL;
3427                 if (copy_from_user(&req_u.req, optval, len))
3428                         return -EFAULT;
3429                 return packet_set_ring(sk, &req_u, 0,
3430                         optname == PACKET_TX_RING);
3431         }
3432         case PACKET_COPY_THRESH:
3433         {
3434                 int val;
3435
3436                 if (optlen != sizeof(val))
3437                         return -EINVAL;
3438                 if (copy_from_user(&val, optval, sizeof(val)))
3439                         return -EFAULT;
3440
3441                 pkt_sk(sk)->copy_thresh = val;
3442                 return 0;
3443         }
3444         case PACKET_VERSION:
3445         {
3446                 int val;
3447
3448                 if (optlen != sizeof(val))
3449                         return -EINVAL;
3450                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3451                         return -EBUSY;
3452                 if (copy_from_user(&val, optval, sizeof(val)))
3453                         return -EFAULT;
3454                 switch (val) {
3455                 case TPACKET_V1:
3456                 case TPACKET_V2:
3457                 case TPACKET_V3:
3458                         po->tp_version = val;
3459                         return 0;
3460                 default:
3461                         return -EINVAL;
3462                 }
3463         }
3464         case PACKET_RESERVE:
3465         {
3466                 unsigned int val;
3467
3468                 if (optlen != sizeof(val))
3469                         return -EINVAL;
3470                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3471                         return -EBUSY;
3472                 if (copy_from_user(&val, optval, sizeof(val)))
3473                         return -EFAULT;
3474                 po->tp_reserve = val;
3475                 return 0;
3476         }
3477         case PACKET_LOSS:
3478         {
3479                 unsigned int val;
3480
3481                 if (optlen != sizeof(val))
3482                         return -EINVAL;
3483                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3484                         return -EBUSY;
3485                 if (copy_from_user(&val, optval, sizeof(val)))
3486                         return -EFAULT;
3487                 po->tp_loss = !!val;
3488                 return 0;
3489         }
3490         case PACKET_AUXDATA:
3491         {
3492                 int val;
3493
3494                 if (optlen < sizeof(val))
3495                         return -EINVAL;
3496                 if (copy_from_user(&val, optval, sizeof(val)))
3497                         return -EFAULT;
3498
3499                 po->auxdata = !!val;
3500                 return 0;
3501         }
3502         case PACKET_ORIGDEV:
3503         {
3504                 int val;
3505
3506                 if (optlen < sizeof(val))
3507                         return -EINVAL;
3508                 if (copy_from_user(&val, optval, sizeof(val)))
3509                         return -EFAULT;
3510
3511                 po->origdev = !!val;
3512                 return 0;
3513         }
3514         case PACKET_VNET_HDR:
3515         {
3516                 int val;
3517
3518                 if (sock->type != SOCK_RAW)
3519                         return -EINVAL;
3520                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3521                         return -EBUSY;
3522                 if (optlen < sizeof(val))
3523                         return -EINVAL;
3524                 if (copy_from_user(&val, optval, sizeof(val)))
3525                         return -EFAULT;
3526
3527                 po->has_vnet_hdr = !!val;
3528                 return 0;
3529         }
3530         case PACKET_TIMESTAMP:
3531         {
3532                 int val;
3533
3534                 if (optlen != sizeof(val))
3535                         return -EINVAL;
3536                 if (copy_from_user(&val, optval, sizeof(val)))
3537                         return -EFAULT;
3538
3539                 po->tp_tstamp = val;
3540                 return 0;
3541         }
3542         case PACKET_FANOUT:
3543         {
3544                 int val;
3545
3546                 if (optlen != sizeof(val))
3547                         return -EINVAL;
