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         if (po->prot_hook.func == tpacket_rcv) {
1305                 spin_lock(&po->sk.sk_receive_queue.lock);
1306                 ret = __packet_rcv_has_room(po, skb);
1307                 spin_unlock(&po->sk.sk_receive_queue.lock);
1308         } else {
1309                 ret = __packet_rcv_has_room(po, skb);
1310         }
1311
1312         has_room = ret == ROOM_NORMAL;
1313         if (po->pressure == has_room)
1314                 xchg(&po->pressure, !has_room);
1315
1316         return ret;
1317 }
1318
1319 static void packet_sock_destruct(struct sock *sk)
1320 {
1321         skb_queue_purge(&sk->sk_error_queue);
1322
1323         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1324         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1325
1326         if (!sock_flag(sk, SOCK_DEAD)) {
1327                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1328                 return;
1329         }
1330
1331         sk_refcnt_debug_dec(sk);
1332 }
1333
1334 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1335 {
1336         int x = atomic_read(&f->rr_cur) + 1;
1337
1338         if (x >= num)
1339                 x = 0;
1340
1341         return x;
1342 }
1343
1344 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1345 {
1346         u32 rxhash;
1347         int i, count = 0;
1348
1349         rxhash = skb_get_hash(skb);
1350         for (i = 0; i < ROLLOVER_HLEN; i++)
1351                 if (po->rollover->history[i] == rxhash)
1352                         count++;
1353
1354         po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1355         return count > (ROLLOVER_HLEN >> 1);
1356 }
1357
1358 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1359                                       struct sk_buff *skb,
1360                                       unsigned int num)
1361 {
1362         return reciprocal_scale(skb_get_hash(skb), num);
1363 }
1364
1365 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1366                                     struct sk_buff *skb,
1367                                     unsigned int num)
1368 {
1369         int cur, old;
1370
1371         cur = atomic_read(&f->rr_cur);
1372         while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1373                                      fanout_rr_next(f, num))) != cur)
1374                 cur = old;
1375         return cur;
1376 }
1377
1378 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1379                                      struct sk_buff *skb,
1380                                      unsigned int num)
1381 {
1382         return smp_processor_id() % num;
1383 }
1384
1385 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1386                                      struct sk_buff *skb,
1387                                      unsigned int num)
1388 {
1389         return prandom_u32_max(num);
1390 }
1391
1392 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1393                                           struct sk_buff *skb,
1394                                           unsigned int idx, bool try_self,
1395                                           unsigned int num)
1396 {
1397         struct packet_sock *po, *po_next;
1398         unsigned int i, j, room = ROOM_NONE;
1399
1400         po = pkt_sk(f->arr[idx]);
1401
1402         if (try_self) {
1403                 room = packet_rcv_has_room(po, skb);
1404                 if (room == ROOM_NORMAL ||
1405                     (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1406                         return idx;
1407         }
1408
1409         i = j = min_t(int, po->rollover->sock, num - 1);
1410         do {
1411                 po_next = pkt_sk(f->arr[i]);
1412                 if (po_next != po && !po_next->pressure &&
1413                     packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1414                         if (i != j)
1415                                 po->rollover->sock = i;
1416                         atomic_long_inc(&po->rollover->num);
1417                         if (room == ROOM_LOW)
1418                                 atomic_long_inc(&po->rollover->num_huge);
1419                         return i;
1420                 }
1421
1422                 if (++i == num)
1423                         i = 0;
1424         } while (i != j);
1425
1426         atomic_long_inc(&po->rollover->num_failed);
1427         return idx;
1428 }
1429
1430 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1431                                     struct sk_buff *skb,
1432                                     unsigned int num)
1433 {
1434         return skb_get_queue_mapping(skb) % num;
1435 }
1436
1437 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1438 {
1439         return f->flags & (flag >> 8);
1440 }
1441
1442 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1443                              struct packet_type *pt, struct net_device *orig_dev)
1444 {
1445         struct packet_fanout *f = pt->af_packet_priv;
1446         unsigned int num = f->num_members;
1447         struct packet_sock *po;
1448         unsigned int idx;
1449
1450         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1451             !num) {
1452                 kfree_skb(skb);
1453                 return 0;
1454         }
1455
1456         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1457                 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1458                 if (!skb)
1459                         return 0;
1460         }
1461         switch (f->type) {
1462         case PACKET_FANOUT_HASH:
1463         default:
1464                 idx = fanout_demux_hash(f, skb, num);
1465                 break;
1466         case PACKET_FANOUT_LB:
1467                 idx = fanout_demux_lb(f, skb, num);
1468                 break;
1469         case PACKET_FANOUT_CPU:
1470                 idx = fanout_demux_cpu(f, skb, num);
1471                 break;
1472         case PACKET_FANOUT_RND:
1473                 idx = fanout_demux_rnd(f, skb, num);
1474                 break;
1475         case PACKET_FANOUT_QM:
1476                 idx = fanout_demux_qm(f, skb, num);
1477                 break;
1478         case PACKET_FANOUT_ROLLOVER:
1479                 idx = fanout_demux_rollover(f, skb, 0, false, num);
1480                 break;
1481         }
1482
1483         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1484                 idx = fanout_demux_rollover(f, skb, idx, true, num);
1485
1486         po = pkt_sk(f->arr[idx]);
1487         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1488 }
1489
1490 DEFINE_MUTEX(fanout_mutex);
1491 EXPORT_SYMBOL_GPL(fanout_mutex);
1492 static LIST_HEAD(fanout_list);
1493
1494 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1495 {
1496         struct packet_fanout *f = po->fanout;
1497
1498         spin_lock(&f->lock);
1499         f->arr[f->num_members] = sk;
1500         smp_wmb();
1501         f->num_members++;
1502         spin_unlock(&f->lock);
1503 }
1504
1505 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1506 {
1507         struct packet_fanout *f = po->fanout;
1508         int i;
1509
1510         spin_lock(&f->lock);
1511         for (i = 0; i < f->num_members; i++) {
1512                 if (f->arr[i] == sk)
1513                         break;
1514         }
1515         BUG_ON(i >= f->num_members);
1516         f->arr[i] = f->arr[f->num_members - 1];
1517         f->num_members--;
1518         spin_unlock(&f->lock);
1519 }
1520
1521 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1522 {
1523         if (ptype->af_packet_priv == (void *)((struct packet_sock *)sk)->fanout)
1524                 return true;
1525
1526         return false;
1527 }
1528
1529 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1530 {
1531         struct packet_sock *po = pkt_sk(sk);
1532         struct packet_fanout *f, *match;
1533         u8 type = type_flags & 0xff;
1534         u8 flags = type_flags >> 8;
1535         int err;
1536
1537         switch (type) {
1538         case PACKET_FANOUT_ROLLOVER:
1539                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1540                         return -EINVAL;
1541         case PACKET_FANOUT_HASH:
1542         case PACKET_FANOUT_LB:
1543         case PACKET_FANOUT_CPU:
1544         case PACKET_FANOUT_RND:
1545         case PACKET_FANOUT_QM:
1546                 break;
1547         default:
1548                 return -EINVAL;
1549         }
1550
1551         if (!po->running)
1552                 return -EINVAL;
1553
1554         if (po->fanout)
1555                 return -EALREADY;
1556
1557         if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER) {
1558                 po->rollover = kzalloc(sizeof(*po->rollover), GFP_KERNEL);
1559                 if (!po->rollover)
1560                         return -ENOMEM;
1561                 atomic_long_set(&po->rollover->num, 0);
1562                 atomic_long_set(&po->rollover->num_huge, 0);
1563                 atomic_long_set(&po->rollover->num_failed, 0);
1564         }
1565
1566         mutex_lock(&fanout_mutex);
1567         match = NULL;
1568         list_for_each_entry(f, &fanout_list, list) {
1569                 if (f->id == id &&
1570                     read_pnet(&f->net) == sock_net(sk)) {
1571                         match = f;
1572                         break;
1573                 }
1574         }
1575         err = -EINVAL;
1576         if (match && match->flags != flags)
1577                 goto out;
1578         if (!match) {
1579                 err = -ENOMEM;
1580                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1581                 if (!match)
1582                         goto out;
1583                 write_pnet(&match->net, sock_net(sk));
1584                 match->id = id;
1585                 match->type = type;
1586                 match->flags = flags;
1587                 atomic_set(&match->rr_cur, 0);
1588                 INIT_LIST_HEAD(&match->list);
1589                 spin_lock_init(&match->lock);
1590                 atomic_set(&match->sk_ref, 0);
1591                 match->prot_hook.type = po->prot_hook.type;
1592                 match->prot_hook.dev = po->prot_hook.dev;
1593                 match->prot_hook.func = packet_rcv_fanout;
1594                 match->prot_hook.af_packet_priv = match;
1595                 match->prot_hook.id_match = match_fanout_group;
1596                 dev_add_pack(&match->prot_hook);
1597                 list_add(&match->list, &fanout_list);
1598         }
1599         err = -EINVAL;
1600         if (match->type == type &&
1601             match->prot_hook.type == po->prot_hook.type &&
1602             match->prot_hook.dev == po->prot_hook.dev) {
1603                 err = -ENOSPC;
1604                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1605                         __dev_remove_pack(&po->prot_hook);
1606                         po->fanout = match;
1607                         atomic_inc(&match->sk_ref);
1608                         __fanout_link(sk, po);
1609                         err = 0;
1610                 }
1611         }
1612 out:
1613         mutex_unlock(&fanout_mutex);
1614         if (err) {
1615                 kfree(po->rollover);
1616                 po->rollover = NULL;
1617         }
1618         return err;
1619 }
1620
1621 static void fanout_release(struct sock *sk)
1622 {
1623         struct packet_sock *po = pkt_sk(sk);
1624         struct packet_fanout *f;
1625
1626         f = po->fanout;
1627         if (!f)
1628                 return;
1629
1630         mutex_lock(&fanout_mutex);
1631         po->fanout = NULL;
1632
1633         if (atomic_dec_and_test(&f->sk_ref)) {
1634                 list_del(&f->list);
1635                 dev_remove_pack(&f->prot_hook);
1636                 kfree(f);
1637         }
1638         mutex_unlock(&fanout_mutex);
1639
1640         kfree(po->rollover);
1641 }
1642
1643 static const struct proto_ops packet_ops;
1644
1645 static const struct proto_ops packet_ops_spkt;
1646
1647 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1648                            struct packet_type *pt, struct net_device *orig_dev)
1649 {
1650         struct sock *sk;
1651         struct sockaddr_pkt *spkt;
1652
1653         /*
1654          *      When we registered the protocol we saved the socket in the data
1655          *      field for just this event.
1656          */
1657
1658         sk = pt->af_packet_priv;
1659
1660         /*
1661          *      Yank back the headers [hope the device set this
1662          *      right or kerboom...]
1663          *
1664          *      Incoming packets have ll header pulled,
1665          *      push it back.
1666          *
1667          *      For outgoing ones skb->data == skb_mac_header(skb)
1668          *      so that this procedure is noop.
1669          */
1670
1671         if (skb->pkt_type == PACKET_LOOPBACK)
1672                 goto out;
1673
1674         if (!net_eq(dev_net(dev), sock_net(sk)))
1675                 goto out;
1676
1677         skb = skb_share_check(skb, GFP_ATOMIC);
1678         if (skb == NULL)
1679                 goto oom;
1680
1681         /* drop any routing info */
1682         skb_dst_drop(skb);
1683
1684         /* drop conntrack reference */
1685         nf_reset(skb);
1686
1687         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1688
1689         skb_push(skb, skb->data - skb_mac_header(skb));
1690
1691         /*
1692          *      The SOCK_PACKET socket receives _all_ frames.
