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 net_device __rcu *cached_dev;
 299         struct packet_type      prot_hook ____cacheline_aligned_in_smp;
 300 };
 301
 302 #define PACKET_FANOUT_MAX       256
 303
 304 struct packet_fanout {
 305 #ifdef CONFIG_NET_NS
 306         struct net              *net;
 307 #endif
 308         unsigned int            num_members;
 309         u16                     id;
 310         u8                      type;
 311         u8                      defrag;
 312         atomic_t                rr_cur;
 313         struct list_head        list;
 314         struct sock             *arr[PACKET_FANOUT_MAX];
 315         spinlock_t              lock;
 316         atomic_t                sk_ref;
 317         struct packet_type      prot_hook ____cacheline_aligned_in_smp;
 318 };
 319
 320 struct packet_skb_cb {
 321         unsigned int origlen;
 322         union {
 323                 struct sockaddr_pkt pkt;
 324                 struct sockaddr_ll ll;
 325         } sa;
 326 };
 327
 328 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 329
 330 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 331 #define GET_PBLOCK_DESC(x, bid) \
 332         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 333 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 334         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 335 #define GET_NEXT_PRB_BLK_NUM(x) \
 336         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 337         ((x)->kactive_blk_num+1) : 0)
 338
 339 static struct packet_sock *pkt_sk(struct sock *sk)
 340 {
 341         return (struct packet_sock *)sk;
 342 }
 343
 344 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 345 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 346
 347 /* register_prot_hook must be invoked with the po->bind_lock held,
 348  * or from a context in which asynchronous accesses to the packet
 349  * socket is not possible (packet_create()).
 350  */
 351 static void register_prot_hook(struct sock *sk)
 352 {
 353         struct packet_sock *po = pkt_sk(sk);
 354
 355         if (!po->running) {
 356                 if (po->fanout) {
 357                         __fanout_link(sk, po);
 358                 } else {
 359                         dev_add_pack(&po->prot_hook);
 360                         rcu_assign_pointer(po->cached_dev, po->prot_hook.dev);
 361                 }
 362
 363                 sock_hold(sk);
 364                 po->running = 1;
 365         }
 366 }
 367
 368 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
 369  * held.   If the sync parameter is true, we will temporarily drop
 370  * the po->bind_lock and do a synchronize_net to make sure no
 371  * asynchronous packet processing paths still refer to the elements
 372  * of po->prot_hook.  If the sync parameter is false, it is the
 373  * callers responsibility to take care of this.
 374  */
 375 static void __unregister_prot_hook(struct sock *sk, bool sync)
 376 {
 377         struct packet_sock *po = pkt_sk(sk);
 378
 379         po->running = 0;
 380         if (po->fanout) {
 381                 __fanout_unlink(sk, po);
 382         } else {
 383                 __dev_remove_pack(&po->prot_hook);
 384                 RCU_INIT_POINTER(po->cached_dev, NULL);
 385         }
 386
 387         __sock_put(sk);
 388
 389         if (sync) {
 390                 spin_unlock(&po->bind_lock);
 391                 synchronize_net();
 392                 spin_lock(&po->bind_lock);
 393         }
 394 }
 395
 396 static void unregister_prot_hook(struct sock *sk, bool sync)
 397 {
 398         struct packet_sock *po = pkt_sk(sk);
 399
 400         if (po->running)
 401                 __unregister_prot_hook(sk, sync);
 402 }
 403
 404 static inline __pure struct page *pgv_to_page(void *addr)
 405 {
 406         if (is_vmalloc_addr(addr))
 407                 return vmalloc_to_page(addr);
 408         return virt_to_page(addr);
 409 }
 410
 411 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 412 {
 413         union {
 414                 struct tpacket_hdr *h1;
 415                 struct tpacket2_hdr *h2;
 416                 void *raw;
 417         } h;
 418
 419         h.raw = frame;
 420         switch (po->tp_version) {
 421         case TPACKET_V1:
 422                 h.h1->tp_status = status;
 423                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 424                 break;
 425         case TPACKET_V2:
 426                 h.h2->tp_status = status;
 427                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 428                 break;
 429         case TPACKET_V3:
 430         default:
 431                 WARN(1, "TPACKET version not supported.\n");
 432                 BUG();
 433         }
 434
 435         smp_wmb();
 436 }
 437
 438 static int __packet_get_status(struct packet_sock *po, void *frame)
 439 {
 440         union {
 441                 struct tpacket_hdr *h1;
 442                 struct tpacket2_hdr *h2;
 443                 void *raw;
 444         } h;
 445
 446         smp_rmb();
 447
 448         h.raw = frame;
 449         switch (po->tp_version) {
 450         case TPACKET_V1:
 451                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 452                 return h.h1->tp_status;
 453         case TPACKET_V2:
 454                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 455                 return h.h2->tp_status;
 456         case TPACKET_V3:
 457         default:
 458                 WARN(1, "TPACKET version not supported.\n");
 459                 BUG();
 460                 return 0;
 461         }
 462 }
 463
