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/system.h>
  77 #include <asm/uaccess.h>
  78 #include <asm/ioctls.h>
  79 #include <asm/page.h>
  80 #include <asm/cacheflush.h>
  81 #include <asm/io.h>
  82 #include <linux/proc_fs.h>
  83 #include <linux/seq_file.h>
  84 #include <linux/poll.h>
  85 #include <linux/module.h>
  86 #include <linux/init.h>
  87 #include <linux/mutex.h>
  88 #include <linux/if_vlan.h>
  89 #include <linux/virtio_net.h>
  90 #include <linux/errqueue.h>
  91 #include <linux/net_tstamp.h>
  92
  93 #ifdef CONFIG_INET
  94 #include <net/inet_common.h>
  95 #endif
  96
  97 /*
  98    Assumptions:
  99    - if device has no dev->hard_header routine, it adds and removes ll header
 100      inside itself. In this case ll header is invisible outside of device,
 101      but higher levels still should reserve dev->hard_header_len.
 102      Some devices are enough clever to reallocate skb, when header
 103      will not fit to reserved space (tunnel), another ones are silly
 104      (PPP).
 105    - packet socket receives packets with pulled ll header,
 106      so that SOCK_RAW should push it back.
 107
 108 On receive:
 109 -----------
 110
 111 Incoming, dev->hard_header!=NULL
 112    mac_header -> ll header
 113    data       -> data
 114
 115 Outgoing, dev->hard_header!=NULL
 116    mac_header -> ll header
 117    data       -> ll header
 118
 119 Incoming, dev->hard_header==NULL
 120    mac_header -> UNKNOWN position. It is very likely, that it points to ll
 121                  header.  PPP makes it, that is wrong, because introduce
 122                  assymetry between rx and tx paths.
 123    data       -> data
 124
 125 Outgoing, dev->hard_header==NULL
 126    mac_header -> data. ll header is still not built!
 127    data       -> data
 128
 129 Resume
 130   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 131
 132
 133 On transmit:
 134 ------------
 135
 136 dev->hard_header != NULL
 137    mac_header -> ll header
 138    data       -> ll header
 139
 140 dev->hard_header == NULL (ll header is added by device, we cannot control it)
 141    mac_header -> data
 142    data       -> data
 143
 144    We should set nh.raw on output to correct posistion,
 145    packet classifier depends on it.
 146  */
 147
 148 /* Private packet socket structures. */
 149
 150 struct packet_mclist {
 151         struct packet_mclist    *next;
 152         int                     ifindex;
 153         int                     count;
 154         unsigned short          type;
 155         unsigned short          alen;
 156         unsigned char           addr[MAX_ADDR_LEN];
 157 };
 158 /* identical to struct packet_mreq except it has
 159  * a longer address field.
 160  */
 161 struct packet_mreq_max {
 162         int             mr_ifindex;
 163         unsigned short  mr_type;
 164         unsigned short  mr_alen;
 165         unsigned char   mr_address[MAX_ADDR_LEN];
 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
 172 #define V3_ALIGNMENT    (8)
 173
 174 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 175
 176 #define BLK_PLUS_PRIV(sz_of_priv) \
 177         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 178
 179 /* kbdq - kernel block descriptor queue */
 180 struct tpacket_kbdq_core {
 181         struct pgv      *pkbdq;
 182         unsigned int    feature_req_word;
 183         unsigned int    hdrlen;
 184         unsigned char   reset_pending_on_curr_blk;
 185         unsigned char   delete_blk_timer;
 186         unsigned short  kactive_blk_num;
 187         unsigned short  blk_sizeof_priv;
 188
 189         /* last_kactive_blk_num:
 190          * trick to see if user-space has caught up
 191          * in order to avoid refreshing timer when every single pkt arrives.
 192          */
 193         unsigned short  last_kactive_blk_num;
 194
 195         char            *pkblk_start;
 196         char            *pkblk_end;
 197         int             kblk_size;
 198         unsigned int    knum_blocks;
 199         uint64_t        knxt_seq_num;
 200         char            *prev;
 201         char            *nxt_offset;
 202         struct sk_buff  *skb;
 203
 204         atomic_t        blk_fill_in_prog;
 205
 206         /* Default is set to 8ms */
 207 #define DEFAULT_PRB_RETIRE_TOV  (8)
 208
 209         unsigned short  retire_blk_tov;
 210         unsigned short  version;
 211         unsigned long   tov_in_jiffies;
 212
 213         /* timer to retire an outstanding block */
 214         struct timer_list retire_blk_timer;
 215 };
 216
 217 #define PGV_FROM_VMALLOC 1
 218 struct pgv {
 219         char *buffer;
 220 };
 221
 222 struct packet_ring_buffer {
 223         struct pgv              *pg_vec;
 224         unsigned int            head;
 225         unsigned int            frames_per_block;
 226         unsigned int            frame_size;
 227         unsigned int            frame_max;
 228
 229         unsigned int            pg_vec_order;
 230         unsigned int            pg_vec_pages;
 231         unsigned int            pg_vec_len;
 232
 233         struct tpacket_kbdq_core        prb_bdqc;
 234         atomic_t                pending;
 235 };
 236
 237 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
 238 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
 239 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
 240 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
 241 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
 242 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
 243 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
 244
 245 struct packet_sock;
 246 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
 247
 248 static void *packet_previous_frame(struct packet_sock *po,
 249                 struct packet_ring_buffer *rb,
 250                 int status);
 251 static void packet_increment_head(struct packet_ring_buffer *buff);
 252 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
 253                         struct tpacket_block_desc *);
 254 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 255                         struct packet_sock *);
 256 static void prb_retire_current_block(struct tpacket_kbdq_core *,
 257                 struct packet_sock *, unsigned int status);
 258 static int prb_queue_frozen(struct tpacket_kbdq_core *);
 259 static void prb_open_block(struct tpacket_kbdq_core *,
 260                 struct tpacket_block_desc *);
 261 static void prb_retire_rx_blk_timer_expired(unsigned long);
 262 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 263 static void prb_init_blk_timer(struct packet_sock *,
 264                 struct tpacket_kbdq_core *,
 265                 void (*func) (unsigned long));
 266 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 267 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 268                 struct tpacket3_hdr *);
 269 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 270                 struct tpacket3_hdr *);
 271 static void packet_flush_mclist(struct sock *sk);
 272
 273 struct packet_fanout;
 274 struct packet_sock {
 275         /* struct sock has to be the first member of packet_sock */
 276         struct sock             sk;
 277         struct packet_fanout    *fanout;
 278         struct tpacket_stats    stats;
 279         union  tpacket_stats_u  stats_u;
 280         struct packet_ring_buffer       rx_ring;
 281         struct packet_ring_buffer       tx_ring;
 282         int                     copy_thresh;
 283         spinlock_t              bind_lock;
 284         struct mutex            pg_vec_lock;
 285         unsigned int            running:1,      /* prot_hook is attached*/
 286                                 auxdata:1,
 287                                 origdev:1,
 288                                 has_vnet_hdr:1;
 289         int                     ifindex;        /* bound device         */
 290         __be16                  num;
 291         struct packet_mclist    *mclist;
 292         atomic_t                mapped;
 293         enum tpacket_versions   tp_version;
 294         unsigned int            tp_hdrlen;
 295         unsigned int            tp_reserve;
 296         unsigned int            tp_loss:1;
 297         unsigned int            tp_tstamp;
 298         struct packet_type      prot_hook ____cacheline_aligned_in_smp;
 299 };
 300
 301 #define PACKET_FANOUT_MAX       256
 302
 303 struct packet_fanout {
 304 #ifdef CONFIG_NET_NS
 305         struct net              *net;
 306 #endif
 307         unsigned int            num_members;
 308         u16                     id;
 309         u8                      type;
 310         u8                      defrag;
 311         atomic_t                rr_cur;
 312         struct list_head        list;
 313         struct sock             *arr[PACKET_FANOUT_MAX];
 314         spinlock_t              lock;
 315         atomic_t                sk_ref;
 316         struct packet_type      prot_hook ____cacheline_aligned_in_smp;
 317 };
 318
 319 struct packet_skb_cb {
 320         unsigned int origlen;
 321         union {
 322                 struct sockaddr_pkt pkt;
 323                 struct sockaddr_ll ll;
 324         } sa;
 325 };
 326
 327 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 328
 329 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 330 #define GET_PBLOCK_DESC(x, bid) \
 331         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 332 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 333         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 334 #define GET_NEXT_PRB_BLK_NUM(x) \
 335         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 336         ((x)->kactive_blk_num+1) : 0)
 337
 338 static inline struct packet_sock *pkt_sk(struct sock *sk)
 339 {
 340         return (struct packet_sock *)sk;
 341 }
 342
 343 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 344 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 345
 346 /* register_prot_hook must be invoked with the po->bind_lock held,
 347  * or from a context in which asynchronous accesses to the packet
 348  * socket is not possible (packet_create()).
 349  */
 350 static void register_prot_hook(struct sock *sk)
 351 {
 352         struct packet_sock *po = pkt_sk(sk);
 353         if (!po->running) {
 354                 if (po->fanout)
 355                         __fanout_link(sk, po);
 356                 else
 357                         dev_add_pack(&po->prot_hook);
 358                 sock_hold(sk);
 359                 po->running = 1;
 360         }
 361 }
 362
 363 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
 364  * held.   If the sync parameter is true, we will temporarily drop
 365  * the po->bind_lock and do a synchronize_net to make sure no
 366  * asynchronous packet processing paths still refer to the elements
 367  * of po->prot_hook.  If the sync parameter is false, it is the
 368  * callers responsibility to take care of this.
 369  */
 370 static void __unregister_prot_hook(struct sock *sk, bool sync)
 371 {
 372         struct packet_sock *po = pkt_sk(sk);
 373
 374         po->running = 0;
 375         if (po->fanout)
 376                 __fanout_unlink(sk, po);
 377         else
 378                 __dev_remove_pack(&po->prot_hook);
 379         __sock_put(sk);
 380
 381         if (sync) {
 382                 spin_unlock(&po->bind_lock);
 383                 synchronize_net();
 384                 spin_lock(&po->bind_lock);
 385         }
 386 }
 387
 388 static void unregister_prot_hook(struct sock *sk, bool sync)
 389 {
 390         struct packet_sock *po = pkt_sk(sk);
 391
 392         if (po->running)
 393                 __unregister_prot_hook(sk, sync);
 394 }
 395
 396 static inline __pure struct page *pgv_to_page(void *addr)
 397 {
 398         if (is_vmalloc_addr(addr))
 399                 return vmalloc_to_page(addr);
 400         return virt_to_page(addr);
 401 }
 402
 403 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 404 {
 405         union {
 406                 struct tpacket_hdr *h1;
 407                 struct tpacket2_hdr *h2;
 408                 void *raw;
 409         } h;
 410
 411         h.raw = frame;
 412         switch (po->tp_version) {
 413         case TPACKET_V1:
 414                 h.h1->tp_status = status;
 415                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 416                 break;
 417         case TPACKET_V2:
 418                 h.h2->tp_status = status;
 419                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 420                 break;
 421         case TPACKET_V3:
 422         default:
 423                 WARN(1, "TPACKET version not supported.\n");
 424                 BUG();
 425         }
 426
 427         smp_wmb();
 428 }
 429
 430 static int __packet_get_status(struct packet_sock *po, void *frame)
 431 {
 432         union {
 433                 struct tpacket_hdr *h1;
 434                 struct tpacket2_hdr *h2;
 435                 void *raw;
 436         } h;
 437
 438         smp_rmb();
 439
 440         h.raw = frame;
 441         switch (po->tp_version) {
 442         case TPACKET_V1:
 443                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 444                 return h.h1->tp_status;
 445         case TPACKET_V2:
 446                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 447                 return h.h2->tp_status;
 448         case TPACKET_V3:
 449         default:
 450                 WARN(1, "TPACKET version not supported.\n");
 451                 BUG();
 452                 return 0;
 453         }
 454 }
 455
 456 static void *packet_lookup_frame(struct packet_sock *po,
 457                 struct packet_ring_buffer *rb,
 458                 unsigned int position,
 459                 int status)
 460 {
 461         unsigned int pg_vec_pos, frame_offset;
 462         union {
 463                 struct tpacket_hdr *h1;
 464                 struct tpacket2_hdr *h2;
 465                 void *raw;
 466         } h;
 467
 468         pg_vec_pos = position / rb->frames_per_block;
 469         frame_offset = position % rb->frames_per_block;
 470
 471         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 472                 (frame_offset * rb->frame_size);
 473
 474         if (status != __packet_get_status(po, h.raw))
 475                 return NULL;
 476
 477         return h.raw;
 478 }
 479
 480 static inline void *packet_current_frame(struct packet_sock *po,
 481                 struct packet_ring_buffer *rb,
 482                 int status)
 483 {
 484         return packet_lookup_frame(po, rb, rb->head, status);
 485 }
 486
 487 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 488 {
 489         del_timer_sync(&pkc->retire_blk_timer);
 490 }
 491
 492 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 493                 int tx_ring,
 494                 struct sk_buff_head *rb_queue)
 495 {
 496         struct tpacket_kbdq_core *pkc;
 497
 498         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 499
 500         spin_lock(&rb_queue->lock);
 501         pkc->delete_blk_timer = 1;
 502         spin_unlock(&rb_queue->lock);
 503
 504         prb_del_retire_blk_timer(pkc);
 505 }
 506
 507 static void prb_init_blk_timer(struct packet_sock *po,
 508                 struct tpacket_kbdq_core *pkc,
 509                 void (*func) (unsigned long))
 510 {
 511         init_timer(&pkc->retire_blk_timer);
 512         pkc->retire_blk_timer.data = (long)po;
 513         pkc->retire_blk_timer.function = func;
 514         pkc->retire_blk_timer.expires = jiffies;
 515 }
 516
 517 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
 518 {
 519         struct tpacket_kbdq_core *pkc;
 520
 521         if (tx_ring)
 522                 BUG();
 523
 524         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 525         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
 526 }
 527
 528 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 529                                 int blk_size_in_bytes)
 530 {
 531         struct net_device *dev;
 532         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 533         struct ethtool_cmd ecmd;
 534         int err;
 535
 536         rtnl_lock();
 537         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 538         if (unlikely(!dev)) {
 539                 rtnl_unlock();
 540                 return DEFAULT_PRB_RETIRE_TOV;
 541         }
 542         err = __ethtool_get_settings(dev, &ecmd);
 543         rtnl_unlock();
 544         if (!err) {
 545                 switch (ecmd.speed) {
 546                 case SPEED_10000:
 547                         msec = 1;
 548                         div = 10000/1000;
 549                         break;
 550                 case SPEED_1000:
 551                         msec = 1;
 552                         div = 1000/1000;
 553                         break;
 554                 /*
 555                  * If the link speed is so slow you don't really
 556                  * need to worry about perf anyways
 557                  */
 558                 case SPEED_100:
 559                 case SPEED_10:
 560                 default:
 561                         return DEFAULT_PRB_RETIRE_TOV;
 562                 }
 563         }
 564
 565         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 566
 567         if (div)
 568                 mbits /= div;
 569
 570         tmo = mbits * msec;
 571
 572         if (div)
 573                 return tmo+1;
 574         return tmo;
 575 }
 576
 577 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 578                         union tpacket_req_u *req_u)
 579 {
 580         p1->feature_req_word = req_u->req3.tp_feature_req_word;
 581 }
 582
 583 static void init_prb_bdqc(struct packet_sock *po,
 584                         struct packet_ring_buffer *rb,
 585                         struct pgv *pg_vec,
 586                         union tpacket_req_u *req_u, int tx_ring)
 587 {
 588         struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
 589         struct tpacket_block_desc *pbd;
 590
 591         memset(p1, 0x0, sizeof(*p1));
 592
 593         p1->knxt_seq_num = 1;
 594         p1->pkbdq = pg_vec;
 595         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 596         p1->pkblk_start = (char *)pg_vec[0].buffer;
 597         p1->kblk_size = req_u->req3.tp_block_size;
 598         p1->knum_blocks = req_u->req3.tp_block_nr;
 599         p1->hdrlen = po->tp_hdrlen;
 600         p1->version = po->tp_version;
 601         p1->last_kactive_blk_num = 0;
 602         po->stats_u.stats3.tp_freeze_q_cnt = 0;
 603         if (req_u->req3.tp_retire_blk_tov)
 604                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 605         else
 606                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 607                                                 req_u->req3.tp_block_size);
 608         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 609         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 610
 611         prb_init_ft_ops(p1, req_u);
 612         prb_setup_retire_blk_timer(po, tx_ring);
 613         prb_open_block(p1, pbd);
 614 }
 615
 616 /*  Do NOT update the last_blk_num first.
