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
  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         unsigned int origlen;
 220         union {
 221                 struct sockaddr_pkt pkt;
 222                 struct sockaddr_ll ll;
 223         } sa;
 224 };
 225
 226 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 227
 228 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 229 #define GET_PBLOCK_DESC(x, bid) \
 230         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 231 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 232         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 233 #define GET_NEXT_PRB_BLK_NUM(x) \
 234         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 235         ((x)->kactive_blk_num+1) : 0)
 236
 237 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 238 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 239
 240 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
 241 {
 242         struct net_device *dev;
 243
 244         rcu_read_lock();
 245         dev = rcu_dereference(po->cached_dev);
 246         if (likely(dev))
 247                 dev_hold(dev);
 248         rcu_read_unlock();
 249
 250         return dev;
 251 }
 252
 253 static void packet_cached_dev_assign(struct packet_sock *po,
 254                                      struct net_device *dev)
 255 {
 256         rcu_assign_pointer(po->cached_dev, dev);
 257 }
 258
 259 static void packet_cached_dev_reset(struct packet_sock *po)
 260 {
 261         RCU_INIT_POINTER(po->cached_dev, NULL);
 262 }
 263
 264 /* register_prot_hook must be invoked with the po->bind_lock held,
 265  * or from a context in which asynchronous accesses to the packet
 266  * socket is not possible (packet_create()).
 267  */
 268 static void register_prot_hook(struct sock *sk)
 269 {
 270         struct packet_sock *po = pkt_sk(sk);
 271
 272         if (!po->running) {
 273                 if (po->fanout)
 274                         __fanout_link(sk, po);
 275                 else
 276                         dev_add_pack(&po->prot_hook);
 277
 278                 sock_hold(sk);
 279                 po->running = 1;
 280         }
 281 }
 282
 283 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
 284  * held.   If the sync parameter is true, we will temporarily drop
 285  * the po->bind_lock and do a synchronize_net to make sure no
 286  * asynchronous packet processing paths still refer to the elements
 287  * of po->prot_hook.  If the sync parameter is false, it is the
 288  * callers responsibility to take care of this.
 289  */
 290 static void __unregister_prot_hook(struct sock *sk, bool sync)
 291 {
 292         struct packet_sock *po = pkt_sk(sk);
 293
 294         po->running = 0;
 295
 296         if (po->fanout)
 297                 __fanout_unlink(sk, po);
 298         else
 299                 __dev_remove_pack(&po->prot_hook);
 300
 301         __sock_put(sk);
 302
 303         if (sync) {
 304                 spin_unlock(&po->bind_lock);
 305                 synchronize_net();
 306                 spin_lock(&po->bind_lock);
 307         }
 308 }
 309
 310 static void unregister_prot_hook(struct sock *sk, bool sync)
 311 {
 312         struct packet_sock *po = pkt_sk(sk);
 313
 314         if (po->running)
 315                 __unregister_prot_hook(sk, sync);
 316 }
 317
 318 static inline __pure struct page *pgv_to_page(void *addr)
 319 {
 320         if (is_vmalloc_addr(addr))
 321                 return vmalloc_to_page(addr);
 322         return virt_to_page(addr);
 323 }
 324
 325 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 326 {
 327         union tpacket_uhdr h;
 328
 329         h.raw = frame;
 330         switch (po->tp_version) {
 331         case TPACKET_V1:
 332                 h.h1->tp_status = status;
 333                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 334                 break;
 335         case TPACKET_V2:
 336                 h.h2->tp_status = status;
 337                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 338                 break;
 339         case TPACKET_V3:
 340         default:
 341                 WARN(1, "TPACKET version not supported.\n");
 342                 BUG();
 343         }
 344
 345         smp_wmb();
 346 }
 347
 348 static int __packet_get_status(struct packet_sock *po, void *frame)
 349 {
 350         union tpacket_uhdr h;
 351
 352         smp_rmb();
 353
 354         h.raw = frame;
 355         switch (po->tp_version) {
 356         case TPACKET_V1:
 357                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 358                 return h.h1->tp_status;
 359         case TPACKET_V2:
 360                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 361                 return h.h2->tp_status;
 362         case TPACKET_V3:
 363         default:
 364                 WARN(1, "TPACKET version not supported.\n");
 365                 BUG();
 366                 return 0;
 367         }
 368 }
 369
 370 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
 371                                    unsigned int flags)
 372 {
 373         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 374
 375         if (shhwtstamps) {
 376                 if ((flags & SOF_TIMESTAMPING_SYS_HARDWARE) &&
 377                     ktime_to_timespec_cond(shhwtstamps->syststamp, ts))
 378                         return TP_STATUS_TS_SYS_HARDWARE;
 379                 if ((flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
 380                     ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
 381                         return TP_STATUS_TS_RAW_HARDWARE;
 382         }
 383
 384         if (ktime_to_timespec_cond(skb->tstamp, ts))
 385                 return TP_STATUS_TS_SOFTWARE;
 386
 387         return 0;
 388 }
 389
 390 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
 391                                     struct sk_buff *skb)
 392 {
 393         union tpacket_uhdr h;
 394         struct timespec ts;
 395         __u32 ts_status;
 396
 397         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
 398                 return 0;
 399
 400         h.raw = frame;
 401         switch (po->tp_version) {
 402         case TPACKET_V1:
 403                 h.h1->tp_sec = ts.tv_sec;
 404                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
 405                 break;
 406         case TPACKET_V2:
 407                 h.h2->tp_sec = ts.tv_sec;
 408                 h.h2->tp_nsec = ts.tv_nsec;
 409                 break;
 410         case TPACKET_V3:
 411         default:
 412                 WARN(1, "TPACKET version not supported.\n");
 413                 BUG();
 414         }
 415
 416         /* one flush is safe, as both fields always lie on the same cacheline */
 417         flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
 418         smp_wmb();
 419
 420         return ts_status;
 421 }
 422
 423 static void *packet_lookup_frame(struct packet_sock *po,
 424                 struct packet_ring_buffer *rb,
 425                 unsigned int position,
 426                 int status)
 427 {
 428         unsigned int pg_vec_pos, frame_offset;
 429         union tpacket_uhdr h;
 430
 431         pg_vec_pos = position / rb->frames_per_block;
 432         frame_offset = position % rb->frames_per_block;
 433
 434         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 435                 (frame_offset * rb->frame_size);
 436
 437         if (status != __packet_get_status(po, h.raw))
 438                 return NULL;
 439
 440         return h.raw;
 441 }
 442
 443 static void *packet_current_frame(struct packet_sock *po,
 444                 struct packet_ring_buffer *rb,
 445                 int status)
 446 {
 447         return packet_lookup_frame(po, rb, rb->head, status);
 448 }
 449
 450 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 451 {
 452         del_timer_sync(&pkc->retire_blk_timer);
 453 }
 454
 455 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 456                 int tx_ring,
 457                 struct sk_buff_head *rb_queue)
 458 {
 459         struct tpacket_kbdq_core *pkc;
 460
 461         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 462
 463         spin_lock_bh(&rb_queue->lock);
 464         pkc->delete_blk_timer = 1;
 465         spin_unlock_bh(&rb_queue->lock);
 466
 467         prb_del_retire_blk_timer(pkc);
 468 }
 469
 470 static void prb_init_blk_timer(struct packet_sock *po,
 471                 struct tpacket_kbdq_core *pkc,
 472                 void (*func) (unsigned long))
 473 {
 474         init_timer(&pkc->retire_blk_timer);
 475         pkc->retire_blk_timer.data = (long)po;
 476         pkc->retire_blk_timer.function = func;
 477         pkc->retire_blk_timer.expires = jiffies;
 478 }
 479
 480 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
 481 {
 482         struct tpacket_kbdq_core *pkc;
 483
 484         if (tx_ring)
 485                 BUG();
 486
 487         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 488         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
 489 }
 490
 491 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 492                                 int blk_size_in_bytes)
 493 {
 494         struct net_device *dev;
 495         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 496         struct ethtool_cmd ecmd;
 497         int err;
 498         u32 speed;
 499
 500         rtnl_lock();
 501         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 502         if (unlikely(!dev)) {
 503                 rtnl_unlock();
 504                 return DEFAULT_PRB_RETIRE_TOV;
 505         }
 506         err = __ethtool_get_settings(dev, &ecmd);
 507         speed = ethtool_cmd_speed(&ecmd);
 508         rtnl_unlock();
 509         if (!err) {
 510                 /*
 511                  * If the link speed is so slow you don't really
 512                  * need to worry about perf anyways
 513                  */
 514                 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
 515                         return DEFAULT_PRB_RETIRE_TOV;
 516                 } else {
 517                         msec = 1;
 518                         div = speed / 1000;
 519                 }
 520         }
 521
 522         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 523
 524         if (div)
 525                 mbits /= div;
 526
 527         tmo = mbits * msec;
 528
 529         if (div)
 530                 return tmo+1;
 531         return tmo;
 532 }
 533
 534 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 535                         union tpacket_req_u *req_u)
 536 {
 537         p1->feature_req_word = req_u->req3.tp_feature_req_word;
 538 }
 539
 540 static void init_prb_bdqc(struct packet_sock *po,
 541                         struct packet_ring_buffer *rb,
 542                         struct pgv *pg_vec,
 543                         union tpacket_req_u *req_u, int tx_ring)
 544 {
 545         struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
 546         struct tpacket_block_desc *pbd;
 547
 548         memset(p1, 0x0, sizeof(*p1));
 549
 550         p1->knxt_seq_num = 1;
 551         p1->pkbdq = pg_vec;
 552         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 553         p1->pkblk_start = pg_vec[0].buffer;
 554         p1->kblk_size = req_u->req3.tp_block_size;
 555         p1->knum_blocks = req_u->req3.tp_block_nr;
 556         p1->hdrlen = po->tp_hdrlen;
 557         p1->version = po->tp_version;
 558         p1->last_kactive_blk_num = 0;
 559         po->stats.stats3.tp_freeze_q_cnt = 0;
 560         if (req_u->req3.tp_retire_blk_tov)
 561                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 562         else
 563                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 564                                                 req_u->req3.tp_block_size);
 565         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 566         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 567
 568         prb_init_ft_ops(p1, req_u);
 569         prb_setup_retire_blk_timer(po, tx_ring);
 570         prb_open_block(p1, pbd);
 571 }
 572
 573 /*  Do NOT update the last_blk_num first.
 574  *  Assumes sk_buff_head lock is held.
 575  */
 576 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 577 {
 578         mod_timer(&pkc->retire_blk_timer,
 579                         jiffies + pkc->tov_in_jiffies);
 580         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 581 }
 582
 583 /*
 584  * Timer logic:
 585  * 1) We refresh the timer only when we open a block.
 586  *    By doing this we don't waste cycles refreshing the timer
 587  *        on packet-by-packet basis.
 588  *
 589  * With a 1MB block-size, on a 1Gbps line, it will take
 590  * i) ~8 ms to fill a block + ii) memcpy etc.
 591  * In this cut we are not accounting for the memcpy time.
 592  *
 593  * So, if the user sets the 'tmo' to 10ms then the timer
 594  * will never fire while the block is still getting filled
 595  * (which is what we want). However, the user could choose
 596  * to close a block early and that's fine.
 597  *
 598  * But when the timer does fire, we check whether or not to refresh it.
 599  * Since the tmo granularity is in msecs, it is not too expensive
 600  * to refresh the timer, lets say every '8' msecs.
 601  * Either the user can set the 'tmo' or we can derive it based on
 602  * a) line-speed and b) block-size.
 603  * prb_calc_retire_blk_tmo() calculates the tmo.
 604  *
 605  */
 606 static void prb_retire_rx_blk_timer_expired(unsigned long data)
 607 {
 608         struct packet_sock *po = (struct packet_sock *)data;
 609         struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
 610         unsigned int frozen;
 611         struct tpacket_block_desc *pbd;
 612
 613         spin_lock(&po->sk.sk_receive_queue.lock);
 614
 615         frozen = prb_queue_frozen(pkc);
 616         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 617
 618         if (unlikely(pkc->delete_blk_timer))
 619                 goto out;
 620
 621         /* We only need to plug the race when the block is partially filled.
 622          * tpacket_rcv:
 623          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 624          *              copy_bits() is in progress ...
 625          *              timer fires on other cpu:
 626          *              we can't retire the current block because copy_bits
 627          *              is in progress.
 628          *
 629          */
 630         if (BLOCK_NUM_PKTS(pbd)) {
 631                 while (atomic_read(&pkc->blk_fill_in_prog)) {
 632                         /* Waiting for skb_copy_bits to finish... */
 633                         cpu_relax();
 634                 }
 635         }
 636
 637         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 638                 if (!frozen) {
 639                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 640                         if (!prb_dispatch_next_block(pkc, po))
 641                                 goto refresh_timer;
 642                         else
 643                                 goto out;
 644                 } else {
 645                         /* Case 1. Queue was frozen because user-space was
 646                          *         lagging behind.
 647                          */
 648                         if (prb_curr_blk_in_use(pkc, pbd)) {
 649                                 /*
 650                                  * Ok, user-space is still behind.
 651                                  * So just refresh the timer.
 652                                  */
 653                                 goto refresh_timer;
 654                         } else {
 655                                /* Case 2. queue was frozen,user-space caught up,
 656                                 * now the link went idle && the timer fired.
 657                                 * We don't have a block to close.So we open this
 658                                 * block and restart the timer.
 659                                 * opening a block thaws the queue,restarts timer
 660                                 * Thawing/timer-refresh is a side effect.