 464 static void *packet_lookup_frame(struct packet_sock *po,
 465                 struct packet_ring_buffer *rb,
 466                 unsigned int position,
 467                 int status)
 468 {
 469         unsigned int pg_vec_pos, frame_offset;
 470         union {
 471                 struct tpacket_hdr *h1;
 472                 struct tpacket2_hdr *h2;
 473                 void *raw;
 474         } h;
 475
 476         pg_vec_pos = position / rb->frames_per_block;
 477         frame_offset = position % rb->frames_per_block;
 478
 479         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 480                 (frame_offset * rb->frame_size);
 481
 482         if (status != __packet_get_status(po, h.raw))
 483                 return NULL;
 484
 485         return h.raw;
 486 }
 487
 488 static void *packet_current_frame(struct packet_sock *po,
 489                 struct packet_ring_buffer *rb,
 490                 int status)
 491 {
 492         return packet_lookup_frame(po, rb, rb->head, status);
 493 }
 494
 495 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 496 {
 497         del_timer_sync(&pkc->retire_blk_timer);
 498 }
 499
 500 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 501                 int tx_ring,
 502                 struct sk_buff_head *rb_queue)
 503 {
 504         struct tpacket_kbdq_core *pkc;
 505
 506         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 507
 508         spin_lock_bh(&rb_queue->lock);
 509         pkc->delete_blk_timer = 1;
 510         spin_unlock_bh(&rb_queue->lock);
 511
 512         prb_del_retire_blk_timer(pkc);
 513 }
 514
 515 static void prb_init_blk_timer(struct packet_sock *po,
 516                 struct tpacket_kbdq_core *pkc,
 517                 void (*func) (unsigned long))
 518 {
 519         init_timer(&pkc->retire_blk_timer);
 520         pkc->retire_blk_timer.data = (long)po;
 521         pkc->retire_blk_timer.function = func;
 522         pkc->retire_blk_timer.expires = jiffies;
 523 }
 524
 525 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
 526 {
 527         struct tpacket_kbdq_core *pkc;
 528
 529         if (tx_ring)
 530                 BUG();
 531
 532         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 533         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
 534 }
 535
 536 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 537                                 int blk_size_in_bytes)
 538 {
 539         struct net_device *dev;
 540         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 541         struct ethtool_cmd ecmd;
 542         int err;
 543
 544         rtnl_lock();
 545         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 546         if (unlikely(!dev)) {
 547                 rtnl_unlock();
 548                 return DEFAULT_PRB_RETIRE_TOV;
 549         }
 550         err = __ethtool_get_settings(dev, &ecmd);
 551         rtnl_unlock();
 552         if (!err) {
 553                 switch (ecmd.speed) {
 554                 case SPEED_10000:
 555                         msec = 1;
 556                         div = 10000/1000;
 557                         break;
 558                 case SPEED_1000:
 559                         msec = 1;
 560                         div = 1000/1000;
 561                         break;
 562                 /*
 563                  * If the link speed is so slow you don't really
 564                  * need to worry about perf anyways
 565                  */
 566                 case SPEED_100:
 567                 case SPEED_10:
 568                 default:
 569                         return DEFAULT_PRB_RETIRE_TOV;
 570                 }
 571         }
 572
 573         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 574
 575         if (div)
 576                 mbits /= div;
 577
 578         tmo = mbits * msec;
 579
 580         if (div)
 581                 return tmo+1;
 582         return tmo;
 583 }
 584
 585 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 586                         union tpacket_req_u *req_u)
 587 {
 588         p1->feature_req_word = req_u->req3.tp_feature_req_word;
 589 }
 590
 591 static void init_prb_bdqc(struct packet_sock *po,
 592                         struct packet_ring_buffer *rb,
 593                         struct pgv *pg_vec,
 594                         union tpacket_req_u *req_u, int tx_ring)
 595 {
 596         struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
 597         struct tpacket_block_desc *pbd;
 598
 599         memset(p1, 0x0, sizeof(*p1));
 600
 601         p1->knxt_seq_num = 1;
 602         p1->pkbdq = pg_vec;
 603         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 604         p1->pkblk_start = (char *)pg_vec[0].buffer;
 605         p1->kblk_size = req_u->req3.tp_block_size;
 606         p1->knum_blocks = req_u->req3.tp_block_nr;
 607         p1->hdrlen = po->tp_hdrlen;
 608         p1->version = po->tp_version;
 609         p1->last_kactive_blk_num = 0;
 610         po->stats_u.stats3.tp_freeze_q_cnt = 0;
 611         if (req_u->req3.tp_retire_blk_tov)
 612                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 613         else
 614                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 615                                                 req_u->req3.tp_block_size);
 616         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 617         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 618
 619         prb_init_ft_ops(p1, req_u);
 620         prb_setup_retire_blk_timer(po, tx_ring);
 621         prb_open_block(p1, pbd);
 622 }
 623
 624 /*  Do NOT update the last_blk_num first.