3548                 if (copy_from_user(&val, optval, sizeof(val)))
3549                         return -EFAULT;
3550
3551                 return fanout_add(sk, val & 0xffff, val >> 16);
3552         }
3553         case PACKET_TX_HAS_OFF:
3554         {
3555                 unsigned int val;
3556
3557                 if (optlen != sizeof(val))
3558                         return -EINVAL;
3559                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3560                         return -EBUSY;
3561                 if (copy_from_user(&val, optval, sizeof(val)))
3562                         return -EFAULT;
3563                 po->tp_tx_has_off = !!val;
3564                 return 0;
3565         }
3566         case PACKET_QDISC_BYPASS:
3567         {
3568                 int val;
3569
3570                 if (optlen != sizeof(val))
3571                         return -EINVAL;
3572                 if (copy_from_user(&val, optval, sizeof(val)))
3573                         return -EFAULT;
3574
3575                 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3576                 return 0;
3577         }
3578         default:
3579                 return -ENOPROTOOPT;
3580         }
3581 }
3582
3583 static int packet_getsockopt(struct socket *sock, int level, int optname,
3584                              char __user *optval, int __user *optlen)
3585 {
3586         int len;
3587         int val, lv = sizeof(val);
3588         struct sock *sk = sock->sk;
3589         struct packet_sock *po = pkt_sk(sk);
3590         void *data = &val;
3591         union tpacket_stats_u st;
3592         struct tpacket_rollover_stats rstats;
3593
3594         if (level != SOL_PACKET)
3595                 return -ENOPROTOOPT;
3596
3597         if (get_user(len, optlen))
3598                 return -EFAULT;
3599
3600         if (len < 0)
3601                 return -EINVAL;
3602
3603         switch (optname) {
3604         case PACKET_STATISTICS:
3605                 spin_lock_bh(&sk->sk_receive_queue.lock);
3606                 memcpy(&st, &po->stats, sizeof(st));
3607                 memset(&po->stats, 0, sizeof(po->stats));
3608                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3609
3610                 if (po->tp_version == TPACKET_V3) {
3611                         lv = sizeof(struct tpacket_stats_v3);
3612                         st.stats3.tp_packets += st.stats3.tp_drops;
3613                         data = &st.stats3;
3614                 } else {
3615                         lv = sizeof(struct tpacket_stats);
3616                         st.stats1.tp_packets += st.stats1.tp_drops;
3617                         data = &st.stats1;
3618                 }
3619
3620                 break;
3621         case PACKET_AUXDATA:
3622                 val = po->auxdata;
3623                 break;
3624         case PACKET_ORIGDEV:
3625                 val = po->origdev;
3626                 break;
3627         case PACKET_VNET_HDR:
3628                 val = po->has_vnet_hdr;
3629                 break;
3630         case PACKET_VERSION:
3631                 val = po->tp_version;
3632                 break;
3633         case PACKET_HDRLEN:
3634                 if (len > sizeof(int))
3635                         len = sizeof(int);
3636                 if (copy_from_user(&val, optval, len))
3637                         return -EFAULT;
3638                 switch (val) {
3639                 case TPACKET_V1:
3640                         val = sizeof(struct tpacket_hdr);
3641                         break;
3642                 case TPACKET_V2:
3643                         val = sizeof(struct tpacket2_hdr);
3644                         break;
3645                 case TPACKET_V3:
3646                         val = sizeof(struct tpacket3_hdr);
3647                         break;
3648                 default:
3649                         return -EINVAL;
3650                 }
3651                 break;
3652         case PACKET_RESERVE:
3653                 val = po->tp_reserve;
3654                 break;
3655         case PACKET_LOSS:
3656                 val = po->tp_loss;
3657                 break;
3658         case PACKET_TIMESTAMP:
3659                 val = po->tp_tstamp;
3660                 break;
3661         case PACKET_FANOUT:
3662                 val = (po->fanout ?