 661                                 */
 662                                 prb_open_block(pkc, pbd);
 663                                 goto out;
 664                         }
 665                 }
 666         }
 667
 668 refresh_timer:
 669         _prb_refresh_rx_retire_blk_timer(pkc);
 670
 671 out:
 672         spin_unlock(&po->sk.sk_receive_queue.lock);
 673 }
 674
 675 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 676                 struct tpacket_block_desc *pbd1, __u32 status)
 677 {
 678         /* Flush everything minus the block header */
 679
 680 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 681         u8 *start, *end;
 682
 683         start = (u8 *)pbd1;
 684
 685         /* Skip the block header(we know header WILL fit in 4K) */
 686         start += PAGE_SIZE;
 687
 688         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 689         for (; start < end; start += PAGE_SIZE)
 690                 flush_dcache_page(pgv_to_page(start));
 691
 692         smp_wmb();
 693 #endif
 694
 695         /* Now update the block status. */
 696
 697         BLOCK_STATUS(pbd1) = status;
 698
 699         /* Flush the block header */
 700
 701 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 702         start = (u8 *)pbd1;
 703         flush_dcache_page(pgv_to_page(start));
 704
 705         smp_wmb();
 706 #endif
 707 }
 708
 709 /*
 710  * Side effect:
 711  *
 712  * 1) flush the block
 713  * 2) Increment active_blk_num
 714  *
 715  * Note:We DONT refresh the timer on purpose.
 716  *      Because almost always the next block will be opened.
 717  */
 718 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 719                 struct tpacket_block_desc *pbd1,
 720                 struct packet_sock *po, unsigned int stat)
 721 {
 722         __u32 status = TP_STATUS_USER | stat;
 723
 724         struct tpacket3_hdr *last_pkt;
 725         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 726
 727         if (po->stats.stats3.tp_drops)
 728                 status |= TP_STATUS_LOSING;
 729
 730         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 731         last_pkt->tp_next_offset = 0;
 732
 733         /* Get the ts of the last pkt */
 734         if (BLOCK_NUM_PKTS(pbd1)) {
 735                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 736                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 737         } else {
 738                 /* Ok, we tmo'd - so get the current time */
 739                 struct timespec ts;
 740                 getnstimeofday(&ts);
 741                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 742                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 743         }
 744
 745         smp_wmb();
 746
 747         /* Flush the block */
 748         prb_flush_block(pkc1, pbd1, status);
 749
 750         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 751 }
 752
 753 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 754 {
 755         pkc->reset_pending_on_curr_blk = 0;
 756 }
 757
 758 /*
 759  * Side effect of opening a block:
 760  *
 761  * 1) prb_queue is thawed.
 762  * 2) retire_blk_timer is refreshed.
 763  *
 764  */
 765 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 766         struct tpacket_block_desc *pbd1)
 767 {
 768         struct timespec ts;
 769         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 770
 771         smp_rmb();
 772
 773         /* We could have just memset this but we will lose the
 774          * flexibility of making the priv area sticky
 775          */
 776
 777         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 778         BLOCK_NUM_PKTS(pbd1) = 0;
 779         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 780
 781         getnstimeofday(&ts);
 782
 783         h1->ts_first_pkt.ts_sec = ts.tv_sec;
 784         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 785
 786         pkc1->pkblk_start = (char *)pbd1;
 787         pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 788
 789         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 790         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 791
 792         pbd1->version = pkc1->version;
 793         pkc1->prev = pkc1->nxt_offset;
 794         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 795
 796         prb_thaw_queue(pkc1);
 797         _prb_refresh_rx_retire_blk_timer(pkc1);
 798
 799         smp_wmb();
 800 }
 801
 802 /*
 803  * Queue freeze logic:
 804  * 1) Assume tp_block_nr = 8 blocks.
 805  * 2) At time 't0', user opens Rx ring.
 806  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 807  * 4) user-space is either sleeping or processing block '0'.
 808  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 809  *    it will close block-7,loop around and try to fill block '0'.
 810  *    call-flow:
 811  *    __packet_lookup_frame_in_block
 812  *      prb_retire_current_block()
 813  *      prb_dispatch_next_block()
 814  *        |->(BLOCK_STATUS == USER) evaluates to true
 815  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 816  * 6) Now there are two cases:
 817  *    6.1) Link goes idle right after the queue is frozen.
 818  *         But remember, the last open_block() refreshed the timer.
 819  *         When this timer expires,it will refresh itself so that we can
 820  *         re-open block-0 in near future.
 821  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 822  *         case and __packet_lookup_frame_in_block will check if block-0
 823  *         is free and can now be re-used.
 824  */
 825 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 826                                   struct packet_sock *po)
 827 {
 828         pkc->reset_pending_on_curr_blk = 1;
 829         po->stats.stats3.tp_freeze_q_cnt++;
 830 }
 831
 832 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 833
 834 /*
 835  * If the next block is free then we will dispatch it
 836  * and return a good offset.
 837  * Else, we will freeze the queue.
 838  * So, caller must check the return value.
 839  */
 840 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 841                 struct packet_sock *po)
 842 {
 843         struct tpacket_block_desc *pbd;
 844
 845         smp_rmb();
 846
 847         /* 1. Get current block num */
 848         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 849
 850         /* 2. If this block is currently in_use then freeze the queue */
 851         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 852                 prb_freeze_queue(pkc, po);
 853                 return NULL;
 854         }
 855
 856         /*
 857          * 3.
 858          * open this block and return the offset where the first packet
 859          * needs to get stored.
 860          */
 861         prb_open_block(pkc, pbd);
 862         return (void *)pkc->nxt_offset;
 863 }
 864
 865 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 866                 struct packet_sock *po, unsigned int status)
 867 {
 868         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 869
 870         /* retire/close the current block */
 871         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 872                 /*
 873                  * Plug the case where copy_bits() is in progress on
 874                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 875                  * have space to copy the pkt in the current block and
 876                  * called prb_retire_current_block()
 877                  *
 878                  * We don't need to worry about the TMO case because
 879                  * the timer-handler already handled this case.
 880                  */
 881                 if (!(status & TP_STATUS_BLK_TMO)) {
 882                         while (atomic_read(&pkc->blk_fill_in_prog)) {
 883                                 /* Waiting for skb_copy_bits to finish... */
 884                                 cpu_relax();
 885                         }
 886                 }
 887                 prb_close_block(pkc, pbd, po, status);
 888                 return;
 889         }
 890 }
 891
 892 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
 893                                       struct tpacket_block_desc *pbd)
 894 {
 895         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 896 }
 897
 898 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 899 {
 900         return pkc->reset_pending_on_curr_blk;
 901 }
 902
 903 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 904 {
 905         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 906         atomic_dec(&pkc->blk_fill_in_prog);
 907 }
 908
 909 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 910                         struct tpacket3_hdr *ppd)
 911 {
 912         ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
 913 }
 914
 915 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 916                         struct tpacket3_hdr *ppd)
 917 {
 918         ppd->hv1.tp_rxhash = 0;
 919 }
 920
 921 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
 922                         struct tpacket3_hdr *ppd)
 923 {
 924         if (vlan_tx_tag_present(pkc->skb)) {
 925                 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
 926                 ppd->tp_status = TP_STATUS_VLAN_VALID;
 927         } else {
 928                 ppd->hv1.tp_vlan_tci = 0;
 929                 ppd->tp_status = TP_STATUS_AVAILABLE;
 930         }
 931 }
 932
 933 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
 934                         struct tpacket3_hdr *ppd)
 935 {
 936         prb_fill_vlan_info(pkc, ppd);
 937
 938         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
 939                 prb_fill_rxhash(pkc, ppd);
 940         else
 941                 prb_clear_rxhash(pkc, ppd);
 942 }
 943
 944 static void prb_fill_curr_block(char *curr,
 945                                 struct tpacket_kbdq_core *pkc,
 946                                 struct tpacket_block_desc *pbd,
 947                                 unsigned int len)
 948 {
 949         struct tpacket3_hdr *ppd;
 950
 951         ppd  = (struct tpacket3_hdr *)curr;
 952         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
 953         pkc->prev = curr;
 954         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
 955         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
 956         BLOCK_NUM_PKTS(pbd) += 1;
 957         atomic_inc(&pkc->blk_fill_in_prog);
 958         prb_run_all_ft_ops(pkc, ppd);
 959 }
 960
 961 /* Assumes caller has the sk->rx_queue.lock */
 962 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
 963                                             struct sk_buff *skb,
 964                                                 int status,
 965                                             unsigned int len
 966                                             )
 967 {
 968         struct tpacket_kbdq_core *pkc;
 969         struct tpacket_block_desc *pbd;
 970         char *curr, *end;
 971
 972         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 973         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 974
 975         /* Queue is frozen when user space is lagging behind */
 976         if (prb_queue_frozen(pkc)) {
 977                 /*
 978                  * Check if that last block which caused the queue to freeze,
 979                  * is still in_use by user-space.
 980                  */
 981                 if (prb_curr_blk_in_use(pkc, pbd)) {
 982                         /* Can't record this packet */
 983                         return NULL;
 984                 } else {
 985                         /*
 986                          * Ok, the block was released by user-space.
 987                          * Now let's open that block.
 988                          * opening a block also thaws the queue.
 989                          * Thawing is a side effect.
 990                          */
 991                         prb_open_block(pkc, pbd);
 992                 }
 993         }
 994
 995         smp_mb();
 996         curr = pkc->nxt_offset;
 997         pkc->skb = skb;
 998         end = (char *)pbd + pkc->kblk_size;
 999
1000         /* first try the current block */
1001         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1002                 prb_fill_curr_block(curr, pkc, pbd, len);
1003                 return (void *)curr;
1004         }
1005
1006         /* Ok, close the current block */
1007         prb_retire_current_block(pkc, po, 0);
1008
1009         /* Now, try to dispatch the next block */
1010         curr = (char *)prb_dispatch_next_block(pkc, po);
1011         if (curr) {
1012                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1013                 prb_fill_curr_block(curr, pkc, pbd, len);
1014                 return (void *)curr;
1015         }
1016
1017         /*
1018          * No free blocks are available.user_space hasn't caught up yet.
1019          * Queue was just frozen and now this packet will get dropped.
1020          */
1021         return NULL;
1022 }
1023
1024 static void *packet_current_rx_frame(struct packet_sock *po,
1025                                             struct sk_buff *skb,
1026                                             int status, unsigned int len)
1027 {
1028         char *curr = NULL;
1029         switch (po->tp_version) {
1030         case TPACKET_V1:
1031         case TPACKET_V2:
1032                 curr = packet_lookup_frame(po, &po->rx_ring,
1033                                         po->rx_ring.head, status);
1034                 return curr;
1035         case TPACKET_V3:
1036                 return __packet_lookup_frame_in_block(po, skb, status, len);
1037         default:
1038                 WARN(1, "TPACKET version not supported\n");
1039                 BUG();
1040                 return NULL;
1041         }
1042 }
1043
1044 static void *prb_lookup_block(struct packet_sock *po,
1045                                      struct packet_ring_buffer *rb,
1046                                      unsigned int idx,
1047                                      int status)
1048 {
1049         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1050         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1051
1052         if (status != BLOCK_STATUS(pbd))
1053                 return NULL;
1054         return pbd;
1055 }
1056
1057 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1058 {
1059         unsigned int prev;
1060         if (rb->prb_bdqc.kactive_blk_num)
1061                 prev = rb->prb_bdqc.kactive_blk_num-1;
1062         else
1063                 prev = rb->prb_bdqc.knum_blocks-1;
1064         return prev;
1065 }
1066
1067 /* Assumes caller has held the rx_queue.lock */
1068 static void *__prb_previous_block(struct packet_sock *po,
1069                                          struct packet_ring_buffer *rb,
1070                                          int status)
1071 {
1072         unsigned int previous = prb_previous_blk_num(rb);
1073         return prb_lookup_block(po, rb, previous, status);
1074 }
1075
1076 static void *packet_previous_rx_frame(struct packet_sock *po,
1077                                              struct packet_ring_buffer *rb,
1078                                              int status)
1079 {
1080         if (po->tp_version <= TPACKET_V2)
1081                 return packet_previous_frame(po, rb, status);
1082
1083         return __prb_previous_block(po, rb, status);
1084 }
1085
1086 static void packet_increment_rx_head(struct packet_sock *po,
1087                                             struct packet_ring_buffer *rb)
1088 {
1089         switch (po->tp_version) {
1090         case TPACKET_V1:
1091         case TPACKET_V2:
1092                 return packet_increment_head(rb);
1093         case TPACKET_V3:
1094         default:
1095                 WARN(1, "TPACKET version not supported.\n");
1096                 BUG();
1097                 return;
1098         }
1099 }
1100
1101 static void *packet_previous_frame(struct packet_sock *po,
1102                 struct packet_ring_buffer *rb,
1103                 int status)
1104 {
1105         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1106         return packet_lookup_frame(po, rb, previous, status);
1107 }
1108
1109 static void packet_increment_head(struct packet_ring_buffer *buff)
1110 {
1111         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1112 }
1113
1114 static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1115 {
1116         struct sock *sk = &po->sk;
1117         bool has_room;
1118
1119         if (po->prot_hook.func != tpacket_rcv)
1120                 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1121                         <= sk->sk_rcvbuf;
1122
1123         spin_lock(&sk->sk_receive_queue.lock);
1124         if (po->tp_version == TPACKET_V3)
1125                 has_room = prb_lookup_block(po, &po->rx_ring,
1126                                             po->rx_ring.prb_bdqc.kactive_blk_num,
1127                                             TP_STATUS_KERNEL);
1128         else
1129                 has_room = packet_lookup_frame(po, &po->rx_ring,
1130                                                po->rx_ring.head,
1131                                                TP_STATUS_KERNEL);
1132         spin_unlock(&sk->sk_receive_queue.lock);
1133
1134         return has_room;
1135 }
1136
1137 static void packet_sock_destruct(struct sock *sk)
1138 {
1139         skb_queue_purge(&sk->sk_error_queue);
1140
1141         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1142         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1143