 625  *  Assumes sk_buff_head lock is held.
 626  */
 627 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 628 {
 629         mod_timer(&pkc->retire_blk_timer,
 630                         jiffies + pkc->tov_in_jiffies);
 631         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 632 }
 633
 634 /*
 635  * Timer logic:
 636  * 1) We refresh the timer only when we open a block.
 637  *    By doing this we don't waste cycles refreshing the timer
 638  *        on packet-by-packet basis.
 639  *
 640  * With a 1MB block-size, on a 1Gbps line, it will take
 641  * i) ~8 ms to fill a block + ii) memcpy etc.
 642  * In this cut we are not accounting for the memcpy time.
 643  *
 644  * So, if the user sets the 'tmo' to 10ms then the timer
 645  * will never fire while the block is still getting filled
 646  * (which is what we want). However, the user could choose
 647  * to close a block early and that's fine.
 648  *
 649  * But when the timer does fire, we check whether or not to refresh it.
 650  * Since the tmo granularity is in msecs, it is not too expensive
 651  * to refresh the timer, lets say every '8' msecs.
 652  * Either the user can set the 'tmo' or we can derive it based on
 653  * a) line-speed and b) block-size.
 654  * prb_calc_retire_blk_tmo() calculates the tmo.
 655  *
 656  */
 657 static void prb_retire_rx_blk_timer_expired(unsigned long data)
 658 {
 659         struct packet_sock *po = (struct packet_sock *)data;
 660         struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
 661         unsigned int frozen;
 662         struct tpacket_block_desc *pbd;
 663
 664         spin_lock(&po->sk.sk_receive_queue.lock);
 665
 666         frozen = prb_queue_frozen(pkc);
 667         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 668
 669         if (unlikely(pkc->delete_blk_timer))
 670                 goto out;
 671
 672         /* We only need to plug the race when the block is partially filled.
 673          * tpacket_rcv:
 674          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 675          *              copy_bits() is in progress ...
 676          *              timer fires on other cpu:
 677          *              we can't retire the current block because copy_bits
 678          *              is in progress.
 679          *
 680          */
 681         if (BLOCK_NUM_PKTS(pbd)) {
 682                 while (atomic_read(&pkc->blk_fill_in_prog)) {
 683                         /* Waiting for skb_copy_bits to finish... */
 684                         cpu_relax();
 685                 }
 686         }
 687
 688         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 689                 if (!frozen) {
 690                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 691                         if (!prb_dispatch_next_block(pkc, po))
 692                                 goto refresh_timer;
 693                         else
 694                                 goto out;
 695                 } else {
 696                         /* Case 1. Queue was frozen because user-space was
 697                          *         lagging behind.
 698                          */
 699                         if (prb_curr_blk_in_use(pkc, pbd)) {
 700                                 /*
 701                                  * Ok, user-space is still behind.
 702                                  * So just refresh the timer.
 703                                  */
 704                                 goto refresh_timer;
 705                         } else {
 706                                /* Case 2. queue was frozen,user-space caught up,
 707                                 * now the link went idle && the timer fired.
 708                                 * We don't have a block to close.So we open this
 709                                 * block and restart the timer.
 710                                 * opening a block thaws the queue,restarts timer
 711                                 * Thawing/timer-refresh is a side effect.