 617  *  Assumes sk_buff_head lock is held.
 618  */
 619 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 620 {
 621         mod_timer(&pkc->retire_blk_timer,
 622                         jiffies + pkc->tov_in_jiffies);
 623         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 624 }
 625
 626 /*
 627  * Timer logic:
 628  * 1) We refresh the timer only when we open a block.
 629  *    By doing this we don't waste cycles refreshing the timer
 630  *        on packet-by-packet basis.
 631  *
 632  * With a 1MB block-size, on a 1Gbps line, it will take
 633  * i) ~8 ms to fill a block + ii) memcpy etc.
 634  * In this cut we are not accounting for the memcpy time.
 635  *
 636  * So, if the user sets the 'tmo' to 10ms then the timer
 637  * will never fire while the block is still getting filled
 638  * (which is what we want). However, the user could choose
 639  * to close a block early and that's fine.
 640  *
 641  * But when the timer does fire, we check whether or not to refresh it.
 642  * Since the tmo granularity is in msecs, it is not too expensive
 643  * to refresh the timer, lets say every '8' msecs.
 644  * Either the user can set the 'tmo' or we can derive it based on
 645  * a) line-speed and b) block-size.
 646  * prb_calc_retire_blk_tmo() calculates the tmo.
 647  *
 648  */
 649 static void prb_retire_rx_blk_timer_expired(unsigned long data)
 650 {
 651         struct packet_sock *po = (struct packet_sock *)data;
 652         struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
 653         unsigned int frozen;
 654         struct tpacket_block_desc *pbd;
 655
 656         spin_lock(&po->sk.sk_receive_queue.lock);
 657
 658         frozen = prb_queue_frozen(pkc);
 659         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 660
 661         if (unlikely(pkc->delete_blk_timer))
 662                 goto out;
 663
 664         /* We only need to plug the race when the block is partially filled.
 665          * tpacket_rcv:
 666          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 667          *              copy_bits() is in progress ...
 668          *              timer fires on other cpu:
 669          *              we can't retire the current block because copy_bits
 670          *              is in progress.
 671          *
 672          */
 673         if (BLOCK_NUM_PKTS(pbd)) {
 674                 while (atomic_read(&pkc->blk_fill_in_prog)) {
 675                         /* Waiting for skb_copy_bits to finish... */
 676                         cpu_relax();
 677                 }
 678         }
 679
 680         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 681                 if (!frozen) {
 682                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 683                         if (!prb_dispatch_next_block(pkc, po))
 684                                 goto refresh_timer;
 685                         else
 686                                 goto out;
 687                 } else {
 688                         /* Case 1. Queue was frozen because user-space was
 689                          *         lagging behind.
 690                          */
 691                         if (prb_curr_blk_in_use(pkc, pbd)) {
 692                                 /*
 693                                  * Ok, user-space is still behind.
 694                                  * So just refresh the timer.
 695                                  */
 696                                 goto refresh_timer;
 697                         } else {
 698                                /* Case 2. queue was frozen,user-space caught up,
 699                                 * now the link went idle && the timer fired.
 700                                 * We don't have a block to close.So we open this
 701                                 * block and restart the timer.
 702                                 * opening a block thaws the queue,restarts timer
 703                                 * Thawing/timer-refresh is a side effect.
 704                                 */
 705                                 prb_open_block(pkc, pbd);
 706                                 goto out;
 707                         }
 708                 }
 709         }
 710
 711 refresh_timer:
 712         _prb_refresh_rx_retire_blk_timer(pkc);
 713
 714 out:
 715         spin_unlock(&po->sk.sk_receive_queue.lock);
 716 }
 717
 718 static inline void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 719                 struct tpacket_block_desc *pbd1, __u32 status)
 720 {
 721         /* Flush everything minus the block header */
 722
 723 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 724         u8 *start, *end;
 725
 726         start = (u8 *)pbd1;
 727
 728         /* Skip the block header(we know header WILL fit in 4K) */
 729         start += PAGE_SIZE;
 730
 731         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 732         for (; start < end; start += PAGE_SIZE)
 733                 flush_dcache_page(pgv_to_page(start));
 734
 735         smp_wmb();
 736 #endif
 737
 738         /* Now update the block status. */
 739
 740         BLOCK_STATUS(pbd1) = status;
 741
 742         /* Flush the block header */
 743
 744 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 745         start = (u8 *)pbd1;
 746         flush_dcache_page(pgv_to_page(start));
 747
 748         smp_wmb();
 749 #endif
 750 }
 751
 752 /*
 753  * Side effect:
 754  *
 755  * 1) flush the block
 756  * 2) Increment active_blk_num
 757  *
 758  * Note:We DONT refresh the timer on purpose.
 759  *      Because almost always the next block will be opened.
 760  */
 761 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 762                 struct tpacket_block_desc *pbd1,
 763                 struct packet_sock *po, unsigned int stat)
 764 {
 765         __u32 status = TP_STATUS_USER | stat;
 766
 767         struct tpacket3_hdr *last_pkt;
 768         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 769
 770         if (po->stats.tp_drops)
 771                 status |= TP_STATUS_LOSING;
 772
 773         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 774         last_pkt->tp_next_offset = 0;
 775
 776         /* Get the ts of the last pkt */
 777         if (BLOCK_NUM_PKTS(pbd1)) {
 778                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 779                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 780         } else {
 781                 /* Ok, we tmo'd - so get the current time */
 782                 struct timespec ts;
 783                 getnstimeofday(&ts);
 784                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 785                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 786         }
 787
 788         smp_wmb();
 789
 790         /* Flush the block */
 791         prb_flush_block(pkc1, pbd1, status);
 792
 793         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 794 }
 795
 796 static inline void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 797 {
 798         pkc->reset_pending_on_curr_blk = 0;
 799 }
 800
 801 /*
 802  * Side effect of opening a block:
 803  *
 804  * 1) prb_queue is thawed.
 805  * 2) retire_blk_timer is refreshed.
 806  *
 807  */
 808 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 809         struct tpacket_block_desc *pbd1)
 810 {
 811         struct timespec ts;
 812         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 813
 814         smp_rmb();
 815
 816         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd1))) {
 817
 818                 /* We could have just memset this but we will lose the
 819                  * flexibility of making the priv area sticky
 820                  */
 821                 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 822                 BLOCK_NUM_PKTS(pbd1) = 0;
 823                 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 824                 getnstimeofday(&ts);
 825                 h1->ts_first_pkt.ts_sec = ts.tv_sec;
 826                 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 827                 pkc1->pkblk_start = (char *)pbd1;
 828                 pkc1->nxt_offset = (char *)(pkc1->pkblk_start +
 829                 BLK_PLUS_PRIV(pkc1->blk_sizeof_priv));
 830                 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 831                 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 832                 pbd1->version = pkc1->version;
 833                 pkc1->prev = pkc1->nxt_offset;
 834                 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 835                 prb_thaw_queue(pkc1);
 836                 _prb_refresh_rx_retire_blk_timer(pkc1);
 837
 838                 smp_wmb();
 839
 840                 return;
 841         }
 842
 843         WARN(1, "ERROR block:%p is NOT FREE status:%d kactive_blk_num:%d\n",
 844                 pbd1, BLOCK_STATUS(pbd1), pkc1->kactive_blk_num);
 845         dump_stack();
 846         BUG();
 847 }
 848
 849 /*
 850  * Queue freeze logic:
 851  * 1) Assume tp_block_nr = 8 blocks.
 852  * 2) At time 't0', user opens Rx ring.
 853  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 854  * 4) user-space is either sleeping or processing block '0'.
 855  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 856  *    it will close block-7,loop around and try to fill block '0'.
 857  *    call-flow:
 858  *    __packet_lookup_frame_in_block
 859  *      prb_retire_current_block()
 860  *      prb_dispatch_next_block()
 861  *        |->(BLOCK_STATUS == USER) evaluates to true
 862  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 863  * 6) Now there are two cases:
 864  *    6.1) Link goes idle right after the queue is frozen.
 865  *         But remember, the last open_block() refreshed the timer.
 866  *         When this timer expires,it will refresh itself so that we can
 867  *         re-open block-0 in near future.
 868  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 869  *         case and __packet_lookup_frame_in_block will check if block-0
 870  *         is free and can now be re-used.
 871  */
 872 static inline void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 873                                   struct packet_sock *po)
 874 {
 875         pkc->reset_pending_on_curr_blk = 1;
 876         po->stats_u.stats3.tp_freeze_q_cnt++;
 877 }
 878
 879 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 880
 881 /*
 882  * If the next block is free then we will dispatch it
 883  * and return a good offset.
 884  * Else, we will freeze the queue.
 885  * So, caller must check the return value.
 886  */
 887 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 888                 struct packet_sock *po)
 889 {
 890         struct tpacket_block_desc *pbd;
 891
 892         smp_rmb();
 893
 894         /* 1. Get current block num */
 895         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 896
 897         /* 2. If this block is currently in_use then freeze the queue */
 898         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 899                 prb_freeze_queue(pkc, po);
 900                 return NULL;
 901         }
 902
 903         /*
 904          * 3.
 905          * open this block and return the offset where the first packet
 906          * needs to get stored.
 907          */
 908         prb_open_block(pkc, pbd);
 909         return (void *)pkc->nxt_offset;
 910 }
 911
 912 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 913                 struct packet_sock *po, unsigned int status)
 914 {
 915         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 916
 917         /* retire/close the current block */
 918         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 919                 /*
 920                  * Plug the case where copy_bits() is in progress on
 921                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 922                  * have space to copy the pkt in the current block and
 923                  * called prb_retire_current_block()
 924                  *
 925                  * We don't need to worry about the TMO case because
 926                  * the timer-handler already handled this case.
 927                  */
 928                 if (!(status & TP_STATUS_BLK_TMO)) {
 929                         while (atomic_read(&pkc->blk_fill_in_prog)) {
 930                                 /* Waiting for skb_copy_bits to finish... */
 931                                 cpu_relax();
 932                         }
 933                 }
 934                 prb_close_block(pkc, pbd, po, status);
 935                 return;
 936         }
 937
 938         WARN(1, "ERROR-pbd[%d]:%p\n", pkc->kactive_blk_num, pbd);
 939         dump_stack();
 940         BUG();
 941 }
 942
 943 static inline int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
 944                                       struct tpacket_block_desc *pbd)
 945 {
 946         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 947 }
 948
 949 static inline int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 950 {
 951         return pkc->reset_pending_on_curr_blk;
 952 }
 953
 954 static inline void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 955 {
 956         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 957         atomic_dec(&pkc->blk_fill_in_prog);
 958 }
 959
 960 static inline void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 961                         struct tpacket3_hdr *ppd)
 962 {
 963         ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
 964 }
 965
 966 static inline void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 967                         struct tpacket3_hdr *ppd)
 968 {
 969         ppd->hv1.tp_rxhash = 0;
 970 }
 971
 972 static inline void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
 973                         struct tpacket3_hdr *ppd)
 974 {
 975         if (vlan_tx_tag_present(pkc->skb)) {
 976                 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
 977                 ppd->tp_status = TP_STATUS_VLAN_VALID;
 978         } else {
 979                 ppd->hv1.tp_vlan_tci = ppd->tp_status = 0;
 980         }
 981 }
 982
 983 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
 984                         struct tpacket3_hdr *ppd)
 985 {
 986         prb_fill_vlan_info(pkc, ppd);
 987
 988         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
 989                 prb_fill_rxhash(pkc, ppd);
 990         else
 991                 prb_clear_rxhash(pkc, ppd);
 992 }
 993
 994 static inline void prb_fill_curr_block(char *curr,
 995                                 struct tpacket_kbdq_core *pkc,
 996                                 struct tpacket_block_desc *pbd,
 997                                 unsigned int len)
 998 {
 999         struct tpacket3_hdr *ppd;
1000
1001         ppd  = (struct tpacket3_hdr *)curr;
1002         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1003         pkc->prev = curr;
1004         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1005         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1006         BLOCK_NUM_PKTS(pbd) += 1;
1007         atomic_inc(&pkc->blk_fill_in_prog);
1008         prb_run_all_ft_ops(pkc, ppd);
1009 }
1010
1011 /* Assumes caller has the sk->rx_queue.lock */
1012 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1013                                             struct sk_buff *skb,
1014                                                 int status,
1015                                             unsigned int len
1016                                             )
1017 {
1018         struct tpacket_kbdq_core *pkc;
1019         struct tpacket_block_desc *pbd;
1020         char *curr, *end;
1021
1022         pkc = GET_PBDQC_FROM_RB(((struct packet_ring_buffer *)&po->rx_ring));
1023         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1024
1025         /* Queue is frozen when user space is lagging behind */
1026         if (prb_queue_frozen(pkc)) {
1027                 /*
1028                  * Check if that last block which caused the queue to freeze,
1029                  * is still in_use by user-space.
1030                  */
1031                 if (prb_curr_blk_in_use(pkc, pbd)) {
1032                         /* Can't record this packet */
1033                         return NULL;
1034                 } else {
1035                         /*
1036                          * Ok, the block was released by user-space.
1037                          * Now let's open that block.
1038                          * opening a block also thaws the queue.