3663                        ((u32)po->fanout->id |
3664                         ((u32)po->fanout->type << 16) |
3665                         ((u32)po->fanout->flags << 24)) :
3666                        0);
3667                 break;
3668         case PACKET_ROLLOVER_STATS:
3669                 if (!po->rollover)
3670                         return -EINVAL;
3671                 rstats.tp_all = atomic_long_read(&po->rollover->num);
3672                 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3673                 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3674                 data = &rstats;
3675                 lv = sizeof(rstats);
3676                 break;
3677         case PACKET_TX_HAS_OFF:
3678                 val = po->tp_tx_has_off;
3679                 break;
3680         case PACKET_QDISC_BYPASS:
3681                 val = packet_use_direct_xmit(po);
3682                 break;
3683         default:
3684                 return -ENOPROTOOPT;
3685         }
3686
3687         if (len > lv)
3688                 len = lv;
3689         if (put_user(len, optlen))
3690                 return -EFAULT;
3691         if (copy_to_user(optval, data, len))
3692                 return -EFAULT;
3693         return 0;
3694 }
3695
3696
3697 static int packet_notifier(struct notifier_block *this,
3698                            unsigned long msg, void *ptr)
3699 {
3700         struct sock *sk;
3701         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3702         struct net *net = dev_net(dev);
3703
3704         rcu_read_lock();
3705         sk_for_each_rcu(sk, &net->packet.sklist) {
3706                 struct packet_sock *po = pkt_sk(sk);
3707
3708                 switch (msg) {
3709                 case NETDEV_UNREGISTER:
3710                         if (po->mclist)
3711                                 packet_dev_mclist_delete(dev, &po->mclist);
3712                         /* fallthrough */
3713
3714                 case NETDEV_DOWN:
3715                         if (dev->ifindex == po->ifindex) {
3716                                 spin_lock(&po->bind_lock);
3717                                 if (po->running) {
3718                                         __unregister_prot_hook(sk, false);
3719                                         sk->sk_err = ENETDOWN;
3720                                         if (!sock_flag(sk, SOCK_DEAD))
3721                                                 sk->sk_error_report(sk);
3722                                 }
3723                                 if (msg == NETDEV_UNREGISTER) {
3724                                         packet_cached_dev_reset(po);
3725                                         po->ifindex = -1;
3726                                         if (po->prot_hook.dev)
3727                                                 dev_put(po->prot_hook.dev);
3728                                         po->prot_hook.dev = NULL;
3729                                 }
3730                                 spin_unlock(&po->bind_lock);
3731                         }
3732                         break;
3733                 case NETDEV_UP:
3734                         if (dev->ifindex == po->ifindex) {
3735                                 spin_lock(&po->bind_lock);
3736                                 if (po->num)
3737                                         register_prot_hook(sk);
3738                                 spin_unlock(&po->bind_lock);
3739                         }
3740                         break;
3741                 }
3742         }
3743         rcu_read_unlock();
3744         return NOTIFY_DONE;
3745 }
3746
3747
3748 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3749                         unsigned long arg)
3750 {
3751         struct sock *sk = sock->sk;
3752
3753         switch (cmd) {
3754         case SIOCOUTQ:
3755         {
3756                 int amount = sk_wmem_alloc_get(sk);
3757
3758                 return put_user(amount, (int __user *)arg);
3759         }
3760         case SIOCINQ:
3761         {
3762                 struct sk_buff *skb;
3763                 int amount = 0;
3764
3765                 spin_lock_bh(&sk->sk_receive_queue.lock);
3766                 skb = skb_peek(&sk->sk_receive_queue);
3767                 if (skb)
3768                         amount = skb->len;
3769                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3770                 return put_user(amount, (int __user *)arg);
3771         }
3772         case SIOCGSTAMP:
3773                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3774         case SIOCGSTAMPNS:
3775                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3776
3777 #ifdef CONFIG_INET
3778         case SIOCADDRT:
3779         case SIOCDELRT:
3780         case SIOCDARP:
3781         case SIOCGARP:
3782         case SIOCSARP:
3783         case SIOCGIFADDR:
3784         case SIOCSIFADDR:
3785         case SIOCGIFBRDADDR:
3786         case SIOCSIFBRDADDR:
3787         case SIOCGIFNETMASK:
3788         case SIOCSIFNETMASK:
3789         case SIOCGIFDSTADDR:
3790         case SIOCSIFDSTADDR:
3791         case SIOCSIFFLAGS:
3792                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3793 #endif
3794
3795         default:
3796                 return -ENOIOCTLCMD;
3797         }
3798         return 0;
3799 }
3800
3801 static unsigned int packet_poll(struct file *file, struct socket *sock,
3802                                 poll_table *wait)
3803 {
3804         struct sock *sk = sock->sk;
3805         struct packet_sock *po = pkt_sk(sk);
3806         unsigned int mask = datagram_poll(file, sock, wait);
3807
3808         spin_lock_bh(&sk->sk_receive_queue.lock);
3809         if (po->rx_ring.pg_vec) {
3810                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3811                         TP_STATUS_KERNEL))
3812                         mask |= POLLIN | POLLRDNORM;
3813         }
3814         if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
3815                 po->pressure = 0;
3816         spin_unlock_bh(&sk->sk_receive_queue.lock);
3817         spin_lock_bh(&sk->sk_write_queue.lock);
3818         if (po->tx_ring.pg_vec) {
3819                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3820                         mask |= POLLOUT | POLLWRNORM;
3821         }
3822         spin_unlock_bh(&sk->sk_write_queue.lock);
3823         return mask;
3824 }
3825
3826
3827 /* Dirty? Well, I still did not learn better way to account
3828  * for user mmaps.