1693          */
1694
1695         spkt->spkt_family = dev->type;
1696         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1697         spkt->spkt_protocol = skb->protocol;
1698
1699         /*
1700          *      Charge the memory to the socket. This is done specifically
1701          *      to prevent sockets using all the memory up.
1702          */
1703
1704         if (sock_queue_rcv_skb(sk, skb) == 0)
1705                 return 0;
1706
1707 out:
1708         kfree_skb(skb);
1709 oom:
1710         return 0;
1711 }
1712
1713
1714 /*
1715  *      Output a raw packet to a device layer. This bypasses all the other
1716  *      protocol layers and you must therefore supply it with a complete frame
1717  */
1718
1719 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1720                                size_t len)
1721 {
1722         struct sock *sk = sock->sk;
1723         DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1724         struct sk_buff *skb = NULL;
1725         struct net_device *dev;
1726         __be16 proto = 0;
1727         int err;
1728         int extra_len = 0;
1729
1730         /*
1731          *      Get and verify the address.
1732          */
1733
1734         if (saddr) {
1735                 if (msg->msg_namelen < sizeof(struct sockaddr))
1736                         return -EINVAL;
1737                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1738                         proto = saddr->spkt_protocol;
1739         } else
1740                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1741
1742         /*
1743          *      Find the device first to size check it
1744          */
1745
1746         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1747 retry:
1748         rcu_read_lock();
1749         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1750         err = -ENODEV;
1751         if (dev == NULL)
1752                 goto out_unlock;
1753
1754         err = -ENETDOWN;
1755         if (!(dev->flags & IFF_UP))
1756                 goto out_unlock;
1757
1758         /*
1759          * You may not queue a frame bigger than the mtu. This is the lowest level
1760          * raw protocol and you must do your own fragmentation at this level.
1761          */
1762
1763         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1764                 if (!netif_supports_nofcs(dev)) {
1765                         err = -EPROTONOSUPPORT;
1766                         goto out_unlock;
1767                 }
1768                 extra_len = 4; /* We're doing our own CRC */
1769         }
1770
1771         err = -EMSGSIZE;
1772         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1773                 goto out_unlock;
1774
1775         if (!skb) {
1776                 size_t reserved = LL_RESERVED_SPACE(dev);
1777                 int tlen = dev->needed_tailroom;
1778                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1779
1780                 rcu_read_unlock();
1781                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1782                 if (skb == NULL)
1783                         return -ENOBUFS;
1784                 /* FIXME: Save some space for broken drivers that write a hard
1785                  * header at transmission time by themselves. PPP is the notable
1786                  * one here. This should really be fixed at the driver level.
1787                  */
1788                 skb_reserve(skb, reserved);
1789                 skb_reset_network_header(skb);
1790
1791                 /* Try to align data part correctly */
1792                 if (hhlen) {
1793                         skb->data -= hhlen;
1794                         skb->tail -= hhlen;
1795                         if (len < hhlen)
1796                                 skb_reset_network_header(skb);
1797                 }
1798                 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1799                 if (err)
1800                         goto out_free;
1801                 goto retry;
1802         }
1803
1804         if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1805                 /* Earlier code assumed this would be a VLAN pkt,
1806                  * double-check this now that we have the actual
1807                  * packet in hand.
1808                  */
1809                 struct ethhdr *ehdr;
1810                 skb_reset_mac_header(skb);
1811                 ehdr = eth_hdr(skb);
1812                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1813                         err = -EMSGSIZE;
1814                         goto out_unlock;
1815                 }
1816         }
1817
1818         skb->protocol = proto;
1819         skb->dev = dev;
1820         skb->priority = sk->sk_priority;
1821         skb->mark = sk->sk_mark;
1822
1823         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1824
1825         if (unlikely(extra_len == 4))
1826                 skb->no_fcs = 1;
1827
1828         skb_probe_transport_header(skb, 0);
1829
1830         dev_queue_xmit(skb);
1831         rcu_read_unlock();
1832         return len;
1833
1834 out_unlock:
1835         rcu_read_unlock();
1836 out_free:
1837         kfree_skb(skb);
1838         return err;
1839 }
1840
1841 static unsigned int run_filter(const struct sk_buff *skb,
1842                                       const struct sock *sk,
1843                                       unsigned int res)
1844 {
1845         struct sk_filter *filter;
1846
1847         rcu_read_lock();
1848         filter = rcu_dereference(sk->sk_filter);
1849         if (filter != NULL)
1850                 res = SK_RUN_FILTER(filter, skb);
1851         rcu_read_unlock();
1852
1853         return res;
1854 }
1855
1856 /*
1857  * This function makes lazy skb cloning in hope that most of packets
1858  * are discarded by BPF.
1859  *
1860  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1861  * and skb->cb are mangled. It works because (and until) packets
1862  * falling here are owned by current CPU. Output packets are cloned
1863  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1864  * sequencially, so that if we return skb to original state on exit,
1865  * we will not harm anyone.
1866  */
1867
1868 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1869                       struct packet_type *pt, struct net_device *orig_dev)
1870 {
1871         struct sock *sk;
1872         struct sockaddr_ll *sll;
1873         struct packet_sock *po;
1874         u8 *skb_head = skb->data;
1875         int skb_len = skb->len;
1876         unsigned int snaplen, res;
1877
1878         if (skb->pkt_type == PACKET_LOOPBACK)
1879                 goto drop;
1880
1881         sk = pt->af_packet_priv;
1882         po = pkt_sk(sk);
1883
1884         if (!net_eq(dev_net(dev), sock_net(sk)))
1885                 goto drop;
1886
1887         skb->dev = dev;
1888
1889         if (dev->header_ops) {
1890                 /* The device has an explicit notion of ll header,
1891                  * exported to higher levels.
1892                  *
1893                  * Otherwise, the device hides details of its frame
1894                  * structure, so that corresponding packet head is
1895                  * never delivered to user.
1896                  */
1897                 if (sk->sk_type != SOCK_DGRAM)
1898                         skb_push(skb, skb->data - skb_mac_header(skb));
1899                 else if (skb->pkt_type == PACKET_OUTGOING) {
1900                         /* Special case: outgoing packets have ll header at head */
1901                         skb_pull(skb, skb_network_offset(skb));
1902                 }
1903         }
1904
1905         snaplen = skb->len;
1906
1907         res = run_filter(skb, sk, snaplen);
1908         if (!res)
1909                 goto drop_n_restore;
1910         if (snaplen > res)
1911                 snaplen = res;
1912
1913         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1914                 goto drop_n_acct;
1915
1916         if (skb_shared(skb)) {
1917                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1918                 if (nskb == NULL)
1919                         goto drop_n_acct;
1920
1921                 if (skb_head != skb->data) {
1922                         skb->data = skb_head;
1923                         skb->len = skb_len;
1924                 }
1925                 consume_skb(skb);
1926                 skb = nskb;
1927         }
1928
1929         sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
1930
1931         sll = &PACKET_SKB_CB(skb)->sa.ll;
1932         sll->sll_hatype = dev->type;
1933         sll->sll_pkttype = skb->pkt_type;
1934         if (unlikely(po->origdev))
1935                 sll->sll_ifindex = orig_dev->ifindex;
1936         else
1937                 sll->sll_ifindex = dev->ifindex;
1938
1939         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1940
1941         /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
1942          * Use their space for storing the original skb length.
1943          */
1944         PACKET_SKB_CB(skb)->sa.origlen = skb->len;
1945
1946         if (pskb_trim(skb, snaplen))
1947                 goto drop_n_acct;
1948
1949         skb_set_owner_r(skb, sk);
1950         skb->dev = NULL;
1951         skb_dst_drop(skb);
1952
1953         /* drop conntrack reference */
1954         nf_reset(skb);
1955
1956         spin_lock(&sk->sk_receive_queue.lock);
1957         po->stats.stats1.tp_packets++;
1958         sock_skb_set_dropcount(sk, skb);
1959         __skb_queue_tail(&sk->sk_receive_queue, skb);
1960         spin_unlock(&sk->sk_receive_queue.lock);
1961         sk->sk_data_ready(sk);
1962         return 0;
1963
1964 drop_n_acct:
1965         spin_lock(&sk->sk_receive_queue.lock);
1966         po->stats.stats1.tp_drops++;
1967         atomic_inc(&sk->sk_drops);
1968         spin_unlock(&sk->sk_receive_queue.lock);
1969
1970 drop_n_restore:
1971         if (skb_head != skb->data && skb_shared(skb)) {
1972                 skb->data = skb_head;
1973                 skb->len = skb_len;
1974         }
1975 drop:
1976         consume_skb(skb);
1977         return 0;
1978 }
1979
1980 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1981                        struct packet_type *pt, struct net_device *orig_dev)
1982 {
1983         struct sock *sk;
1984         struct packet_sock *po;
1985         struct sockaddr_ll *sll;
1986         union tpacket_uhdr h;
1987         u8 *skb_head = skb->data;
1988         int skb_len = skb->len;
1989         unsigned int snaplen, res;
1990         unsigned long status = TP_STATUS_USER;
1991         unsigned short macoff, netoff, hdrlen;
1992         struct sk_buff *copy_skb = NULL;
1993         struct timespec ts;
1994         __u32 ts_status;
1995
1996         /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
1997          * We may add members to them until current aligned size without forcing
1998          * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
1999          */
2000         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2001         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2002
2003         if (skb->pkt_type == PACKET_LOOPBACK)
2004                 goto drop;
2005
2006         sk = pt->af_packet_priv;
2007         po = pkt_sk(sk);
2008
2009         if (!net_eq(dev_net(dev), sock_net(sk)))
2010                 goto drop;
2011
2012         if (dev->header_ops) {
2013                 if (sk->sk_type != SOCK_DGRAM)
2014                         skb_push(skb, skb->data - skb_mac_header(skb));
2015                 else if (skb->pkt_type == PACKET_OUTGOING) {
2016                         /* Special case: outgoing packets have ll header at head */
2017                         skb_pull(skb, skb_network_offset(skb));
2018                 }
2019         }
2020
2021         snaplen = skb->len;
2022
2023         res = run_filter(skb, sk, snaplen);
2024         if (!res)
2025                 goto drop_n_restore;
2026
2027         if (skb->ip_summed == CHECKSUM_PARTIAL)
2028                 status |= TP_STATUS_CSUMNOTREADY;
2029         else if (skb->pkt_type != PACKET_OUTGOING &&
2030                  (skb->ip_summed == CHECKSUM_COMPLETE ||
2031                   skb_csum_unnecessary(skb)))
2032                 status |= TP_STATUS_CSUM_VALID;
2033
2034         if (snaplen > res)
2035                 snaplen = res;
2036
2037         if (sk->sk_type == SOCK_DGRAM) {
2038                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2039                                   po->tp_reserve;
2040         } else {
2041                 unsigned int maclen = skb_network_offset(skb);
2042                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2043                                        (maclen < 16 ? 16 : maclen)) +
2044                         po->tp_reserve;
2045                 macoff = netoff - maclen;
2046         }
2047         if (po->tp_version <= TPACKET_V2) {
2048                 if (macoff + snaplen > po->rx_ring.frame_size) {
2049                         if (po->copy_thresh &&
2050                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2051                                 if (skb_shared(skb)) {
2052                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
2053                                 } else {
2054                                         copy_skb = skb_get(skb);
2055                                         skb_head = skb->data;
2056                                 }
2057                                 if (copy_skb)
2058                                         skb_set_owner_r(copy_skb, sk);
2059                         }
2060                         snaplen = po->rx_ring.frame_size - macoff;
2061                         if ((int)snaplen < 0)
2062                                 snaplen = 0;
2063                 }
2064         } else if (unlikely(macoff + snaplen >
2065                             GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2066                 u32 nval;
2067
2068                 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2069                 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2070                             snaplen, nval, macoff);
2071                 snaplen = nval;
2072                 if (unlikely((int)snaplen < 0)) {
2073                         snaplen = 0;
2074                         macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2075                 }
2076         }
2077         spin_lock(&sk->sk_receive_queue.lock);
2078         h.raw = packet_current_rx_frame(po, skb,
2079                                         TP_STATUS_KERNEL, (macoff+snaplen));
2080         if (!h.raw)
2081                 goto ring_is_full;
2082         if (po->tp_version <= TPACKET_V2) {
2083                 packet_increment_rx_head(po, &po->rx_ring);
2084         /*
2085          * LOSING will be reported till you read the stats,
2086          * because it's COR - Clear On Read.