1039                          * Thawing is a side effect.
1040                          */
1041                         prb_open_block(pkc, pbd);
1042                 }
1043         }
1044
1045         smp_mb();
1046         curr = pkc->nxt_offset;
1047         pkc->skb = skb;
1048         end = (char *) ((char *)pbd + pkc->kblk_size);
1049
1050         /* first try the current block */
1051         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1052                 prb_fill_curr_block(curr, pkc, pbd, len);
1053                 return (void *)curr;
1054         }
1055
1056         /* Ok, close the current block */
1057         prb_retire_current_block(pkc, po, 0);
1058
1059         /* Now, try to dispatch the next block */
1060         curr = (char *)prb_dispatch_next_block(pkc, po);
1061         if (curr) {
1062                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1063                 prb_fill_curr_block(curr, pkc, pbd, len);
1064                 return (void *)curr;
1065         }
1066
1067         /*
1068          * No free blocks are available.user_space hasn't caught up yet.
1069          * Queue was just frozen and now this packet will get dropped.
1070          */
1071         return NULL;
1072 }
1073
1074 static inline void *packet_current_rx_frame(struct packet_sock *po,
1075                                             struct sk_buff *skb,
1076                                             int status, unsigned int len)
1077 {
1078         char *curr = NULL;
1079         switch (po->tp_version) {
1080         case TPACKET_V1:
1081         case TPACKET_V2:
1082                 curr = packet_lookup_frame(po, &po->rx_ring,
1083                                         po->rx_ring.head, status);
1084                 return curr;
1085         case TPACKET_V3:
1086                 return __packet_lookup_frame_in_block(po, skb, status, len);
1087         default:
1088                 WARN(1, "TPACKET version not supported\n");
1089                 BUG();
1090                 return 0;
1091         }
1092 }
1093
1094 static inline void *prb_lookup_block(struct packet_sock *po,
1095                                      struct packet_ring_buffer *rb,
1096                                      unsigned int previous,
1097                                      int status)
1098 {
1099         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1100         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, previous);
1101
1102         if (status != BLOCK_STATUS(pbd))
1103                 return NULL;
1104         return pbd;
1105 }
1106
1107 static inline int prb_previous_blk_num(struct packet_ring_buffer *rb)
1108 {
1109         unsigned int prev;
1110         if (rb->prb_bdqc.kactive_blk_num)
1111                 prev = rb->prb_bdqc.kactive_blk_num-1;
1112         else
1113                 prev = rb->prb_bdqc.knum_blocks-1;
1114         return prev;
1115 }
1116
1117 /* Assumes caller has held the rx_queue.lock */
1118 static inline void *__prb_previous_block(struct packet_sock *po,
1119                                          struct packet_ring_buffer *rb,
1120                                          int status)
1121 {
1122         unsigned int previous = prb_previous_blk_num(rb);
1123         return prb_lookup_block(po, rb, previous, status);
1124 }
1125
1126 static inline void *packet_previous_rx_frame(struct packet_sock *po,
1127                                              struct packet_ring_buffer *rb,
1128                                              int status)
1129 {
1130         if (po->tp_version <= TPACKET_V2)
1131                 return packet_previous_frame(po, rb, status);
1132
1133         return __prb_previous_block(po, rb, status);
1134 }
1135
1136 static inline void packet_increment_rx_head(struct packet_sock *po,
1137                                             struct packet_ring_buffer *rb)
1138 {
1139         switch (po->tp_version) {
1140         case TPACKET_V1:
1141         case TPACKET_V2:
1142                 return packet_increment_head(rb);
1143         case TPACKET_V3:
1144         default:
1145                 WARN(1, "TPACKET version not supported.\n");
1146                 BUG();
1147                 return;
1148         }
1149 }
1150
1151 static inline void *packet_previous_frame(struct packet_sock *po,
1152                 struct packet_ring_buffer *rb,
1153                 int status)
1154 {
1155         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1156         return packet_lookup_frame(po, rb, previous, status);
1157 }
1158
1159 static inline void packet_increment_head(struct packet_ring_buffer *buff)
1160 {
1161         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1162 }
1163
1164 static void packet_sock_destruct(struct sock *sk)
1165 {
1166         skb_queue_purge(&sk->sk_error_queue);
1167
1168         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1169         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1170
1171         if (!sock_flag(sk, SOCK_DEAD)) {
1172                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1173                 return;
1174         }
1175
1176         sk_refcnt_debug_dec(sk);
1177 }
1178
1179 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1180 {
1181         int x = atomic_read(&f->rr_cur) + 1;
1182
1183         if (x >= num)
1184                 x = 0;
1185
1186         return x;
1187 }
1188
1189 static struct sock *fanout_demux_hash(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1190 {
1191         u32 idx, hash = skb->rxhash;
1192
1193         idx = ((u64)hash * num) >> 32;
1194
1195         return f->arr[idx];
1196 }
1197
1198 static struct sock *fanout_demux_lb(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1199 {
1200         int cur, old;
1201
1202         cur = atomic_read(&f->rr_cur);
1203         while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1204                                      fanout_rr_next(f, num))) != cur)
1205                 cur = old;
1206         return f->arr[cur];
1207 }
1208
1209 static struct sock *fanout_demux_cpu(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1210 {
1211         unsigned int cpu = smp_processor_id();
1212
1213         return f->arr[cpu % num];
1214 }
1215
1216 static struct sk_buff *fanout_check_defrag(struct sk_buff *skb)
1217 {
1218 #ifdef CONFIG_INET
1219         const struct iphdr *iph;
1220         u32 len;
1221
1222         if (skb->protocol != htons(ETH_P_IP))
1223                 return skb;
1224
1225         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1226                 return skb;
1227
1228         iph = ip_hdr(skb);
1229         if (iph->ihl < 5 || iph->version != 4)
1230                 return skb;
1231         if (!pskb_may_pull(skb, iph->ihl*4))
1232                 return skb;
1233         iph = ip_hdr(skb);
1234         len = ntohs(iph->tot_len);
1235         if (skb->len < len || len < (iph->ihl * 4))
1236                 return skb;
1237
1238         if (ip_is_fragment(ip_hdr(skb))) {
1239                 skb = skb_share_check(skb, GFP_ATOMIC);
1240                 if (skb) {
1241                         if (pskb_trim_rcsum(skb, len))
1242                                 return skb;
1243                         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
1244                         if (ip_defrag(skb, IP_DEFRAG_AF_PACKET))
1245                                 return NULL;
1246                         skb->rxhash = 0;
1247                 }
1248         }
1249 #endif
1250         return skb;
1251 }
1252
1253 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1254                              struct packet_type *pt, struct net_device *orig_dev)
1255 {
1256         struct packet_fanout *f = pt->af_packet_priv;
1257         unsigned int num = f->num_members;
1258         struct packet_sock *po;
1259         struct sock *sk;
1260
1261         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1262             !num) {
1263                 kfree_skb(skb);
1264                 return 0;
1265         }
1266
1267         switch (f->type) {
1268         case PACKET_FANOUT_HASH:
1269         default:
1270                 if (f->defrag) {
1271                         skb = fanout_check_defrag(skb);
1272                         if (!skb)
1273                                 return 0;
1274                 }
1275                 skb_get_rxhash(skb);
1276                 sk = fanout_demux_hash(f, skb, num);
1277                 break;
1278         case PACKET_FANOUT_LB:
1279                 sk = fanout_demux_lb(f, skb, num);
1280                 break;
1281         case PACKET_FANOUT_CPU:
1282                 sk = fanout_demux_cpu(f, skb, num);
1283                 break;
1284         }
1285
1286         po = pkt_sk(sk);
1287
1288         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1289 }
1290
1291 static DEFINE_MUTEX(fanout_mutex);
1292 static LIST_HEAD(fanout_list);
1293
1294 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1295 {
1296         struct packet_fanout *f = po->fanout;
1297
1298         spin_lock(&f->lock);
1299         f->arr[f->num_members] = sk;
1300         smp_wmb();
1301         f->num_members++;
1302         spin_unlock(&f->lock);
1303 }
1304
1305 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1306 {
1307         struct packet_fanout *f = po->fanout;
1308         int i;
1309
1310         spin_lock(&f->lock);
1311         for (i = 0; i < f->num_members; i++) {
1312                 if (f->arr[i] == sk)
1313                         break;
1314         }
1315         BUG_ON(i >= f->num_members);
1316         f->arr[i] = f->arr[f->num_members - 1];
1317         f->num_members--;
1318         spin_unlock(&f->lock);
1319 }
1320
1321 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1322 {
1323         struct packet_sock *po = pkt_sk(sk);
1324         struct packet_fanout *f, *match;
1325         u8 type = type_flags & 0xff;
1326         u8 defrag = (type_flags & PACKET_FANOUT_FLAG_DEFRAG) ? 1 : 0;
1327         int err;
1328
1329         switch (type) {
1330         case PACKET_FANOUT_HASH:
1331         case PACKET_FANOUT_LB:
1332         case PACKET_FANOUT_CPU:
1333                 break;
1334         default:
1335                 return -EINVAL;
1336         }
1337
1338         if (!po->running)
1339                 return -EINVAL;
1340
1341         if (po->fanout)
1342                 return -EALREADY;
1343
1344         mutex_lock(&fanout_mutex);
1345         match = NULL;
1346         list_for_each_entry(f, &fanout_list, list) {
1347                 if (f->id == id &&
1348                     read_pnet(&f->net) == sock_net(sk)) {
1349                         match = f;
1350                         break;
1351                 }
1352         }
1353         err = -EINVAL;
1354         if (match && match->defrag != defrag)
1355                 goto out;
1356         if (!match) {
1357                 err = -ENOMEM;
1358                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1359                 if (!match)
1360                         goto out;
1361                 write_pnet(&match->net, sock_net(sk));
1362                 match->id = id;
1363                 match->type = type;
1364                 match->defrag = defrag;
1365                 atomic_set(&match->rr_cur, 0);
1366                 INIT_LIST_HEAD(&match->list);
1367                 spin_lock_init(&match->lock);
1368                 atomic_set(&match->sk_ref, 0);
1369                 match->prot_hook.type = po->prot_hook.type;
1370                 match->prot_hook.dev = po->prot_hook.dev;
1371                 match->prot_hook.func = packet_rcv_fanout;
1372                 match->prot_hook.af_packet_priv = match;
1373                 dev_add_pack(&match->prot_hook);
1374                 list_add(&match->list, &fanout_list);
1375         }
1376         err = -EINVAL;
1377         if (match->type == type &&
1378             match->prot_hook.type == po->prot_hook.type &&
1379             match->prot_hook.dev == po->prot_hook.dev) {
1380                 err = -ENOSPC;
1381                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1382                         __dev_remove_pack(&po->prot_hook);
1383                         po->fanout = match;
1384                         atomic_inc(&match->sk_ref);
1385                         __fanout_link(sk, po);
1386                         err = 0;
1387                 }
1388         }
1389 out:
1390         mutex_unlock(&fanout_mutex);
1391         return err;
1392 }
1393
1394 static void fanout_release(struct sock *sk)
1395 {
1396         struct packet_sock *po = pkt_sk(sk);
1397         struct packet_fanout *f;
1398
1399         f = po->fanout;
1400         if (!f)
1401                 return;
1402
1403         po->fanout = NULL;
1404
1405         mutex_lock(&fanout_mutex);
1406         if (atomic_dec_and_test(&f->sk_ref)) {
1407                 list_del(&f->list);
1408                 dev_remove_pack(&f->prot_hook);
1409                 kfree(f);
1410         }
1411         mutex_unlock(&fanout_mutex);
1412 }
1413
1414 static const struct proto_ops packet_ops;
1415
1416 static const struct proto_ops packet_ops_spkt;
1417
1418 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1419                            struct packet_type *pt, struct net_device *orig_dev)
1420 {
1421         struct sock *sk;
1422         struct sockaddr_pkt *spkt;
1423
1424         /*
1425          *      When we registered the protocol we saved the socket in the data
1426          *      field for just this event.
1427          */
1428
1429         sk = pt->af_packet_priv;
1430
1431         /*
1432          *      Yank back the headers [hope the device set this
1433          *      right or kerboom...]
1434          *
1435          *      Incoming packets have ll header pulled,
1436          *      push it back.
1437          *
1438          *      For outgoing ones skb->data == skb_mac_header(skb)
1439          *      so that this procedure is noop.
1440          */
1441
1442         if (skb->pkt_type == PACKET_LOOPBACK)
1443                 goto out;
1444
1445         if (!net_eq(dev_net(dev), sock_net(sk)))
1446                 goto out;
1447
1448         skb = skb_share_check(skb, GFP_ATOMIC);
1449         if (skb == NULL)
1450                 goto oom;
1451
1452         /* drop any routing info */
1453         skb_dst_drop(skb);
1454
1455         /* drop conntrack reference */
1456         nf_reset(skb);
1457
1458         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1459
1460         skb_push(skb, skb->data - skb_mac_header(skb));
1461
1462         /*
1463          *      The SOCK_PACKET socket receives _all_ frames.
1464          */
1465
1466         spkt->spkt_family = dev->type;
1467         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1468         spkt->spkt_protocol = skb->protocol;
1469
1470         /*
1471          *      Charge the memory to the socket. This is done specifically
1472          *      to prevent sockets using all the memory up.
1473          */
1474
1475         if (sock_queue_rcv_skb(sk, skb) == 0)
1476                 return 0;
1477
1478 out:
1479         kfree_skb(skb);
1480 oom:
1481         return 0;
1482 }
1483
1484
1485 /*
1486  *      Output a raw packet to a device layer. This bypasses all the other
1487  *      protocol layers and you must therefore supply it with a complete frame
1488  */
1489
1490 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1491                                struct msghdr *msg, size_t len)
1492 {
1493         struct sock *sk = sock->sk;
1494         struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1495         struct sk_buff *skb = NULL;
1496         struct net_device *dev;
1497         __be16 proto = 0;
1498         int err;
1499
1500         /*
1501          *      Get and verify the address.
1502          */
1503
1504         if (saddr) {
1505                 if (msg->msg_namelen < sizeof(struct sockaddr))
1506                         return -EINVAL;
1507                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1508                         proto = saddr->spkt_protocol;
1509         } else
1510                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1511
1512         /*
1513          *      Find the device first to size check it
1514          */
1515
1516         saddr->spkt_device[13] = 0;
1517 retry:
1518         rcu_read_lock();
1519         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1520         err = -ENODEV;
1521         if (dev == NULL)
1522                 goto out_unlock;
1523
1524         err = -ENETDOWN;
1525         if (!(dev->flags & IFF_UP))
1526                 goto out_unlock;
1527
1528         /*
1529          * You may not queue a frame bigger than the mtu. This is the lowest level
1530          * raw protocol and you must do your own fragmentation at this level.