1144         if (!sock_flag(sk, SOCK_DEAD)) {
1145                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1146                 return;
1147         }
1148
1149         sk_refcnt_debug_dec(sk);
1150 }
1151
1152 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1153 {
1154         int x = atomic_read(&f->rr_cur) + 1;
1155
1156         if (x >= num)
1157                 x = 0;
1158
1159         return x;
1160 }
1161
1162 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1163                                       struct sk_buff *skb,
1164                                       unsigned int num)
1165 {
1166         return (((u64)skb->rxhash) * num) >> 32;
1167 }
1168
1169 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1170                                     struct sk_buff *skb,
1171                                     unsigned int num)
1172 {
1173         int cur, old;
1174
1175         cur = atomic_read(&f->rr_cur);
1176         while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1177                                      fanout_rr_next(f, num))) != cur)
1178                 cur = old;
1179         return cur;
1180 }
1181
1182 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1183                                      struct sk_buff *skb,
1184                                      unsigned int num)
1185 {
1186         return smp_processor_id() % num;
1187 }
1188
1189 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1190                                           struct sk_buff *skb,
1191                                           unsigned int idx, unsigned int skip,
1192                                           unsigned int num)
1193 {
1194         unsigned int i, j;
1195
1196         i = j = min_t(int, f->next[idx], num - 1);
1197         do {
1198                 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1199                         if (i != j)
1200                                 f->next[idx] = i;
1201                         return i;
1202                 }
1203                 if (++i == num)
1204                         i = 0;
1205         } while (i != j);
1206
1207         return idx;
1208 }
1209
1210 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1211 {
1212         return f->flags & (flag >> 8);
1213 }
1214
1215 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1216                              struct packet_type *pt, struct net_device *orig_dev)
1217 {
1218         struct packet_fanout *f = pt->af_packet_priv;
1219         unsigned int num = f->num_members;
1220         struct packet_sock *po;
1221         unsigned int idx;
1222
1223         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1224             !num) {
1225                 kfree_skb(skb);
1226                 return 0;
1227         }
1228
1229         switch (f->type) {
1230         case PACKET_FANOUT_HASH:
1231         default:
1232                 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1233                         skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1234                         if (!skb)
1235                                 return 0;
1236                 }
1237                 skb_get_rxhash(skb);
1238                 idx = fanout_demux_hash(f, skb, num);
1239                 break;
1240         case PACKET_FANOUT_LB:
1241                 idx = fanout_demux_lb(f, skb, num);
1242                 break;
1243         case PACKET_FANOUT_CPU:
1244                 idx = fanout_demux_cpu(f, skb, num);
1245                 break;
1246         case PACKET_FANOUT_ROLLOVER:
1247                 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1248                 break;
1249         }
1250
1251         po = pkt_sk(f->arr[idx]);
1252         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1253             unlikely(!packet_rcv_has_room(po, skb))) {
1254                 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1255                 po = pkt_sk(f->arr[idx]);
1256         }
1257
1258         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1259 }
1260
1261 DEFINE_MUTEX(fanout_mutex);
1262 EXPORT_SYMBOL_GPL(fanout_mutex);
1263 static LIST_HEAD(fanout_list);
1264
1265 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1266 {
1267         struct packet_fanout *f = po->fanout;
1268
1269         spin_lock(&f->lock);
1270         f->arr[f->num_members] = sk;
1271         smp_wmb();
1272         f->num_members++;
1273         spin_unlock(&f->lock);
1274 }
1275
1276 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1277 {
1278         struct packet_fanout *f = po->fanout;
1279         int i;
1280
1281         spin_lock(&f->lock);
1282         for (i = 0; i < f->num_members; i++) {
1283                 if (f->arr[i] == sk)
1284                         break;
1285         }
1286         BUG_ON(i >= f->num_members);
1287         f->arr[i] = f->arr[f->num_members - 1];
1288         f->num_members--;
1289         spin_unlock(&f->lock);
1290 }
1291
1292 static bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1293 {
1294         if (ptype->af_packet_priv == (void*)((struct packet_sock *)sk)->fanout)
1295                 return true;
1296
1297         return false;
1298 }
1299
1300 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1301 {
1302         struct packet_sock *po = pkt_sk(sk);
1303         struct packet_fanout *f, *match;
1304         u8 type = type_flags & 0xff;
1305         u8 flags = type_flags >> 8;
1306         int err;
1307
1308         switch (type) {
1309         case PACKET_FANOUT_ROLLOVER:
1310                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1311                         return -EINVAL;
1312         case PACKET_FANOUT_HASH:
1313         case PACKET_FANOUT_LB:
1314         case PACKET_FANOUT_CPU:
1315                 break;
1316         default:
1317                 return -EINVAL;
1318         }
1319
1320         if (!po->running)
1321                 return -EINVAL;
1322
1323         if (po->fanout)
1324                 return -EALREADY;
1325
1326         mutex_lock(&fanout_mutex);
1327         match = NULL;
1328         list_for_each_entry(f, &fanout_list, list) {
1329                 if (f->id == id &&
1330                     read_pnet(&f->net) == sock_net(sk)) {
1331                         match = f;
1332                         break;
1333                 }
1334         }
1335         err = -EINVAL;
1336         if (match && match->flags != flags)
1337                 goto out;
1338         if (!match) {
1339                 err = -ENOMEM;
1340                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1341                 if (!match)
1342                         goto out;
1343                 write_pnet(&match->net, sock_net(sk));
1344                 match->id = id;
1345                 match->type = type;
1346                 match->flags = flags;
1347                 atomic_set(&match->rr_cur, 0);
1348                 INIT_LIST_HEAD(&match->list);
1349                 spin_lock_init(&match->lock);
1350                 atomic_set(&match->sk_ref, 0);
1351                 match->prot_hook.type = po->prot_hook.type;
1352                 match->prot_hook.dev = po->prot_hook.dev;
1353                 match->prot_hook.func = packet_rcv_fanout;
1354                 match->prot_hook.af_packet_priv = match;
1355                 match->prot_hook.id_match = match_fanout_group;
1356                 dev_add_pack(&match->prot_hook);
1357                 list_add(&match->list, &fanout_list);
1358         }
1359         err = -EINVAL;
1360         if (match->type == type &&
1361             match->prot_hook.type == po->prot_hook.type &&
1362             match->prot_hook.dev == po->prot_hook.dev) {
1363                 err = -ENOSPC;
1364                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1365                         __dev_remove_pack(&po->prot_hook);
1366                         po->fanout = match;
1367                         atomic_inc(&match->sk_ref);
1368                         __fanout_link(sk, po);
1369                         err = 0;
1370                 }
1371         }
1372 out:
1373         mutex_unlock(&fanout_mutex);
1374         return err;
1375 }
1376
1377 static void fanout_release(struct sock *sk)
1378 {
1379         struct packet_sock *po = pkt_sk(sk);
1380         struct packet_fanout *f;
1381
1382         f = po->fanout;
1383         if (!f)
1384                 return;
1385
1386         mutex_lock(&fanout_mutex);
1387         po->fanout = NULL;
1388
1389         if (atomic_dec_and_test(&f->sk_ref)) {
1390                 list_del(&f->list);
1391                 dev_remove_pack(&f->prot_hook);
1392                 kfree(f);
1393         }
1394         mutex_unlock(&fanout_mutex);
1395 }
1396
1397 static const struct proto_ops packet_ops;
1398
1399 static const struct proto_ops packet_ops_spkt;
1400
1401 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1402                            struct packet_type *pt, struct net_device *orig_dev)
1403 {
1404         struct sock *sk;
1405         struct sockaddr_pkt *spkt;
1406
1407         /*
1408          *      When we registered the protocol we saved the socket in the data
1409          *      field for just this event.
1410          */
1411
1412         sk = pt->af_packet_priv;
1413
1414         /*
1415          *      Yank back the headers [hope the device set this
1416          *      right or kerboom...]
1417          *
1418          *      Incoming packets have ll header pulled,
1419          *      push it back.
1420          *
1421          *      For outgoing ones skb->data == skb_mac_header(skb)
1422          *      so that this procedure is noop.
1423          */
1424
1425         if (skb->pkt_type == PACKET_LOOPBACK)
1426                 goto out;
1427
1428         if (!net_eq(dev_net(dev), sock_net(sk)))
1429                 goto out;
1430
1431         skb = skb_share_check(skb, GFP_ATOMIC);
1432         if (skb == NULL)
1433                 goto oom;
1434
1435         /* drop any routing info */
1436         skb_dst_drop(skb);
1437
1438         /* drop conntrack reference */
1439         nf_reset(skb);
1440
1441         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1442
1443         skb_push(skb, skb->data - skb_mac_header(skb));
1444
1445         /*
1446          *      The SOCK_PACKET socket receives _all_ frames.
1447          */
1448
1449         spkt->spkt_family = dev->type;
1450         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1451         spkt->spkt_protocol = skb->protocol;
1452
1453         /*
1454          *      Charge the memory to the socket. This is done specifically
1455          *      to prevent sockets using all the memory up.
1456          */
1457
1458         if (sock_queue_rcv_skb(sk, skb) == 0)
1459                 return 0;
1460
1461 out:
1462         kfree_skb(skb);
1463 oom:
1464         return 0;
1465 }
1466
1467
1468 /*
1469  *      Output a raw packet to a device layer. This bypasses all the other
1470  *      protocol layers and you must therefore supply it with a complete frame
1471  */
1472
1473 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1474                                struct msghdr *msg, size_t len)
1475 {
1476         struct sock *sk = sock->sk;
1477         struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1478         struct sk_buff *skb = NULL;
1479         struct net_device *dev;
1480         __be16 proto = 0;
1481         int err;
1482         int extra_len = 0;
1483
1484         /*
1485          *      Get and verify the address.
1486          */
1487
1488         if (saddr) {
1489                 if (msg->msg_namelen < sizeof(struct sockaddr))
1490                         return -EINVAL;
1491                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1492                         proto = saddr->spkt_protocol;
1493         } else
1494                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1495
1496         /*
1497          *      Find the device first to size check it
1498          */
1499
1500         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1501 retry:
1502         rcu_read_lock();
1503         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1504         err = -ENODEV;
1505         if (dev == NULL)
1506                 goto out_unlock;
1507
1508         err = -ENETDOWN;
1509         if (!(dev->flags & IFF_UP))
1510                 goto out_unlock;
1511
1512         /*
1513          * You may not queue a frame bigger than the mtu. This is the lowest level
1514          * raw protocol and you must do your own fragmentation at this level.
1515          */
1516
1517         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1518                 if (!netif_supports_nofcs(dev)) {
1519                         err = -EPROTONOSUPPORT;
1520                         goto out_unlock;
1521                 }
1522                 extra_len = 4; /* We're doing our own CRC */
1523         }
1524
1525         err = -EMSGSIZE;
1526         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1527                 goto out_unlock;
1528
1529         if (!skb) {
1530                 size_t reserved = LL_RESERVED_SPACE(dev);
1531                 int tlen = dev->needed_tailroom;
1532                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1533
1534                 rcu_read_unlock();
1535                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1536                 if (skb == NULL)
1537                         return -ENOBUFS;
1538                 /* FIXME: Save some space for broken drivers that write a hard
1539                  * header at transmission time by themselves. PPP is the notable
1540                  * one here. This should really be fixed at the driver level.
1541                  */
1542                 skb_reserve(skb, reserved);
1543                 skb_reset_network_header(skb);
1544
1545                 /* Try to align data part correctly */
1546                 if (hhlen) {
1547                         skb->data -= hhlen;
1548                         skb->tail -= hhlen;
1549                         if (len < hhlen)
1550                                 skb_reset_network_header(skb);
1551                 }
1552                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1553                 if (err)
1554                         goto out_free;
1555                 goto retry;
1556         }
1557
1558         if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1559                 /* Earlier code assumed this would be a VLAN pkt,
1560                  * double-check this now that we have the actual
1561                  * packet in hand.
1562                  */
1563                 struct ethhdr *ehdr;
1564                 skb_reset_mac_header(skb);
1565                 ehdr = eth_hdr(skb);
1566                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1567                         err = -EMSGSIZE;
1568                         goto out_unlock;
1569                 }
1570         }
1571
1572         skb->protocol = proto;
1573         skb->dev = dev;
1574         skb->priority = sk->sk_priority;
1575         skb->mark = sk->sk_mark;
1576
1577         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1578
1579         if (unlikely(extra_len == 4))
1580                 skb->no_fcs = 1;
1581
1582         skb_probe_transport_header(skb, 0);
1583
1584         dev_queue_xmit(skb);
1585         rcu_read_unlock();
1586         return len;
1587
1588 out_unlock:
1589         rcu_read_unlock();
1590 out_free:
1591         kfree_skb(skb);
1592         return err;
1593 }
1594
1595 static unsigned int run_filter(const struct sk_buff *skb,
1596                                       const struct sock *sk,
1597                                       unsigned int res)
1598 {
1599         struct sk_filter *filter;
1600
1601         rcu_read_lock();
1602         filter = rcu_dereference(sk->sk_filter);
1603         if (filter != NULL)
1604                 res = SK_RUN_FILTER(filter, skb);
1605         rcu_read_unlock();
1606
1607         return res;
1608 }
1609
1610 /*
1611  * This function makes lazy skb cloning in hope that most of packets
1612  * are discarded by BPF.
1613  *
1614  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1615  * and skb->cb are mangled. It works because (and until) packets
1616  * falling here are owned by current CPU. Output packets are cloned
1617  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1618  * sequencially, so that if we return skb to original state on exit,
1619  * we will not harm anyone.
1620  */
1621
1622 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1623                       struct packet_type *pt, struct net_device *orig_dev)
1624 {
1625         struct sock *sk;
1626         struct sockaddr_ll *sll;
1627         struct packet_sock *po;
1628         u8 *skb_head = skb->data;
1629         int skb_len = skb->len;
1630         unsigned int snaplen, res;
1631
1632         if (skb->pkt_type == PACKET_LOOPBACK)
1633                 goto drop;
1634
1635         sk = pt->af_packet_priv;
1636         po = pkt_sk(sk);
1637
1638         if (!net_eq(dev_net(dev), sock_net(sk)))
1639                 goto drop;
1640
1641         skb->dev = dev;
1642
1643         if (dev->header_ops) {
1644                 /* The device has an explicit notion of ll header,
1645                  * exported to higher levels.
1646                  *
1647                  * Otherwise, the device hides details of its frame
1648                  * structure, so that corresponding packet head is
1649                  * never delivered to user.