2087          * Anyways, moving it for V1/V2 only as V3 doesn't need this
2088          * at packet level.
2089          */
2090                 if (po->stats.stats1.tp_drops)
2091                         status |= TP_STATUS_LOSING;
2092         }
2093         po->stats.stats1.tp_packets++;
2094         if (copy_skb) {
2095                 status |= TP_STATUS_COPY;
2096                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2097         }
2098         spin_unlock(&sk->sk_receive_queue.lock);
2099
2100         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2101
2102         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2103                 getnstimeofday(&ts);
2104
2105         status |= ts_status;
2106
2107         switch (po->tp_version) {
2108         case TPACKET_V1:
2109                 h.h1->tp_len = skb->len;
2110                 h.h1->tp_snaplen = snaplen;
2111                 h.h1->tp_mac = macoff;
2112                 h.h1->tp_net = netoff;
2113                 h.h1->tp_sec = ts.tv_sec;
2114                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2115                 hdrlen = sizeof(*h.h1);
2116                 break;
2117         case TPACKET_V2:
2118                 h.h2->tp_len = skb->len;
2119                 h.h2->tp_snaplen = snaplen;
2120                 h.h2->tp_mac = macoff;
2121                 h.h2->tp_net = netoff;
2122                 h.h2->tp_sec = ts.tv_sec;
2123                 h.h2->tp_nsec = ts.tv_nsec;
2124                 if (skb_vlan_tag_present(skb)) {
2125                         h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2126                         h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2127                         status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2128                 } else {
2129                         h.h2->tp_vlan_tci = 0;
2130                         h.h2->tp_vlan_tpid = 0;
2131                 }
2132                 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2133                 hdrlen = sizeof(*h.h2);
2134                 break;
2135         case TPACKET_V3:
2136                 /* tp_nxt_offset,vlan are already populated above.
2137                  * So DONT clear those fields here
2138                  */
2139                 h.h3->tp_status |= status;
2140                 h.h3->tp_len = skb->len;
2141                 h.h3->tp_snaplen = snaplen;
2142                 h.h3->tp_mac = macoff;
2143                 h.h3->tp_net = netoff;
2144                 h.h3->tp_sec  = ts.tv_sec;
2145                 h.h3->tp_nsec = ts.tv_nsec;
2146                 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2147                 hdrlen = sizeof(*h.h3);
2148                 break;
2149         default:
2150                 BUG();
2151         }
2152
2153         sll = h.raw + TPACKET_ALIGN(hdrlen);
2154         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2155         sll->sll_family = AF_PACKET;
2156         sll->sll_hatype = dev->type;
2157         sll->sll_protocol = skb->protocol;
2158         sll->sll_pkttype = skb->pkt_type;
2159         if (unlikely(po->origdev))
2160                 sll->sll_ifindex = orig_dev->ifindex;
2161         else
2162                 sll->sll_ifindex = dev->ifindex;
2163
2164         smp_mb();
2165
2166 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2167         if (po->tp_version <= TPACKET_V2) {
2168                 u8 *start, *end;
2169
2170                 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2171                                         macoff + snaplen);
2172
2173                 for (start = h.raw; start < end; start += PAGE_SIZE)
2174                         flush_dcache_page(pgv_to_page(start));
2175         }
2176         smp_wmb();
2177 #endif
2178
2179         if (po->tp_version <= TPACKET_V2) {
2180                 __packet_set_status(po, h.raw, status);
2181                 sk->sk_data_ready(sk);
2182         } else {
2183                 prb_clear_blk_fill_status(&po->rx_ring);
2184         }
2185
2186 drop_n_restore:
2187         if (skb_head != skb->data && skb_shared(skb)) {
2188                 skb->data = skb_head;
2189                 skb->len = skb_len;
2190         }
2191 drop:
2192         kfree_skb(skb);
2193         return 0;
2194
2195 ring_is_full:
2196         po->stats.stats1.tp_drops++;
2197         spin_unlock(&sk->sk_receive_queue.lock);
2198
2199         sk->sk_data_ready(sk);
2200         kfree_skb(copy_skb);
2201         goto drop_n_restore;
2202 }
2203
2204 static void tpacket_destruct_skb(struct sk_buff *skb)
2205 {
2206         struct packet_sock *po = pkt_sk(skb->sk);
2207
2208         if (likely(po->tx_ring.pg_vec)) {
2209                 void *ph;
2210                 __u32 ts;
2211
2212                 ph = skb_shinfo(skb)->destructor_arg;
2213                 packet_dec_pending(&po->tx_ring);
2214
2215                 ts = __packet_set_timestamp(po, ph, skb);
2216                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2217         }
2218
2219         sock_wfree(skb);
2220 }
2221
2222 static bool ll_header_truncated(const struct net_device *dev, int len)
2223 {
2224         /* net device doesn't like empty head */
2225         if (unlikely(len <= dev->hard_header_len)) {
2226                 net_warn_ratelimited("%s: packet size is too short (%d <= %d)\n",
2227                                      current->comm, len, dev->hard_header_len);
2228                 return true;
2229         }
2230
2231         return false;
2232 }
2233
2234 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2235                 void *frame, struct net_device *dev, int size_max,
2236                 __be16 proto, unsigned char *addr, int hlen)
2237 {
2238         union tpacket_uhdr ph;
2239         int to_write, offset, len, tp_len, nr_frags, len_max;
2240         struct socket *sock = po->sk.sk_socket;
2241         struct page *page;
2242         void *data;
2243         int err;
2244
2245         ph.raw = frame;
2246
2247         skb->protocol = proto;
2248         skb->dev = dev;
2249         skb->priority = po->sk.sk_priority;
2250         skb->mark = po->sk.sk_mark;
2251         sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2252         skb_shinfo(skb)->destructor_arg = ph.raw;
2253
2254         switch (po->tp_version) {
2255         case TPACKET_V2:
2256                 tp_len = ph.h2->tp_len;
2257                 break;
2258         default:
2259                 tp_len = ph.h1->tp_len;
2260                 break;
2261         }
2262         if (unlikely(tp_len > size_max)) {
2263                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2264                 return -EMSGSIZE;
2265         }
2266
2267         skb_reserve(skb, hlen);
2268         skb_reset_network_header(skb);
2269
2270         if (!packet_use_direct_xmit(po))
2271                 skb_probe_transport_header(skb, 0);
2272         if (unlikely(po->tp_tx_has_off)) {
2273                 int off_min, off_max, off;
2274                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2275                 off_max = po->tx_ring.frame_size - tp_len;
2276                 if (sock->type == SOCK_DGRAM) {
2277                         switch (po->tp_version) {
2278                         case TPACKET_V2:
2279                                 off = ph.h2->tp_net;
2280                                 break;
2281                         default:
2282                                 off = ph.h1->tp_net;
2283                                 break;
2284                         }
2285                 } else {
2286                         switch (po->tp_version) {
2287                         case TPACKET_V2:
2288                                 off = ph.h2->tp_mac;
2289                                 break;
2290                         default:
2291                                 off = ph.h1->tp_mac;
2292                                 break;
2293                         }
2294                 }
2295                 if (unlikely((off < off_min) || (off_max < off)))
2296                         return -EINVAL;
2297                 data = ph.raw + off;
2298         } else {
2299                 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2300         }
2301         to_write = tp_len;
2302
2303         if (sock->type == SOCK_DGRAM) {
2304                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2305                                 NULL, tp_len);
2306                 if (unlikely(err < 0))
2307                         return -EINVAL;
2308         } else if (dev->hard_header_len) {
2309                 if (ll_header_truncated(dev, tp_len))
2310                         return -EINVAL;
2311
2312                 skb_push(skb, dev->hard_header_len);
2313                 err = skb_store_bits(skb, 0, data,
2314                                 dev->hard_header_len);
2315                 if (unlikely(err))
2316                         return err;
2317
2318                 data += dev->hard_header_len;
2319                 to_write -= dev->hard_header_len;
2320         }
2321
2322         offset = offset_in_page(data);
2323         len_max = PAGE_SIZE - offset;
2324         len = ((to_write > len_max) ? len_max : to_write);
2325
2326         skb->data_len = to_write;
2327         skb->len += to_write;
2328         skb->truesize += to_write;
2329         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2330
2331         while (likely(to_write)) {
2332                 nr_frags = skb_shinfo(skb)->nr_frags;
2333
2334                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2335                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2336                                MAX_SKB_FRAGS);
2337                         return -EFAULT;
2338                 }
2339
2340                 page = pgv_to_page(data);
2341                 data += len;
2342                 flush_dcache_page(page);
2343                 get_page(page);
2344                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2345                 to_write -= len;
2346                 offset = 0;
2347                 len_max = PAGE_SIZE;
2348                 len = ((to_write > len_max) ? len_max : to_write);
2349         }
2350
2351         return tp_len;
2352 }
2353
2354 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2355 {
2356         struct sk_buff *skb;
2357         struct net_device *dev;
2358         __be16 proto;
2359         int err, reserve = 0;
2360         void *ph;
2361         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2362         bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2363         int tp_len, size_max;
2364         unsigned char *addr;
2365         int len_sum = 0;
2366         int status = TP_STATUS_AVAILABLE;
2367         int hlen, tlen;
2368
2369         mutex_lock(&po->pg_vec_lock);
2370
2371         if (likely(saddr == NULL)) {
2372                 dev     = packet_cached_dev_get(po);
2373                 proto   = po->num;
2374                 addr    = NULL;
2375         } else {
2376                 err = -EINVAL;
2377                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2378                         goto out;
2379                 if (msg->msg_namelen < (saddr->sll_halen
2380                                         + offsetof(struct sockaddr_ll,
2381                                                 sll_addr)))
2382                         goto out;
2383                 proto   = saddr->sll_protocol;
2384                 addr    = saddr->sll_addr;
2385                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2386         }
2387
2388         err = -ENXIO;
2389         if (unlikely(dev == NULL))
2390                 goto out;
2391         err = -ENETDOWN;
2392         if (unlikely(!(dev->flags & IFF_UP)))
2393                 goto out_put;
2394
2395         reserve = dev->hard_header_len + VLAN_HLEN;
2396         size_max = po->tx_ring.frame_size
2397                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2398
2399         if (size_max > dev->mtu + reserve)
2400                 size_max = dev->mtu + reserve;
2401
2402         do {
2403                 ph = packet_current_frame(po, &po->tx_ring,
2404                                           TP_STATUS_SEND_REQUEST);
2405                 if (unlikely(ph == NULL)) {
2406                         if (need_wait && need_resched())
2407                                 schedule();
2408                         continue;
2409                 }
2410
2411                 status = TP_STATUS_SEND_REQUEST;
2412                 hlen = LL_RESERVED_SPACE(dev);
2413                 tlen = dev->needed_tailroom;
2414                 skb = sock_alloc_send_skb(&po->sk,
2415                                 hlen + tlen + sizeof(struct sockaddr_ll),
2416                                 !need_wait, &err);
2417
2418                 if (unlikely(skb == NULL)) {
2419                         /* we assume the socket was initially writeable ... */
2420                         if (likely(len_sum > 0))
2421                                 err = len_sum;
2422                         goto out_status;
2423                 }
2424                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2425                                           addr, hlen);
2426                 if (tp_len > dev->mtu + dev->hard_header_len) {
2427                         struct ethhdr *ehdr;
2428                         /* Earlier code assumed this would be a VLAN pkt,
2429                          * double-check this now that we have the actual
2430                          * packet in hand.