3829  */
3830
3831 static void packet_mm_open(struct vm_area_struct *vma)
3832 {
3833         struct file *file = vma->vm_file;
3834         struct socket *sock = file->private_data;
3835         struct sock *sk = sock->sk;
3836
3837         if (sk)
3838                 atomic_inc(&pkt_sk(sk)->mapped);
3839 }
3840
3841 static void packet_mm_close(struct vm_area_struct *vma)
3842 {
3843         struct file *file = vma->vm_file;
3844         struct socket *sock = file->private_data;
3845         struct sock *sk = sock->sk;
3846
3847         if (sk)
3848                 atomic_dec(&pkt_sk(sk)->mapped);
3849 }
3850
3851 static const struct vm_operations_struct packet_mmap_ops = {
3852         .open   =       packet_mm_open,
3853         .close  =       packet_mm_close,
3854 };
3855
3856 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3857                         unsigned int len)
3858 {
3859         int i;
3860
3861         for (i = 0; i < len; i++) {
3862                 if (likely(pg_vec[i].buffer)) {
3863                         if (is_vmalloc_addr(pg_vec[i].buffer))
3864                                 vfree(pg_vec[i].buffer);
3865                         else
3866                                 free_pages((unsigned long)pg_vec[i].buffer,
3867                                            order);
3868                         pg_vec[i].buffer = NULL;
3869                 }
3870         }
3871         kfree(pg_vec);
3872 }
3873
3874 static char *alloc_one_pg_vec_page(unsigned long order)
3875 {
3876         char *buffer;
3877         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3878                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3879
3880         buffer = (char *) __get_free_pages(gfp_flags, order);
3881         if (buffer)
3882                 return buffer;
3883
3884         /* __get_free_pages failed, fall back to vmalloc */
3885         buffer = vzalloc((1 << order) * PAGE_SIZE);
3886         if (buffer)
3887                 return buffer;
3888
3889         /* vmalloc failed, lets dig into swap here */
3890         gfp_flags &= ~__GFP_NORETRY;
3891         buffer = (char *) __get_free_pages(gfp_flags, order);
3892         if (buffer)
3893                 return buffer;
3894
3895         /* complete and utter failure */
3896         return NULL;
3897 }
3898
3899 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3900 {
3901         unsigned int block_nr = req->tp_block_nr;
3902         struct pgv *pg_vec;
3903         int i;
3904
3905         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3906         if (unlikely(!pg_vec))
3907                 goto out;
3908
3909         for (i = 0; i < block_nr; i++) {
3910                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3911                 if (unlikely(!pg_vec[i].buffer))
3912                         goto out_free_pgvec;
3913         }
3914
3915 out:
3916         return pg_vec;
3917
3918 out_free_pgvec:
3919         free_pg_vec(pg_vec, order, block_nr);
3920         pg_vec = NULL;
3921         goto out;
3922 }
3923
3924 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3925                 int closing, int tx_ring)
3926 {
3927         struct pgv *pg_vec = NULL;
3928         struct packet_sock *po = pkt_sk(sk);
3929         int was_running, order = 0;
3930         struct packet_ring_buffer *rb;
3931         struct sk_buff_head *rb_queue;
3932         __be16 num;
3933         int err = -EINVAL;
3934         /* Added to avoid minimal code churn */
3935         struct tpacket_req *req = &req_u->req;
3936
3937         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3938         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3939                 WARN(1, "Tx-ring is not supported.\n");
3940                 goto out;
3941         }
3942
3943         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3944         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3945
3946         err = -EBUSY;
3947         if (!closing) {
3948                 if (atomic_read(&po->mapped))
3949                         goto out;
3950                 if (packet_read_pending(rb))
3951                         goto out;
3952         }
3953
3954         if (req->tp_block_nr) {
3955                 /* Sanity tests and some calculations */