1531          */
1532
1533         err = -EMSGSIZE;
1534         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN)
1535                 goto out_unlock;
1536
1537         if (!skb) {
1538                 size_t reserved = LL_RESERVED_SPACE(dev);
1539                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1540
1541                 rcu_read_unlock();
1542                 skb = sock_wmalloc(sk, len + reserved, 0, GFP_KERNEL);
1543                 if (skb == NULL)
1544                         return -ENOBUFS;
1545                 /* FIXME: Save some space for broken drivers that write a hard
1546                  * header at transmission time by themselves. PPP is the notable
1547                  * one here. This should really be fixed at the driver level.
1548                  */
1549                 skb_reserve(skb, reserved);
1550                 skb_reset_network_header(skb);
1551
1552                 /* Try to align data part correctly */
1553                 if (hhlen) {
1554                         skb->data -= hhlen;
1555                         skb->tail -= hhlen;
1556                         if (len < hhlen)
1557                                 skb_reset_network_header(skb);
1558                 }
1559                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1560                 if (err)
1561                         goto out_free;
1562                 goto retry;
1563         }
1564
1565         if (len > (dev->mtu + dev->hard_header_len)) {
1566                 /* Earlier code assumed this would be a VLAN pkt,
1567                  * double-check this now that we have the actual
1568                  * packet in hand.
1569                  */
1570                 struct ethhdr *ehdr;
1571                 skb_reset_mac_header(skb);
1572                 ehdr = eth_hdr(skb);
1573                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1574                         err = -EMSGSIZE;
1575                         goto out_unlock;
1576                 }
1577         }
1578
1579         skb->protocol = proto;
1580         skb->dev = dev;
1581         skb->priority = sk->sk_priority;
1582         skb->mark = sk->sk_mark;
1583         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1584         if (err < 0)
1585                 goto out_unlock;
1586
1587         dev_queue_xmit(skb);
1588         rcu_read_unlock();
1589         return len;
1590
1591 out_unlock:
1592         rcu_read_unlock();
1593 out_free:
1594         kfree_skb(skb);
1595         return err;
1596 }
1597
1598 static inline unsigned int run_filter(const struct sk_buff *skb,
1599                                       const struct sock *sk,
1600                                       unsigned int res)
1601 {
1602         struct sk_filter *filter;
1603
1604         rcu_read_lock();
1605         filter = rcu_dereference(sk->sk_filter);
1606         if (filter != NULL)
1607                 res = SK_RUN_FILTER(filter, skb);
1608         rcu_read_unlock();
1609
1610         return res;
1611 }
1612
1613 /*
1614  * This function makes lazy skb cloning in hope that most of packets
1615  * are discarded by BPF.
1616  *
1617  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1618  * and skb->cb are mangled. It works because (and until) packets
1619  * falling here are owned by current CPU. Output packets are cloned
1620  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1621  * sequencially, so that if we return skb to original state on exit,
1622  * we will not harm anyone.
1623  */
1624
1625 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1626                       struct packet_type *pt, struct net_device *orig_dev)
1627 {
1628         struct sock *sk;
1629         struct sockaddr_ll *sll;
1630         struct packet_sock *po;
1631         u8 *skb_head = skb->data;
1632         int skb_len = skb->len;
1633         unsigned int snaplen, res;
1634
1635         if (skb->pkt_type == PACKET_LOOPBACK)
1636                 goto drop;
1637
1638         sk = pt->af_packet_priv;
1639         po = pkt_sk(sk);
1640
1641         if (!net_eq(dev_net(dev), sock_net(sk)))
1642                 goto drop;
1643
1644         skb->dev = dev;
1645
1646         if (dev->header_ops) {
1647                 /* The device has an explicit notion of ll header,
1648                  * exported to higher levels.
1649                  *
1650                  * Otherwise, the device hides details of its frame
1651                  * structure, so that corresponding packet head is
1652                  * never delivered to user.
1653                  */
1654                 if (sk->sk_type != SOCK_DGRAM)
1655                         skb_push(skb, skb->data - skb_mac_header(skb));
1656                 else if (skb->pkt_type == PACKET_OUTGOING) {
1657                         /* Special case: outgoing packets have ll header at head */
1658                         skb_pull(skb, skb_network_offset(skb));
1659                 }
1660         }
1661
1662         snaplen = skb->len;
1663
1664         res = run_filter(skb, sk, snaplen);
1665         if (!res)
1666                 goto drop_n_restore;
1667         if (snaplen > res)
1668                 snaplen = res;
1669
1670         if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
1671             (unsigned)sk->sk_rcvbuf)
1672                 goto drop_n_acct;
1673
1674         if (skb_shared(skb)) {
1675                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1676                 if (nskb == NULL)
1677                         goto drop_n_acct;
1678
1679                 if (skb_head != skb->data) {
1680                         skb->data = skb_head;
1681                         skb->len = skb_len;
1682                 }
1683                 kfree_skb(skb);
1684                 skb = nskb;
1685         }
1686
1687         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1688                      sizeof(skb->cb));
1689
1690         sll = &PACKET_SKB_CB(skb)->sa.ll;
1691         sll->sll_family = AF_PACKET;
1692         sll->sll_hatype = dev->type;
1693         sll->sll_protocol = skb->protocol;
1694         sll->sll_pkttype = skb->pkt_type;
1695         if (unlikely(po->origdev))
1696                 sll->sll_ifindex = orig_dev->ifindex;
1697         else
1698                 sll->sll_ifindex = dev->ifindex;
1699
1700         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1701
1702         PACKET_SKB_CB(skb)->origlen = skb->len;
1703
1704         if (pskb_trim(skb, snaplen))
1705                 goto drop_n_acct;
1706
1707         skb_set_owner_r(skb, sk);
1708         skb->dev = NULL;
1709         skb_dst_drop(skb);
1710
1711         /* drop conntrack reference */
1712         nf_reset(skb);
1713
1714         spin_lock(&sk->sk_receive_queue.lock);
1715         po->stats.tp_packets++;
1716         skb->dropcount = atomic_read(&sk->sk_drops);
1717         __skb_queue_tail(&sk->sk_receive_queue, skb);
1718         spin_unlock(&sk->sk_receive_queue.lock);
1719         sk->sk_data_ready(sk, skb->len);
1720         return 0;
1721
1722 drop_n_acct:
1723         po->stats.tp_drops = atomic_inc_return(&sk->sk_drops);
1724
1725 drop_n_restore:
1726         if (skb_head != skb->data && skb_shared(skb)) {
1727                 skb->data = skb_head;
1728                 skb->len = skb_len;
1729         }
1730 drop:
1731         consume_skb(skb);
1732         return 0;
1733 }
1734
1735 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1736                        struct packet_type *pt, struct net_device *orig_dev)
1737 {
1738         struct sock *sk;
1739         struct packet_sock *po;
1740         struct sockaddr_ll *sll;
1741         union {
1742                 struct tpacket_hdr *h1;
1743                 struct tpacket2_hdr *h2;
1744                 struct tpacket3_hdr *h3;
1745                 void *raw;
1746         } h;
1747         u8 *skb_head = skb->data;
1748         int skb_len = skb->len;
1749         unsigned int snaplen, res;
1750         unsigned long status = TP_STATUS_USER;
1751         unsigned short macoff, netoff, hdrlen;
1752         struct sk_buff *copy_skb = NULL;
1753         struct timeval tv;
1754         struct timespec ts;
1755         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1756
1757         if (skb->pkt_type == PACKET_LOOPBACK)
1758                 goto drop;
1759
1760         sk = pt->af_packet_priv;
1761         po = pkt_sk(sk);
1762
1763         if (!net_eq(dev_net(dev), sock_net(sk)))
1764                 goto drop;
1765
1766         if (dev->header_ops) {
1767                 if (sk->sk_type != SOCK_DGRAM)
1768                         skb_push(skb, skb->data - skb_mac_header(skb));
1769                 else if (skb->pkt_type == PACKET_OUTGOING) {
1770                         /* Special case: outgoing packets have ll header at head */
1771                         skb_pull(skb, skb_network_offset(skb));
1772                 }
1773         }
1774
1775         if (skb->ip_summed == CHECKSUM_PARTIAL)
1776                 status |= TP_STATUS_CSUMNOTREADY;
1777
1778         snaplen = skb->len;
1779
1780         res = run_filter(skb, sk, snaplen);
1781         if (!res)
1782                 goto drop_n_restore;
1783         if (snaplen > res)
1784                 snaplen = res;
1785
1786         if (sk->sk_type == SOCK_DGRAM) {
1787                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1788                                   po->tp_reserve;
1789         } else {
1790                 unsigned maclen = skb_network_offset(skb);
1791                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1792                                        (maclen < 16 ? 16 : maclen)) +
1793                         po->tp_reserve;
1794                 macoff = netoff - maclen;
1795         }
1796         if (po->tp_version <= TPACKET_V2) {
1797                 if (macoff + snaplen > po->rx_ring.frame_size) {
1798                         if (po->copy_thresh &&
1799                                 atomic_read(&sk->sk_rmem_alloc) + skb->truesize
1800                                 < (unsigned)sk->sk_rcvbuf) {
1801                                 if (skb_shared(skb)) {
1802                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1803                                 } else {
1804                                         copy_skb = skb_get(skb);
1805                                         skb_head = skb->data;
1806                                 }
1807                                 if (copy_skb)
1808                                         skb_set_owner_r(copy_skb, sk);
1809                         }
1810                         snaplen = po->rx_ring.frame_size - macoff;
1811                         if ((int)snaplen < 0)
1812                                 snaplen = 0;
1813                 }
1814         }
1815         spin_lock(&sk->sk_receive_queue.lock);
1816         h.raw = packet_current_rx_frame(po, skb,
1817                                         TP_STATUS_KERNEL, (macoff+snaplen));
1818         if (!h.raw)
1819                 goto ring_is_full;
1820         if (po->tp_version <= TPACKET_V2) {
1821                 packet_increment_rx_head(po, &po->rx_ring);
1822         /*
1823          * LOSING will be reported till you read the stats,
1824          * because it's COR - Clear On Read.
1825          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1826          * at packet level.
1827          */
1828                 if (po->stats.tp_drops)
1829                         status |= TP_STATUS_LOSING;
1830         }
1831         po->stats.tp_packets++;
1832         if (copy_skb) {
1833                 status |= TP_STATUS_COPY;
1834                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1835         }
1836         spin_unlock(&sk->sk_receive_queue.lock);
1837
1838         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1839
1840         switch (po->tp_version) {
1841         case TPACKET_V1:
1842                 h.h1->tp_len = skb->len;
1843                 h.h1->tp_snaplen = snaplen;
1844                 h.h1->tp_mac = macoff;
1845                 h.h1->tp_net = netoff;
1846                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1847                                 && shhwtstamps->syststamp.tv64)
1848                         tv = ktime_to_timeval(shhwtstamps->syststamp);
1849                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1850                                 && shhwtstamps->hwtstamp.tv64)
1851                         tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1852                 else if (skb->tstamp.tv64)
1853                         tv = ktime_to_timeval(skb->tstamp);
1854                 else
1855                         do_gettimeofday(&tv);
1856                 h.h1->tp_sec = tv.tv_sec;
1857                 h.h1->tp_usec = tv.tv_usec;
1858                 hdrlen = sizeof(*h.h1);
1859                 break;
1860         case TPACKET_V2:
1861                 h.h2->tp_len = skb->len;
1862                 h.h2->tp_snaplen = snaplen;
1863                 h.h2->tp_mac = macoff;
1864                 h.h2->tp_net = netoff;
1865                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1866                                 && shhwtstamps->syststamp.tv64)
1867                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1868                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1869                                 && shhwtstamps->hwtstamp.tv64)
1870                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1871                 else if (skb->tstamp.tv64)
1872                         ts = ktime_to_timespec(skb->tstamp);
1873                 else
1874                         getnstimeofday(&ts);
1875                 h.h2->tp_sec = ts.tv_sec;
1876                 h.h2->tp_nsec = ts.tv_nsec;
1877                 if (vlan_tx_tag_present(skb)) {
1878                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1879                         status |= TP_STATUS_VLAN_VALID;
1880                 } else {
1881                         h.h2->tp_vlan_tci = 0;
1882                 }
1883                 h.h2->tp_padding = 0;
1884                 hdrlen = sizeof(*h.h2);
1885                 break;
1886         case TPACKET_V3:
1887                 /* tp_nxt_offset,vlan are already populated above.