1650                  */
1651                 if (sk->sk_type != SOCK_DGRAM)
1652                         skb_push(skb, skb->data - skb_mac_header(skb));
1653                 else if (skb->pkt_type == PACKET_OUTGOING) {
1654                         /* Special case: outgoing packets have ll header at head */
1655                         skb_pull(skb, skb_network_offset(skb));
1656                 }
1657         }
1658
1659         snaplen = skb->len;
1660
1661         res = run_filter(skb, sk, snaplen);
1662         if (!res)
1663                 goto drop_n_restore;
1664         if (snaplen > res)
1665                 snaplen = res;
1666
1667         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1668                 goto drop_n_acct;
1669
1670         if (skb_shared(skb)) {
1671                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1672                 if (nskb == NULL)
1673                         goto drop_n_acct;
1674
1675                 if (skb_head != skb->data) {
1676                         skb->data = skb_head;
1677                         skb->len = skb_len;
1678                 }
1679                 consume_skb(skb);
1680                 skb = nskb;
1681         }
1682
1683         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1684                      sizeof(skb->cb));
1685
1686         sll = &PACKET_SKB_CB(skb)->sa.ll;
1687         sll->sll_family = AF_PACKET;
1688         sll->sll_hatype = dev->type;
1689         sll->sll_protocol = skb->protocol;
1690         sll->sll_pkttype = skb->pkt_type;
1691         if (unlikely(po->origdev))
1692                 sll->sll_ifindex = orig_dev->ifindex;
1693         else
1694                 sll->sll_ifindex = dev->ifindex;
1695
1696         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1697
1698         PACKET_SKB_CB(skb)->origlen = skb->len;
1699
1700         if (pskb_trim(skb, snaplen))
1701                 goto drop_n_acct;
1702
1703         skb_set_owner_r(skb, sk);
1704         skb->dev = NULL;
1705         skb_dst_drop(skb);
1706
1707         /* drop conntrack reference */
1708         nf_reset(skb);
1709
1710         spin_lock(&sk->sk_receive_queue.lock);
1711         po->stats.stats1.tp_packets++;
1712         skb->dropcount = atomic_read(&sk->sk_drops);
1713         __skb_queue_tail(&sk->sk_receive_queue, skb);
1714         spin_unlock(&sk->sk_receive_queue.lock);
1715         sk->sk_data_ready(sk, skb->len);
1716         return 0;
1717
1718 drop_n_acct:
1719         spin_lock(&sk->sk_receive_queue.lock);
1720         po->stats.stats1.tp_drops++;
1721         atomic_inc(&sk->sk_drops);
1722         spin_unlock(&sk->sk_receive_queue.lock);
1723
1724 drop_n_restore:
1725         if (skb_head != skb->data && skb_shared(skb)) {
1726                 skb->data = skb_head;
1727                 skb->len = skb_len;
1728         }
1729 drop:
1730         consume_skb(skb);
1731         return 0;
1732 }
1733
1734 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1735                        struct packet_type *pt, struct net_device *orig_dev)
1736 {
1737         struct sock *sk;
1738         struct packet_sock *po;
1739         struct sockaddr_ll *sll;
1740         union tpacket_uhdr h;
1741         u8 *skb_head = skb->data;
1742         int skb_len = skb->len;
1743         unsigned int snaplen, res;
1744         unsigned long status = TP_STATUS_USER;
1745         unsigned short macoff, netoff, hdrlen;
1746         struct sk_buff *copy_skb = NULL;
1747         struct timespec ts;
1748         __u32 ts_status;
1749
1750         if (skb->pkt_type == PACKET_LOOPBACK)
1751                 goto drop;
1752
1753         sk = pt->af_packet_priv;
1754         po = pkt_sk(sk);
1755
1756         if (!net_eq(dev_net(dev), sock_net(sk)))
1757                 goto drop;
1758
1759         if (dev->header_ops) {
1760                 if (sk->sk_type != SOCK_DGRAM)
1761                         skb_push(skb, skb->data - skb_mac_header(skb));
1762                 else if (skb->pkt_type == PACKET_OUTGOING) {
1763                         /* Special case: outgoing packets have ll header at head */
1764                         skb_pull(skb, skb_network_offset(skb));
1765                 }
1766         }
1767
1768         if (skb->ip_summed == CHECKSUM_PARTIAL)
1769                 status |= TP_STATUS_CSUMNOTREADY;
1770
1771         snaplen = skb->len;
1772
1773         res = run_filter(skb, sk, snaplen);
1774         if (!res)
1775                 goto drop_n_restore;
1776         if (snaplen > res)
1777                 snaplen = res;
1778
1779         if (sk->sk_type == SOCK_DGRAM) {
1780                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1781                                   po->tp_reserve;
1782         } else {
1783                 unsigned int maclen = skb_network_offset(skb);
1784                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1785                                        (maclen < 16 ? 16 : maclen)) +
1786                         po->tp_reserve;
1787                 macoff = netoff - maclen;
1788         }
1789         if (po->tp_version <= TPACKET_V2) {
1790                 if (macoff + snaplen > po->rx_ring.frame_size) {
1791                         if (po->copy_thresh &&
1792                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1793                                 if (skb_shared(skb)) {
1794                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1795                                 } else {
1796                                         copy_skb = skb_get(skb);
1797                                         skb_head = skb->data;
1798                                 }
1799                                 if (copy_skb)
1800                                         skb_set_owner_r(copy_skb, sk);
1801                         }
1802                         snaplen = po->rx_ring.frame_size - macoff;
1803                         if ((int)snaplen < 0)
1804                                 snaplen = 0;
1805                 }
1806         }
1807         spin_lock(&sk->sk_receive_queue.lock);
1808         h.raw = packet_current_rx_frame(po, skb,
1809                                         TP_STATUS_KERNEL, (macoff+snaplen));
1810         if (!h.raw)
1811                 goto ring_is_full;
1812         if (po->tp_version <= TPACKET_V2) {
1813                 packet_increment_rx_head(po, &po->rx_ring);
1814         /*
1815          * LOSING will be reported till you read the stats,
1816          * because it's COR - Clear On Read.
1817          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1818          * at packet level.
1819          */
1820                 if (po->stats.stats1.tp_drops)
1821                         status |= TP_STATUS_LOSING;
1822         }
1823         po->stats.stats1.tp_packets++;
1824         if (copy_skb) {
1825                 status |= TP_STATUS_COPY;
1826                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1827         }
1828         spin_unlock(&sk->sk_receive_queue.lock);
1829
1830         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1831
1832         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
1833                 getnstimeofday(&ts);
1834
1835         status |= ts_status;
1836
1837         switch (po->tp_version) {
1838         case TPACKET_V1:
1839                 h.h1->tp_len = skb->len;
1840                 h.h1->tp_snaplen = snaplen;
1841                 h.h1->tp_mac = macoff;
1842                 h.h1->tp_net = netoff;
1843                 h.h1->tp_sec = ts.tv_sec;
1844                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
1845                 hdrlen = sizeof(*h.h1);
1846                 break;
1847         case TPACKET_V2:
1848                 h.h2->tp_len = skb->len;
1849                 h.h2->tp_snaplen = snaplen;
1850                 h.h2->tp_mac = macoff;
1851                 h.h2->tp_net = netoff;
1852                 h.h2->tp_sec = ts.tv_sec;
1853                 h.h2->tp_nsec = ts.tv_nsec;
1854                 if (vlan_tx_tag_present(skb)) {
1855                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1856                         status |= TP_STATUS_VLAN_VALID;
1857                 } else {
1858                         h.h2->tp_vlan_tci = 0;
1859                 }
1860                 h.h2->tp_padding = 0;
1861                 hdrlen = sizeof(*h.h2);
1862                 break;
1863         case TPACKET_V3:
1864                 /* tp_nxt_offset,vlan are already populated above.
1865                  * So DONT clear those fields here
1866                  */
1867                 h.h3->tp_status |= status;
1868                 h.h3->tp_len = skb->len;
1869                 h.h3->tp_snaplen = snaplen;
1870                 h.h3->tp_mac = macoff;
1871                 h.h3->tp_net = netoff;
1872                 h.h3->tp_sec  = ts.tv_sec;
1873                 h.h3->tp_nsec = ts.tv_nsec;
1874                 hdrlen = sizeof(*h.h3);
1875                 break;
1876         default:
1877                 BUG();
1878         }
1879
1880         sll = h.raw + TPACKET_ALIGN(hdrlen);
1881         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1882         sll->sll_family = AF_PACKET;
1883         sll->sll_hatype = dev->type;
1884         sll->sll_protocol = skb->protocol;
1885         sll->sll_pkttype = skb->pkt_type;
1886         if (unlikely(po->origdev))
1887                 sll->sll_ifindex = orig_dev->ifindex;
1888         else
1889                 sll->sll_ifindex = dev->ifindex;
1890
1891         smp_mb();
1892 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1893         {
1894                 u8 *start, *end;
1895
1896                 if (po->tp_version <= TPACKET_V2) {
1897                         end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1898                                 + macoff + snaplen);
1899                         for (start = h.raw; start < end; start += PAGE_SIZE)
1900                                 flush_dcache_page(pgv_to_page(start));
1901                 }
1902                 smp_wmb();
1903         }
1904 #endif
1905         if (po->tp_version <= TPACKET_V2)
1906                 __packet_set_status(po, h.raw, status);
1907         else
1908                 prb_clear_blk_fill_status(&po->rx_ring);
1909
1910         sk->sk_data_ready(sk, 0);
1911
1912 drop_n_restore:
1913         if (skb_head != skb->data && skb_shared(skb)) {
1914                 skb->data = skb_head;
1915                 skb->len = skb_len;
1916         }
1917 drop:
1918         kfree_skb(skb);
1919         return 0;
1920
1921 ring_is_full:
1922         po->stats.stats1.tp_drops++;
1923         spin_unlock(&sk->sk_receive_queue.lock);
1924
1925         sk->sk_data_ready(sk, 0);
1926         kfree_skb(copy_skb);
1927         goto drop_n_restore;
1928 }
1929
1930 static void tpacket_destruct_skb(struct sk_buff *skb)
1931 {
1932         struct packet_sock *po = pkt_sk(skb->sk);
1933         void *ph;
1934
1935         if (likely(po->tx_ring.pg_vec)) {
1936                 __u32 ts;
1937
1938                 ph = skb_shinfo(skb)->destructor_arg;
1939                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1940                 atomic_dec(&po->tx_ring.pending);
1941
1942                 ts = __packet_set_timestamp(po, ph, skb);
1943                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
1944         }
1945
1946         sock_wfree(skb);
1947 }
1948
1949 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1950                 void *frame, struct net_device *dev, int size_max,
1951                 __be16 proto, unsigned char *addr, int hlen)
1952 {
1953         union tpacket_uhdr ph;
1954         int to_write, offset, len, tp_len, nr_frags, len_max;
1955         struct socket *sock = po->sk.sk_socket;
1956         struct page *page;
1957         void *data;
1958         int err;
1959
1960         ph.raw = frame;
1961
1962         skb->protocol = proto;
1963         skb->dev = dev;
1964         skb->priority = po->sk.sk_priority;
1965         skb->mark = po->sk.sk_mark;
1966         sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
1967         skb_shinfo(skb)->destructor_arg = ph.raw;
1968
1969         switch (po->tp_version) {
1970         case TPACKET_V2:
1971                 tp_len = ph.h2->tp_len;
1972                 break;
1973         default:
1974                 tp_len = ph.h1->tp_len;
1975                 break;
1976         }
1977         if (unlikely(tp_len > size_max)) {
1978                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1979                 return -EMSGSIZE;
1980         }
1981
1982         skb_reserve(skb, hlen);
1983         skb_reset_network_header(skb);
1984         skb_probe_transport_header(skb, 0);
1985
1986         if (po->tp_tx_has_off) {
1987                 int off_min, off_max, off;
1988                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
1989                 off_max = po->tx_ring.frame_size - tp_len;
1990                 if (sock->type == SOCK_DGRAM) {
1991                         switch (po->tp_version) {
1992                         case TPACKET_V2:
1993                                 off = ph.h2->tp_net;
1994                                 break;
1995                         default:
1996                                 off = ph.h1->tp_net;
1997                                 break;
1998                         }
1999                 } else {
2000                         switch (po->tp_version) {
2001                         case TPACKET_V2:
2002                                 off = ph.h2->tp_mac;
2003                                 break;
2004                         default:
2005                                 off = ph.h1->tp_mac;
2006                                 break;
2007                         }
2008                 }
2009                 if (unlikely((off < off_min) || (off_max < off)))
2010                         return -EINVAL;
2011                 data = ph.raw + off;
2012         } else {
2013                 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2014         }
2015         to_write = tp_len;
2016
2017         if (sock->type == SOCK_DGRAM) {
2018                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2019                                 NULL, tp_len);
2020                 if (unlikely(err < 0))
2021                         return -EINVAL;
2022         } else if (dev->hard_header_len) {
2023                 /* net device doesn't like empty head */
2024                 if (unlikely(tp_len <= dev->hard_header_len)) {
2025                         pr_err("packet size is too short (%d < %d)\n",
2026                                tp_len, dev->hard_header_len);
2027                         return -EINVAL;
2028                 }
2029
2030                 skb_push(skb, dev->hard_header_len);
2031                 err = skb_store_bits(skb, 0, data,
2032                                 dev->hard_header_len);
2033                 if (unlikely(err))
2034                         return err;
2035
2036                 data += dev->hard_header_len;
2037                 to_write -= dev->hard_header_len;
2038         }
2039
2040         offset = offset_in_page(data);
2041         len_max = PAGE_SIZE - offset;
2042         len = ((to_write > len_max) ? len_max : to_write);
2043
2044         skb->data_len = to_write;
2045         skb->len += to_write;
2046         skb->truesize += to_write;
2047         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2048
2049         while (likely(to_write)) {
2050                 nr_frags = skb_shinfo(skb)->nr_frags;
2051
2052                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2053                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2054                                MAX_SKB_FRAGS);
2055                         return -EFAULT;
2056                 }
2057
2058                 page = pgv_to_page(data);
2059                 data += len;
2060                 flush_dcache_page(page);
2061                 get_page(page);
2062                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2063                 to_write -= len;
2064                 offset = 0;
2065                 len_max = PAGE_SIZE;
2066                 len = ((to_write > len_max) ? len_max : to_write);
2067         }
2068
2069         return tp_len;
2070 }
2071
2072 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2073 {
2074         struct sk_buff *skb;
2075         struct net_device *dev;
2076         __be16 proto;
2077         int err, reserve = 0;
2078         void *ph;
2079         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2080         int tp_len, size_max;
2081         unsigned char *addr;
2082         int len_sum = 0;
2083         int status = TP_STATUS_AVAILABLE;
2084         int hlen, tlen;
2085
2086         mutex_lock(&po->pg_vec_lock);
2087
2088         if (likely(saddr == NULL)) {
2089                 dev     = packet_cached_dev_get(po);
2090                 proto   = po->num;
2091                 addr    = NULL;
2092         } else {
2093                 err = -EINVAL;
2094                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2095                         goto out;
2096                 if (msg->msg_namelen < (saddr->sll_halen
2097                                         + offsetof(struct sockaddr_ll,