2431                          */
2432
2433                         skb_reset_mac_header(skb);
2434                         ehdr = eth_hdr(skb);
2435                         if (ehdr->h_proto != htons(ETH_P_8021Q))
2436                                 tp_len = -EMSGSIZE;
2437                 }
2438                 if (unlikely(tp_len < 0)) {
2439                         if (po->tp_loss) {
2440                                 __packet_set_status(po, ph,
2441                                                 TP_STATUS_AVAILABLE);
2442                                 packet_increment_head(&po->tx_ring);
2443                                 kfree_skb(skb);
2444                                 continue;
2445                         } else {
2446                                 status = TP_STATUS_WRONG_FORMAT;
2447                                 err = tp_len;
2448                                 goto out_status;
2449                         }
2450                 }
2451
2452                 packet_pick_tx_queue(dev, skb);
2453
2454                 skb->destructor = tpacket_destruct_skb;
2455                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2456                 packet_inc_pending(&po->tx_ring);
2457
2458                 status = TP_STATUS_SEND_REQUEST;
2459                 err = po->xmit(skb);
2460                 if (unlikely(err > 0)) {
2461                         err = net_xmit_errno(err);
2462                         if (err && __packet_get_status(po, ph) ==
2463                                    TP_STATUS_AVAILABLE) {
2464                                 /* skb was destructed already */
2465                                 skb = NULL;
2466                                 goto out_status;
2467                         }
2468                         /*
2469                          * skb was dropped but not destructed yet;
2470                          * let's treat it like congestion or err < 0
2471                          */
2472                         err = 0;
2473                 }
2474                 packet_increment_head(&po->tx_ring);
2475                 len_sum += tp_len;
2476         } while (likely((ph != NULL) ||
2477                 /* Note: packet_read_pending() might be slow if we have
2478                  * to call it as it's per_cpu variable, but in fast-path
2479                  * we already short-circuit the loop with the first
2480                  * condition, and luckily don't have to go that path
2481                  * anyway.
2482                  */
2483                  (need_wait && packet_read_pending(&po->tx_ring))));
2484
2485         err = len_sum;
2486         goto out_put;
2487
2488 out_status:
2489         __packet_set_status(po, ph, status);
2490         kfree_skb(skb);
2491 out_put:
2492         dev_put(dev);
2493 out:
2494         mutex_unlock(&po->pg_vec_lock);
2495         return err;
2496 }
2497
2498 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2499                                         size_t reserve, size_t len,
2500                                         size_t linear, int noblock,
2501                                         int *err)
2502 {
2503         struct sk_buff *skb;
2504
2505         /* Under a page?  Don't bother with paged skb. */
2506         if (prepad + len < PAGE_SIZE || !linear)
2507                 linear = len;
2508
2509         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2510                                    err, 0);
2511         if (!skb)
2512                 return NULL;
2513
2514         skb_reserve(skb, reserve);
2515         skb_put(skb, linear);
2516         skb->data_len = len - linear;
2517         skb->len += len - linear;
2518
2519         return skb;
2520 }
2521
2522 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2523 {
2524         struct sock *sk = sock->sk;
2525         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2526         struct sk_buff *skb;
2527         struct net_device *dev;
2528         __be16 proto;
2529         unsigned char *addr;
2530         int err, reserve = 0;
2531         struct virtio_net_hdr vnet_hdr = { 0 };
2532         int offset = 0;
2533         int vnet_hdr_len;
2534         struct packet_sock *po = pkt_sk(sk);
2535         unsigned short gso_type = 0;
2536         int hlen, tlen;
2537         int extra_len = 0;
2538         ssize_t n;
2539
2540         /*
2541          *      Get and verify the address.
2542          */
2543
2544         if (likely(saddr == NULL)) {
2545                 dev     = packet_cached_dev_get(po);
2546                 proto   = po->num;
2547                 addr    = NULL;
2548         } else {
2549                 err = -EINVAL;
2550                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2551                         goto out;
2552                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2553                         goto out;
2554                 proto   = saddr->sll_protocol;
2555                 addr    = saddr->sll_addr;
2556                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2557         }
2558
2559         err = -ENXIO;
2560         if (unlikely(dev == NULL))
2561                 goto out_unlock;
2562         err = -ENETDOWN;
2563         if (unlikely(!(dev->flags & IFF_UP)))
2564                 goto out_unlock;
2565
2566         if (sock->type == SOCK_RAW)
2567                 reserve = dev->hard_header_len;
2568         if (po->has_vnet_hdr) {
2569                 vnet_hdr_len = sizeof(vnet_hdr);
2570
2571                 err = -EINVAL;
2572                 if (len < vnet_hdr_len)
2573                         goto out_unlock;
2574
2575                 len -= vnet_hdr_len;
2576
2577                 err = -EFAULT;
2578                 n = copy_from_iter(&vnet_hdr, vnet_hdr_len, &msg->msg_iter);
2579                 if (n != vnet_hdr_len)
2580                         goto out_unlock;
2581
2582                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2583                     (__virtio16_to_cpu(false, vnet_hdr.csum_start) +
2584                      __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2 >
2585                       __virtio16_to_cpu(false, vnet_hdr.hdr_len)))
2586                         vnet_hdr.hdr_len = __cpu_to_virtio16(false,
2587                                  __virtio16_to_cpu(false, vnet_hdr.csum_start) +
2588                                 __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2);
2589
2590                 err = -EINVAL;
2591                 if (__virtio16_to_cpu(false, vnet_hdr.hdr_len) > len)
2592                         goto out_unlock;
2593
2594                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2595                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2596                         case VIRTIO_NET_HDR_GSO_TCPV4:
2597                                 gso_type = SKB_GSO_TCPV4;
2598                                 break;
2599                         case VIRTIO_NET_HDR_GSO_TCPV6:
2600                                 gso_type = SKB_GSO_TCPV6;
2601                                 break;
2602                         case VIRTIO_NET_HDR_GSO_UDP:
2603                                 gso_type = SKB_GSO_UDP;
2604                                 break;
2605                         default:
2606                                 goto out_unlock;
2607                         }
2608
2609                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2610                                 gso_type |= SKB_GSO_TCP_ECN;
2611
2612                         if (vnet_hdr.gso_size == 0)
2613                                 goto out_unlock;
2614
2615                 }
2616         }
2617
2618         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2619                 if (!netif_supports_nofcs(dev)) {
2620                         err = -EPROTONOSUPPORT;
2621                         goto out_unlock;
2622                 }
2623                 extra_len = 4; /* We're doing our own CRC */
2624         }
2625
2626         err = -EMSGSIZE;
2627         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2628                 goto out_unlock;
2629
2630         err = -ENOBUFS;
2631         hlen = LL_RESERVED_SPACE(dev);
2632         tlen = dev->needed_tailroom;
2633         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
2634                                __virtio16_to_cpu(false, vnet_hdr.hdr_len),
2635                                msg->msg_flags & MSG_DONTWAIT, &err);
2636         if (skb == NULL)
2637                 goto out_unlock;
2638
2639         skb_set_network_header(skb, reserve);
2640
2641         err = -EINVAL;
2642         if (sock->type == SOCK_DGRAM) {
2643                 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2644                 if (unlikely(offset < 0))
2645                         goto out_free;
2646         } else {
2647                 if (ll_header_truncated(dev, len))
2648                         goto out_free;
2649         }
2650
2651         /* Returns -EFAULT on error */
2652         err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2653         if (err)
2654                 goto out_free;
2655
2656         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2657
2658         if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2659                 /* Earlier code assumed this would be a VLAN pkt,
2660                  * double-check this now that we have the actual
2661                  * packet in hand.
2662                  */
2663                 struct ethhdr *ehdr;
2664                 skb_reset_mac_header(skb);
2665                 ehdr = eth_hdr(skb);
2666                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2667                         err = -EMSGSIZE;
2668                         goto out_free;
2669                 }
2670         }
2671
2672         skb->protocol = proto;
2673         skb->dev = dev;
2674         skb->priority = sk->sk_priority;
2675         skb->mark = sk->sk_mark;
2676
2677         packet_pick_tx_queue(dev, skb);
2678
2679         if (po->has_vnet_hdr) {
2680                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2681                         u16 s = __virtio16_to_cpu(false, vnet_hdr.csum_start);
2682                         u16 o = __virtio16_to_cpu(false, vnet_hdr.csum_offset);
2683                         if (!skb_partial_csum_set(skb, s, o)) {
2684                                 err = -EINVAL;
2685                                 goto out_free;
2686                         }
2687                 }
2688
2689                 skb_shinfo(skb)->gso_size =
2690                         __virtio16_to_cpu(false, vnet_hdr.gso_size);
2691                 skb_shinfo(skb)->gso_type = gso_type;
2692
2693                 /* Header must be checked, and gso_segs computed. */
2694                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2695                 skb_shinfo(skb)->gso_segs = 0;
2696
2697                 len += vnet_hdr_len;
2698         }
2699
2700         if (!packet_use_direct_xmit(po))
2701                 skb_probe_transport_header(skb, reserve);
2702         if (unlikely(extra_len == 4))
2703                 skb->no_fcs = 1;
2704
2705         err = po->xmit(skb);
2706         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2707                 goto out_unlock;
2708
2709         dev_put(dev);
2710
2711         return len;
2712
2713 out_free:
2714         kfree_skb(skb);
2715 out_unlock:
2716         if (dev)
2717                 dev_put(dev);
2718 out:
2719         return err;
2720 }
2721
2722 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2723 {
2724         struct sock *sk = sock->sk;
2725         struct packet_sock *po = pkt_sk(sk);
2726
2727         if (po->tx_ring.pg_vec)
2728                 return tpacket_snd(po, msg);
2729         else
2730                 return packet_snd(sock, msg, len);
2731 }
2732
2733 /*
2734  *      Close a PACKET socket. This is fairly simple. We immediately go
2735  *      to 'closed' state and remove our protocol entry in the device list.
2736  */
2737
2738 static int packet_release(struct socket *sock)
2739 {
2740         struct sock *sk = sock->sk;
2741         struct packet_sock *po;
2742         struct net *net;
2743         union tpacket_req_u req_u;
2744
2745         if (!sk)
2746                 return 0;
2747
2748         net = sock_net(sk);
2749         po = pkt_sk(sk);
2750
2751         mutex_lock(&net->packet.sklist_lock);
2752         sk_del_node_init_rcu(sk);
2753         mutex_unlock(&net->packet.sklist_lock);
2754
2755         preempt_disable();
2756         sock_prot_inuse_add(net, sk->sk_prot, -1);
2757         preempt_enable();
2758
2759         spin_lock(&po->bind_lock);
2760         unregister_prot_hook(sk, false);
2761         packet_cached_dev_reset(po);
2762
2763         if (po->prot_hook.dev) {
2764                 dev_put(po->prot_hook.dev);
2765                 po->prot_hook.dev = NULL;
2766         }
2767         spin_unlock(&po->bind_lock);
2768
2769         packet_flush_mclist(sk);
2770
2771         if (po->rx_ring.pg_vec) {
2772                 memset(&req_u, 0, sizeof(req_u));
2773                 packet_set_ring(sk, &req_u, 1, 0);
2774         }
2775
2776         if (po->tx_ring.pg_vec) {
2777                 memset(&req_u, 0, sizeof(req_u));
2778                 packet_set_ring(sk, &req_u, 1, 1);
2779         }
2780
2781         fanout_release(sk);
2782
2783         synchronize_net();
2784         /*
2785          *      Now the socket is dead. No more input will appear.