3956                 err = -EBUSY;
3957                 if (unlikely(rb->pg_vec))
3958                         goto out;
3959
3960                 switch (po->tp_version) {
3961                 case TPACKET_V1:
3962                         po->tp_hdrlen = TPACKET_HDRLEN;
3963                         break;
3964                 case TPACKET_V2:
3965                         po->tp_hdrlen = TPACKET2_HDRLEN;
3966                         break;
3967                 case TPACKET_V3:
3968                         po->tp_hdrlen = TPACKET3_HDRLEN;
3969                         break;
3970                 }
3971
3972                 err = -EINVAL;
3973                 if (unlikely((int)req->tp_block_size <= 0))
3974                         goto out;
3975                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3976                         goto out;
3977                 if (po->tp_version >= TPACKET_V3 &&
3978                     (int)(req->tp_block_size -
3979                           BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
3980                         goto out;
3981                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3982                                         po->tp_reserve))
3983                         goto out;
3984                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3985                         goto out;
3986
3987                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3988                 if (unlikely(rb->frames_per_block <= 0))
3989                         goto out;
3990                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3991                                         req->tp_frame_nr))
3992                         goto out;
3993
3994                 err = -ENOMEM;
3995                 order = get_order(req->tp_block_size);
3996                 pg_vec = alloc_pg_vec(req, order);
3997                 if (unlikely(!pg_vec))
3998                         goto out;
3999                 switch (po->tp_version) {
4000                 case TPACKET_V3:
4001                 /* Transmit path is not supported. We checked
4002                  * it above but just being paranoid
4003                  */
4004                         if (!tx_ring)
4005                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
4006                         break;
4007                 default:
4008                         break;
4009                 }
4010         }
4011         /* Done */
4012         else {
4013                 err = -EINVAL;
4014                 if (unlikely(req->tp_frame_nr))
4015                         goto out;
4016         }
4017
4018         lock_sock(sk);
4019
4020         /* Detach socket from network */
4021         spin_lock(&po->bind_lock);
4022         was_running = po->running;
4023         num = po->num;
4024         if (was_running) {
4025                 po->num = 0;
4026                 __unregister_prot_hook(sk, false);
4027         }
4028         spin_unlock(&po->bind_lock);
4029
4030         synchronize_net();
4031
4032         err = -EBUSY;
4033         mutex_lock(&po->pg_vec_lock);
4034         if (closing || atomic_read(&po->mapped) == 0) {
4035                 err = 0;
4036                 spin_lock_bh(&rb_queue->lock);
4037                 swap(rb->pg_vec, pg_vec);
4038                 rb->frame_max = (req->tp_frame_nr - 1);
4039                 rb->head = 0;
4040                 rb->frame_size = req->tp_frame_size;
4041                 spin_unlock_bh(&rb_queue->lock);
4042
4043                 swap(rb->pg_vec_order, order);
4044                 swap(rb->pg_vec_len, req->tp_block_nr);
4045
4046                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4047                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4048                                                 tpacket_rcv : packet_rcv;
4049                 skb_queue_purge(rb_queue);
4050                 if (atomic_read(&po->mapped))
4051                         pr_err("packet_mmap: vma is busy: %d\n",
4052                                atomic_read(&po->mapped));
4053         }
4054         mutex_unlock(&po->pg_vec_lock);
4055
4056         spin_lock(&po->bind_lock);
4057         if (was_running) {
4058                 po->num = num;