2098                                                 sll_addr)))
2099                         goto out;
2100                 proto   = saddr->sll_protocol;
2101                 addr    = saddr->sll_addr;
2102                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2103         }
2104
2105         err = -ENXIO;
2106         if (unlikely(dev == NULL))
2107                 goto out;
2108         err = -ENETDOWN;
2109         if (unlikely(!(dev->flags & IFF_UP)))
2110                 goto out_put;
2111
2112         reserve = dev->hard_header_len;
2113
2114         size_max = po->tx_ring.frame_size
2115                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2116
2117         if (size_max > dev->mtu + reserve)
2118                 size_max = dev->mtu + reserve;
2119
2120         do {
2121                 ph = packet_current_frame(po, &po->tx_ring,
2122                                 TP_STATUS_SEND_REQUEST);
2123
2124                 if (unlikely(ph == NULL)) {
2125                         schedule();
2126                         continue;
2127                 }
2128
2129                 status = TP_STATUS_SEND_REQUEST;
2130                 hlen = LL_RESERVED_SPACE(dev);
2131                 tlen = dev->needed_tailroom;
2132                 skb = sock_alloc_send_skb(&po->sk,
2133                                 hlen + tlen + sizeof(struct sockaddr_ll),
2134                                 0, &err);
2135
2136                 if (unlikely(skb == NULL))
2137                         goto out_status;
2138
2139                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2140                                 addr, hlen);
2141
2142                 if (unlikely(tp_len < 0)) {
2143                         if (po->tp_loss) {
2144                                 __packet_set_status(po, ph,
2145                                                 TP_STATUS_AVAILABLE);
2146                                 packet_increment_head(&po->tx_ring);
2147                                 kfree_skb(skb);
2148                                 continue;
2149                         } else {
2150                                 status = TP_STATUS_WRONG_FORMAT;
2151                                 err = tp_len;
2152                                 goto out_status;
2153                         }
2154                 }
2155
2156                 skb->destructor = tpacket_destruct_skb;
2157                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2158                 atomic_inc(&po->tx_ring.pending);
2159
2160                 status = TP_STATUS_SEND_REQUEST;
2161                 err = dev_queue_xmit(skb);
2162                 if (unlikely(err > 0)) {
2163                         err = net_xmit_errno(err);
2164                         if (err && __packet_get_status(po, ph) ==
2165                                    TP_STATUS_AVAILABLE) {
2166                                 /* skb was destructed already */
2167                                 skb = NULL;
2168                                 goto out_status;
2169                         }
2170                         /*
2171                          * skb was dropped but not destructed yet;
2172                          * let's treat it like congestion or err < 0
2173                          */
2174                         err = 0;
2175                 }
2176                 packet_increment_head(&po->tx_ring);
2177                 len_sum += tp_len;
2178         } while (likely((ph != NULL) ||
2179                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2180                          (atomic_read(&po->tx_ring.pending))))
2181                 );
2182
2183         err = len_sum;
2184         goto out_put;
2185
2186 out_status:
2187         __packet_set_status(po, ph, status);
2188         kfree_skb(skb);
2189 out_put:
2190         dev_put(dev);
2191 out:
2192         mutex_unlock(&po->pg_vec_lock);
2193         return err;
2194 }
2195
2196 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2197                                         size_t reserve, size_t len,
2198                                         size_t linear, int noblock,
2199                                         int *err)
2200 {
2201         struct sk_buff *skb;
2202
2203         /* Under a page?  Don't bother with paged skb. */
2204         if (prepad + len < PAGE_SIZE || !linear)
2205                 linear = len;
2206
2207         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2208                                    err);
2209         if (!skb)
2210                 return NULL;
2211
2212         skb_reserve(skb, reserve);
2213         skb_put(skb, linear);
2214         skb->data_len = len - linear;
2215         skb->len += len - linear;
2216
2217         return skb;
2218 }
2219
2220 static int packet_snd(struct socket *sock,
2221                           struct msghdr *msg, size_t len)
2222 {
2223         struct sock *sk = sock->sk;
2224         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2225         struct sk_buff *skb;
2226         struct net_device *dev;
2227         __be16 proto;
2228         unsigned char *addr;
2229         int err, reserve = 0;
2230         struct virtio_net_hdr vnet_hdr = { 0 };
2231         int offset = 0;
2232         int vnet_hdr_len;
2233         struct packet_sock *po = pkt_sk(sk);
2234         unsigned short gso_type = 0;
2235         int hlen, tlen;
2236         int extra_len = 0;
2237
2238         /*
2239          *      Get and verify the address.
2240          */
2241
2242         if (likely(saddr == NULL)) {
2243                 dev     = packet_cached_dev_get(po);
2244                 proto   = po->num;
2245                 addr    = NULL;
2246         } else {
2247                 err = -EINVAL;
2248                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2249                         goto out;
2250                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2251                         goto out;
2252                 proto   = saddr->sll_protocol;
2253                 addr    = saddr->sll_addr;
2254                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2255         }
2256
2257         err = -ENXIO;
2258         if (unlikely(dev == NULL))
2259                 goto out_unlock;
2260         err = -ENETDOWN;
2261         if (unlikely(!(dev->flags & IFF_UP)))
2262                 goto out_unlock;
2263
2264         if (sock->type == SOCK_RAW)
2265                 reserve = dev->hard_header_len;
2266         if (po->has_vnet_hdr) {
2267                 vnet_hdr_len = sizeof(vnet_hdr);
2268
2269                 err = -EINVAL;
2270                 if (len < vnet_hdr_len)
2271                         goto out_unlock;
2272
2273                 len -= vnet_hdr_len;
2274
2275                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2276                                        vnet_hdr_len);
2277                 if (err < 0)
2278                         goto out_unlock;
2279
2280                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2281                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2282                       vnet_hdr.hdr_len))
2283                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
2284                                                  vnet_hdr.csum_offset + 2;
2285
2286                 err = -EINVAL;
2287                 if (vnet_hdr.hdr_len > len)
2288                         goto out_unlock;
2289
2290                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2291                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2292                         case VIRTIO_NET_HDR_GSO_TCPV4:
2293                                 gso_type = SKB_GSO_TCPV4;
2294                                 break;
2295                         case VIRTIO_NET_HDR_GSO_TCPV6:
2296                                 gso_type = SKB_GSO_TCPV6;
2297                                 break;
2298                         case VIRTIO_NET_HDR_GSO_UDP:
2299                                 gso_type = SKB_GSO_UDP;
2300                                 break;
2301                         default:
2302                                 goto out_unlock;
2303                         }
2304
2305                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2306                                 gso_type |= SKB_GSO_TCP_ECN;
2307
2308                         if (vnet_hdr.gso_size == 0)
2309                                 goto out_unlock;
2310
2311                 }
2312         }
2313
2314         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2315                 if (!netif_supports_nofcs(dev)) {
2316                         err = -EPROTONOSUPPORT;
2317                         goto out_unlock;
2318                 }
2319                 extra_len = 4; /* We're doing our own CRC */
2320         }
2321
2322         err = -EMSGSIZE;
2323         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2324                 goto out_unlock;
2325
2326         err = -ENOBUFS;
2327         hlen = LL_RESERVED_SPACE(dev);
2328         tlen = dev->needed_tailroom;
2329         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2330                                msg->msg_flags & MSG_DONTWAIT, &err);
2331         if (skb == NULL)
2332                 goto out_unlock;
2333
2334         skb_set_network_header(skb, reserve);
2335
2336         err = -EINVAL;
2337         if (sock->type == SOCK_DGRAM &&
2338             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2339                 goto out_free;
2340
2341         /* Returns -EFAULT on error */
2342         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2343         if (err)
2344                 goto out_free;
2345
2346         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2347
2348         if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2349                 /* Earlier code assumed this would be a VLAN pkt,
2350                  * double-check this now that we have the actual
2351                  * packet in hand.
2352                  */
2353                 struct ethhdr *ehdr;
2354                 skb_reset_mac_header(skb);
2355                 ehdr = eth_hdr(skb);
2356                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2357                         err = -EMSGSIZE;
2358                         goto out_free;
2359                 }
2360         }
2361
2362         skb->protocol = proto;
2363         skb->dev = dev;
2364         skb->priority = sk->sk_priority;
2365         skb->mark = sk->sk_mark;
2366
2367         if (po->has_vnet_hdr) {
2368                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2369                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2370                                                   vnet_hdr.csum_offset)) {
2371                                 err = -EINVAL;
2372                                 goto out_free;
2373                         }
2374                 }
2375
2376                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2377                 skb_shinfo(skb)->gso_type = gso_type;
2378
2379                 /* Header must be checked, and gso_segs computed. */
2380                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2381                 skb_shinfo(skb)->gso_segs = 0;
2382
2383                 len += vnet_hdr_len;
2384         }
2385
2386         skb_probe_transport_header(skb, reserve);
2387
2388         if (unlikely(extra_len == 4))
2389                 skb->no_fcs = 1;
2390
2391         /*
2392          *      Now send it
2393          */
2394
2395         err = dev_queue_xmit(skb);
2396         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2397                 goto out_unlock;
2398
2399         dev_put(dev);
2400
2401         return len;
2402
2403 out_free:
2404         kfree_skb(skb);
2405 out_unlock:
2406         if (dev)
2407                 dev_put(dev);
2408 out:
2409         return err;
2410 }
2411
2412 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2413                 struct msghdr *msg, size_t len)
2414 {
2415         struct sock *sk = sock->sk;
2416         struct packet_sock *po = pkt_sk(sk);
2417         if (po->tx_ring.pg_vec)
2418                 return tpacket_snd(po, msg);
2419         else
2420                 return packet_snd(sock, msg, len);
2421 }
2422
2423 /*
2424  *      Close a PACKET socket. This is fairly simple. We immediately go
2425  *      to 'closed' state and remove our protocol entry in the device list.
2426  */
2427
2428 static int packet_release(struct socket *sock)
2429 {
2430         struct sock *sk = sock->sk;
2431         struct packet_sock *po;
2432         struct net *net;
2433         union tpacket_req_u req_u;
2434
2435         if (!sk)
2436                 return 0;
2437
2438         net = sock_net(sk);
2439         po = pkt_sk(sk);
2440
2441         mutex_lock(&net->packet.sklist_lock);
2442         sk_del_node_init_rcu(sk);
2443         mutex_unlock(&net->packet.sklist_lock);
2444
2445         preempt_disable();
2446         sock_prot_inuse_add(net, sk->sk_prot, -1);
2447         preempt_enable();
2448
2449         spin_lock(&po->bind_lock);
2450         unregister_prot_hook(sk, false);
2451         packet_cached_dev_reset(po);
2452
2453         if (po->prot_hook.dev) {
2454                 dev_put(po->prot_hook.dev);
2455                 po->prot_hook.dev = NULL;
2456         }
2457         spin_unlock(&po->bind_lock);
2458
2459         packet_flush_mclist(sk);
2460
2461         if (po->rx_ring.pg_vec) {
2462                 memset(&req_u, 0, sizeof(req_u));
2463                 packet_set_ring(sk, &req_u, 1, 0);
2464         }
2465
2466         if (po->tx_ring.pg_vec) {
2467                 memset(&req_u, 0, sizeof(req_u));
2468                 packet_set_ring(sk, &req_u, 1, 1);
2469         }
2470
2471         fanout_release(sk);
2472
2473         synchronize_net();
2474         /*
2475          *      Now the socket is dead. No more input will appear.
2476          */
2477         sock_orphan(sk);
2478         sock->sk = NULL;
2479
2480         /* Purge queues */
2481
2482         skb_queue_purge(&sk->sk_receive_queue);
2483         sk_refcnt_debug_release(sk);
2484
2485         sock_put(sk);
2486         return 0;
2487 }
2488
2489 /*
2490  *      Attach a packet hook.
2491  */
2492
2493 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2494 {
2495         struct packet_sock *po = pkt_sk(sk);
2496
2497         if (po->fanout) {
2498                 if (dev)
2499                         dev_put(dev);
2500
2501                 return -EINVAL;
2502         }
2503
2504         lock_sock(sk);
2505
2506         spin_lock(&po->bind_lock);
2507         unregister_prot_hook(sk, true);
2508
2509         po->num = protocol;
2510         po->prot_hook.type = protocol;
2511         if (po->prot_hook.dev)
2512                 dev_put(po->prot_hook.dev);
2513
2514         po->prot_hook.dev = dev;
2515         po->ifindex = dev ? dev->ifindex : 0;
2516
2517         packet_cached_dev_assign(po, dev);
2518
2519         if (protocol == 0)
2520                 goto out_unlock;
2521
2522         if (!dev || (dev->flags & IFF_UP)) {
2523                 register_prot_hook(sk);
2524         } else {
2525                 sk->sk_err = ENETDOWN;
2526                 if (!sock_flag(sk, SOCK_DEAD))
2527                         sk->sk_error_report(sk);
2528         }
2529
2530 out_unlock:
2531         spin_unlock(&po->bind_lock);
2532         release_sock(sk);
2533         return 0;
2534 }
2535
2536 /*
2537  *      Bind a packet socket to a device
2538  */
2539
2540 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2541                             int addr_len)
2542 {
2543         struct sock *sk = sock->sk;
2544         char name[15];
2545         struct net_device *dev;
2546         int err = -ENODEV;
2547
2548         /*
2549          *      Check legality
2550          */
2551
2552         if (addr_len != sizeof(struct sockaddr))
2553                 return -EINVAL;
2554         strlcpy(name, uaddr->sa_data, sizeof(name));
2555
2556         dev = dev_get_by_name(sock_net(sk), name);
2557         if (dev)
2558                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2559         return err;
2560 }
2561
2562 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2563 {
2564         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2565         struct sock *sk = sock->sk;
2566         struct net_device *dev = NULL;
2567         int err;
2568
2569
2570         /*
2571          *      Check legality
2572          */
2573
2574         if (addr_len < sizeof(struct sockaddr_ll))
2575                 return -EINVAL;
2576         if (sll->sll_family != AF_PACKET)
2577                 return -EINVAL;
2578
2579         if (sll->sll_ifindex) {
2580                 err = -ENODEV;
2581                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2582                 if (dev == NULL)
2583                         goto out;
2584         }
2585         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2586
2587 out:
2588         return err;
2589 }
2590
2591 static struct proto packet_proto = {
2592         .name     = "PACKET",
2593         .owner    = THIS_MODULE,
2594         .obj_size = sizeof(struct packet_sock),
2595 };
2596
2597 /*
2598  *      Create a packet of type SOCK_PACKET.