 712                                 */
 713                                 prb_open_block(pkc, pbd);
 714                                 goto out;
 715                         }
 716                 }
 717         }
 718
 719 refresh_timer:
 720         _prb_refresh_rx_retire_blk_timer(pkc);
 721
 722 out:
 723         spin_unlock(&po->sk.sk_receive_queue.lock);
 724 }
 725
 726 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 727                 struct tpacket_block_desc *pbd1, __u32 status)
 728 {
 729         /* Flush everything minus the block header */
 730
 731 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 732         u8 *start, *end;
 733
 734         start = (u8 *)pbd1;
 735
 736         /* Skip the block header(we know header WILL fit in 4K) */
 737         start += PAGE_SIZE;
 738
 739         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 740         for (; start < end; start += PAGE_SIZE)
 741                 flush_dcache_page(pgv_to_page(start));
 742
 743         smp_wmb();
 744 #endif
 745
 746         /* Now update the block status. */
 747
 748         BLOCK_STATUS(pbd1) = status;
 749
 750         /* Flush the block header */
 751
 752 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 753         start = (u8 *)pbd1;
 754         flush_dcache_page(pgv_to_page(start));
 755
 756         smp_wmb();
 757 #endif
 758 }
 759
 760 /*
 761  * Side effect:
 762  *
 763  * 1) flush the block
 764  * 2) Increment active_blk_num
 765  *
 766  * Note:We DONT refresh the timer on purpose.
 767  *      Because almost always the next block will be opened.
 768  */
 769 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 770                 struct tpacket_block_desc *pbd1,
 771                 struct packet_sock *po, unsigned int stat)
 772 {
 773         __u32 status = TP_STATUS_USER | stat;
 774
 775         struct tpacket3_hdr *last_pkt;
 776         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 777
 778         if (po->stats.tp_drops)
 779                 status |= TP_STATUS_LOSING;
 780
 781         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 782         last_pkt->tp_next_offset = 0;
 783
 784         /* Get the ts of the last pkt */
 785         if (BLOCK_NUM_PKTS(pbd1)) {
 786                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 787                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 788         } else {
 789                 /* Ok, we tmo'd - so get the current time */
 790                 struct timespec ts;
 791                 getnstimeofday(&ts);
 792                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 793                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 794         }
 795
 796         smp_wmb();
 797
 798         /* Flush the block */
 799         prb_flush_block(pkc1, pbd1, status);
 800
 801         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 802 }
 803
 804 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 805 {
 806         pkc->reset_pending_on_curr_blk = 0;
 807 }
 808
 809 /*
 810  * Side effect of opening a block:
 811  *
 812  * 1) prb_queue is thawed.
 813  * 2) retire_blk_timer is refreshed.
 814  *
 815  */
 816 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 817         struct tpacket_block_desc *pbd1)
 818 {
 819         struct timespec ts;
 820         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 821
 822         smp_rmb();
 823
 824         /* We could have just memset this but we will lose the
 825          * flexibility of making the priv area sticky
 826          */
 827         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 828         BLOCK_NUM_PKTS(pbd1) = 0;
 829         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 830         getnstimeofday(&ts);
 831         h1->ts_first_pkt.ts_sec = ts.tv_sec;
 832         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 833         pkc1->pkblk_start = (char *)pbd1;
 834         pkc1->nxt_offset = (char *)(pkc1->pkblk_start +
 835                                     BLK_PLUS_PRIV(pkc1->blk_sizeof_priv));
 836         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 837         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 838         pbd1->version = pkc1->version;
 839         pkc1->prev = pkc1->nxt_offset;
 840         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 841         prb_thaw_queue(pkc1);
 842         _prb_refresh_rx_retire_blk_timer(pkc1);
 843
 844         smp_wmb();
 845 }
 846
 847 /*
 848  * Queue freeze logic:
 849  * 1) Assume tp_block_nr = 8 blocks.
 850  * 2) At time 't0', user opens Rx ring.
 851  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 852  * 4) user-space is either sleeping or processing block '0'.
 853  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 854  *    it will close block-7,loop around and try to fill block '0'.
 855  *    call-flow:
 856  *    __packet_lookup_frame_in_block
 857  *      prb_retire_current_block()
 858  *      prb_dispatch_next_block()
 859  *        |->(BLOCK_STATUS == USER) evaluates to true
 860  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 861  * 6) Now there are two cases:
 862  *    6.1) Link goes idle right after the queue is frozen.
 863  *         But remember, the last open_block() refreshed the timer.
 864  *         When this timer expires,it will refresh itself so that we can
 865  *         re-open block-0 in near future.
 866  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 867  *         case and __packet_lookup_frame_in_block will check if block-0
 868  *         is free and can now be re-used.
 869  */
 870 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 871                                   struct packet_sock *po)
 872 {
 873         pkc->reset_pending_on_curr_blk = 1;
 874         po->stats_u.stats3.tp_freeze_q_cnt++;
 875 }
 876
 877 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 878
 879 /*
 880  * If the next block is free then we will dispatch it
 881  * and return a good offset.