4059                 register_prot_hook(sk);
4060         }
4061         spin_unlock(&po->bind_lock);
4062         if (closing && (po->tp_version > TPACKET_V2)) {
4063                 /* Because we don't support block-based V3 on tx-ring */
4064                 if (!tx_ring)
4065                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
4066         }
4067         release_sock(sk);
4068
4069         if (pg_vec)
4070                 free_pg_vec(pg_vec, order, req->tp_block_nr);
4071 out:
4072         return err;
4073 }
4074
4075 static int packet_mmap(struct file *file, struct socket *sock,
4076                 struct vm_area_struct *vma)
4077 {
4078         struct sock *sk = sock->sk;
4079         struct packet_sock *po = pkt_sk(sk);
4080         unsigned long size, expected_size;
4081         struct packet_ring_buffer *rb;
4082         unsigned long start;
4083         int err = -EINVAL;
4084         int i;
4085
4086         if (vma->vm_pgoff)
4087                 return -EINVAL;
4088
4089         mutex_lock(&po->pg_vec_lock);
4090
4091         expected_size = 0;
4092         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4093                 if (rb->pg_vec) {
4094                         expected_size += rb->pg_vec_len
4095                                                 * rb->pg_vec_pages
4096                                                 * PAGE_SIZE;
4097                 }
4098         }
4099
4100         if (expected_size == 0)
4101                 goto out;
4102
4103         size = vma->vm_end - vma->vm_start;
4104         if (size != expected_size)
4105                 goto out;
4106
4107         start = vma->vm_start;
4108         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4109                 if (rb->pg_vec == NULL)
4110                         continue;
4111
4112                 for (i = 0; i < rb->pg_vec_len; i++) {
4113                         struct page *page;
4114                         void *kaddr = rb->pg_vec[i].buffer;
4115                         int pg_num;
4116
4117                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4118                                 page = pgv_to_page(kaddr);
4119                                 err = vm_insert_page(vma, start, page);
4120                                 if (unlikely(err))
4121                                         goto out;
4122                                 start += PAGE_SIZE;
4123                                 kaddr += PAGE_SIZE;
4124                         }
4125                 }
4126         }
4127
4128         atomic_inc(&po->mapped);
4129         vma->vm_ops = &packet_mmap_ops;
4130         err = 0;
4131
4132 out:
4133         mutex_unlock(&po->pg_vec_lock);
4134         return err;
4135 }
4136
4137 static const struct proto_ops packet_ops_spkt = {
4138         .family =       PF_PACKET,
4139         .owner =        THIS_MODULE,
4140         .release =      packet_release,
4141         .bind =         packet_bind_spkt,
4142         .connect =      sock_no_connect,
4143         .socketpair =   sock_no_socketpair,
4144         .accept =       sock_no_accept,
4145         .getname =      packet_getname_spkt,
4146         .poll =         datagram_poll,
4147         .ioctl =        packet_ioctl,
4148         .listen =       sock_no_listen,
4149         .shutdown =     sock_no_shutdown,
4150         .setsockopt =   sock_no_setsockopt,
4151         .getsockopt =   sock_no_getsockopt,
4152         .sendmsg =      packet_sendmsg_spkt,
4153         .recvmsg =      packet_recvmsg,
4154         .mmap =         sock_no_mmap,
4155         .sendpage =     sock_no_sendpage,
4156 };
4157
4158 static const struct proto_ops packet_ops = {
4159         .family =       PF_PACKET,
4160         .owner =        THIS_MODULE,
4161         .release =      packet_release,
4162         .bind =         packet_bind,
4163         .connect =      sock_no_connect,
4164         .socketpair =   sock_no_socketpair,
4165         .accept =       sock_no_accept,
4166         .getname =      packet_getname,
4167         .poll =         packet_poll,
4168         .ioctl =        packet_ioctl,
4169         .listen =       sock_no_listen,
4170         .shutdown =     sock_no_shutdown,
4171         .setsockopt =   packet_setsockopt,