2599  */
2600
2601 static int packet_create(struct net *net, struct socket *sock, int protocol,
2602                          int kern)
2603 {
2604         struct sock *sk;
2605         struct packet_sock *po;
2606         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2607         int err;
2608
2609         if (!ns_capable(net->user_ns, CAP_NET_RAW))
2610                 return -EPERM;
2611         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2612             sock->type != SOCK_PACKET)
2613                 return -ESOCKTNOSUPPORT;
2614
2615         sock->state = SS_UNCONNECTED;
2616
2617         err = -ENOBUFS;
2618         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2619         if (sk == NULL)
2620                 goto out;
2621
2622         sock->ops = &packet_ops;
2623         if (sock->type == SOCK_PACKET)
2624                 sock->ops = &packet_ops_spkt;
2625
2626         sock_init_data(sock, sk);
2627
2628         po = pkt_sk(sk);
2629         sk->sk_family = PF_PACKET;
2630         po->num = proto;
2631
2632         packet_cached_dev_reset(po);
2633
2634         sk->sk_destruct = packet_sock_destruct;
2635         sk_refcnt_debug_inc(sk);
2636
2637         /*
2638          *      Attach a protocol block
2639          */
2640
2641         spin_lock_init(&po->bind_lock);
2642         mutex_init(&po->pg_vec_lock);
2643         po->prot_hook.func = packet_rcv;
2644
2645         if (sock->type == SOCK_PACKET)
2646                 po->prot_hook.func = packet_rcv_spkt;
2647
2648         po->prot_hook.af_packet_priv = sk;
2649
2650         if (proto) {
2651                 po->prot_hook.type = proto;
2652                 register_prot_hook(sk);
2653         }
2654
2655         mutex_lock(&net->packet.sklist_lock);
2656         sk_add_node_rcu(sk, &net->packet.sklist);
2657         mutex_unlock(&net->packet.sklist_lock);
2658
2659         preempt_disable();
2660         sock_prot_inuse_add(net, &packet_proto, 1);
2661         preempt_enable();
2662
2663         return 0;
2664 out:
2665         return err;
2666 }
2667
2668 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2669 {
2670         struct sock_exterr_skb *serr;
2671         struct sk_buff *skb, *skb2;
2672         int copied, err;
2673
2674         err = -EAGAIN;
2675         skb = skb_dequeue(&sk->sk_error_queue);
2676         if (skb == NULL)
2677                 goto out;
2678
2679         copied = skb->len;
2680         if (copied > len) {
2681                 msg->msg_flags |= MSG_TRUNC;
2682                 copied = len;
2683         }
2684         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2685         if (err)
2686                 goto out_free_skb;
2687
2688         sock_recv_timestamp(msg, sk, skb);
2689
2690         serr = SKB_EXT_ERR(skb);
2691         put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2692                  sizeof(serr->ee), &serr->ee);
2693
2694         msg->msg_flags |= MSG_ERRQUEUE;
2695         err = copied;
2696
2697         /* Reset and regenerate socket error */
2698         spin_lock_bh(&sk->sk_error_queue.lock);
2699         sk->sk_err = 0;
2700         if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2701                 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2702                 spin_unlock_bh(&sk->sk_error_queue.lock);
2703                 sk->sk_error_report(sk);
2704         } else
2705                 spin_unlock_bh(&sk->sk_error_queue.lock);
2706
2707 out_free_skb:
2708         kfree_skb(skb);
2709 out:
2710         return err;
2711 }
2712
2713 /*
2714  *      Pull a packet from our receive queue and hand it to the user.
2715  *      If necessary we block.
2716  */
2717
2718 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2719                           struct msghdr *msg, size_t len, int flags)
2720 {
2721         struct sock *sk = sock->sk;
2722         struct sk_buff *skb;
2723         int copied, err;
2724         int vnet_hdr_len = 0;
2725
2726         err = -EINVAL;
2727         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2728                 goto out;
2729
2730 #if 0
2731         /* What error should we return now? EUNATTACH? */
2732         if (pkt_sk(sk)->ifindex < 0)
2733                 return -ENODEV;
2734 #endif
2735
2736         if (flags & MSG_ERRQUEUE) {
2737                 err = packet_recv_error(sk, msg, len);
2738                 goto out;
2739         }
2740
2741         /*
2742          *      Call the generic datagram receiver. This handles all sorts
2743          *      of horrible races and re-entrancy so we can forget about it
2744          *      in the protocol layers.
2745          *
2746          *      Now it will return ENETDOWN, if device have just gone down,
2747          *      but then it will block.
2748          */
2749
2750         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2751
2752         /*
2753          *      An error occurred so return it. Because skb_recv_datagram()
2754          *      handles the blocking we don't see and worry about blocking
2755          *      retries.
2756          */
2757
2758         if (skb == NULL)
2759                 goto out;
2760
2761         if (pkt_sk(sk)->has_vnet_hdr) {
2762                 struct virtio_net_hdr vnet_hdr = { 0 };
2763
2764                 err = -EINVAL;
2765                 vnet_hdr_len = sizeof(vnet_hdr);
2766                 if (len < vnet_hdr_len)
2767                         goto out_free;
2768
2769                 len -= vnet_hdr_len;
2770
2771                 if (skb_is_gso(skb)) {
2772                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2773
2774                         /* This is a hint as to how much should be linear. */
2775                         vnet_hdr.hdr_len = skb_headlen(skb);
2776                         vnet_hdr.gso_size = sinfo->gso_size;
2777                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2778                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2779                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2780                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2781                         else if (sinfo->gso_type & SKB_GSO_UDP)
2782                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2783                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2784                                 goto out_free;
2785                         else
2786                                 BUG();
2787                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2788                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2789                 } else
2790                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2791
2792                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2793                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2794                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2795                         vnet_hdr.csum_offset = skb->csum_offset;
2796                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2797                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2798                 } /* else everything is zero */
2799
2800                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2801                                      vnet_hdr_len);
2802                 if (err < 0)
2803                         goto out_free;
2804         }
2805
2806         /* You lose any data beyond the buffer you gave. If it worries
2807          * a user program they can ask the device for its MTU
2808          * anyway.
2809          */
2810         copied = skb->len;
2811         if (copied > len) {
2812                 copied = len;
2813                 msg->msg_flags |= MSG_TRUNC;
2814         }
2815
2816         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2817         if (err)
2818                 goto out_free;
2819
2820         sock_recv_ts_and_drops(msg, sk, skb);
2821
2822         if (msg->msg_name) {
2823                 /* If the address length field is there to be filled
2824                  * in, we fill it in now.
2825                  */
2826                 if (sock->type == SOCK_PACKET) {
2827                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
2828                 } else {
2829                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2830                         msg->msg_namelen = sll->sll_halen +
2831                                 offsetof(struct sockaddr_ll, sll_addr);
2832                 }
2833                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2834                        msg->msg_namelen);
2835         }
2836
2837         if (pkt_sk(sk)->auxdata) {
2838                 struct tpacket_auxdata aux;
2839
2840                 aux.tp_status = TP_STATUS_USER;
2841                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2842                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2843                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2844                 aux.tp_snaplen = skb->len;
2845                 aux.tp_mac = 0;
2846                 aux.tp_net = skb_network_offset(skb);
2847                 if (vlan_tx_tag_present(skb)) {
2848                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2849                         aux.tp_status |= TP_STATUS_VLAN_VALID;
2850                 } else {
2851                         aux.tp_vlan_tci = 0;
2852                 }
2853                 aux.tp_padding = 0;
2854                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2855         }
2856
2857         /*
2858          *      Free or return the buffer as appropriate. Again this
2859          *      hides all the races and re-entrancy issues from us.
2860          */
2861         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2862
2863 out_free:
2864         skb_free_datagram(sk, skb);
2865 out:
2866         return err;
2867 }
2868
2869 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2870                                int *uaddr_len, int peer)
2871 {
2872         struct net_device *dev;
2873         struct sock *sk = sock->sk;
2874
2875         if (peer)
2876                 return -EOPNOTSUPP;
2877
2878         uaddr->sa_family = AF_PACKET;
2879         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
2880         rcu_read_lock();
2881         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2882         if (dev)
2883                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
2884         rcu_read_unlock();
2885         *uaddr_len = sizeof(*uaddr);
2886
2887         return 0;
2888 }
2889
2890 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2891                           int *uaddr_len, int peer)
2892 {
2893         struct net_device *dev;
2894         struct sock *sk = sock->sk;
2895         struct packet_sock *po = pkt_sk(sk);
2896         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2897
2898         if (peer)
2899                 return -EOPNOTSUPP;
2900
2901         sll->sll_family = AF_PACKET;
2902         sll->sll_ifindex = po->ifindex;
2903         sll->sll_protocol = po->num;
2904         sll->sll_pkttype = 0;
2905         rcu_read_lock();
2906         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2907         if (dev) {
2908                 sll->sll_hatype = dev->type;
2909                 sll->sll_halen = dev->addr_len;
2910                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2911         } else {
2912                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
2913                 sll->sll_halen = 0;
2914         }
2915         rcu_read_unlock();
2916         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2917
2918         return 0;
2919 }
2920
2921 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2922                          int what)
2923 {
2924         switch (i->type) {
2925         case PACKET_MR_MULTICAST:
2926                 if (i->alen != dev->addr_len)
2927                         return -EINVAL;
2928                 if (what > 0)
2929                         return dev_mc_add(dev, i->addr);
2930                 else
2931                         return dev_mc_del(dev, i->addr);
2932                 break;
2933         case PACKET_MR_PROMISC:
2934                 return dev_set_promiscuity(dev, what);
2935                 break;
2936         case PACKET_MR_ALLMULTI:
2937                 return dev_set_allmulti(dev, what);
2938                 break;
2939         case PACKET_MR_UNICAST:
2940                 if (i->alen != dev->addr_len)
2941                         return -EINVAL;
2942                 if (what > 0)
2943                         return dev_uc_add(dev, i->addr);
2944                 else
2945                         return dev_uc_del(dev, i->addr);
2946                 break;
2947         default:
2948                 break;
2949         }
2950         return 0;
2951 }
2952
2953 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2954 {
2955         for ( ; i; i = i->next) {
2956                 if (i->ifindex == dev->ifindex)
2957                         packet_dev_mc(dev, i, what);
2958         }
2959 }
2960
2961 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2962 {
2963         struct packet_sock *po = pkt_sk(sk);
2964         struct packet_mclist *ml, *i;
2965         struct net_device *dev;
2966         int err;
2967
2968         rtnl_lock();
2969
2970         err = -ENODEV;
2971         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2972         if (!dev)
2973                 goto done;
2974
2975         err = -EINVAL;
2976         if (mreq->mr_alen > dev->addr_len)
2977                 goto done;
2978
2979         err = -ENOBUFS;
2980         i = kmalloc(sizeof(*i), GFP_KERNEL);
2981         if (i == NULL)
2982                 goto done;
2983
2984         err = 0;
2985         for (ml = po->mclist; ml; ml = ml->next) {
2986                 if (ml->ifindex == mreq->mr_ifindex &&
2987                     ml->type == mreq->mr_type &&
2988                     ml->alen == mreq->mr_alen &&
2989                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2990                         ml->count++;
2991                         /* Free the new element ... */
2992                         kfree(i);
2993                         goto done;
2994                 }
2995         }
2996
2997         i->type = mreq->mr_type;
2998         i->ifindex = mreq->mr_ifindex;
2999         i->alen = mreq->mr_alen;
3000         memcpy(i->addr, mreq->mr_address, i->alen);
3001         i->count = 1;
3002         i->next = po->mclist;
3003         po->mclist = i;
3004         err = packet_dev_mc(dev, i, 1);
3005         if (err) {
3006                 po->mclist = i->next;
3007                 kfree(i);
3008         }
3009
3010 done:
3011         rtnl_unlock();
3012         return err;
3013 }
3014
3015 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3016 {
3017         struct packet_mclist *ml, **mlp;
3018
3019         rtnl_lock();
3020
3021         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3022                 if (ml->ifindex == mreq->mr_ifindex &&
3023                     ml->type == mreq->mr_type &&
3024                     ml->alen == mreq->mr_alen &&
3025                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3026                         if (--ml->count == 0) {
3027                                 struct net_device *dev;
3028                                 *mlp = ml->next;
3029                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3030                                 if (dev)
3031                                         packet_dev_mc(dev, ml, -1);
3032                                 kfree(ml);
3033                         }
3034                         rtnl_unlock();
3035                         return 0;
3036                 }
3037         }
3038         rtnl_unlock();
3039         return -EADDRNOTAVAIL;
3040 }
3041
3042 static void packet_flush_mclist(struct sock *sk)
3043 {
3044         struct packet_sock *po = pkt_sk(sk);
3045         struct packet_mclist *ml;
3046
3047         if (!po->mclist)
3048                 return;
3049
3050         rtnl_lock();