1888                  * So DONT clear those fields here
1889                  */
1890                 h.h3->tp_status |= status;
1891                 h.h3->tp_len = skb->len;
1892                 h.h3->tp_snaplen = snaplen;
1893                 h.h3->tp_mac = macoff;
1894                 h.h3->tp_net = netoff;
1895                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1896                                 && shhwtstamps->syststamp.tv64)
1897                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1898                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1899                                 && shhwtstamps->hwtstamp.tv64)
1900                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1901                 else if (skb->tstamp.tv64)
1902                         ts = ktime_to_timespec(skb->tstamp);
1903                 else
1904                         getnstimeofday(&ts);
1905                 h.h3->tp_sec  = ts.tv_sec;
1906                 h.h3->tp_nsec = ts.tv_nsec;
1907                 hdrlen = sizeof(*h.h3);
1908                 break;
1909         default:
1910                 BUG();
1911         }
1912
1913         sll = h.raw + TPACKET_ALIGN(hdrlen);
1914         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1915         sll->sll_family = AF_PACKET;
1916         sll->sll_hatype = dev->type;
1917         sll->sll_protocol = skb->protocol;
1918         sll->sll_pkttype = skb->pkt_type;
1919         if (unlikely(po->origdev))
1920                 sll->sll_ifindex = orig_dev->ifindex;
1921         else
1922                 sll->sll_ifindex = dev->ifindex;
1923
1924         smp_mb();
1925 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1926         {
1927                 u8 *start, *end;
1928
1929                 if (po->tp_version <= TPACKET_V2) {
1930                         end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1931                                 + macoff + snaplen);
1932                         for (start = h.raw; start < end; start += PAGE_SIZE)
1933                                 flush_dcache_page(pgv_to_page(start));
1934                 }
1935                 smp_wmb();
1936         }
1937 #endif
1938         if (po->tp_version <= TPACKET_V2)
1939                 __packet_set_status(po, h.raw, status);
1940         else
1941                 prb_clear_blk_fill_status(&po->rx_ring);
1942
1943         sk->sk_data_ready(sk, 0);
1944
1945 drop_n_restore:
1946         if (skb_head != skb->data && skb_shared(skb)) {
1947                 skb->data = skb_head;
1948                 skb->len = skb_len;
1949         }
1950 drop:
1951         kfree_skb(skb);
1952         return 0;
1953
1954 ring_is_full:
1955         po->stats.tp_drops++;
1956         spin_unlock(&sk->sk_receive_queue.lock);
1957
1958         sk->sk_data_ready(sk, 0);
1959         kfree_skb(copy_skb);
1960         goto drop_n_restore;
1961 }
1962
1963 static void tpacket_destruct_skb(struct sk_buff *skb)
1964 {
1965         struct packet_sock *po = pkt_sk(skb->sk);
1966         void *ph;
1967
1968         BUG_ON(skb == NULL);
1969
1970         if (likely(po->tx_ring.pg_vec)) {
1971                 ph = skb_shinfo(skb)->destructor_arg;
1972                 BUG_ON(__packet_get_status(po, ph) != TP_STATUS_SENDING);
1973                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1974                 atomic_dec(&po->tx_ring.pending);
1975                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1976         }
1977
1978         sock_wfree(skb);
1979 }
1980
1981 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1982                 void *frame, struct net_device *dev, int size_max,
1983                 __be16 proto, unsigned char *addr)
1984 {
1985         union {
1986                 struct tpacket_hdr *h1;
1987                 struct tpacket2_hdr *h2;
1988                 void *raw;
1989         } ph;
1990         int to_write, offset, len, tp_len, nr_frags, len_max;
1991         struct socket *sock = po->sk.sk_socket;
1992         struct page *page;
1993         void *data;
1994         int err;
1995
1996         ph.raw = frame;
1997
1998         skb->protocol = proto;
1999         skb->dev = dev;
2000         skb->priority = po->sk.sk_priority;
2001         skb->mark = po->sk.sk_mark;
2002         skb_shinfo(skb)->destructor_arg = ph.raw;
2003
2004         switch (po->tp_version) {
2005         case TPACKET_V2:
2006                 tp_len = ph.h2->tp_len;
2007                 break;
2008         default:
2009                 tp_len = ph.h1->tp_len;
2010                 break;
2011         }
2012         if (unlikely(tp_len > size_max)) {
2013                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2014                 return -EMSGSIZE;
2015         }
2016
2017         skb_reserve(skb, LL_RESERVED_SPACE(dev));
2018         skb_reset_network_header(skb);
2019
2020         data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2021         to_write = tp_len;
2022
2023         if (sock->type == SOCK_DGRAM) {
2024                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2025                                 NULL, tp_len);
2026                 if (unlikely(err < 0))
2027                         return -EINVAL;
2028         } else if (dev->hard_header_len) {
2029                 /* net device doesn't like empty head */
2030                 if (unlikely(tp_len <= dev->hard_header_len)) {
2031                         pr_err("packet size is too short (%d < %d)\n",
2032                                tp_len, dev->hard_header_len);
2033                         return -EINVAL;
2034                 }
2035
2036                 skb_push(skb, dev->hard_header_len);
2037                 err = skb_store_bits(skb, 0, data,
2038                                 dev->hard_header_len);
2039                 if (unlikely(err))
2040                         return err;
2041
2042                 data += dev->hard_header_len;
2043                 to_write -= dev->hard_header_len;
2044         }
2045
2046         err = -EFAULT;
2047         offset = offset_in_page(data);
2048         len_max = PAGE_SIZE - offset;
2049         len = ((to_write > len_max) ? len_max : to_write);
2050
2051         skb->data_len = to_write;
2052         skb->len += to_write;
2053         skb->truesize += to_write;
2054         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2055
2056         while (likely(to_write)) {
2057                 nr_frags = skb_shinfo(skb)->nr_frags;
2058
2059                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2060                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2061                                MAX_SKB_FRAGS);
2062                         return -EFAULT;
2063                 }
2064
2065                 page = pgv_to_page(data);
2066                 data += len;
2067                 flush_dcache_page(page);
2068                 get_page(page);
2069                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2070                 to_write -= len;
2071                 offset = 0;
2072                 len_max = PAGE_SIZE;
2073                 len = ((to_write > len_max) ? len_max : to_write);
2074         }
2075
2076         return tp_len;
2077 }
2078
2079 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2080 {
2081         struct sk_buff *skb;
2082         struct net_device *dev;
2083         __be16 proto;
2084         bool need_rls_dev = false;
2085         int err, reserve = 0;
2086         void *ph;
2087         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2088         int tp_len, size_max;
2089         unsigned char *addr;
2090         int len_sum = 0;
2091         int status = 0;
2092
2093         mutex_lock(&po->pg_vec_lock);
2094
2095         err = -EBUSY;
2096         if (saddr == NULL) {
2097                 dev = po->prot_hook.dev;
2098                 proto   = po->num;
2099                 addr    = NULL;
2100         } else {
2101                 err = -EINVAL;
2102                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2103                         goto out;
2104                 if (msg->msg_namelen < (saddr->sll_halen
2105                                         + offsetof(struct sockaddr_ll,
2106                                                 sll_addr)))
2107                         goto out;
2108                 proto   = saddr->sll_protocol;
2109                 addr    = saddr->sll_addr;
2110                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2111                 need_rls_dev = true;
2112         }
2113
2114         err = -ENXIO;
2115         if (unlikely(dev == NULL))
2116                 goto out;
2117
2118         reserve = dev->hard_header_len;
2119
2120         err = -ENETDOWN;
2121         if (unlikely(!(dev->flags & IFF_UP)))
2122                 goto out_put;
2123
2124         size_max = po->tx_ring.frame_size
2125                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2126
2127         if (size_max > dev->mtu + reserve)
2128                 size_max = dev->mtu + reserve;
2129
2130         do {
2131                 ph = packet_current_frame(po, &po->tx_ring,
2132                                 TP_STATUS_SEND_REQUEST);
2133
2134                 if (unlikely(ph == NULL)) {
2135                         schedule();
2136                         continue;
2137                 }
2138
2139                 status = TP_STATUS_SEND_REQUEST;
2140                 skb = sock_alloc_send_skb(&po->sk,
2141                                 LL_ALLOCATED_SPACE(dev)
2142                                 + sizeof(struct sockaddr_ll),
2143                                 0, &err);
2144
2145                 if (unlikely(skb == NULL))
2146                         goto out_status;
2147
2148                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2149                                 addr);
2150
2151                 if (unlikely(tp_len < 0)) {
2152                         if (po->tp_loss) {
2153                                 __packet_set_status(po, ph,
2154                                                 TP_STATUS_AVAILABLE);
2155                                 packet_increment_head(&po->tx_ring);
2156                                 kfree_skb(skb);
2157                                 continue;
2158                         } else {
2159                                 status = TP_STATUS_WRONG_FORMAT;
2160                                 err = tp_len;
2161                                 goto out_status;
2162                         }
2163                 }
2164
2165                 skb->destructor = tpacket_destruct_skb;
2166                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2167                 atomic_inc(&po->tx_ring.pending);
2168
2169                 status = TP_STATUS_SEND_REQUEST;
2170                 err = dev_queue_xmit(skb);
2171                 if (unlikely(err > 0)) {
2172                         err = net_xmit_errno(err);
2173                         if (err && __packet_get_status(po, ph) ==
2174                                    TP_STATUS_AVAILABLE) {
2175                                 /* skb was destructed already */
2176                                 skb = NULL;
2177                                 goto out_status;
2178                         }
2179                         /*
2180                          * skb was dropped but not destructed yet;
2181                          * let's treat it like congestion or err < 0
2182                          */
2183                         err = 0;
2184                 }
2185                 packet_increment_head(&po->tx_ring);
2186                 len_sum += tp_len;
2187         } while (likely((ph != NULL) ||
2188                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2189                          (atomic_read(&po->tx_ring.pending))))
2190                 );
2191
2192         err = len_sum;
2193         goto out_put;
2194
2195 out_status:
2196         __packet_set_status(po, ph, status);
2197         kfree_skb(skb);
2198 out_put:
2199         if (need_rls_dev)
2200                 dev_put(dev);
2201 out:
2202         mutex_unlock(&po->pg_vec_lock);
2203         return err;
2204 }
2205
2206 static inline struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2207                                                size_t reserve, size_t len,
2208                                                size_t linear, int noblock,
2209                                                int *err)
2210 {
2211         struct sk_buff *skb;
2212
2213         /* Under a page?  Don't bother with paged skb. */
2214         if (prepad + len < PAGE_SIZE || !linear)
2215                 linear = len;
2216
2217         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2218                                    err);
2219         if (!skb)
2220                 return NULL;
2221
2222         skb_reserve(skb, reserve);
2223         skb_put(skb, linear);
2224         skb->data_len = len - linear;
2225         skb->len += len - linear;
2226
2227         return skb;
2228 }
2229
2230 static int packet_snd(struct socket *sock,
2231                           struct msghdr *msg, size_t len)
2232 {
2233         struct sock *sk = sock->sk;
2234         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2235         struct sk_buff *skb;
2236         struct net_device *dev;
2237         __be16 proto;
2238         bool need_rls_dev = false;
2239         unsigned char *addr;
2240         int err, reserve = 0;
2241         struct virtio_net_hdr vnet_hdr = { 0 };
2242         int offset = 0;
2243         int vnet_hdr_len;
2244         struct packet_sock *po = pkt_sk(sk);
2245         unsigned short gso_type = 0;
2246
2247         /*
2248          *      Get and verify the address.
2249          */
2250
2251         if (saddr == NULL) {
2252                 dev = po->prot_hook.dev;
2253                 proto   = po->num;
2254                 addr    = NULL;
2255         } else {
2256                 err = -EINVAL;
2257                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2258                         goto out;
2259                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2260                         goto out;
2261                 proto   = saddr->sll_protocol;
2262                 addr    = saddr->sll_addr;
2263                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2264                 need_rls_dev = true;
2265         }
2266
2267         err = -ENXIO;
2268         if (dev == NULL)
2269                 goto out_unlock;
2270         if (sock->type == SOCK_RAW)
2271                 reserve = dev->hard_header_len;
2272
2273         err = -ENETDOWN;
2274         if (!(dev->flags & IFF_UP))
2275                 goto out_unlock;
2276
2277         if (po->has_vnet_hdr) {
2278                 vnet_hdr_len = sizeof(vnet_hdr);
2279
2280                 err = -EINVAL;
2281                 if (len < vnet_hdr_len)
2282                         goto out_unlock;
2283
2284                 len -= vnet_hdr_len;
2285
2286                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2287                                        vnet_hdr_len);
2288                 if (err < 0)
2289                         goto out_unlock;
2290
2291                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2292                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2293                       vnet_hdr.hdr_len))
2294                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
2295                                                  vnet_hdr.csum_offset + 2;
2296
2297                 err = -EINVAL;
2298                 if (vnet_hdr.hdr_len > len)
2299                         goto out_unlock;
2300
2301                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2302                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2303                         case VIRTIO_NET_HDR_GSO_TCPV4:
2304                                 gso_type = SKB_GSO_TCPV4;
2305                                 break;
2306                         case VIRTIO_NET_HDR_GSO_TCPV6:
2307                                 gso_type = SKB_GSO_TCPV6;
2308                                 break;
2309                         case VIRTIO_NET_HDR_GSO_UDP:
2310                                 gso_type = SKB_GSO_UDP;
2311                                 break;
2312                         default:
2313                                 goto out_unlock;
2314                         }
2315
2316                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2317                                 gso_type |= SKB_GSO_TCP_ECN;
2318
2319                         if (vnet_hdr.gso_size == 0)
2320                                 goto out_unlock;
2321
2322                 }
2323         }
2324
2325         err = -EMSGSIZE;
2326         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN))
2327                 goto out_unlock;
2328
2329         err = -ENOBUFS;
2330         skb = packet_alloc_skb(sk, LL_ALLOCATED_SPACE(dev),
2331                                LL_RESERVED_SPACE(dev), len, vnet_hdr.hdr_len,
2332                                msg->msg_flags & MSG_DONTWAIT, &err);
2333         if (skb == NULL)
2334                 goto out_unlock;
2335
2336         skb_set_network_header(skb, reserve);
2337
2338         err = -EINVAL;
2339         if (sock->type == SOCK_DGRAM &&
2340             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2341                 goto out_free;
2342
2343         /* Returns -EFAULT on error */
2344         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2345         if (err)
2346                 goto out_free;
2347         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2348         if (err < 0)
2349                 goto out_free;
2350
2351         if (!gso_type && (len > dev->mtu + reserve)) {
2352                 /* Earlier code assumed this would be a VLAN pkt,
2353                  * double-check this now that we have the actual
2354                  * packet in hand.
2355                  */
2356                 struct ethhdr *ehdr;
2357                 skb_reset_mac_header(skb);
2358                 ehdr = eth_hdr(skb);
2359                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2360                         err = -EMSGSIZE;
2361                         goto out_free;
2362                 }
2363         }
2364
2365         skb->protocol = proto;
2366         skb->dev = dev;
2367         skb->priority = sk->sk_priority;
2368         skb->mark = sk->sk_mark;
2369
2370         if (po->has_vnet_hdr) {
2371                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2372                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2373                                                   vnet_hdr.csum_offset)) {
2374                                 err = -EINVAL;
2375                                 goto out_free;
2376                         }
2377                 }
2378
2379                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2380                 skb_shinfo(skb)->gso_type = gso_type;
2381
2382                 /* Header must be checked, and gso_segs computed. */
2383                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2384                 skb_shinfo(skb)->gso_segs = 0;
2385
2386                 len += vnet_hdr_len;
2387         }
2388
2389         /*
2390          *      Now send it
2391          */
2392
2393         err = dev_queue_xmit(skb);
2394         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2395                 goto out_unlock;
2396
2397         if (need_rls_dev)
2398                 dev_put(dev);
2399
2400         return len;
2401
2402 out_free:
2403         kfree_skb(skb);
2404 out_unlock:
2405         if (dev && need_rls_dev)
2406                 dev_put(dev);
2407 out:
2408         return err;
2409 }
2410
2411 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2412                 struct msghdr *msg, size_t len)
2413 {
2414         struct sock *sk = sock->sk;
2415         struct packet_sock *po = pkt_sk(sk);
2416         if (po->tx_ring.pg_vec)
2417                 return tpacket_snd(po, msg);
2418         else
2419                 return packet_snd(sock, msg, len);
2420 }
2421
2422 /*
2423  *      Close a PACKET socket. This is fairly simple. We immediately go
2424  *      to 'closed' state and remove our protocol entry in the device list.
2425  */
2426
2427 static int packet_release(struct socket *sock)
2428 {
2429         struct sock *sk = sock->sk;
2430         struct packet_sock *po;
2431         struct net *net;
2432         union tpacket_req_u req_u;
2433
2434         if (!sk)
2435                 return 0;
2436
2437         net = sock_net(sk);
2438         po = pkt_sk(sk);
2439
2440         spin_lock_bh(&net->packet.sklist_lock);
2441         sk_del_node_init_rcu(sk);
2442         sock_prot_inuse_add(net, sk->sk_prot, -1);
2443         spin_unlock_bh(&net->packet.sklist_lock);
2444
2445         spin_lock(&po->bind_lock);
2446         unregister_prot_hook(sk, false);
2447         if (po->prot_hook.dev) {
2448                 dev_put(po->prot_hook.dev);
2449                 po->prot_hook.dev = NULL;
2450         }
2451         spin_unlock(&po->bind_lock);
2452
2453         packet_flush_mclist(sk);
2454
2455         memset(&req_u, 0, sizeof(req_u));
2456
2457         if (po->rx_ring.pg_vec)
2458                 packet_set_ring(sk, &req_u, 1, 0);
2459
2460         if (po->tx_ring.pg_vec)
2461                 packet_set_ring(sk, &req_u, 1, 1);
2462
2463         fanout_release(sk);
2464
2465         synchronize_net();
2466         /*
2467          *      Now the socket is dead. No more input will appear.