2786          */
2787         sock_orphan(sk);
2788         sock->sk = NULL;
2789
2790         /* Purge queues */
2791
2792         skb_queue_purge(&sk->sk_receive_queue);
2793         packet_free_pending(po);
2794         sk_refcnt_debug_release(sk);
2795
2796         sock_put(sk);
2797         return 0;
2798 }
2799
2800 /*
2801  *      Attach a packet hook.
2802  */
2803
2804 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 proto)
2805 {
2806         struct packet_sock *po = pkt_sk(sk);
2807         const struct net_device *dev_curr;
2808         __be16 proto_curr;
2809         bool need_rehook;
2810
2811         if (po->fanout) {
2812                 if (dev)
2813                         dev_put(dev);
2814
2815                 return -EINVAL;
2816         }
2817
2818         lock_sock(sk);
2819         spin_lock(&po->bind_lock);
2820
2821         proto_curr = po->prot_hook.type;
2822         dev_curr = po->prot_hook.dev;
2823
2824         need_rehook = proto_curr != proto || dev_curr != dev;
2825
2826         if (need_rehook) {
2827                 unregister_prot_hook(sk, true);
2828
2829                 po->num = proto;
2830                 po->prot_hook.type = proto;
2831
2832                 if (po->prot_hook.dev)
2833                         dev_put(po->prot_hook.dev);
2834
2835                 po->prot_hook.dev = dev;
2836
2837                 po->ifindex = dev ? dev->ifindex : 0;
2838                 packet_cached_dev_assign(po, dev);
2839         }
2840
2841         if (proto == 0 || !need_rehook)
2842                 goto out_unlock;
2843
2844         if (!dev || (dev->flags & IFF_UP)) {
2845                 register_prot_hook(sk);
2846         } else {
2847                 sk->sk_err = ENETDOWN;
2848                 if (!sock_flag(sk, SOCK_DEAD))
2849                         sk->sk_error_report(sk);
2850         }
2851
2852 out_unlock:
2853         spin_unlock(&po->bind_lock);
2854         release_sock(sk);
2855         return 0;
2856 }
2857
2858 /*
2859  *      Bind a packet socket to a device
2860  */
2861
2862 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2863                             int addr_len)
2864 {
2865         struct sock *sk = sock->sk;
2866         char name[15];
2867         struct net_device *dev;
2868         int err = -ENODEV;
2869
2870         /*
2871          *      Check legality
2872          */
2873
2874         if (addr_len != sizeof(struct sockaddr))
2875                 return -EINVAL;
2876         strlcpy(name, uaddr->sa_data, sizeof(name));
2877
2878         dev = dev_get_by_name(sock_net(sk), name);
2879         if (dev)
2880                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2881         return err;
2882 }
2883
2884 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2885 {
2886         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2887         struct sock *sk = sock->sk;
2888         struct net_device *dev = NULL;
2889         int err;
2890
2891
2892         /*
2893          *      Check legality
2894          */
2895
2896         if (addr_len < sizeof(struct sockaddr_ll))
2897                 return -EINVAL;
2898         if (sll->sll_family != AF_PACKET)
2899                 return -EINVAL;
2900
2901         if (sll->sll_ifindex) {
2902                 err = -ENODEV;
2903                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2904                 if (dev == NULL)
2905                         goto out;
2906         }
2907         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2908
2909 out:
2910         return err;
2911 }
2912
2913 static struct proto packet_proto = {
2914         .name     = "PACKET",
2915         .owner    = THIS_MODULE,
2916         .obj_size = sizeof(struct packet_sock),
2917 };
2918
2919 /*
2920  *      Create a packet of type SOCK_PACKET.
2921  */
2922
2923 static int packet_create(struct net *net, struct socket *sock, int protocol,
2924                          int kern)
2925 {
2926         struct sock *sk;
2927         struct packet_sock *po;
2928         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2929         int err;
2930
2931         if (!ns_capable(net->user_ns, CAP_NET_RAW))
2932                 return -EPERM;
2933         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2934             sock->type != SOCK_PACKET)
2935                 return -ESOCKTNOSUPPORT;
2936
2937         sock->state = SS_UNCONNECTED;
2938
2939         err = -ENOBUFS;
2940         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
2941         if (sk == NULL)
2942                 goto out;
2943
2944         sock->ops = &packet_ops;
2945         if (sock->type == SOCK_PACKET)
2946                 sock->ops = &packet_ops_spkt;
2947
2948         sock_init_data(sock, sk);
2949
2950         po = pkt_sk(sk);
2951         sk->sk_family = PF_PACKET;
2952         po->num = proto;
2953         po->xmit = dev_queue_xmit;
2954
2955         err = packet_alloc_pending(po);
2956         if (err)
2957                 goto out2;
2958
2959         packet_cached_dev_reset(po);
2960
2961         sk->sk_destruct = packet_sock_destruct;
2962         sk_refcnt_debug_inc(sk);
2963
2964         /*
2965          *      Attach a protocol block
2966          */
2967
2968         spin_lock_init(&po->bind_lock);
2969         mutex_init(&po->pg_vec_lock);
2970         po->rollover = NULL;
2971         po->prot_hook.func = packet_rcv;
2972
2973         if (sock->type == SOCK_PACKET)
2974                 po->prot_hook.func = packet_rcv_spkt;
2975
2976         po->prot_hook.af_packet_priv = sk;
2977
2978         if (proto) {
2979                 po->prot_hook.type = proto;
2980                 register_prot_hook(sk);
2981         }
2982
2983         mutex_lock(&net->packet.sklist_lock);
2984         sk_add_node_rcu(sk, &net->packet.sklist);
2985         mutex_unlock(&net->packet.sklist_lock);
2986
2987         preempt_disable();
2988         sock_prot_inuse_add(net, &packet_proto, 1);
2989         preempt_enable();
2990
2991         return 0;
2992 out2:
2993         sk_free(sk);
2994 out:
2995         return err;
2996 }
2997
2998 /*
2999  *      Pull a packet from our receive queue and hand it to the user.
3000  *      If necessary we block.
3001  */
3002
3003 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3004                           int flags)
3005 {
3006         struct sock *sk = sock->sk;
3007         struct sk_buff *skb;
3008         int copied, err;
3009         int vnet_hdr_len = 0;
3010         unsigned int origlen = 0;
3011
3012         err = -EINVAL;
3013         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3014                 goto out;
3015
3016 #if 0
3017         /* What error should we return now? EUNATTACH? */
3018         if (pkt_sk(sk)->ifindex < 0)
3019                 return -ENODEV;
3020 #endif
3021
3022         if (flags & MSG_ERRQUEUE) {
3023                 err = sock_recv_errqueue(sk, msg, len,
3024                                          SOL_PACKET, PACKET_TX_TIMESTAMP);
3025                 goto out;
3026         }
3027
3028         /*
3029          *      Call the generic datagram receiver. This handles all sorts
3030          *      of horrible races and re-entrancy so we can forget about it
3031          *      in the protocol layers.
3032          *
3033          *      Now it will return ENETDOWN, if device have just gone down,
3034          *      but then it will block.
3035          */
3036
3037         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3038
3039         /*
3040          *      An error occurred so return it. Because skb_recv_datagram()
3041          *      handles the blocking we don't see and worry about blocking
3042          *      retries.
3043          */
3044
3045         if (skb == NULL)
3046                 goto out;
3047
3048         if (pkt_sk(sk)->pressure)
3049                 packet_rcv_has_room(pkt_sk(sk), NULL);
3050
3051         if (pkt_sk(sk)->has_vnet_hdr) {
3052                 struct virtio_net_hdr vnet_hdr = { 0 };
3053
3054                 err = -EINVAL;
3055                 vnet_hdr_len = sizeof(vnet_hdr);
3056                 if (len < vnet_hdr_len)
3057                         goto out_free;
3058
3059                 len -= vnet_hdr_len;
3060
3061                 if (skb_is_gso(skb)) {
3062                         struct skb_shared_info *sinfo = skb_shinfo(skb);
3063
3064                         /* This is a hint as to how much should be linear. */
3065                         vnet_hdr.hdr_len =
3066                                 __cpu_to_virtio16(false, skb_headlen(skb));
3067                         vnet_hdr.gso_size =
3068                                 __cpu_to_virtio16(false, sinfo->gso_size);
3069                         if (sinfo->gso_type & SKB_GSO_TCPV4)
3070                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
3071                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
3072                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
3073                         else if (sinfo->gso_type & SKB_GSO_UDP)
3074                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
3075                         else if (sinfo->gso_type & SKB_GSO_FCOE)
3076                                 goto out_free;
3077                         else
3078                                 BUG();
3079                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
3080                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
3081                 } else
3082                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
3083
3084                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3085                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
3086                         vnet_hdr.csum_start = __cpu_to_virtio16(false,
3087                                           skb_checksum_start_offset(skb));
3088                         vnet_hdr.csum_offset = __cpu_to_virtio16(false,
3089                                                          skb->csum_offset);
3090                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3091                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
3092                 } /* else everything is zero */
3093
3094                 err = memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_len);
3095                 if (err < 0)
3096                         goto out_free;
3097         }
3098
3099         /* You lose any data beyond the buffer you gave. If it worries
3100          * a user program they can ask the device for its MTU
3101          * anyway.
3102          */
3103         copied = skb->len;
3104         if (copied > len) {
3105                 copied = len;
3106                 msg->msg_flags |= MSG_TRUNC;
3107         }
3108
3109         err = skb_copy_datagram_msg(skb, 0, msg, copied);
3110         if (err)
3111                 goto out_free;
3112
3113         if (sock->type != SOCK_PACKET) {
3114                 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3115
3116                 /* Original length was stored in sockaddr_ll fields */
3117                 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3118                 sll->sll_family = AF_PACKET;
3119                 sll->sll_protocol = skb->protocol;
3120         }
3121
3122         sock_recv_ts_and_drops(msg, sk, skb);
3123
3124         if (msg->msg_name) {
3125                 /* If the address length field is there to be filled
3126                  * in, we fill it in now.
3127                  */
3128                 if (sock->type == SOCK_PACKET) {
3129                         __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3130                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
3131                 } else {
3132                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3133
3134                         msg->msg_namelen = sll->sll_halen +
3135                                 offsetof(struct sockaddr_ll, sll_addr);
3136                 }
3137                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3138                        msg->msg_namelen);
3139         }
3140
3141         if (pkt_sk(sk)->auxdata) {
3142                 struct tpacket_auxdata aux;
3143
3144                 aux.tp_status = TP_STATUS_USER;
3145                 if (skb->ip_summed == CHECKSUM_PARTIAL)
3146                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3147                 else if (skb->pkt_type != PACKET_OUTGOING &&
3148                          (skb->ip_summed == CHECKSUM_COMPLETE ||
3149                           skb_csum_unnecessary(skb)))
3150                         aux.tp_status |= TP_STATUS_CSUM_VALID;
3151
3152                 aux.tp_len = origlen;
3153                 aux.tp_snaplen = skb->len;
3154                 aux.tp_mac = 0;
3155                 aux.tp_net = skb_network_offset(skb);
3156                 if (skb_vlan_tag_present(skb)) {
3157                         aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3158                         aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3159                         aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3160                 } else {
3161                         aux.tp_vlan_tci = 0;
3162                         aux.tp_vlan_tpid = 0;
3163                 }
3164                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3165         }
3166
3167         /*
3168          *      Free or return the buffer as appropriate. Again this
3169          *      hides all the races and re-entrancy issues from us.