3051         while ((ml = po->mclist) != NULL) {
3052                 struct net_device *dev;
3053
3054                 po->mclist = ml->next;
3055                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3056                 if (dev != NULL)
3057                         packet_dev_mc(dev, ml, -1);
3058                 kfree(ml);
3059         }
3060         rtnl_unlock();
3061 }
3062
3063 static int
3064 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3065 {
3066         struct sock *sk = sock->sk;
3067         struct packet_sock *po = pkt_sk(sk);
3068         int ret;
3069
3070         if (level != SOL_PACKET)
3071                 return -ENOPROTOOPT;
3072
3073         switch (optname) {
3074         case PACKET_ADD_MEMBERSHIP:
3075         case PACKET_DROP_MEMBERSHIP:
3076         {
3077                 struct packet_mreq_max mreq;
3078                 int len = optlen;
3079                 memset(&mreq, 0, sizeof(mreq));
3080                 if (len < sizeof(struct packet_mreq))
3081                         return -EINVAL;
3082                 if (len > sizeof(mreq))
3083                         len = sizeof(mreq);
3084                 if (copy_from_user(&mreq, optval, len))
3085                         return -EFAULT;
3086                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3087                         return -EINVAL;
3088                 if (optname == PACKET_ADD_MEMBERSHIP)
3089                         ret = packet_mc_add(sk, &mreq);
3090                 else
3091                         ret = packet_mc_drop(sk, &mreq);
3092                 return ret;
3093         }
3094
3095         case PACKET_RX_RING:
3096         case PACKET_TX_RING:
3097         {
3098                 union tpacket_req_u req_u;
3099                 int len;
3100
3101                 switch (po->tp_version) {
3102                 case TPACKET_V1:
3103                 case TPACKET_V2:
3104                         len = sizeof(req_u.req);
3105                         break;
3106                 case TPACKET_V3:
3107                 default:
3108                         len = sizeof(req_u.req3);
3109                         break;
3110                 }
3111                 if (optlen < len)
3112                         return -EINVAL;
3113                 if (pkt_sk(sk)->has_vnet_hdr)
3114                         return -EINVAL;
3115                 if (copy_from_user(&req_u.req, optval, len))
3116                         return -EFAULT;
3117                 return packet_set_ring(sk, &req_u, 0,
3118                         optname == PACKET_TX_RING);
3119         }
3120         case PACKET_COPY_THRESH:
3121         {
3122                 int val;
3123
3124                 if (optlen != sizeof(val))
3125                         return -EINVAL;
3126                 if (copy_from_user(&val, optval, sizeof(val)))
3127                         return -EFAULT;
3128
3129                 pkt_sk(sk)->copy_thresh = val;
3130                 return 0;
3131         }
3132         case PACKET_VERSION:
3133         {
3134                 int val;
3135
3136                 if (optlen != sizeof(val))
3137                         return -EINVAL;
3138                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3139                         return -EBUSY;
3140                 if (copy_from_user(&val, optval, sizeof(val)))
3141                         return -EFAULT;
3142                 switch (val) {
3143                 case TPACKET_V1:
3144                 case TPACKET_V2:
3145                 case TPACKET_V3:
3146                         po->tp_version = val;
3147                         return 0;
3148                 default:
3149                         return -EINVAL;
3150                 }
3151         }
3152         case PACKET_RESERVE:
3153         {
3154                 unsigned int val;
3155
3156                 if (optlen != sizeof(val))
3157                         return -EINVAL;
3158                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3159                         return -EBUSY;
3160                 if (copy_from_user(&val, optval, sizeof(val)))
3161                         return -EFAULT;
3162                 po->tp_reserve = val;
3163                 return 0;
3164         }
3165         case PACKET_LOSS:
3166         {
3167                 unsigned int val;
3168
3169                 if (optlen != sizeof(val))
3170                         return -EINVAL;
3171                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3172                         return -EBUSY;
3173                 if (copy_from_user(&val, optval, sizeof(val)))
3174                         return -EFAULT;
3175                 po->tp_loss = !!val;
3176                 return 0;
3177         }
3178         case PACKET_AUXDATA:
3179         {
3180                 int val;
3181
3182                 if (optlen < sizeof(val))
3183                         return -EINVAL;
3184                 if (copy_from_user(&val, optval, sizeof(val)))
3185                         return -EFAULT;
3186
3187                 po->auxdata = !!val;
3188                 return 0;
3189         }
3190         case PACKET_ORIGDEV:
3191         {
3192                 int val;
3193
3194                 if (optlen < sizeof(val))
3195                         return -EINVAL;
3196                 if (copy_from_user(&val, optval, sizeof(val)))
3197                         return -EFAULT;
3198
3199                 po->origdev = !!val;
3200                 return 0;
3201         }
3202         case PACKET_VNET_HDR:
3203         {
3204                 int val;
3205
3206                 if (sock->type != SOCK_RAW)
3207                         return -EINVAL;
3208                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3209                         return -EBUSY;
3210                 if (optlen < sizeof(val))
3211                         return -EINVAL;
3212                 if (copy_from_user(&val, optval, sizeof(val)))
3213                         return -EFAULT;
3214
3215                 po->has_vnet_hdr = !!val;
3216                 return 0;
3217         }
3218         case PACKET_TIMESTAMP:
3219         {
3220                 int val;
3221
3222                 if (optlen != sizeof(val))
3223                         return -EINVAL;
3224                 if (copy_from_user(&val, optval, sizeof(val)))
3225                         return -EFAULT;
3226
3227                 po->tp_tstamp = val;
3228                 return 0;
3229         }
3230         case PACKET_FANOUT:
3231         {
3232                 int val;
3233
3234                 if (optlen != sizeof(val))
3235                         return -EINVAL;
3236                 if (copy_from_user(&val, optval, sizeof(val)))
3237                         return -EFAULT;
3238
3239                 return fanout_add(sk, val & 0xffff, val >> 16);
3240         }
3241         case PACKET_TX_HAS_OFF:
3242         {
3243                 unsigned int val;
3244
3245                 if (optlen != sizeof(val))
3246                         return -EINVAL;
3247                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3248                         return -EBUSY;
3249                 if (copy_from_user(&val, optval, sizeof(val)))
3250                         return -EFAULT;
3251                 po->tp_tx_has_off = !!val;
3252                 return 0;
3253         }
3254         default:
3255                 return -ENOPROTOOPT;
3256         }
3257 }
3258
3259 static int packet_getsockopt(struct socket *sock, int level, int optname,
3260                              char __user *optval, int __user *optlen)
3261 {
3262         int len;
3263         int val, lv = sizeof(val);
3264         struct sock *sk = sock->sk;
3265         struct packet_sock *po = pkt_sk(sk);
3266         void *data = &val;
3267         union tpacket_stats_u st;
3268
3269         if (level != SOL_PACKET)
3270                 return -ENOPROTOOPT;
3271
3272         if (get_user(len, optlen))
3273                 return -EFAULT;
3274
3275         if (len < 0)
3276                 return -EINVAL;
3277
3278         switch (optname) {
3279         case PACKET_STATISTICS:
3280                 spin_lock_bh(&sk->sk_receive_queue.lock);
3281                 memcpy(&st, &po->stats, sizeof(st));
3282                 memset(&po->stats, 0, sizeof(po->stats));
3283                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3284
3285                 if (po->tp_version == TPACKET_V3) {
3286                         lv = sizeof(struct tpacket_stats_v3);
3287                         st.stats3.tp_packets += st.stats3.tp_drops;
3288                         data = &st.stats3;
3289                 } else {
3290                         lv = sizeof(struct tpacket_stats);
3291                         st.stats1.tp_packets += st.stats1.tp_drops;
3292                         data = &st.stats1;
3293                 }
3294
3295                 break;
3296         case PACKET_AUXDATA:
3297                 val = po->auxdata;
3298                 break;
3299         case PACKET_ORIGDEV:
3300                 val = po->origdev;
3301                 break;
3302         case PACKET_VNET_HDR:
3303                 val = po->has_vnet_hdr;
3304                 break;
3305         case PACKET_VERSION:
3306                 val = po->tp_version;
3307                 break;
3308         case PACKET_HDRLEN:
3309                 if (len > sizeof(int))
3310                         len = sizeof(int);
3311                 if (copy_from_user(&val, optval, len))
3312                         return -EFAULT;
3313                 switch (val) {
3314                 case TPACKET_V1:
3315                         val = sizeof(struct tpacket_hdr);
3316                         break;
3317                 case TPACKET_V2:
3318                         val = sizeof(struct tpacket2_hdr);
3319                         break;
3320                 case TPACKET_V3:
3321                         val = sizeof(struct tpacket3_hdr);
3322                         break;
3323                 default:
3324                         return -EINVAL;
3325                 }
3326                 break;
3327         case PACKET_RESERVE:
3328                 val = po->tp_reserve;
3329                 break;
3330         case PACKET_LOSS:
3331                 val = po->tp_loss;
3332                 break;
3333         case PACKET_TIMESTAMP:
3334                 val = po->tp_tstamp;
3335                 break;
3336         case PACKET_FANOUT:
3337                 val = (po->fanout ?
3338                        ((u32)po->fanout->id |
3339                         ((u32)po->fanout->type << 16) |
3340                         ((u32)po->fanout->flags << 24)) :
3341                        0);
3342                 break;
3343         case PACKET_TX_HAS_OFF:
3344                 val = po->tp_tx_has_off;
3345                 break;
3346         default:
3347                 return -ENOPROTOOPT;
3348         }
3349
3350         if (len > lv)
3351                 len = lv;
3352         if (put_user(len, optlen))
3353                 return -EFAULT;
3354         if (copy_to_user(optval, data, len))
3355                 return -EFAULT;
3356         return 0;
3357 }
3358
3359
3360 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3361 {
3362         struct sock *sk;
3363         struct net_device *dev = data;
3364         struct net *net = dev_net(dev);
3365
3366         rcu_read_lock();
3367         sk_for_each_rcu(sk, &net->packet.sklist) {
3368                 struct packet_sock *po = pkt_sk(sk);
3369
3370                 switch (msg) {
3371                 case NETDEV_UNREGISTER:
3372                         if (po->mclist)
3373                                 packet_dev_mclist(dev, po->mclist, -1);
3374                         /* fallthrough */
3375
3376                 case NETDEV_DOWN:
3377                         if (dev->ifindex == po->ifindex) {
3378                                 spin_lock(&po->bind_lock);
3379                                 if (po->running) {
3380                                         __unregister_prot_hook(sk, false);
3381                                         sk->sk_err = ENETDOWN;
3382                                         if (!sock_flag(sk, SOCK_DEAD))
3383                                                 sk->sk_error_report(sk);
3384                                 }
3385                                 if (msg == NETDEV_UNREGISTER) {
3386                                         packet_cached_dev_reset(po);
3387                                         po->ifindex = -1;
3388                                         if (po->prot_hook.dev)
3389                                                 dev_put(po->prot_hook.dev);
3390                                         po->prot_hook.dev = NULL;
3391                                 }
3392                                 spin_unlock(&po->bind_lock);
3393                         }
3394                         break;
3395                 case NETDEV_UP:
3396                         if (dev->ifindex == po->ifindex) {
3397                                 spin_lock(&po->bind_lock);
3398                                 if (po->num)
3399                                         register_prot_hook(sk);
3400                                 spin_unlock(&po->bind_lock);
3401                         }
3402                         break;
3403                 }
3404         }
3405         rcu_read_unlock();
3406         return NOTIFY_DONE;
3407 }
3408
3409
3410 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3411                         unsigned long arg)
3412 {
3413         struct sock *sk = sock->sk;
3414
3415         switch (cmd) {
3416         case SIOCOUTQ:
3417         {
3418                 int amount = sk_wmem_alloc_get(sk);
3419
3420                 return put_user(amount, (int __user *)arg);
3421         }
3422         case SIOCINQ:
3423         {
3424                 struct sk_buff *skb;
3425                 int amount = 0;
3426
3427                 spin_lock_bh(&sk->sk_receive_queue.lock);
3428                 skb = skb_peek(&sk->sk_receive_queue);
3429                 if (skb)
3430                         amount = skb->len;
3431                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3432                 return put_user(amount, (int __user *)arg);
3433         }
3434         case SIOCGSTAMP:
3435                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3436         case SIOCGSTAMPNS:
3437                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3438
3439 #ifdef CONFIG_INET
3440         case SIOCADDRT:
3441         case SIOCDELRT:
3442         case SIOCDARP:
3443         case SIOCGARP:
3444         case SIOCSARP:
3445         case SIOCGIFADDR:
3446         case SIOCSIFADDR:
3447         case SIOCGIFBRDADDR:
3448         case SIOCSIFBRDADDR:
3449         case SIOCGIFNETMASK:
3450         case SIOCSIFNETMASK:
3451         case SIOCGIFDSTADDR:
3452         case SIOCSIFDSTADDR:
3453         case SIOCSIFFLAGS:
3454                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3455 #endif
3456
3457         default:
3458                 return -ENOIOCTLCMD;
3459         }
3460         return 0;
3461 }
3462
3463 static unsigned int packet_poll(struct file *file, struct socket *sock,
3464                                 poll_table *wait)
3465 {
3466         struct sock *sk = sock->sk;
3467         struct packet_sock *po = pkt_sk(sk);
3468         unsigned int mask = datagram_poll(file, sock, wait);
3469
3470         spin_lock_bh(&sk->sk_receive_queue.lock);
3471         if (po->rx_ring.pg_vec) {
3472                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3473                         TP_STATUS_KERNEL))
3474                         mask |= POLLIN | POLLRDNORM;
3475         }
3476         spin_unlock_bh(&sk->sk_receive_queue.lock);
3477         spin_lock_bh(&sk->sk_write_queue.lock);
3478         if (po->tx_ring.pg_vec) {
3479                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3480                         mask |= POLLOUT | POLLWRNORM;
3481         }
3482         spin_unlock_bh(&sk->sk_write_queue.lock);
3483         return mask;
3484 }
3485
3486
3487 /* Dirty? Well, I still did not learn better way to account
3488  * for user mmaps.