4172         .getsockopt =   packet_getsockopt,
4173         .sendmsg =      packet_sendmsg,
4174         .recvmsg =      packet_recvmsg,
4175         .mmap =         packet_mmap,
4176         .sendpage =     sock_no_sendpage,
4177 };
4178
4179 static const struct net_proto_family packet_family_ops = {
4180         .family =       PF_PACKET,
4181         .create =       packet_create,
4182         .owner  =       THIS_MODULE,
4183 };
4184
4185 static struct notifier_block packet_netdev_notifier = {
4186         .notifier_call =        packet_notifier,
4187 };
4188
4189 #ifdef CONFIG_PROC_FS
4190
4191 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4192         __acquires(RCU)
4193 {
4194         struct net *net = seq_file_net(seq);
4195
4196         rcu_read_lock();
4197         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4198 }
4199
4200 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4201 {
4202         struct net *net = seq_file_net(seq);
4203         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4204 }
4205
4206 static void packet_seq_stop(struct seq_file *seq, void *v)
4207         __releases(RCU)
4208 {
4209         rcu_read_unlock();
4210 }
4211
4212 static int packet_seq_show(struct seq_file *seq, void *v)
4213 {
4214         if (v == SEQ_START_TOKEN)
4215                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
4216         else {
4217                 struct sock *s = sk_entry(v);
4218                 const struct packet_sock *po = pkt_sk(s);
4219
4220                 seq_printf(seq,
4221                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4222                            s,
4223                            atomic_read(&s->sk_refcnt),
4224                            s->sk_type,
4225                            ntohs(po->num),
4226                            po->ifindex,
4227                            po->running,
4228                            atomic_read(&s->sk_rmem_alloc),
4229                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4230                            sock_i_ino(s));
4231         }
4232
4233         return 0;
4234 }
4235
4236 static const struct seq_operations packet_seq_ops = {
4237         .start  = packet_seq_start,
4238         .next   = packet_seq_next,
4239         .stop   = packet_seq_stop,
4240         .show   = packet_seq_show,
4241 };
4242
4243 static int packet_seq_open(struct inode *inode, struct file *file)
4244 {
4245         return seq_open_net(inode, file, &packet_seq_ops,
4246                             sizeof(struct seq_net_private));
4247 }
4248
4249 static const struct file_operations packet_seq_fops = {
4250         .owner          = THIS_MODULE,
4251         .open           = packet_seq_open,
4252         .read           = seq_read,
4253         .llseek         = seq_lseek,
4254         .release        = seq_release_net,
4255 };
4256
4257 #endif
4258
4259 static int __net_init packet_net_init(struct net *net)
4260 {
4261         mutex_init(&net->packet.sklist_lock);
4262         INIT_HLIST_HEAD(&net->packet.sklist);
4263
4264         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4265                 return -ENOMEM;
4266
4267         return 0;
4268 }
4269
4270 static void __net_exit packet_net_exit(struct net *net)
4271 {
4272         remove_proc_entry("packet", net->proc_net);
4273 }
4274
4275 static struct pernet_operations packet_net_ops = {
4276         .init = packet_net_init,
4277         .exit = packet_net_exit,
4278 };
4279
4280
4281 static void __exit packet_exit(void)
4282 {
4283         unregister_netdevice_notifier(&packet_netdev_notifier);
4284         unregister_pernet_subsys(&packet_net_ops);
4285         sock_unregister(PF_PACKET);
4286         proto_unregister(&packet_proto);
4287 }
4288
4289 static int __init packet_init(void)
4290 {
4291         int rc = proto_register(&packet_proto, 0);
4292
4293         if (rc != 0)
4294                 goto out;
4295
4296         sock_register(&packet_family_ops);
4297         register_pernet_subsys(&packet_net_ops);
4298         register_netdevice_notifier(&packet_netdev_notifier);
4299 out:
4300         return rc;
4301 }
4302
4303 module_init(packet_init);
4304 module_exit(packet_exit);
4305 MODULE_LICENSE("GPL");
4306 MODULE_ALIAS_NETPROTO(PF_PACKET);