3170          */
3171         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3172
3173 out_free:
3174         skb_free_datagram(sk, skb);
3175 out:
3176         return err;
3177 }
3178
3179 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3180                                int *uaddr_len, int peer)
3181 {
3182         struct net_device *dev;
3183         struct sock *sk = sock->sk;
3184
3185         if (peer)
3186                 return -EOPNOTSUPP;
3187
3188         uaddr->sa_family = AF_PACKET;
3189         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3190         rcu_read_lock();
3191         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3192         if (dev)
3193                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3194         rcu_read_unlock();
3195         *uaddr_len = sizeof(*uaddr);
3196
3197         return 0;
3198 }
3199
3200 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3201                           int *uaddr_len, int peer)
3202 {
3203         struct net_device *dev;
3204         struct sock *sk = sock->sk;
3205         struct packet_sock *po = pkt_sk(sk);
3206         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3207
3208         if (peer)
3209                 return -EOPNOTSUPP;
3210
3211         sll->sll_family = AF_PACKET;
3212         sll->sll_ifindex = po->ifindex;
3213         sll->sll_protocol = po->num;
3214         sll->sll_pkttype = 0;
3215         rcu_read_lock();
3216         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3217         if (dev) {
3218                 sll->sll_hatype = dev->type;
3219                 sll->sll_halen = dev->addr_len;
3220                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3221         } else {
3222                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
3223                 sll->sll_halen = 0;
3224         }
3225         rcu_read_unlock();
3226         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3227
3228         return 0;
3229 }
3230
3231 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3232                          int what)
3233 {
3234         switch (i->type) {
3235         case PACKET_MR_MULTICAST:
3236                 if (i->alen != dev->addr_len)
3237                         return -EINVAL;
3238                 if (what > 0)
3239                         return dev_mc_add(dev, i->addr);
3240                 else
3241                         return dev_mc_del(dev, i->addr);
3242                 break;
3243         case PACKET_MR_PROMISC:
3244                 return dev_set_promiscuity(dev, what);
3245         case PACKET_MR_ALLMULTI:
3246                 return dev_set_allmulti(dev, what);
3247         case PACKET_MR_UNICAST:
3248                 if (i->alen != dev->addr_len)
3249                         return -EINVAL;
3250                 if (what > 0)
3251                         return dev_uc_add(dev, i->addr);
3252                 else
3253                         return dev_uc_del(dev, i->addr);
3254                 break;
3255         default:
3256                 break;
3257         }
3258         return 0;
3259 }
3260
3261 static void packet_dev_mclist_delete(struct net_device *dev,
3262                                      struct packet_mclist **mlp)
3263 {
3264         struct packet_mclist *ml;
3265
3266         while ((ml = *mlp) != NULL) {
3267                 if (ml->ifindex == dev->ifindex) {
3268                         packet_dev_mc(dev, ml, -1);
3269                         *mlp = ml->next;
3270                         kfree(ml);
3271                 } else
3272                         mlp = &ml->next;
3273         }
3274 }
3275
3276 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3277 {
3278         struct packet_sock *po = pkt_sk(sk);
3279         struct packet_mclist *ml, *i;
3280         struct net_device *dev;
3281         int err;
3282
3283         rtnl_lock();
3284
3285         err = -ENODEV;
3286         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3287         if (!dev)
3288                 goto done;
3289
3290         err = -EINVAL;
3291         if (mreq->mr_alen > dev->addr_len)
3292                 goto done;
3293
3294         err = -ENOBUFS;
3295         i = kmalloc(sizeof(*i), GFP_KERNEL);
3296         if (i == NULL)
3297                 goto done;
3298
3299         err = 0;
3300         for (ml = po->mclist; ml; ml = ml->next) {
3301                 if (ml->ifindex == mreq->mr_ifindex &&
3302                     ml->type == mreq->mr_type &&
3303                     ml->alen == mreq->mr_alen &&
3304                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3305                         ml->count++;
3306                         /* Free the new element ... */
3307                         kfree(i);
3308                         goto done;
3309                 }
3310         }
3311
3312         i->type = mreq->mr_type;
3313         i->ifindex = mreq->mr_ifindex;
3314         i->alen = mreq->mr_alen;
3315         memcpy(i->addr, mreq->mr_address, i->alen);
3316         i->count = 1;
3317         i->next = po->mclist;
3318         po->mclist = i;
3319         err = packet_dev_mc(dev, i, 1);
3320         if (err) {
3321                 po->mclist = i->next;
3322                 kfree(i);
3323         }
3324
3325 done:
3326         rtnl_unlock();
3327         return err;
3328 }
3329
3330 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3331 {
3332         struct packet_mclist *ml, **mlp;
3333
3334         rtnl_lock();
3335
3336         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3337                 if (ml->ifindex == mreq->mr_ifindex &&
3338                     ml->type == mreq->mr_type &&
3339                     ml->alen == mreq->mr_alen &&
3340                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3341                         if (--ml->count == 0) {
3342                                 struct net_device *dev;
3343                                 *mlp = ml->next;
3344                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3345                                 if (dev)
3346                                         packet_dev_mc(dev, ml, -1);
3347                                 kfree(ml);
3348                         }
3349                         break;
3350                 }
3351         }
3352         rtnl_unlock();
3353         return 0;
3354 }
3355
3356 static void packet_flush_mclist(struct sock *sk)
3357 {
3358         struct packet_sock *po = pkt_sk(sk);
3359         struct packet_mclist *ml;
3360
3361         if (!po->mclist)
3362                 return;
3363
3364         rtnl_lock();
3365         while ((ml = po->mclist) != NULL) {
3366                 struct net_device *dev;
3367
3368                 po->mclist = ml->next;
3369                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3370                 if (dev != NULL)
3371                         packet_dev_mc(dev, ml, -1);
3372                 kfree(ml);
3373         }
3374         rtnl_unlock();
3375 }
3376
3377 static int
3378 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3379 {
3380         struct sock *sk = sock->sk;
3381         struct packet_sock *po = pkt_sk(sk);
3382         int ret;
3383
3384         if (level != SOL_PACKET)
3385                 return -ENOPROTOOPT;
3386
3387         switch (optname) {
3388         case PACKET_ADD_MEMBERSHIP:
3389         case PACKET_DROP_MEMBERSHIP:
3390         {
3391                 struct packet_mreq_max mreq;
3392                 int len = optlen;
3393                 memset(&mreq, 0, sizeof(mreq));
3394                 if (len < sizeof(struct packet_mreq))
3395                         return -EINVAL;
3396                 if (len > sizeof(mreq))
3397                         len = sizeof(mreq);
3398                 if (copy_from_user(&mreq, optval, len))
3399                         return -EFAULT;
3400                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3401                         return -EINVAL;
3402                 if (optname == PACKET_ADD_MEMBERSHIP)
3403                         ret = packet_mc_add(sk, &mreq);
3404                 else
3405                         ret = packet_mc_drop(sk, &mreq);
3406                 return ret;
3407         }
3408
3409         case PACKET_RX_RING:
3410         case PACKET_TX_RING:
3411         {
3412                 union tpacket_req_u req_u;
3413                 int len;
3414
3415                 switch (po->tp_version) {
3416                 case TPACKET_V1:
3417                 case TPACKET_V2:
3418                         len = sizeof(req_u.req);
3419                         break;
3420                 case TPACKET_V3:
3421                 default:
3422                         len = sizeof(req_u.req3);
3423                         break;
3424                 }
3425                 if (optlen < len)
3426                         return -EINVAL;
3427                 if (pkt_sk(sk)->has_vnet_hdr)
3428                         return -EINVAL;
3429                 if (copy_from_user(&req_u.req, optval, len))
3430                         return -EFAULT;
3431                 return packet_set_ring(sk, &req_u, 0,
3432                         optname == PACKET_TX_RING);
3433         }
3434         case PACKET_COPY_THRESH:
3435         {
3436                 int val;
3437
3438                 if (optlen != sizeof(val))
3439                         return -EINVAL;
3440                 if (copy_from_user(&val, optval, sizeof(val)))
3441                         return -EFAULT;
3442
3443                 pkt_sk(sk)->copy_thresh = val;
3444                 return 0;
3445         }
3446         case PACKET_VERSION:
3447         {
3448                 int val;
3449
3450                 if (optlen != sizeof(val))
3451                         return -EINVAL;
3452                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3453                         return -EBUSY;
3454                 if (copy_from_user(&val, optval, sizeof(val)))
3455                         return -EFAULT;
3456                 switch (val) {
3457                 case TPACKET_V1:
3458                 case TPACKET_V2:
3459                 case TPACKET_V3:
3460                         po->tp_version = val;
3461                         return 0;
3462                 default:
3463                         return -EINVAL;
3464                 }
3465         }
3466         case PACKET_RESERVE:
3467         {
3468                 unsigned int val;
3469
3470                 if (optlen != sizeof(val))
3471                         return -EINVAL;
3472                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3473                         return -EBUSY;
3474                 if (copy_from_user(&val, optval, sizeof(val)))
3475                         return -EFAULT;
3476                 po->tp_reserve = val;
3477                 return 0;
3478         }
3479         case PACKET_LOSS:
3480         {
3481                 unsigned int val;
3482
3483                 if (optlen != sizeof(val))
3484                         return -EINVAL;
3485                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3486                         return -EBUSY;
3487                 if (copy_from_user(&val, optval, sizeof(val)))
3488                         return -EFAULT;
3489                 po->tp_loss = !!val;
3490                 return 0;
3491         }
3492         case PACKET_AUXDATA:
3493         {
3494                 int val;
3495
3496                 if (optlen < sizeof(val))
3497                         return -EINVAL;
3498                 if (copy_from_user(&val, optval, sizeof(val)))
3499                         return -EFAULT;
3500
3501                 po->auxdata = !!val;
3502                 return 0;
3503         }
3504         case PACKET_ORIGDEV:
3505         {
3506                 int val;
3507
3508                 if (optlen < sizeof(val))
3509                         return -EINVAL;
3510                 if (copy_from_user(&val, optval, sizeof(val)))
3511                         return -EFAULT;
3512
3513                 po->origdev = !!val;
3514                 return 0;
3515         }
3516         case PACKET_VNET_HDR:
3517         {
3518                 int val;
3519
3520                 if (sock->type != SOCK_RAW)
3521                         return -EINVAL;
3522                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3523                         return -EBUSY;
3524                 if (optlen < sizeof(val))
3525                         return -EINVAL;
3526                 if (copy_from_user(&val, optval, sizeof(val)))
3527                         return -EFAULT;
3528
3529                 po->has_vnet_hdr = !!val;
3530                 return 0;
3531         }
3532         case PACKET_TIMESTAMP:
3533         {
3534                 int val;
3535
3536                 if (optlen != sizeof(val))
3537                         return -EINVAL;
3538                 if (copy_from_user(&val, optval, sizeof(val)))
3539                         return -EFAULT;
3540
3541                 po->tp_tstamp = val;
3542                 return 0;
3543         }
3544         case PACKET_FANOUT:
3545         {
3546                 int val;
3547
3548                 if (optlen != sizeof(val))
3549                         return -EINVAL;
3550                 if (copy_from_user(&val, optval, sizeof(val)))
3551                         return -EFAULT;
3552
3553                 return fanout_add(sk, val & 0xffff, val >> 16);
3554         }
3555         case PACKET_TX_HAS_OFF:
3556         {
3557                 unsigned int val;
3558
3559                 if (optlen != sizeof(val))
3560                         return -EINVAL;
3561                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3562                         return -EBUSY;
3563                 if (copy_from_user(&val, optval, sizeof(val)))
3564                         return -EFAULT;
3565                 po->tp_tx_has_off = !!val;
3566                 return 0;
3567         }
3568         case PACKET_QDISC_BYPASS:
3569         {
3570                 int val;
3571
3572                 if (optlen != sizeof(val))
3573                         return -EINVAL;
3574                 if (copy_from_user(&val, optval, sizeof(val)))
3575                         return -EFAULT;
3576
3577                 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3578                 return 0;
3579         }
3580         default:
3581                 return -ENOPROTOOPT;
3582         }
3583 }
3584
3585 static int packet_getsockopt(struct socket *sock, int level, int optname,
3586                              char __user *optval, int __user *optlen)
3587 {
3588         int len;
3589         int val, lv = sizeof(val);
3590         struct sock *sk = sock->sk;
3591         struct packet_sock *po = pkt_sk(sk);
3592         void *data = &val;
3593         union tpacket_stats_u st;
3594         struct tpacket_rollover_stats rstats;
3595
3596         if (level != SOL_PACKET)
3597                 return -ENOPROTOOPT;
3598
3599         if (get_user(len, optlen))
3600                 return -EFAULT;
3601
3602         if (len < 0)
3603                 return -EINVAL;
3604
3605         switch (optname) {
3606         case PACKET_STATISTICS:
3607                 spin_lock_bh(&sk->sk_receive_queue.lock);
3608                 memcpy(&st, &po->stats, sizeof(st));
3609                 memset(&po->stats, 0, sizeof(po->stats));
3610                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3611
3612                 if (po->tp_version == TPACKET_V3) {
3613                         lv = sizeof(struct tpacket_stats_v3);
3614                         st.stats3.tp_packets += st.stats3.tp_drops;
3615                         data = &st.stats3;
3616                 } else {
3617                         lv = sizeof(struct tpacket_stats);
3618                         st.stats1.tp_packets += st.stats1.tp_drops;
3619                         data = &st.stats1;
3620                 }
3621
3622                 break;
3623         case PACKET_AUXDATA:
3624                 val = po->auxdata;
3625                 break;
3626         case PACKET_ORIGDEV:
3627                 val = po->origdev;
3628                 break;
3629         case PACKET_VNET_HDR:
3630                 val = po->has_vnet_hdr;
3631                 break;
3632         case PACKET_VERSION:
3633                 val = po->tp_version;
3634                 break;
3635         case PACKET_HDRLEN:
3636                 if (len > sizeof(int))
3637                         len = sizeof(int);
3638                 if (copy_from_user(&val, optval, len))
3639                         return -EFAULT;
3640                 switch (val) {
3641                 case TPACKET_V1:
3642                         val = sizeof(struct tpacket_hdr);
3643                         break;
3644                 case TPACKET_V2:
3645                         val = sizeof(struct tpacket2_hdr);
3646                         break;
3647                 case TPACKET_V3:
3648                         val = sizeof(struct tpacket3_hdr);
3649                         break;
3650                 default:
3651                         return -EINVAL;
3652                 }
3653                 break;
3654         case PACKET_RESERVE:
3655                 val = po->tp_reserve;
3656                 break;
3657         case PACKET_LOSS:
3658                 val = po->tp_loss;
3659                 break;
3660         case PACKET_TIMESTAMP:
3661                 val = po->tp_tstamp;
3662                 break;
3663         case PACKET_FANOUT:
3664                 val = (po->fanout ?