2468          */
2469         sock_orphan(sk);
2470         sock->sk = NULL;
2471
2472         /* Purge queues */
2473
2474         skb_queue_purge(&sk->sk_receive_queue);
2475         sk_refcnt_debug_release(sk);
2476
2477         sock_put(sk);
2478         return 0;
2479 }
2480
2481 /*
2482  *      Attach a packet hook.
2483  */
2484
2485 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2486 {
2487         struct packet_sock *po = pkt_sk(sk);
2488
2489         if (po->fanout)
2490                 return -EINVAL;
2491
2492         lock_sock(sk);
2493
2494         spin_lock(&po->bind_lock);
2495         unregister_prot_hook(sk, true);
2496         po->num = protocol;
2497         po->prot_hook.type = protocol;
2498         if (po->prot_hook.dev)
2499                 dev_put(po->prot_hook.dev);
2500         po->prot_hook.dev = dev;
2501
2502         po->ifindex = dev ? dev->ifindex : 0;
2503
2504         if (protocol == 0)
2505                 goto out_unlock;
2506
2507         if (!dev || (dev->flags & IFF_UP)) {
2508                 register_prot_hook(sk);
2509         } else {
2510                 sk->sk_err = ENETDOWN;
2511                 if (!sock_flag(sk, SOCK_DEAD))
2512                         sk->sk_error_report(sk);
2513         }
2514
2515 out_unlock:
2516         spin_unlock(&po->bind_lock);
2517         release_sock(sk);
2518         return 0;
2519 }
2520
2521 /*
2522  *      Bind a packet socket to a device
2523  */
2524
2525 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2526                             int addr_len)
2527 {
2528         struct sock *sk = sock->sk;
2529         char name[15];
2530         struct net_device *dev;
2531         int err = -ENODEV;
2532
2533         /*
2534          *      Check legality
2535          */
2536
2537         if (addr_len != sizeof(struct sockaddr))
2538                 return -EINVAL;
2539         strlcpy(name, uaddr->sa_data, sizeof(name));
2540
2541         dev = dev_get_by_name(sock_net(sk), name);
2542         if (dev)
2543                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2544         return err;
2545 }
2546
2547 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2548 {
2549         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2550         struct sock *sk = sock->sk;
2551         struct net_device *dev = NULL;
2552         int err;
2553
2554
2555         /*
2556          *      Check legality
2557          */
2558
2559         if (addr_len < sizeof(struct sockaddr_ll))
2560                 return -EINVAL;
2561         if (sll->sll_family != AF_PACKET)
2562                 return -EINVAL;
2563
2564         if (sll->sll_ifindex) {
2565                 err = -ENODEV;
2566                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2567                 if (dev == NULL)
2568                         goto out;
2569         }
2570         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2571
2572 out:
2573         return err;
2574 }
2575
2576 static struct proto packet_proto = {
2577         .name     = "PACKET",
2578         .owner    = THIS_MODULE,
2579         .obj_size = sizeof(struct packet_sock),
2580 };
2581
2582 /*
2583  *      Create a packet of type SOCK_PACKET.
2584  */
2585
2586 static int packet_create(struct net *net, struct socket *sock, int protocol,
2587                          int kern)
2588 {
2589         struct sock *sk;
2590         struct packet_sock *po;
2591         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2592         int err;
2593
2594         if (!capable(CAP_NET_RAW))
2595                 return -EPERM;
2596         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2597             sock->type != SOCK_PACKET)
2598                 return -ESOCKTNOSUPPORT;
2599
2600         sock->state = SS_UNCONNECTED;
2601
2602         err = -ENOBUFS;
2603         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2604         if (sk == NULL)
2605                 goto out;
2606
2607         sock->ops = &packet_ops;
2608         if (sock->type == SOCK_PACKET)
2609                 sock->ops = &packet_ops_spkt;
2610
2611         sock_init_data(sock, sk);
2612
2613         po = pkt_sk(sk);
2614         sk->sk_family = PF_PACKET;
2615         po->num = proto;
2616
2617         sk->sk_destruct = packet_sock_destruct;
2618         sk_refcnt_debug_inc(sk);
2619
2620         /*
2621          *      Attach a protocol block
2622          */
2623
2624         spin_lock_init(&po->bind_lock);
2625         mutex_init(&po->pg_vec_lock);
2626         po->prot_hook.func = packet_rcv;
2627
2628         if (sock->type == SOCK_PACKET)
2629                 po->prot_hook.func = packet_rcv_spkt;
2630
2631         po->prot_hook.af_packet_priv = sk;
2632
2633         if (proto) {
2634                 po->prot_hook.type = proto;
2635                 register_prot_hook(sk);
2636         }
2637
2638         spin_lock_bh(&net->packet.sklist_lock);
2639         sk_add_node_rcu(sk, &net->packet.sklist);
2640         sock_prot_inuse_add(net, &packet_proto, 1);
2641         spin_unlock_bh(&net->packet.sklist_lock);
2642
2643         return 0;
2644 out:
2645         return err;
2646 }
2647
2648 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2649 {
2650         struct sock_exterr_skb *serr;
2651         struct sk_buff *skb, *skb2;
2652         int copied, err;
2653
2654         err = -EAGAIN;
2655         skb = skb_dequeue(&sk->sk_error_queue);
2656         if (skb == NULL)
2657                 goto out;
2658
2659         copied = skb->len;
2660         if (copied > len) {
2661                 msg->msg_flags |= MSG_TRUNC;
2662                 copied = len;
2663         }
2664         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2665         if (err)
2666                 goto out_free_skb;
2667
2668         sock_recv_timestamp(msg, sk, skb);
2669
2670         serr = SKB_EXT_ERR(skb);
2671         put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2672                  sizeof(serr->ee), &serr->ee);
2673
2674         msg->msg_flags |= MSG_ERRQUEUE;
2675         err = copied;
2676
2677         /* Reset and regenerate socket error */
2678         spin_lock_bh(&sk->sk_error_queue.lock);
2679         sk->sk_err = 0;
2680         if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2681                 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2682                 spin_unlock_bh(&sk->sk_error_queue.lock);
2683                 sk->sk_error_report(sk);
2684         } else
2685                 spin_unlock_bh(&sk->sk_error_queue.lock);
2686
2687 out_free_skb:
2688         kfree_skb(skb);
2689 out:
2690         return err;
2691 }
2692
2693 /*
2694  *      Pull a packet from our receive queue and hand it to the user.
2695  *      If necessary we block.
2696  */
2697
2698 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2699                           struct msghdr *msg, size_t len, int flags)
2700 {
2701         struct sock *sk = sock->sk;
2702         struct sk_buff *skb;
2703         int copied, err;
2704         struct sockaddr_ll *sll;
2705         int vnet_hdr_len = 0;
2706
2707         err = -EINVAL;
2708         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2709                 goto out;
2710
2711 #if 0
2712         /* What error should we return now? EUNATTACH? */
2713         if (pkt_sk(sk)->ifindex < 0)
2714                 return -ENODEV;
2715 #endif
2716
2717         if (flags & MSG_ERRQUEUE) {
2718                 err = packet_recv_error(sk, msg, len);
2719                 goto out;
2720         }
2721
2722         /*
2723          *      Call the generic datagram receiver. This handles all sorts
2724          *      of horrible races and re-entrancy so we can forget about it
2725          *      in the protocol layers.
2726          *
2727          *      Now it will return ENETDOWN, if device have just gone down,
2728          *      but then it will block.
2729          */
2730
2731         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2732
2733         /*
2734          *      An error occurred so return it. Because skb_recv_datagram()
2735          *      handles the blocking we don't see and worry about blocking
2736          *      retries.
2737          */
2738
2739         if (skb == NULL)
2740                 goto out;
2741
2742         if (pkt_sk(sk)->has_vnet_hdr) {
2743                 struct virtio_net_hdr vnet_hdr = { 0 };
2744
2745                 err = -EINVAL;
2746                 vnet_hdr_len = sizeof(vnet_hdr);
2747                 if (len < vnet_hdr_len)
2748                         goto out_free;
2749
2750                 len -= vnet_hdr_len;
2751
2752                 if (skb_is_gso(skb)) {
2753                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2754
2755                         /* This is a hint as to how much should be linear. */
2756                         vnet_hdr.hdr_len = skb_headlen(skb);
2757                         vnet_hdr.gso_size = sinfo->gso_size;
2758                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2759                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2760                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2761                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2762                         else if (sinfo->gso_type & SKB_GSO_UDP)
2763                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2764                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2765                                 goto out_free;
2766                         else
2767                                 BUG();
2768                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2769                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2770                 } else
2771                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2772
2773                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2774                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2775                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2776                         vnet_hdr.csum_offset = skb->csum_offset;
2777                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2778                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2779                 } /* else everything is zero */
2780
2781                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2782                                      vnet_hdr_len);
2783                 if (err < 0)
2784                         goto out_free;
2785         }
2786
2787         /*
2788          *      If the address length field is there to be filled in, we fill
2789          *      it in now.
2790          */
2791
2792         sll = &PACKET_SKB_CB(skb)->sa.ll;
2793         if (sock->type == SOCK_PACKET)
2794                 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2795         else
2796                 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2797
2798         /*
2799          *      You lose any data beyond the buffer you gave. If it worries a
2800          *      user program they can ask the device for its MTU anyway.
2801          */
2802
2803         copied = skb->len;
2804         if (copied > len) {
2805                 copied = len;
2806                 msg->msg_flags |= MSG_TRUNC;
2807         }
2808
2809         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2810         if (err)
2811                 goto out_free;
2812
2813         sock_recv_ts_and_drops(msg, sk, skb);
2814
2815         if (msg->msg_name)
2816                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2817                        msg->msg_namelen);
2818
2819         if (pkt_sk(sk)->auxdata) {
2820                 struct tpacket_auxdata aux;
2821
2822                 aux.tp_status = TP_STATUS_USER;
2823                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2824                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2825                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2826                 aux.tp_snaplen = skb->len;
2827                 aux.tp_mac = 0;
2828                 aux.tp_net = skb_network_offset(skb);
2829                 if (vlan_tx_tag_present(skb)) {
2830                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2831                         aux.tp_status |= TP_STATUS_VLAN_VALID;
2832                 } else {
2833                         aux.tp_vlan_tci = 0;
2834                 }
2835                 aux.tp_padding = 0;
2836                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2837         }
2838
2839         /*
2840          *      Free or return the buffer as appropriate. Again this
2841          *      hides all the races and re-entrancy issues from us.