 882  * Else, we will freeze the queue.
 883  * So, caller must check the return value.
 884  */
 885 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 886                 struct packet_sock *po)
 887 {
 888         struct tpacket_block_desc *pbd;
 889
 890         smp_rmb();
 891
 892         /* 1. Get current block num */
 893         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 894
 895         /* 2. If this block is currently in_use then freeze the queue */
 896         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 897                 prb_freeze_queue(pkc, po);
 898                 return NULL;
 899         }
 900
 901         /*
 902          * 3.
 903          * open this block and return the offset where the first packet
 904          * needs to get stored.
 905          */
 906         prb_open_block(pkc, pbd);
 907         return (void *)pkc->nxt_offset;
 908 }
 909
 910 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 911                 struct packet_sock *po, unsigned int status)
 912 {
 913         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 914
 915         /* retire/close the current block */
 916         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 917                 /*
 918                  * Plug the case where copy_bits() is in progress on
 919                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 920                  * have space to copy the pkt in the current block and
 921                  * called prb_retire_current_block()
 922                  *
 923                  * We don't need to worry about the TMO case because
 924                  * the timer-handler already handled this case.
 925                  */
 926                 if (!(status & TP_STATUS_BLK_TMO)) {
 927                         while (atomic_read(&pkc->blk_fill_in_prog)) {
 928                                 /* Waiting for skb_copy_bits to finish... */
 929                                 cpu_relax();
 930                         }
 931                 }
 932                 prb_close_block(pkc, pbd, po, status);
 933                 return;
 934         }
 935 }
 936
 937 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
 938                                       struct tpacket_block_desc *pbd)
 939 {
 940         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 941 }
 942
 943 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 944 {
 945         return pkc->reset_pending_on_curr_blk;
 946 }
 947
 948 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 949 {
 950         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 951         atomic_dec(&pkc->blk_fill_in_prog);
 952 }
 953
 954 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 955                         struct tpacket3_hdr *ppd)
 956 {
 957         ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
 958 }
 959
 960 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 961                         struct tpacket3_hdr *ppd)
 962 {
 963         ppd->hv1.tp_rxhash = 0;
 964 }
 965
 966 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
 967                         struct tpacket3_hdr *ppd)
 968 {
 969         if (vlan_tx_tag_present(pkc->skb)) {
 970                 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
 971                 ppd->tp_status = TP_STATUS_VLAN_VALID;
 972         } else {
 973                 ppd->hv1.tp_vlan_tci = ppd->tp_status = 0;
 974         }
 975 }
 976
 977 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
 978                         struct tpacket3_hdr *ppd)
 979 {
 980         prb_fill_vlan_info(pkc, ppd);
 981
 982         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
 983                 prb_fill_rxhash(pkc, ppd);
 984         else
 985                 prb_clear_rxhash(pkc, ppd);
 986 }
 987
 988 static void prb_fill_curr_block(char *curr,
 989                                 struct tpacket_kbdq_core *pkc,
 990                                 struct tpacket_block_desc *pbd,
 991                                 unsigned int len)
 992 {
 993         struct tpacket3_hdr *ppd;
 994
 995         ppd  = (struct tpacket3_hdr *)curr;
 996         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
 997         pkc->prev = curr;
 998         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
 999         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1000         BLOCK_NUM_PKTS(pbd) += 1;
1001         atomic_inc(&pkc->blk_fill_in_prog);
1002         prb_run_all_ft_ops(pkc, ppd);
1003 }
1004
1005 /* Assumes caller has the sk->rx_queue.lock */
1006 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1007                                             struct sk_buff *skb,
1008                                                 int status,
1009                                             unsigned int len
1010                                             )
1011 {
1012         struct tpacket_kbdq_core *pkc;
1013         struct tpacket_block_desc *pbd;
1014         char *curr, *end;
1015
1016         pkc = GET_PBDQC_FROM_RB(((struct packet_ring_buffer *)&po->rx_ring));
1017         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1018
1019         /* Queue is frozen when user space is lagging behind */
1020         if (prb_queue_frozen(pkc)) {
1021                 /*
1022                  * Check if that last block which caused the queue to freeze,
1023                  * is still in_use by user-space.
1024                  */
1025                 if (prb_curr_blk_in_use(pkc, pbd)) {
1026                         /* Can't record this packet */
1027                         return NULL;
1028                 } else {
1029                         /*
1030                          * Ok, the block was released by user-space.
1031                          * Now let's open that block.
1032                          * opening a block also thaws the queue.