3665                        ((u32)po->fanout->id |
3666                         ((u32)po->fanout->type << 16) |
3667                         ((u32)po->fanout->flags << 24)) :
3668                        0);
3669                 break;
3670         case PACKET_ROLLOVER_STATS:
3671                 if (!po->rollover)
3672                         return -EINVAL;
3673                 rstats.tp_all = atomic_long_read(&po->rollover->num);
3674                 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3675                 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3676                 data = &rstats;
3677                 lv = sizeof(rstats);
3678                 break;
3679         case PACKET_TX_HAS_OFF:
3680                 val = po->tp_tx_has_off;
3681                 break;
3682         case PACKET_QDISC_BYPASS:
3683                 val = packet_use_direct_xmit(po);
3684                 break;
3685         default:
3686                 return -ENOPROTOOPT;
3687         }
3688
3689         if (len > lv)
3690                 len = lv;
3691         if (put_user(len, optlen))
3692                 return -EFAULT;
3693         if (copy_to_user(optval, data, len))
3694                 return -EFAULT;
3695         return 0;
3696 }
3697
3698
3699 static int packet_notifier(struct notifier_block *this,
3700                            unsigned long msg, void *ptr)
3701 {
3702         struct sock *sk;
3703         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3704         struct net *net = dev_net(dev);
3705
3706         rcu_read_lock();
3707         sk_for_each_rcu(sk, &net->packet.sklist) {
3708                 struct packet_sock *po = pkt_sk(sk);
3709
3710                 switch (msg) {
3711                 case NETDEV_UNREGISTER:
3712                         if (po->mclist)
3713                                 packet_dev_mclist_delete(dev, &po->mclist);
3714                         /* fallthrough */
3715
3716                 case NETDEV_DOWN:
3717                         if (dev->ifindex == po->ifindex) {
3718                                 spin_lock(&po->bind_lock);
3719                                 if (po->running) {
3720                                         __unregister_prot_hook(sk, false);
3721                                         sk->sk_err = ENETDOWN;
3722                                         if (!sock_flag(sk, SOCK_DEAD))
3723                                                 sk->sk_error_report(sk);
3724                                 }
3725                                 if (msg == NETDEV_UNREGISTER) {
3726                                         packet_cached_dev_reset(po);
3727                                         po->ifindex = -1;
3728                                         if (po->prot_hook.dev)
3729                                                 dev_put(po->prot_hook.dev);
3730                                         po->prot_hook.dev = NULL;
3731                                 }
3732                                 spin_unlock(&po->bind_lock);
3733                         }
3734                         break;
3735                 case NETDEV_UP:
3736                         if (dev->ifindex == po->ifindex) {
3737                                 spin_lock(&po->bind_lock);
3738                                 if (po->num)
3739                                         register_prot_hook(sk);
3740                                 spin_unlock(&po->bind_lock);
3741                         }
3742                         break;
3743                 }
3744         }
3745         rcu_read_unlock();
3746         return NOTIFY_DONE;
3747 }
3748
3749
3750 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3751                         unsigned long arg)
3752 {
3753         struct sock *sk = sock->sk;
3754
3755         switch (cmd) {
3756         case SIOCOUTQ:
3757         {
3758                 int amount = sk_wmem_alloc_get(sk);
3759
3760                 return put_user(amount, (int __user *)arg);
3761         }
3762         case SIOCINQ:
3763         {
3764                 struct sk_buff *skb;
3765                 int amount = 0;
3766
3767                 spin_lock_bh(&sk->sk_receive_queue.lock);
3768                 skb = skb_peek(&sk->sk_receive_queue);
3769                 if (skb)
3770                         amount = skb->len;
3771                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3772                 return put_user(amount, (int __user *)arg);
3773         }
3774         case SIOCGSTAMP:
3775                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3776         case SIOCGSTAMPNS:
3777                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3778
3779 #ifdef CONFIG_INET
3780         case SIOCADDRT:
3781         case SIOCDELRT:
3782         case SIOCDARP:
3783         case SIOCGARP:
3784         case SIOCSARP:
3785         case SIOCGIFADDR:
3786         case SIOCSIFADDR:
3787         case SIOCGIFBRDADDR:
3788         case SIOCSIFBRDADDR:
3789         case SIOCGIFNETMASK:
3790         case SIOCSIFNETMASK:
3791         case SIOCGIFDSTADDR:
3792         case SIOCSIFDSTADDR:
3793         case SIOCSIFFLAGS:
3794                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3795 #endif
3796
3797         default:
3798                 return -ENOIOCTLCMD;
3799         }
3800         return 0;
3801 }
3802
3803 static unsigned int packet_poll(struct file *file, struct socket *sock,
3804                                 poll_table *wait)
3805 {
3806         struct sock *sk = sock->sk;
3807         struct packet_sock *po = pkt_sk(sk);
3808         unsigned int mask = datagram_poll(file, sock, wait);
3809
3810         spin_lock_bh(&sk->sk_receive_queue.lock);
3811         if (po->rx_ring.pg_vec) {
3812                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3813                         TP_STATUS_KERNEL))
3814                         mask |= POLLIN | POLLRDNORM;
3815         }
3816         if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
3817                 xchg(&po->pressure, 0);
3818         spin_unlock_bh(&sk->sk_receive_queue.lock);
3819         spin_lock_bh(&sk->sk_write_queue.lock);
3820         if (po->tx_ring.pg_vec) {
3821                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3822                         mask |= POLLOUT | POLLWRNORM;
3823         }
3824         spin_unlock_bh(&sk->sk_write_queue.lock);
3825         return mask;
3826 }
3827
3828
3829 /* Dirty? Well, I still did not learn better way to account
3830  * for user mmaps.