2842          */
2843         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2844
2845 out_free:
2846         skb_free_datagram(sk, skb);
2847 out:
2848         return err;
2849 }
2850
2851 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2852                                int *uaddr_len, int peer)
2853 {
2854         struct net_device *dev;
2855         struct sock *sk = sock->sk;
2856
2857         if (peer)
2858                 return -EOPNOTSUPP;
2859
2860         uaddr->sa_family = AF_PACKET;
2861         rcu_read_lock();
2862         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2863         if (dev)
2864                 strncpy(uaddr->sa_data, dev->name, 14);
2865         else
2866                 memset(uaddr->sa_data, 0, 14);
2867         rcu_read_unlock();
2868         *uaddr_len = sizeof(*uaddr);
2869
2870         return 0;
2871 }
2872
2873 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2874                           int *uaddr_len, int peer)
2875 {
2876         struct net_device *dev;
2877         struct sock *sk = sock->sk;
2878         struct packet_sock *po = pkt_sk(sk);
2879         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2880
2881         if (peer)
2882                 return -EOPNOTSUPP;
2883
2884         sll->sll_family = AF_PACKET;
2885         sll->sll_ifindex = po->ifindex;
2886         sll->sll_protocol = po->num;
2887         sll->sll_pkttype = 0;
2888         rcu_read_lock();
2889         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2890         if (dev) {
2891                 sll->sll_hatype = dev->type;
2892                 sll->sll_halen = dev->addr_len;
2893                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2894         } else {
2895                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
2896                 sll->sll_halen = 0;
2897         }
2898         rcu_read_unlock();
2899         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2900
2901         return 0;
2902 }
2903
2904 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2905                          int what)
2906 {
2907         switch (i->type) {
2908         case PACKET_MR_MULTICAST:
2909                 if (i->alen != dev->addr_len)
2910                         return -EINVAL;
2911                 if (what > 0)
2912                         return dev_mc_add(dev, i->addr);
2913                 else
2914                         return dev_mc_del(dev, i->addr);
2915                 break;
2916         case PACKET_MR_PROMISC:
2917                 return dev_set_promiscuity(dev, what);
2918                 break;
2919         case PACKET_MR_ALLMULTI:
2920                 return dev_set_allmulti(dev, what);
2921                 break;
2922         case PACKET_MR_UNICAST:
2923                 if (i->alen != dev->addr_len)
2924                         return -EINVAL;
2925                 if (what > 0)
2926                         return dev_uc_add(dev, i->addr);
2927                 else
2928                         return dev_uc_del(dev, i->addr);
2929                 break;
2930         default:
2931                 break;
2932         }
2933         return 0;
2934 }
2935
2936 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2937 {
2938         for ( ; i; i = i->next) {
2939                 if (i->ifindex == dev->ifindex)
2940                         packet_dev_mc(dev, i, what);
2941         }
2942 }
2943
2944 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2945 {
2946         struct packet_sock *po = pkt_sk(sk);
2947         struct packet_mclist *ml, *i;
2948         struct net_device *dev;
2949         int err;
2950
2951         rtnl_lock();
2952
2953         err = -ENODEV;
2954         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2955         if (!dev)
2956                 goto done;
2957
2958         err = -EINVAL;
2959         if (mreq->mr_alen > dev->addr_len)
2960                 goto done;
2961
2962         err = -ENOBUFS;
2963         i = kmalloc(sizeof(*i), GFP_KERNEL);
2964         if (i == NULL)
2965                 goto done;
2966
2967         err = 0;
2968         for (ml = po->mclist; ml; ml = ml->next) {
2969                 if (ml->ifindex == mreq->mr_ifindex &&
2970                     ml->type == mreq->mr_type &&
2971                     ml->alen == mreq->mr_alen &&
2972                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2973                         ml->count++;
2974                         /* Free the new element ... */
2975                         kfree(i);
2976                         goto done;
2977                 }
2978         }
2979
2980         i->type = mreq->mr_type;
2981         i->ifindex = mreq->mr_ifindex;
2982         i->alen = mreq->mr_alen;
2983         memcpy(i->addr, mreq->mr_address, i->alen);
2984         i->count = 1;
2985         i->next = po->mclist;
2986         po->mclist = i;
2987         err = packet_dev_mc(dev, i, 1);
2988         if (err) {
2989                 po->mclist = i->next;
2990                 kfree(i);
2991         }
2992
2993 done:
2994         rtnl_unlock();
2995         return err;
2996 }
2997
2998 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2999 {
3000         struct packet_mclist *ml, **mlp;
3001
3002         rtnl_lock();
3003
3004         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3005                 if (ml->ifindex == mreq->mr_ifindex &&
3006                     ml->type == mreq->mr_type &&
3007                     ml->alen == mreq->mr_alen &&
3008                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3009                         if (--ml->count == 0) {
3010                                 struct net_device *dev;
3011                                 *mlp = ml->next;
3012                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3013                                 if (dev)
3014                                         packet_dev_mc(dev, ml, -1);
3015                                 kfree(ml);
3016                         }
3017                         rtnl_unlock();
3018                         return 0;
3019                 }
3020         }
3021         rtnl_unlock();
3022         return -EADDRNOTAVAIL;
3023 }
3024
3025 static void packet_flush_mclist(struct sock *sk)
3026 {
3027         struct packet_sock *po = pkt_sk(sk);
3028         struct packet_mclist *ml;
3029
3030         if (!po->mclist)
3031                 return;
3032
3033         rtnl_lock();
3034         while ((ml = po->mclist) != NULL) {
3035                 struct net_device *dev;
3036
3037                 po->mclist = ml->next;
3038                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3039                 if (dev != NULL)
3040                         packet_dev_mc(dev, ml, -1);
3041                 kfree(ml);
3042         }
3043         rtnl_unlock();
3044 }
3045
3046 static int
3047 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3048 {
3049         struct sock *sk = sock->sk;
3050         struct packet_sock *po = pkt_sk(sk);
3051         int ret;
3052
3053         if (level != SOL_PACKET)
3054                 return -ENOPROTOOPT;
3055
3056         switch (optname) {
3057         case PACKET_ADD_MEMBERSHIP:
3058         case PACKET_DROP_MEMBERSHIP:
3059         {
3060                 struct packet_mreq_max mreq;
3061                 int len = optlen;
3062                 memset(&mreq, 0, sizeof(mreq));
3063                 if (len < sizeof(struct packet_mreq))
3064                         return -EINVAL;
3065                 if (len > sizeof(mreq))
3066                         len = sizeof(mreq);
3067                 if (copy_from_user(&mreq, optval, len))
3068                         return -EFAULT;
3069                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3070                         return -EINVAL;
3071                 if (optname == PACKET_ADD_MEMBERSHIP)
3072                         ret = packet_mc_add(sk, &mreq);
3073                 else
3074                         ret = packet_mc_drop(sk, &mreq);
3075                 return ret;
3076         }
3077
3078         case PACKET_RX_RING:
3079         case PACKET_TX_RING:
3080         {
3081                 union tpacket_req_u req_u;
3082                 int len;
3083
3084                 switch (po->tp_version) {
3085                 case TPACKET_V1:
3086                 case TPACKET_V2:
3087                         len = sizeof(req_u.req);
3088                         break;
3089                 case TPACKET_V3:
3090                 default:
3091                         len = sizeof(req_u.req3);
3092                         break;
3093                 }
3094                 if (optlen < len)
3095                         return -EINVAL;
3096                 if (pkt_sk(sk)->has_vnet_hdr)
3097                         return -EINVAL;
3098                 if (copy_from_user(&req_u.req, optval, len))
3099                         return -EFAULT;
3100                 return packet_set_ring(sk, &req_u, 0,
3101                         optname == PACKET_TX_RING);
3102         }
3103         case PACKET_COPY_THRESH:
3104         {
3105                 int val;
3106
3107                 if (optlen != sizeof(val))
3108                         return -EINVAL;
3109                 if (copy_from_user(&val, optval, sizeof(val)))
3110                         return -EFAULT;
3111
3112                 pkt_sk(sk)->copy_thresh = val;
3113                 return 0;
3114         }
3115         case PACKET_VERSION:
3116         {
3117                 int val;
3118
3119                 if (optlen != sizeof(val))
3120                         return -EINVAL;
3121                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3122                         return -EBUSY;
3123                 if (copy_from_user(&val, optval, sizeof(val)))
3124                         return -EFAULT;
3125                 switch (val) {
3126                 case TPACKET_V1:
3127                 case TPACKET_V2:
3128                 case TPACKET_V3:
3129                         po->tp_version = val;
3130                         return 0;
3131                 default:
3132                         return -EINVAL;
3133                 }
3134         }
3135         case PACKET_RESERVE:
3136         {
3137                 unsigned int val;
3138
3139                 if (optlen != sizeof(val))
3140                         return -EINVAL;
3141                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3142                         return -EBUSY;
3143                 if (copy_from_user(&val, optval, sizeof(val)))
3144                         return -EFAULT;
3145                 po->tp_reserve = val;
3146                 return 0;
3147         }
3148         case PACKET_LOSS:
3149         {
3150                 unsigned int val;
3151
3152                 if (optlen != sizeof(val))
3153                         return -EINVAL;
3154                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3155                         return -EBUSY;
3156                 if (copy_from_user(&val, optval, sizeof(val)))
3157                         return -EFAULT;
3158                 po->tp_loss = !!val;
3159                 return 0;
3160         }
3161         case PACKET_AUXDATA:
3162         {
3163                 int val;
3164
3165                 if (optlen < sizeof(val))
3166                         return -EINVAL;
3167                 if (copy_from_user(&val, optval, sizeof(val)))
3168                         return -EFAULT;
3169
3170                 po->auxdata = !!val;
3171                 return 0;
3172         }
3173         case PACKET_ORIGDEV:
3174         {
3175                 int val;
3176
3177                 if (optlen < sizeof(val))
3178                         return -EINVAL;
3179                 if (copy_from_user(&val, optval, sizeof(val)))
3180                         return -EFAULT;
3181
3182                 po->origdev = !!val;
3183                 return 0;
3184         }
3185         case PACKET_VNET_HDR:
3186         {
3187                 int val;
3188
3189                 if (sock->type != SOCK_RAW)
3190                         return -EINVAL;
3191                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3192                         return -EBUSY;
3193                 if (optlen < sizeof(val))
3194                         return -EINVAL;
3195                 if (copy_from_user(&val, optval, sizeof(val)))
3196                         return -EFAULT;
3197
3198                 po->has_vnet_hdr = !!val;
3199                 return 0;
3200         }
3201         case PACKET_TIMESTAMP:
3202         {
3203                 int val;
3204
3205                 if (optlen != sizeof(val))
3206                         return -EINVAL;
3207                 if (copy_from_user(&val, optval, sizeof(val)))
3208                         return -EFAULT;
3209
3210                 po->tp_tstamp = val;
3211                 return 0;
3212         }
3213         case PACKET_FANOUT:
3214         {
3215                 int val;
3216
3217                 if (optlen != sizeof(val))
3218                         return -EINVAL;
3219                 if (copy_from_user(&val, optval, sizeof(val)))
3220                         return -EFAULT;
3221
3222                 return fanout_add(sk, val & 0xffff, val >> 16);
3223         }
3224         default:
3225                 return -ENOPROTOOPT;
3226         }
3227 }
3228
3229 static int packet_getsockopt(struct socket *sock, int level, int optname,
3230                              char __user *optval, int __user *optlen)
3231 {
3232         int len;
3233         int val;
3234         struct sock *sk = sock->sk;
3235         struct packet_sock *po = pkt_sk(sk);
3236         void *data;
3237         struct tpacket_stats st;
3238         union tpacket_stats_u st_u;
3239
3240         if (level != SOL_PACKET)
3241                 return -ENOPROTOOPT;
3242
3243         if (get_user(len, optlen))
3244                 return -EFAULT;
3245
3246         if (len < 0)
3247                 return -EINVAL;
3248
3249         switch (optname) {
3250         case PACKET_STATISTICS:
3251                 if (po->tp_version == TPACKET_V3) {
3252                         len = sizeof(struct tpacket_stats_v3);
3253                 } else {
3254                         if (len > sizeof(struct tpacket_stats))
3255                                 len = sizeof(struct tpacket_stats);
3256                 }
3257                 spin_lock_bh(&sk->sk_receive_queue.lock);
3258                 if (po->tp_version == TPACKET_V3) {
3259                         memcpy(&st_u.stats3, &po->stats,
3260                         sizeof(struct tpacket_stats));
3261                         st_u.stats3.tp_freeze_q_cnt =
3262                         po->stats_u.stats3.tp_freeze_q_cnt;
3263                         st_u.stats3.tp_packets += po->stats.tp_drops;
3264                         data = &st_u.stats3;
3265                 } else {
3266                         st = po->stats;
3267                         st.tp_packets += st.tp_drops;
3268                         data = &st;
3269                 }
3270                 memset(&po->stats, 0, sizeof(st));
3271                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3272                 break;
3273         case PACKET_AUXDATA:
3274                 if (len > sizeof(int))
3275                         len = sizeof(int);
3276                 val = po->auxdata;
3277
3278                 data = &val;
3279                 break;
3280         case PACKET_ORIGDEV:
3281                 if (len > sizeof(int))
3282                         len = sizeof(int);
3283                 val = po->origdev;
3284
3285                 data = &val;
3286                 break;
3287         case PACKET_VNET_HDR:
3288                 if (len > sizeof(int))
3289                         len = sizeof(int);
3290                 val = po->has_vnet_hdr;
3291
3292                 data = &val;
3293                 break;
3294         case PACKET_VERSION:
3295                 if (len > sizeof(int))
3296                         len = sizeof(int);
3297                 val = po->tp_version;
3298                 data = &val;
3299                 break;
3300         case PACKET_HDRLEN:
3301                 if (len > sizeof(int))
3302                         len = sizeof(int);
3303                 if (copy_from_user(&val, optval, len))
3304                         return -EFAULT;
3305                 switch (val) {
3306                 case TPACKET_V1:
3307                         val = sizeof(struct tpacket_hdr);
3308                         break;
3309                 case TPACKET_V2:
3310                         val = sizeof(struct tpacket2_hdr);
3311                         break;
3312                 case TPACKET_V3:
3313                         val = sizeof(struct tpacket3_hdr);
3314                         break;
3315                 default:
3316                         return -EINVAL;
3317                 }
3318                 data = &val;
3319                 break;
3320         case PACKET_RESERVE:
3321                 if (len > sizeof(unsigned int))
3322                         len = sizeof(unsigned int);
3323                 val = po->tp_reserve;
3324                 data = &val;
3325                 break;
3326         case PACKET_LOSS:
3327                 if (len > sizeof(unsigned int))
3328                         len = sizeof(unsigned int);
3329                 val = po->tp_loss;
3330                 data = &val;
3331                 break;
3332         case PACKET_TIMESTAMP:
3333                 if (len > sizeof(int))
3334                         len = sizeof(int);
3335                 val = po->tp_tstamp;
3336                 data = &val;
3337                 break;
3338         case PACKET_FANOUT:
3339                 if (len > sizeof(int))
3340                         len = sizeof(int);
3341                 val = (po->fanout ?
3342                        ((u32)po->fanout->id |
3343                         ((u32)po->fanout->type << 16)) :
3344                        0);
3345                 data = &val;
3346                 break;
3347         default:
3348                 return -ENOPROTOOPT;
3349         }
3350
3351         if (put_user(len, optlen))
3352                 return -EFAULT;
3353         if (copy_to_user(optval, data, len))
3354                 return -EFAULT;
3355         return 0;
3356 }
3357
3358
3359 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3360 {
3361         struct sock *sk;
3362         struct hlist_node *node;
3363         struct net_device *dev = data;
3364         struct net *net = dev_net(dev);
3365
3366         rcu_read_lock();
3367         sk_for_each_rcu(sk, node, &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                                         po->ifindex = -1;
3387                                         if (po->prot_hook.dev)
3388                                                 dev_put(po->prot_hook.dev);
3389                                         po->prot_hook.dev = NULL;
3390                                 }
3391                                 spin_unlock(&po->bind_lock);
3392                         }
3393                         break;
3394                 case NETDEV_UP:
3395                         if (dev->ifindex == po->ifindex) {
3396                                 spin_lock(&po->bind_lock);
3397                                 if (po->num)
3398                                         register_prot_hook(sk);
3399                                 spin_unlock(&po->bind_lock);
3400                         }
3401                         break;
3402                 }
3403         }
3404         rcu_read_unlock();
3405         return NOTIFY_DONE;
3406 }
3407
3408
3409 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3410                         unsigned long arg)
3411 {
3412         struct sock *sk = sock->sk;
3413
3414         switch (cmd) {
3415         case SIOCOUTQ:
3416         {
3417                 int amount = sk_wmem_alloc_get(sk);
3418
3419                 return put_user(amount, (int __user *)arg);
3420         }
3421         case SIOCINQ:
3422         {
3423                 struct sk_buff *skb;
3424                 int amount = 0;
3425
3426                 spin_lock_bh(&sk->sk_receive_queue.lock);
3427                 skb = skb_peek(&sk->sk_receive_queue);
3428                 if (skb)
3429                         amount = skb->len;
3430                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3431                 return put_user(amount, (int __user *)arg);
3432         }
3433         case SIOCGSTAMP:
3434                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3435         case SIOCGSTAMPNS:
3436                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3437
3438 #ifdef CONFIG_INET
3439         case SIOCADDRT:
3440         case SIOCDELRT:
3441         case SIOCDARP:
3442         case SIOCGARP:
3443         case SIOCSARP:
3444         case SIOCGIFADDR:
3445         case SIOCSIFADDR:
3446         case SIOCGIFBRDADDR:
3447         case SIOCSIFBRDADDR:
3448         case SIOCGIFNETMASK:
3449         case SIOCSIFNETMASK:
3450         case SIOCGIFDSTADDR:
3451         case SIOCSIFDSTADDR:
3452         case SIOCSIFFLAGS:
3453                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3454 #endif
3455
3456         default:
3457                 return -ENOIOCTLCMD;
3458         }
3459         return 0;
3460 }
3461
3462 static unsigned int packet_poll(struct file *file, struct socket *sock,
3463                                 poll_table *wait)
3464 {
3465         struct sock *sk = sock->sk;
3466         struct packet_sock *po = pkt_sk(sk);
3467         unsigned int mask = datagram_poll(file, sock, wait);
3468
3469         spin_lock_bh(&sk->sk_receive_queue.lock);
3470         if (po->rx_ring.pg_vec) {
3471                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3472                         TP_STATUS_KERNEL))
3473                         mask |= POLLIN | POLLRDNORM;
3474         }
3475         spin_unlock_bh(&sk->sk_receive_queue.lock);
3476         spin_lock_bh(&sk->sk_write_queue.lock);
3477         if (po->tx_ring.pg_vec) {
3478                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3479                         mask |= POLLOUT | POLLWRNORM;
3480         }
3481         spin_unlock_bh(&sk->sk_write_queue.lock);
3482         return mask;
3483 }
3484
3485
3486 /* Dirty? Well, I still did not learn better way to account
3487  * for user mmaps.