1033                          * Thawing is a side effect.
1034                          */
1035                         prb_open_block(pkc, pbd);
1036                 }
1037         }
1038
1039         smp_mb();
1040         curr = pkc->nxt_offset;
1041         pkc->skb = skb;
1042         end = (char *) ((char *)pbd + pkc->kblk_size);
1043
1044         /* first try the current block */
1045         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1046                 prb_fill_curr_block(curr, pkc, pbd, len);
1047                 return (void *)curr;
1048         }
1049
1050         /* Ok, close the current block */
1051         prb_retire_current_block(pkc, po, 0);
1052
1053         /* Now, try to dispatch the next block */
1054         curr = (char *)prb_dispatch_next_block(pkc, po);
1055         if (curr) {
1056                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1057                 prb_fill_curr_block(curr, pkc, pbd, len);
1058                 return (void *)curr;
1059         }
1060
1061         /*
1062          * No free blocks are available.user_space hasn't caught up yet.
1063          * Queue was just frozen and now this packet will get dropped.
1064          */
1065         return NULL;
1066 }
1067
1068 static void *packet_current_rx_frame(struct packet_sock *po,
1069                                             struct sk_buff *skb,
1070                                             int status, unsigned int len)
1071 {
1072         char *curr = NULL;
1073         switch (po->tp_version) {
1074         case TPACKET_V1:
1075         case TPACKET_V2:
1076                 curr = packet_lookup_frame(po, &po->rx_ring,
1077                                         po->rx_ring.head, status);
1078                 return curr;
1079         case TPACKET_V3:
1080                 return __packet_lookup_frame_in_block(po, skb, status, len);
1081         default:
1082                 WARN(1, "TPACKET version not supported\n");
1083                 BUG();
1084                 return 0;
1085         }
1086 }
1087
1088 static void *prb_lookup_block(struct packet_sock *po,
1089                                      struct packet_ring_buffer *rb,
1090                                      unsigned int previous,
1091                                      int status)
1092 {
1093         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1094         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, previous);
1095
1096         if (status != BLOCK_STATUS(pbd))
1097                 return NULL;
1098         return pbd;
1099 }
1100
1101 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1102 {
1103         unsigned int prev;
1104         if (rb->prb_bdqc.kactive_blk_num)
1105                 prev = rb->prb_bdqc.kactive_blk_num-1;
1106         else
1107                 prev = rb->prb_bdqc.knum_blocks-1;
1108         return prev;
1109 }
1110
1111 /* Assumes caller has held the rx_queue.lock */
1112 static void *__prb_previous_block(struct packet_sock *po,
1113                                          struct packet_ring_buffer *rb,
1114                                          int status)
1115 {
1116         unsigned int previous = prb_previous_blk_num(rb);
1117         return prb_lookup_block(po, rb, previous, status);
1118 }
1119
1120 static void *packet_previous_rx_frame(struct packet_sock *po,
1121                                              struct packet_ring_buffer *rb,
1122                                              int status)
1123 {
1124         if (po->tp_version <= TPACKET_V2)
1125                 return packet_previous_frame(po, rb, status);
1126
1127         return __prb_previous_block(po, rb, status);
1128 }
1129
1130 static void packet_increment_rx_head(struct packet_sock *po,
1131                                             struct packet_ring_buffer *rb)
1132 {
1133         switch (po->tp_version) {
1134         case TPACKET_V1:
1135         case TPACKET_V2:
1136                 return packet_increment_head(rb);
1137         case TPACKET_V3:
1138         default:
1139                 WARN(1, "TPACKET version not supported.\n");
1140                 BUG();
1141                 return;
1142         }
1143 }
1144
1145 static void *packet_previous_frame(struct packet_sock *po,
1146                 struct packet_ring_buffer *rb,
1147                 int status)
1148 {
1149         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1150         return packet_lookup_frame(po, rb, previous, status);
1151 }
1152
1153 static void packet_increment_head(struct packet_ring_buffer *buff)
1154 {
1155         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1156 }
1157
1158 static void packet_sock_destruct(struct sock *sk)
1159 {
1160         skb_queue_purge(&sk->sk_error_queue);
1161
1162         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1163         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1164
1165         if (!sock_flag(sk, SOCK_DEAD)) {
1166                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1167                 return;
1168         }
1169
1170         sk_refcnt_debug_dec(sk);
1171 }
1172
1173 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1174 {