3831  */
3832
3833 static void packet_mm_open(struct vm_area_struct *vma)
3834 {
3835         struct file *file = vma->vm_file;
3836         struct socket *sock = file->private_data;
3837         struct sock *sk = sock->sk;
3838
3839         if (sk)
3840                 atomic_inc(&pkt_sk(sk)->mapped);
3841 }
3842
3843 static void packet_mm_close(struct vm_area_struct *vma)
3844 {
3845         struct file *file = vma->vm_file;
3846         struct socket *sock = file->private_data;
3847         struct sock *sk = sock->sk;
3848
3849         if (sk)
3850                 atomic_dec(&pkt_sk(sk)->mapped);
3851 }
3852
3853 static const struct vm_operations_struct packet_mmap_ops = {
3854         .open   =       packet_mm_open,
3855         .close  =       packet_mm_close,
3856 };
3857
3858 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3859                         unsigned int len)
3860 {
3861         int i;
3862
3863         for (i = 0; i < len; i++) {
3864                 if (likely(pg_vec[i].buffer)) {
3865                         if (is_vmalloc_addr(pg_vec[i].buffer))
3866                                 vfree(pg_vec[i].buffer);
3867                         else
3868                                 free_pages((unsigned long)pg_vec[i].buffer,
3869                                            order);
3870                         pg_vec[i].buffer = NULL;
3871                 }
3872         }
3873         kfree(pg_vec);
3874 }
3875
3876 static char *alloc_one_pg_vec_page(unsigned long order)
3877 {
3878         char *buffer;
3879         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3880                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3881
3882         buffer = (char *) __get_free_pages(gfp_flags, order);
3883         if (buffer)
3884                 return buffer;
3885
3886         /* __get_free_pages failed, fall back to vmalloc */
3887         buffer = vzalloc((1 << order) * PAGE_SIZE);
3888         if (buffer)
3889                 return buffer;
3890
3891         /* vmalloc failed, lets dig into swap here */
3892         gfp_flags &= ~__GFP_NORETRY;
3893         buffer = (char *) __get_free_pages(gfp_flags, order);
3894         if (buffer)
3895                 return buffer;
3896
3897         /* complete and utter failure */
3898         return NULL;
3899 }
3900
3901 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3902 {
3903         unsigned int block_nr = req->tp_block_nr;
3904         struct pgv *pg_vec;
3905         int i;
3906
3907         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3908         if (unlikely(!pg_vec))
3909                 goto out;
3910
3911         for (i = 0; i < block_nr; i++) {
3912                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3913                 if (unlikely(!pg_vec[i].buffer))
3914                         goto out_free_pgvec;
3915         }
3916
3917 out:
3918         return pg_vec;
3919
3920 out_free_pgvec:
3921         free_pg_vec(pg_vec, order, block_nr);
3922         pg_vec = NULL;
3923         goto out;
3924 }
3925
3926 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3927                 int closing, int tx_ring)
3928 {
3929         struct pgv *pg_vec = NULL;
3930         struct packet_sock *po = pkt_sk(sk);
3931         int was_running, order = 0;
3932         struct packet_ring_buffer *rb;
3933         struct sk_buff_head *rb_queue;
3934         __be16 num;
3935         int err = -EINVAL;
3936         /* Added to avoid minimal code churn */
3937         struct tpacket_req *req = &req_u->req;
3938
3939         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3940         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3941                 WARN(1, "Tx-ring is not supported.\n");
3942                 goto out;
3943         }
3944
3945         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3946         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3947
3948         err = -EBUSY;
3949         if (!closing) {
3950                 if (atomic_read(&po->mapped))
3951                         goto out;
3952                 if (packet_read_pending(rb))
3953                         goto out;
3954         }
3955
3956         if (req->tp_block_nr) {
3957                 /* Sanity tests and some calculations */
3958                 err = -EBUSY;
3959                 if (unlikely(rb->pg_vec))
3960                         goto out;
3961
3962                 switch (po->tp_version) {
3963                 case TPACKET_V1:
3964                         po->tp_hdrlen = TPACKET_HDRLEN;
3965                         break;
3966                 case TPACKET_V2:
3967                         po->tp_hdrlen = TPACKET2_HDRLEN;
3968                         break;
3969                 case TPACKET_V3:
3970                         po->tp_hdrlen = TPACKET3_HDRLEN;
3971                         break;
3972                 }
3973
3974                 err = -EINVAL;
3975                 if (unlikely((int)req->tp_block_size <= 0))
3976                         goto out;
3977                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3978                         goto out;
3979                 if (po->tp_version >= TPACKET_V3 &&
3980                     (int)(req->tp_block_size -
3981                           BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
3982                         goto out;
3983                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3984                                         po->tp_reserve))
3985                         goto out;
3986                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3987                         goto out;
3988
3989                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3990                 if (unlikely(rb->frames_per_block <= 0))
3991                         goto out;
3992                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3993                                         req->tp_frame_nr))
3994                         goto out;
3995
3996                 err = -ENOMEM;
3997                 order = get_order(req->tp_block_size);
3998                 pg_vec = alloc_pg_vec(req, order);
3999                 if (unlikely(!pg_vec))
4000                         goto out;
4001                 switch (po->tp_version) {
4002                 case TPACKET_V3:
4003                 /* Transmit path is not supported. We checked
4004                  * it above but just being paranoid
4005                  */
4006                         if (!tx_ring)
4007                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
4008                         break;
4009                 default:
4010                         break;
4011                 }
4012         }
4013         /* Done */
4014         else {
4015                 err = -EINVAL;
4016                 if (unlikely(req->tp_frame_nr))
4017                         goto out;
4018         }
4019
4020         lock_sock(sk);
4021
4022         /* Detach socket from network */
4023         spin_lock(&po->bind_lock);
4024         was_running = po->running;
4025         num = po->num;
4026         if (was_running) {
4027                 po->num = 0;
4028                 __unregister_prot_hook(sk, false);
4029         }
4030         spin_unlock(&po->bind_lock);
4031
4032         synchronize_net();
4033
4034         err = -EBUSY;
4035         mutex_lock(&po->pg_vec_lock);
4036         if (closing || atomic_read(&po->mapped) == 0) {
4037                 err = 0;
4038                 spin_lock_bh(&rb_queue->lock);
4039                 swap(rb->pg_vec, pg_vec);
4040                 rb->frame_max = (req->tp_frame_nr - 1);
4041                 rb->head = 0;
4042                 rb->frame_size = req->tp_frame_size;
4043                 spin_unlock_bh(&rb_queue->lock);
4044
4045                 swap(rb->pg_vec_order, order);
4046                 swap(rb->pg_vec_len, req->tp_block_nr);
4047
4048                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4049                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4050                                                 tpacket_rcv : packet_rcv;
4051                 skb_queue_purge(rb_queue);
4052                 if (atomic_read(&po->mapped))
4053                         pr_err("packet_mmap: vma is busy: %d\n",
4054                                atomic_read(&po->mapped));
4055         }
4056         mutex_unlock(&po->pg_vec_lock);
4057
4058         spin_lock(&po->bind_lock);
4059         if (was_running) {
4060                 po->num = num;
4061                 register_prot_hook(sk);
4062         }
4063         spin_unlock(&po->bind_lock);
4064         if (closing && (po->tp_version > TPACKET_V2)) {
4065                 /* Because we don't support block-based V3 on tx-ring */
4066                 if (!tx_ring)
4067                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
4068         }
4069         release_sock(sk);
4070
4071         if (pg_vec)
4072                 free_pg_vec(pg_vec, order, req->tp_block_nr);
4073 out:
4074         return err;
4075 }
4076
4077 static int packet_mmap(struct file *file, struct socket *sock,
4078                 struct vm_area_struct *vma)
4079 {
4080         struct sock *sk = sock->sk;
4081         struct packet_sock *po = pkt_sk(sk);
4082         unsigned long size, expected_size;
4083         struct packet_ring_buffer *rb;
4084         unsigned long start;
4085         int err = -EINVAL;
4086         int i;
4087
4088         if (vma->vm_pgoff)
4089                 return -EINVAL;
4090
4091         mutex_lock(&po->pg_vec_lock);
4092
4093         expected_size = 0;
4094         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4095                 if (rb->pg_vec) {
4096                         expected_size += rb->pg_vec_len
4097                                                 * rb->pg_vec_pages
4098                                                 * PAGE_SIZE;
4099                 }
4100         }
4101
4102         if (expected_size == 0)
4103                 goto out;
4104
4105         size = vma->vm_end - vma->vm_start;
4106         if (size != expected_size)
4107                 goto out;
4108
4109         start = vma->vm_start;
4110         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4111                 if (rb->pg_vec == NULL)
4112                         continue;
4113
4114                 for (i = 0; i < rb->pg_vec_len; i++) {
4115                         struct page *page;
4116                         void *kaddr = rb->pg_vec[i].buffer;
4117                         int pg_num;
4118
4119                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4120                                 page = pgv_to_page(kaddr);
4121                                 err = vm_insert_page(vma, start, page);
4122                                 if (unlikely(err))
4123                                         goto out;
4124                                 start += PAGE_SIZE;
4125                                 kaddr += PAGE_SIZE;
4126                         }
4127                 }
4128         }
4129
4130         atomic_inc(&po->mapped);
4131         vma->vm_ops = &packet_mmap_ops;
4132         err = 0;
4133
4134 out:
4135         mutex_unlock(&po->pg_vec_lock);
4136         return err;
4137 }
4138
4139 static const struct proto_ops packet_ops_spkt = {
4140         .family =       PF_PACKET,
4141         .owner =        THIS_MODULE,
4142         .release =      packet_release,
4143         .bind =         packet_bind_spkt,
4144         .connect =      sock_no_connect,
4145         .socketpair =   sock_no_socketpair,
4146         .accept =       sock_no_accept,
4147         .getname =      packet_getname_spkt,
4148         .poll =         datagram_poll,
4149         .ioctl =        packet_ioctl,
4150         .listen =       sock_no_listen,
4151         .shutdown =     sock_no_shutdown,
4152         .setsockopt =   sock_no_setsockopt,
4153         .getsockopt =   sock_no_getsockopt,
4154         .sendmsg =      packet_sendmsg_spkt,
4155         .recvmsg =      packet_recvmsg,
4156         .mmap =         sock_no_mmap,
4157         .sendpage =     sock_no_sendpage,
4158 };
4159
4160 static const struct proto_ops packet_ops = {
4161         .family =       PF_PACKET,
4162         .owner =        THIS_MODULE,
4163         .release =      packet_release,
4164         .bind =         packet_bind,
4165         .connect =      sock_no_connect,
4166         .socketpair =   sock_no_socketpair,
4167         .accept =       sock_no_accept,
4168         .getname =      packet_getname,
4169         .poll =         packet_poll,
4170         .ioctl =        packet_ioctl,
4171         .listen =       sock_no_listen,
4172         .shutdown =     sock_no_shutdown,
4173         .setsockopt =   packet_setsockopt,
4174         .getsockopt =   packet_getsockopt,
4175         .sendmsg =      packet_sendmsg,
4176         .recvmsg =      packet_recvmsg,
4177         .mmap =         packet_mmap,
4178         .sendpage =     sock_no_sendpage,
4179 };
4180
4181 static const struct net_proto_family packet_family_ops = {
4182         .family =       PF_PACKET,
4183         .create =       packet_create,
4184         .owner  =       THIS_MODULE,
4185 };
4186
4187 static struct notifier_block packet_netdev_notifier = {
4188         .notifier_call =        packet_notifier,
4189 };
4190
4191 #ifdef CONFIG_PROC_FS
4192
4193 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4194         __acquires(RCU)
4195 {
4196         struct net *net = seq_file_net(seq);
4197
4198         rcu_read_lock();
4199         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4200 }
4201
4202 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4203 {
4204         struct net *net = seq_file_net(seq);
4205         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4206 }
4207
4208 static void packet_seq_stop(struct seq_file *seq, void *v)
4209         __releases(RCU)
4210 {
4211         rcu_read_unlock();
4212 }
4213
4214 static int packet_seq_show(struct seq_file *seq, void *v)
4215 {
4216         if (v == SEQ_START_TOKEN)
4217                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
4218         else {
4219                 struct sock *s = sk_entry(v);
4220                 const struct packet_sock *po = pkt_sk(s);
4221
4222                 seq_printf(seq,
4223                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4224                            s,
4225                            atomic_read(&s->sk_refcnt),
4226                            s->sk_type,
4227                            ntohs(po->num),
4228                            po->ifindex,
4229                            po->running,
4230                            atomic_read(&s->sk_rmem_alloc),
4231                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4232                            sock_i_ino(s));
4233         }
4234
4235         return 0;
4236 }
4237
4238 static const struct seq_operations packet_seq_ops = {
4239         .start  = packet_seq_start,
4240         .next   = packet_seq_next,
4241         .stop   = packet_seq_stop,
4242         .show   = packet_seq_show,
4243 };
4244
4245 static int packet_seq_open(struct inode *inode, struct file *file)
4246 {
4247         return seq_open_net(inode, file, &packet_seq_ops,
4248                             sizeof(struct seq_net_private));
4249 }
4250
4251 static const struct file_operations packet_seq_fops = {
4252         .owner          = THIS_MODULE,
4253         .open           = packet_seq_open,
4254         .read           = seq_read,
4255         .llseek         = seq_lseek,
4256         .release        = seq_release_net,
4257 };
4258
4259 #endif
4260
4261 static int __net_init packet_net_init(struct net *net)
4262 {
4263         mutex_init(&net->packet.sklist_lock);
4264         INIT_HLIST_HEAD(&net->packet.sklist);
4265
4266         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4267                 return -ENOMEM;
4268
4269         return 0;
4270 }
4271
4272 static void __net_exit packet_net_exit(struct net *net)
4273 {
4274         remove_proc_entry("packet", net->proc_net);
4275 }
4276
4277 static struct pernet_operations packet_net_ops = {
4278         .init = packet_net_init,
4279         .exit = packet_net_exit,
4280 };
4281
4282
4283 static void __exit packet_exit(void)
4284 {
4285         unregister_netdevice_notifier(&packet_netdev_notifier);
4286         unregister_pernet_subsys(&packet_net_ops);
4287         sock_unregister(PF_PACKET);
4288         proto_unregister(&packet_proto);
4289 }
4290
4291 static int __init packet_init(void)
4292 {
4293         int rc = proto_register(&packet_proto, 0);
4294
4295         if (rc != 0)
4296                 goto out;
4297
4298         sock_register(&packet_family_ops);
4299         register_pernet_subsys(&packet_net_ops);
4300         register_netdevice_notifier(&packet_netdev_notifier);
4301 out:
4302         return rc;
4303 }
4304
4305 module_init(packet_init);
4306 module_exit(packet_exit);
4307 MODULE_LICENSE("GPL");
4308 MODULE_ALIAS_NETPROTO(PF_PACKET);