3488  */
3489
3490 static void packet_mm_open(struct vm_area_struct *vma)
3491 {
3492         struct file *file = vma->vm_file;
3493         struct socket *sock = file->private_data;
3494         struct sock *sk = sock->sk;
3495
3496         if (sk)
3497                 atomic_inc(&pkt_sk(sk)->mapped);
3498 }
3499
3500 static void packet_mm_close(struct vm_area_struct *vma)
3501 {
3502         struct file *file = vma->vm_file;
3503         struct socket *sock = file->private_data;
3504         struct sock *sk = sock->sk;
3505
3506         if (sk)
3507                 atomic_dec(&pkt_sk(sk)->mapped);
3508 }
3509
3510 static const struct vm_operations_struct packet_mmap_ops = {
3511         .open   =       packet_mm_open,
3512         .close  =       packet_mm_close,
3513 };
3514
3515 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3516                         unsigned int len)
3517 {
3518         int i;
3519
3520         for (i = 0; i < len; i++) {
3521                 if (likely(pg_vec[i].buffer)) {
3522                         if (is_vmalloc_addr(pg_vec[i].buffer))
3523                                 vfree(pg_vec[i].buffer);
3524                         else
3525                                 free_pages((unsigned long)pg_vec[i].buffer,
3526                                            order);
3527                         pg_vec[i].buffer = NULL;
3528                 }
3529         }
3530         kfree(pg_vec);
3531 }
3532
3533 static inline char *alloc_one_pg_vec_page(unsigned long order)
3534 {
3535         char *buffer = NULL;
3536         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3537                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3538
3539         buffer = (char *) __get_free_pages(gfp_flags, order);
3540
3541         if (buffer)
3542                 return buffer;
3543
3544         /*
3545          * __get_free_pages failed, fall back to vmalloc
3546          */
3547         buffer = vzalloc((1 << order) * PAGE_SIZE);
3548
3549         if (buffer)
3550                 return buffer;
3551
3552         /*
3553          * vmalloc failed, lets dig into swap here
3554          */
3555         gfp_flags &= ~__GFP_NORETRY;
3556         buffer = (char *)__get_free_pages(gfp_flags, order);
3557         if (buffer)
3558                 return buffer;
3559
3560         /*
3561          * complete and utter failure
3562          */
3563         return NULL;
3564 }
3565
3566 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3567 {
3568         unsigned int block_nr = req->tp_block_nr;
3569         struct pgv *pg_vec;
3570         int i;
3571
3572         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3573         if (unlikely(!pg_vec))
3574                 goto out;
3575
3576         for (i = 0; i < block_nr; i++) {
3577                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3578                 if (unlikely(!pg_vec[i].buffer))
3579                         goto out_free_pgvec;
3580         }
3581
3582 out:
3583         return pg_vec;
3584
3585 out_free_pgvec:
3586         free_pg_vec(pg_vec, order, block_nr);
3587         pg_vec = NULL;
3588         goto out;
3589 }
3590
3591 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3592                 int closing, int tx_ring)
3593 {
3594         struct pgv *pg_vec = NULL;
3595         struct packet_sock *po = pkt_sk(sk);
3596         int was_running, order = 0;
3597         struct packet_ring_buffer *rb;
3598         struct sk_buff_head *rb_queue;
3599         __be16 num;
3600         int err = -EINVAL;
3601         /* Added to avoid minimal code churn */
3602         struct tpacket_req *req = &req_u->req;
3603
3604         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3605         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3606                 WARN(1, "Tx-ring is not supported.\n");
3607                 goto out;
3608         }
3609
3610         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3611         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3612
3613         err = -EBUSY;
3614         if (!closing) {
3615                 if (atomic_read(&po->mapped))
3616                         goto out;
3617                 if (atomic_read(&rb->pending))
3618                         goto out;
3619         }
3620
3621         if (req->tp_block_nr) {
3622                 /* Sanity tests and some calculations */
3623                 err = -EBUSY;
3624                 if (unlikely(rb->pg_vec))
3625                         goto out;
3626
3627                 switch (po->tp_version) {
3628                 case TPACKET_V1:
3629                         po->tp_hdrlen = TPACKET_HDRLEN;
3630                         break;
3631                 case TPACKET_V2:
3632                         po->tp_hdrlen = TPACKET2_HDRLEN;
3633                         break;
3634                 case TPACKET_V3:
3635                         po->tp_hdrlen = TPACKET3_HDRLEN;
3636                         break;
3637                 }
3638
3639                 err = -EINVAL;
3640                 if (unlikely((int)req->tp_block_size <= 0))
3641                         goto out;
3642                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3643                         goto out;
3644                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3645                                         po->tp_reserve))
3646                         goto out;
3647                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3648                         goto out;
3649
3650                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3651                 if (unlikely(rb->frames_per_block <= 0))
3652                         goto out;
3653                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3654                                         req->tp_frame_nr))
3655                         goto out;
3656
3657                 err = -ENOMEM;
3658                 order = get_order(req->tp_block_size);
3659                 pg_vec = alloc_pg_vec(req, order);
3660                 if (unlikely(!pg_vec))
3661                         goto out;
3662                 switch (po->tp_version) {
3663                 case TPACKET_V3:
3664                 /* Transmit path is not supported. We checked
3665                  * it above but just being paranoid
3666                  */
3667                         if (!tx_ring)
3668                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3669                                 break;
3670                 default:
3671                         break;
3672                 }
3673         }
3674         /* Done */
3675         else {
3676                 err = -EINVAL;
3677                 if (unlikely(req->tp_frame_nr))
3678                         goto out;
3679         }
3680
3681         lock_sock(sk);
3682
3683         /* Detach socket from network */
3684         spin_lock(&po->bind_lock);
3685         was_running = po->running;
3686         num = po->num;
3687         if (was_running) {
3688                 po->num = 0;
3689                 __unregister_prot_hook(sk, false);
3690         }
3691         spin_unlock(&po->bind_lock);
3692
3693         synchronize_net();
3694
3695         err = -EBUSY;
3696         mutex_lock(&po->pg_vec_lock);
3697         if (closing || atomic_read(&po->mapped) == 0) {
3698                 err = 0;
3699                 spin_lock_bh(&rb_queue->lock);
3700                 swap(rb->pg_vec, pg_vec);
3701                 rb->frame_max = (req->tp_frame_nr - 1);
3702                 rb->head = 0;
3703                 rb->frame_size = req->tp_frame_size;
3704                 spin_unlock_bh(&rb_queue->lock);
3705
3706                 swap(rb->pg_vec_order, order);
3707                 swap(rb->pg_vec_len, req->tp_block_nr);
3708
3709                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3710                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3711                                                 tpacket_rcv : packet_rcv;
3712                 skb_queue_purge(rb_queue);
3713                 if (atomic_read(&po->mapped))
3714                         pr_err("packet_mmap: vma is busy: %d\n",
3715                                atomic_read(&po->mapped));
3716         }
3717         mutex_unlock(&po->pg_vec_lock);
3718
3719         spin_lock(&po->bind_lock);
3720         if (was_running) {
3721                 po->num = num;
3722                 register_prot_hook(sk);
3723         }
3724         spin_unlock(&po->bind_lock);
3725         if (closing && (po->tp_version > TPACKET_V2)) {
3726                 /* Because we don't support block-based V3 on tx-ring */
3727                 if (!tx_ring)
3728                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3729         }
3730         release_sock(sk);
3731
3732         if (pg_vec)
3733                 free_pg_vec(pg_vec, order, req->tp_block_nr);
3734 out:
3735         return err;
3736 }
3737
3738 static int packet_mmap(struct file *file, struct socket *sock,
3739                 struct vm_area_struct *vma)
3740 {
3741         struct sock *sk = sock->sk;
3742         struct packet_sock *po = pkt_sk(sk);
3743         unsigned long size, expected_size;
3744         struct packet_ring_buffer *rb;
3745         unsigned long start;
3746         int err = -EINVAL;
3747         int i;
3748
3749         if (vma->vm_pgoff)
3750                 return -EINVAL;
3751
3752         mutex_lock(&po->pg_vec_lock);
3753
3754         expected_size = 0;
3755         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3756                 if (rb->pg_vec) {
3757                         expected_size += rb->pg_vec_len
3758                                                 * rb->pg_vec_pages
3759                                                 * PAGE_SIZE;
3760                 }
3761         }
3762
3763         if (expected_size == 0)
3764                 goto out;
3765
3766         size = vma->vm_end - vma->vm_start;
3767         if (size != expected_size)
3768                 goto out;
3769
3770         start = vma->vm_start;
3771         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3772                 if (rb->pg_vec == NULL)
3773                         continue;
3774
3775                 for (i = 0; i < rb->pg_vec_len; i++) {
3776                         struct page *page;
3777                         void *kaddr = rb->pg_vec[i].buffer;
3778                         int pg_num;
3779
3780                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3781                                 page = pgv_to_page(kaddr);
3782                                 err = vm_insert_page(vma, start, page);
3783                                 if (unlikely(err))
3784                                         goto out;
3785                                 start += PAGE_SIZE;
3786                                 kaddr += PAGE_SIZE;
3787                         }
3788                 }
3789         }
3790
3791         atomic_inc(&po->mapped);
3792         vma->vm_ops = &packet_mmap_ops;
3793         err = 0;
3794
3795 out:
3796         mutex_unlock(&po->pg_vec_lock);
3797         return err;
3798 }
3799
3800 static const struct proto_ops packet_ops_spkt = {
3801         .family =       PF_PACKET,
3802         .owner =        THIS_MODULE,
3803         .release =      packet_release,
3804         .bind =         packet_bind_spkt,
3805         .connect =      sock_no_connect,
3806         .socketpair =   sock_no_socketpair,
3807         .accept =       sock_no_accept,
3808         .getname =      packet_getname_spkt,
3809         .poll =         datagram_poll,
3810         .ioctl =        packet_ioctl,
3811         .listen =       sock_no_listen,
3812         .shutdown =     sock_no_shutdown,
3813         .setsockopt =   sock_no_setsockopt,
3814         .getsockopt =   sock_no_getsockopt,
3815         .sendmsg =      packet_sendmsg_spkt,
3816         .recvmsg =      packet_recvmsg,
3817         .mmap =         sock_no_mmap,
3818         .sendpage =     sock_no_sendpage,
3819 };
3820
3821 static const struct proto_ops packet_ops = {
3822         .family =       PF_PACKET,
3823         .owner =        THIS_MODULE,
3824         .release =      packet_release,
3825         .bind =         packet_bind,
3826         .connect =      sock_no_connect,
3827         .socketpair =   sock_no_socketpair,
3828         .accept =       sock_no_accept,
3829         .getname =      packet_getname,
3830         .poll =         packet_poll,
3831         .ioctl =        packet_ioctl,
3832         .listen =       sock_no_listen,
3833         .shutdown =     sock_no_shutdown,
3834         .setsockopt =   packet_setsockopt,
3835         .getsockopt =   packet_getsockopt,
3836         .sendmsg =      packet_sendmsg,
3837         .recvmsg =      packet_recvmsg,
3838         .mmap =         packet_mmap,
3839         .sendpage =     sock_no_sendpage,
3840 };
3841
3842 static const struct net_proto_family packet_family_ops = {
3843         .family =       PF_PACKET,
3844         .create =       packet_create,
3845         .owner  =       THIS_MODULE,
3846 };
3847
3848 static struct notifier_block packet_netdev_notifier = {
3849         .notifier_call =        packet_notifier,
3850 };
3851
3852 #ifdef CONFIG_PROC_FS
3853
3854 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3855         __acquires(RCU)
3856 {
3857         struct net *net = seq_file_net(seq);
3858
3859         rcu_read_lock();
3860         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3861 }
3862
3863 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3864 {
3865         struct net *net = seq_file_net(seq);
3866         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3867 }
3868
3869 static void packet_seq_stop(struct seq_file *seq, void *v)
3870         __releases(RCU)
3871 {
3872         rcu_read_unlock();
3873 }
3874
3875 static int packet_seq_show(struct seq_file *seq, void *v)
3876 {
3877         if (v == SEQ_START_TOKEN)
3878                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
3879         else {
3880                 struct sock *s = sk_entry(v);
3881                 const struct packet_sock *po = pkt_sk(s);
3882
3883                 seq_printf(seq,
3884                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
3885                            s,
3886                            atomic_read(&s->sk_refcnt),
3887                            s->sk_type,
3888                            ntohs(po->num),
3889                            po->ifindex,
3890                            po->running,
3891                            atomic_read(&s->sk_rmem_alloc),
3892                            sock_i_uid(s),
3893                            sock_i_ino(s));
3894         }
3895
3896         return 0;
3897 }
3898
3899 static const struct seq_operations packet_seq_ops = {
3900         .start  = packet_seq_start,
3901         .next   = packet_seq_next,
3902         .stop   = packet_seq_stop,
3903         .show   = packet_seq_show,
3904 };
3905
3906 static int packet_seq_open(struct inode *inode, struct file *file)
3907 {
3908         return seq_open_net(inode, file, &packet_seq_ops,
3909                             sizeof(struct seq_net_private));
3910 }
3911
3912 static const struct file_operations packet_seq_fops = {
3913         .owner          = THIS_MODULE,
3914         .open           = packet_seq_open,
3915         .read           = seq_read,
3916         .llseek         = seq_lseek,
3917         .release        = seq_release_net,
3918 };
3919
3920 #endif
3921
3922 static int __net_init packet_net_init(struct net *net)
3923 {
3924         spin_lock_init(&net->packet.sklist_lock);
3925         INIT_HLIST_HEAD(&net->packet.sklist);
3926
3927         if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
3928                 return -ENOMEM;
3929
3930         return 0;
3931 }
3932
3933 static void __net_exit packet_net_exit(struct net *net)
3934 {
3935         proc_net_remove(net, "packet");
3936 }
3937
3938 static struct pernet_operations packet_net_ops = {
3939         .init = packet_net_init,
3940         .exit = packet_net_exit,
3941 };
3942
3943
3944 static void __exit packet_exit(void)
3945 {
3946         unregister_netdevice_notifier(&packet_netdev_notifier);
3947         unregister_pernet_subsys(&packet_net_ops);
3948         sock_unregister(PF_PACKET);
3949         proto_unregister(&packet_proto);
3950 }
3951
3952 static int __init packet_init(void)
3953 {
3954         int rc = proto_register(&packet_proto, 0);
3955
3956         if (rc != 0)
3957                 goto out;
3958
3959         sock_register(&packet_family_ops);
3960         register_pernet_subsys(&packet_net_ops);
3961         register_netdevice_notifier(&packet_netdev_notifier);
3962 out:
3963         return rc;
3964 }
3965
3966 module_init(packet_init);
3967 module_exit(packet_exit);
3968 MODULE_LICENSE("GPL");
3969 MODULE_ALIAS_NETPROTO(PF_PACKET);