1175         int x = atomic_read(&f->rr_cur) + 1;
1176
1177         if (x >= num)
1178                 x = 0;
1179
1180         return x;
1181 }
1182
1183 static struct sock *fanout_demux_hash(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1184 {
1185         u32 idx, hash = skb->rxhash;
1186
1187         idx = ((u64)hash * num) >> 32;
1188
1189         return f->arr[idx];
1190 }
1191
1192 static struct sock *fanout_demux_lb(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1193 {
1194         int cur, old;
1195
1196         cur = atomic_read(&f->rr_cur);
1197         while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1198                                      fanout_rr_next(f, num))) != cur)
1199                 cur = old;
1200         return f->arr[cur];
1201 }
1202
1203 static struct sock *fanout_demux_cpu(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1204 {
1205         unsigned int cpu = smp_processor_id();
1206
1207         return f->arr[cpu % num];
1208 }
1209
1210 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1211                              struct packet_type *pt, struct net_device *orig_dev)
1212 {
1213         struct packet_fanout *f = pt->af_packet_priv;
1214         unsigned int num = f->num_members;
1215         struct packet_sock *po;
1216         struct sock *sk;
1217
1218         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1219             !num) {
1220                 kfree_skb(skb);
1221                 return 0;
1222         }
1223
1224         switch (f->type) {
1225         case PACKET_FANOUT_HASH:
1226         default:
1227                 if (f->defrag) {
1228                         skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1229                         if (!skb)
1230                                 return 0;
1231                 }
1232                 skb_get_rxhash(skb);
1233                 sk = fanout_demux_hash(f, skb, num);
1234                 break;
1235         case PACKET_FANOUT_LB:
1236                 sk = fanout_demux_lb(f, skb, num);
1237                 break;
1238         case PACKET_FANOUT_CPU:
1239                 sk = fanout_demux_cpu(f, skb, num);
1240                 break;
1241         }
1242
1243         po = pkt_sk(sk);
1244
1245         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1246 }
1247
1248 static DEFINE_MUTEX(fanout_mutex);
1249 static LIST_HEAD(fanout_list);
1250
1251 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1252 {
1253         struct packet_fanout *f = po->fanout;
1254
1255         spin_lock(&f->lock);
1256         f->arr[f->num_members] = sk;
1257         smp_wmb();
1258         f->num_members++;
1259         spin_unlock(&f->lock);
1260 }
1261
1262 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1263 {
1264         struct packet_fanout *f = po->fanout;
1265         int i;
1266
1267         spin_lock(&f->lock);
1268         for (i = 0; i < f->num_members; i++) {
1269                 if (f->arr[i] == sk)
1270                         break;
1271         }
1272         BUG_ON(i >= f->num_members);
1273         f->arr[i] = f->arr[f->num_members - 1];
1274         f->num_members--;
1275         spin_unlock(&f->lock);
1276 }
1277
1278 bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1279 {
1280         if (ptype->af_packet_priv == (void*)((struct packet_sock *)sk)->fanout)
1281                 return true;
1282
1283         return false;
1284 }
1285
1286 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1287 {
1288         struct packet_sock *po = pkt_sk(sk);
1289         struct packet_fanout *f, *match;
1290         u8 type = type_flags & 0xff;
1291         u8 defrag = (type_flags & PACKET_FANOUT_FLAG_DEFRAG) ? 1 : 0;
1292         int err;
1293
1294         switch (type) {
1295         case PACKET_FANOUT_HASH:
1296         case PACKET_FANOUT_LB:
1297         case PACKET_FANOUT_CPU:
1298                 break;
1299         default:
1300                 return -EINVAL;
1301         }
1302
1303         if (!po->running)
1304                 return -EINVAL;
1305
1306         if (po->fanout)
1307                 return -EALREADY;
1308
1309         mutex_lock(&fanout_mutex);
1310         match = NULL;
1311         list_for_each_entry(f, &fanout_list, list) {
1312                 if (f->id == id &&
1313                     read_pnet(&f->net) == sock_net(sk)) {
1314                         match = f;
1315                         break;
1316                 }
1317         }
1318         err = -EINVAL;
1319         if (match && match->defrag != defrag)
1320                 goto out;
1321         if (!match) {
1322                 err = -ENOMEM;
1323                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1324                 if (!match)
1325                         goto out;
1326                 write_pnet(&match->net, sock_net(sk));
1327                 match->id = id;
1328                 match->type = type;
1329                 match->defrag = defrag;
1330                 atomic_set(&match->rr_cur, 0);
1331                 INIT_LIST_HEAD(&match->list);
1332                 spin_lock_init(&match->lock);
1333                 atomic_set(&match->sk_ref, 0);
