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