3489  */
3490
3491 static void packet_mm_open(struct vm_area_struct *vma)
3492 {
3493         struct file *file = vma->vm_file;
3494         struct socket *sock = file->private_data;
3495         struct sock *sk = sock->sk;
3496
3497         if (sk)
3498                 atomic_inc(&pkt_sk(sk)->mapped);
3499 }
3500
3501 static void packet_mm_close(struct vm_area_struct *vma)
3502 {
3503         struct file *file = vma->vm_file;
3504         struct socket *sock = file->private_data;
3505         struct sock *sk = sock->sk;
3506
3507         if (sk)
3508                 atomic_dec(&pkt_sk(sk)->mapped);
3509 }
3510
3511 static const struct vm_operations_struct packet_mmap_ops = {
3512         .open   =       packet_mm_open,
3513         .close  =       packet_mm_close,
3514 };
3515
3516 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3517                         unsigned int len)
3518 {
3519         int i;
3520
3521         for (i = 0; i < len; i++) {
3522                 if (likely(pg_vec[i].buffer)) {
3523                         if (is_vmalloc_addr(pg_vec[i].buffer))
3524                                 vfree(pg_vec[i].buffer);
3525                         else
3526                                 free_pages((unsigned long)pg_vec[i].buffer,
3527                                            order);
3528                         pg_vec[i].buffer = NULL;
3529                 }
3530         }
3531         kfree(pg_vec);
3532 }
3533
3534 static char *alloc_one_pg_vec_page(unsigned long order)
3535 {
3536         char *buffer = NULL;
3537         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3538                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3539
3540         buffer = (char *) __get_free_pages(gfp_flags, order);
3541
3542         if (buffer)
3543                 return buffer;
3544
3545         /*
3546          * __get_free_pages failed, fall back to vmalloc
3547          */
3548         buffer = vzalloc((1 << order) * PAGE_SIZE);
3549
3550         if (buffer)
3551                 return buffer;
3552
3553         /*
3554          * vmalloc failed, lets dig into swap here
3555          */
3556         gfp_flags &= ~__GFP_NORETRY;
3557         buffer = (char *)__get_free_pages(gfp_flags, order);
3558         if (buffer)
3559                 return buffer;
3560
3561         /*
3562          * complete and utter failure
3563          */
3564         return NULL;
3565 }
3566
3567 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3568 {
3569         unsigned int block_nr = req->tp_block_nr;
3570         struct pgv *pg_vec;
3571         int i;
3572
3573         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3574         if (unlikely(!pg_vec))
3575                 goto out;
3576
3577         for (i = 0; i < block_nr; i++) {
3578                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3579                 if (unlikely(!pg_vec[i].buffer))
3580                         goto out_free_pgvec;
3581         }
3582
3583 out:
3584         return pg_vec;
3585
3586 out_free_pgvec:
3587         free_pg_vec(pg_vec, order, block_nr);
3588         pg_vec = NULL;
3589         goto out;
3590 }
3591
3592 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3593                 int closing, int tx_ring)
3594 {
3595         struct pgv *pg_vec = NULL;
3596         struct packet_sock *po = pkt_sk(sk);
3597         int was_running, order = 0;
3598         struct packet_ring_buffer *rb;
3599         struct sk_buff_head *rb_queue;
3600         __be16 num;
3601         int err = -EINVAL;
3602         /* Added to avoid minimal code churn */
3603         struct tpacket_req *req = &req_u->req;
3604
3605         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3606         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3607                 WARN(1, "Tx-ring is not supported.\n");
3608                 goto out;
3609         }
3610
3611         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3612         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3613
3614         err = -EBUSY;
3615         if (!closing) {
3616                 if (atomic_read(&po->mapped))
3617                         goto out;
3618                 if (atomic_read(&rb->pending))
3619                         goto out;
3620         }
3621
3622         if (req->tp_block_nr) {
3623                 /* Sanity tests and some calculations */
3624                 err = -EBUSY;
3625                 if (unlikely(rb->pg_vec))
3626                         goto out;
3627
3628                 switch (po->tp_version) {
3629                 case TPACKET_V1:
3630                         po->tp_hdrlen = TPACKET_HDRLEN;
3631                         break;
3632                 case TPACKET_V2:
3633                         po->tp_hdrlen = TPACKET2_HDRLEN;
3634                         break;
3635                 case TPACKET_V3:
3636                         po->tp_hdrlen = TPACKET3_HDRLEN;
3637                         break;
3638                 }
3639
3640                 err = -EINVAL;
3641                 if (unlikely((int)req->tp_block_size <= 0))
3642                         goto out;
3643                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3644                         goto out;
3645                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3646                                         po->tp_reserve))
3647                         goto out;
3648                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3649                         goto out;
3650
3651                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3652                 if (unlikely(rb->frames_per_block <= 0))
3653                         goto out;
3654                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3655                                         req->tp_frame_nr))
3656                         goto out;
3657
3658                 err = -ENOMEM;
3659                 order = get_order(req->tp_block_size);
3660                 pg_vec = alloc_pg_vec(req, order);
3661                 if (unlikely(!pg_vec))
3662                         goto out;
3663                 switch (po->tp_version) {
3664                 case TPACKET_V3:
3665                 /* Transmit path is not supported. We checked
3666                  * it above but just being paranoid
3667                  */
3668                         if (!tx_ring)
3669                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3670                                 break;
3671                 default:
3672                         break;
3673                 }
3674         }
3675         /* Done */
3676         else {
3677                 err = -EINVAL;
3678                 if (unlikely(req->tp_frame_nr))
3679                         goto out;
3680         }
3681
3682         lock_sock(sk);
3683
3684         /* Detach socket from network */
3685         spin_lock(&po->bind_lock);
3686         was_running = po->running;
3687         num = po->num;
3688         if (was_running) {
3689                 po->num = 0;
3690                 __unregister_prot_hook(sk, false);
3691         }
3692         spin_unlock(&po->bind_lock);
3693
3694         synchronize_net();
3695
3696         err = -EBUSY;
3697         mutex_lock(&po->pg_vec_lock);
3698         if (closing || atomic_read(&po->mapped) == 0) {
3699                 err = 0;
3700                 spin_lock_bh(&rb_queue->lock);
3701                 swap(rb->pg_vec, pg_vec);
3702                 rb->frame_max = (req->tp_frame_nr - 1);
3703                 rb->head = 0;
3704                 rb->frame_size = req->tp_frame_size;
3705                 spin_unlock_bh(&rb_queue->lock);
3706
3707                 swap(rb->pg_vec_order, order);
3708                 swap(rb->pg_vec_len, req->tp_block_nr);
3709
3710                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3711                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3712                                                 tpacket_rcv : packet_rcv;
3713                 skb_queue_purge(rb_queue);
3714                 if (atomic_read(&po->mapped))
3715                         pr_err("packet_mmap: vma is busy: %d\n",
3716                                atomic_read(&po->mapped));
3717         }
3718         mutex_unlock(&po->pg_vec_lock);
3719
3720         spin_lock(&po->bind_lock);
3721         if (was_running) {
3722                 po->num = num;
3723                 register_prot_hook(sk);
3724         }
3725         spin_unlock(&po->bind_lock);
3726         if (closing && (po->tp_version > TPACKET_V2)) {
3727                 /* Because we don't support block-based V3 on tx-ring */
3728                 if (!tx_ring)
3729                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3730         }
3731         release_sock(sk);
3732
3733         if (pg_vec)
3734                 free_pg_vec(pg_vec, order, req->tp_block_nr);
3735 out:
3736         return err;
3737 }
3738
3739 static int packet_mmap(struct file *file, struct socket *sock,
3740                 struct vm_area_struct *vma)
3741 {
3742         struct sock *sk = sock->sk;
3743         struct packet_sock *po = pkt_sk(sk);
3744         unsigned long size, expected_size;
3745         struct packet_ring_buffer *rb;
3746         unsigned long start;
3747         int err = -EINVAL;
3748         int i;
3749
3750         if (vma->vm_pgoff)
3751                 return -EINVAL;
3752
3753         mutex_lock(&po->pg_vec_lock);
3754
3755         expected_size = 0;
3756         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3757                 if (rb->pg_vec) {
3758                         expected_size += rb->pg_vec_len
3759                                                 * rb->pg_vec_pages
3760                                                 * PAGE_SIZE;
3761                 }
3762         }
3763
3764         if (expected_size == 0)
3765                 goto out;
3766
3767         size = vma->vm_end - vma->vm_start;
3768         if (size != expected_size)
3769                 goto out;
3770
3771         start = vma->vm_start;
3772         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3773                 if (rb->pg_vec == NULL)
3774                         continue;
3775
3776                 for (i = 0; i < rb->pg_vec_len; i++) {
3777                         struct page *page;
3778                         void *kaddr = rb->pg_vec[i].buffer;
3779                         int pg_num;
3780
3781                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3782                                 page = pgv_to_page(kaddr);
3783                                 err = vm_insert_page(vma, start, page);
3784                                 if (unlikely(err))
3785                                         goto out;
3786                                 start += PAGE_SIZE;
3787                                 kaddr += PAGE_SIZE;
3788                         }
3789                 }
3790         }
3791
3792         atomic_inc(&po->mapped);
3793         vma->vm_ops = &packet_mmap_ops;
3794         err = 0;
3795
3796 out:
3797         mutex_unlock(&po->pg_vec_lock);
3798         return err;
3799 }
3800
3801 static const struct proto_ops packet_ops_spkt = {
3802         .family =       PF_PACKET,
3803         .owner =        THIS_MODULE,
3804         .release =      packet_release,
3805         .bind =         packet_bind_spkt,
3806         .connect =      sock_no_connect,
3807         .socketpair =   sock_no_socketpair,
3808         .accept =       sock_no_accept,
3809         .getname =      packet_getname_spkt,
3810         .poll =         datagram_poll,
3811         .ioctl =        packet_ioctl,
3812         .listen =       sock_no_listen,
3813         .shutdown =     sock_no_shutdown,
3814         .setsockopt =   sock_no_setsockopt,
3815         .getsockopt =   sock_no_getsockopt,
3816         .sendmsg =      packet_sendmsg_spkt,
3817         .recvmsg =      packet_recvmsg,
3818         .mmap =         sock_no_mmap,
3819         .sendpage =     sock_no_sendpage,
3820 };
3821
3822 static const struct proto_ops packet_ops = {
3823         .family =       PF_PACKET,
3824         .owner =        THIS_MODULE,
3825         .release =      packet_release,
3826         .bind =         packet_bind,
3827         .connect =      sock_no_connect,
3828         .socketpair =   sock_no_socketpair,
3829         .accept =       sock_no_accept,
3830         .getname =      packet_getname,
3831         .poll =         packet_poll,
3832         .ioctl =        packet_ioctl,
3833         .listen =       sock_no_listen,
3834         .shutdown =     sock_no_shutdown,
3835         .setsockopt =   packet_setsockopt,
3836         .getsockopt =   packet_getsockopt,
3837         .sendmsg =      packet_sendmsg,
3838         .recvmsg =      packet_recvmsg,
3839         .mmap =         packet_mmap,
3840         .sendpage =     sock_no_sendpage,
3841 };
3842
3843 static const struct net_proto_family packet_family_ops = {
3844         .family =       PF_PACKET,
3845         .create =       packet_create,
3846         .owner  =       THIS_MODULE,
3847 };
3848
3849 static struct notifier_block packet_netdev_notifier = {
3850         .notifier_call =        packet_notifier,
3851 };
3852
3853 #ifdef CONFIG_PROC_FS
3854
3855 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3856         __acquires(RCU)
3857 {
3858         struct net *net = seq_file_net(seq);
3859
3860         rcu_read_lock();
3861         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3862 }
3863
3864 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3865 {
3866         struct net *net = seq_file_net(seq);
3867         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3868 }
3869
3870 static void packet_seq_stop(struct seq_file *seq, void *v)
3871         __releases(RCU)
3872 {
3873         rcu_read_unlock();
3874 }
3875
3876 static int packet_seq_show(struct seq_file *seq, void *v)
3877 {
3878         if (v == SEQ_START_TOKEN)
3879                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
3880         else {
3881                 struct sock *s = sk_entry(v);
3882                 const struct packet_sock *po = pkt_sk(s);
3883
3884                 seq_printf(seq,
3885                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
3886                            s,
3887                            atomic_read(&s->sk_refcnt),
3888                            s->sk_type,
3889                            ntohs(po->num),
3890                            po->ifindex,
3891                            po->running,
3892                            atomic_read(&s->sk_rmem_alloc),
3893                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
3894                            sock_i_ino(s));
3895         }
3896
3897         return 0;
3898 }
3899
3900 static const struct seq_operations packet_seq_ops = {
3901         .start  = packet_seq_start,
3902         .next   = packet_seq_next,
3903         .stop   = packet_seq_stop,
3904         .show   = packet_seq_show,
3905 };
3906
3907 static int packet_seq_open(struct inode *inode, struct file *file)
3908 {
3909         return seq_open_net(inode, file, &packet_seq_ops,
3910                             sizeof(struct seq_net_private));
3911 }
3912
3913 static const struct file_operations packet_seq_fops = {
3914         .owner          = THIS_MODULE,
3915         .open           = packet_seq_open,
3916         .read           = seq_read,
3917         .llseek         = seq_lseek,
3918         .release        = seq_release_net,
3919 };
3920
3921 #endif
3922
3923 static int __net_init packet_net_init(struct net *net)
3924 {
3925         mutex_init(&net->packet.sklist_lock);
3926         INIT_HLIST_HEAD(&net->packet.sklist);
3927
3928         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
3929                 return -ENOMEM;
3930
3931         return 0;
3932 }
3933
3934 static void __net_exit packet_net_exit(struct net *net)
3935 {
3936         remove_proc_entry("packet", net->proc_net);
3937 }
3938
3939 static struct pernet_operations packet_net_ops = {
3940         .init = packet_net_init,
3941         .exit = packet_net_exit,
3942 };
3943
3944
3945 static void __exit packet_exit(void)
3946 {
3947         unregister_netdevice_notifier(&packet_netdev_notifier);
3948         unregister_pernet_subsys(&packet_net_ops);
3949         sock_unregister(PF_PACKET);
3950         proto_unregister(&packet_proto);
3951 }
3952
3953 static int __init packet_init(void)
3954 {
3955         int rc = proto_register(&packet_proto, 0);
3956
3957         if (rc != 0)
3958                 goto out;
3959
3960         sock_register(&packet_family_ops);
3961         register_pernet_subsys(&packet_net_ops);
3962         register_netdevice_notifier(&packet_netdev_notifier);
3963 out:
3964         return rc;
3965 }
3966
3967 module_init(packet_init);
3968 module_exit(packet_exit);
3969 MODULE_LICENSE("GPL");
3970 MODULE_ALIAS_NETPROTO(PF_PACKET);