1334                 match->prot_hook.type = po->prot_hook.type;
1335                 match->prot_hook.dev = po->prot_hook.dev;
1336                 match->prot_hook.func = packet_rcv_fanout;
1337                 match->prot_hook.af_packet_priv = match;
1338                 match->prot_hook.id_match = match_fanout_group;
1339                 dev_add_pack(&match->prot_hook);
1340                 list_add(&match->list, &fanout_list);
1341         }
1342         err = -EINVAL;
1343         if (match->type == type &&
1344             match->prot_hook.type == po->prot_hook.type &&
1345             match->prot_hook.dev == po->prot_hook.dev) {
1346                 err = -ENOSPC;
1347                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1348                         __dev_remove_pack(&po->prot_hook);
1349                         po->fanout = match;
1350                         atomic_inc(&match->sk_ref);
1351                         __fanout_link(sk, po);
1352                         err = 0;
1353                 }
1354         }
1355 out:
1356         mutex_unlock(&fanout_mutex);
1357         return err;
1358 }
1359
1360 static void fanout_release(struct sock *sk)
1361 {
1362         struct packet_sock *po = pkt_sk(sk);
1363         struct packet_fanout *f;
1364
1365         f = po->fanout;
1366         if (!f)
1367                 return;
1368
1369         po->fanout = NULL;
1370
1371         mutex_lock(&fanout_mutex);
1372         if (atomic_dec_and_test(&f->sk_ref)) {
1373                 list_del(&f->list);
1374                 dev_remove_pack(&f->prot_hook);
1375                 kfree(f);
1376         }
1377         mutex_unlock(&fanout_mutex);
1378 }
1379
1380 static const struct proto_ops packet_ops;
1381
1382 static const struct proto_ops packet_ops_spkt;
1383
1384 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1385                            struct packet_type *pt, struct net_device *orig_dev)
1386 {
1387         struct sock *sk;
1388         struct sockaddr_pkt *spkt;
1389
1390         /*
1391          *      When we registered the protocol we saved the socket in the data
1392          *      field for just this event.
1393          */
1394
1395         sk = pt->af_packet_priv;
1396
1397         /*
1398          *      Yank back the headers [hope the device set this
1399          *      right or kerboom...]
1400          *
1401          *      Incoming packets have ll header pulled,
1402          *      push it back.
1403          *
1404          *      For outgoing ones skb->data == skb_mac_header(skb)
1405          *      so that this procedure is noop.
1406          */
1407
1408         if (skb->pkt_type == PACKET_LOOPBACK)
1409                 goto out;
1410
1411         if (!net_eq(dev_net(dev), sock_net(sk)))
1412                 goto out;
1413
1414         skb = skb_share_check(skb, GFP_ATOMIC);
1415         if (skb == NULL)
1416                 goto oom;
1417
1418         /* drop any routing info */
1419         skb_dst_drop(skb);
1420
1421         /* drop conntrack reference */
1422         nf_reset(skb);
1423
1424         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1425
1426         skb_push(skb, skb->data - skb_mac_header(skb));
1427
1428         /*
1429          *      The SOCK_PACKET socket receives _all_ frames.
1430          */
1431
1432         spkt->spkt_family = dev->type;
1433         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1434         spkt->spkt_protocol = skb->protocol;
1435
1436         /*
1437          *      Charge the memory to the socket. This is done specifically
1438          *      to prevent sockets using all the memory up.
1439          */
1440
1441         if (sock_queue_rcv_skb(sk, skb) == 0)
1442                 return 0;
1443
1444 out:
1445         kfree_skb(skb);
1446 oom:
1447         return 0;
1448 }
1449
1450
1451 /*
1452  *      Output a raw packet to a device layer. This bypasses all the other
1453  *      protocol layers and you must therefore supply it with a complete frame
1454  */
1455
1456 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1457                                struct msghdr *msg, size_t len)
1458 {
1459         struct sock *sk = sock->sk;
1460         struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1461         struct sk_buff *skb = NULL;
1462         struct net_device *dev;
1463         __be16 proto = 0;
1464         int err;
1465
1466         /*
1467          *      Get and verify the address.
1468          */
1469
1470         if (saddr) {
1471                 if (msg->msg_namelen < sizeof(struct sockaddr))
1472                         return -EINVAL;
1473                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1474                         proto = saddr->spkt_protocol;
1475         } else
1476                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1477
1478         /*
1479          *      Find the device first to size check it
1480          */
1481
1482         saddr->spkt_device[13] = 0;
1483 retry:
1484         rcu_read_lock();
1485         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1486         err = -ENODEV;
1487         if (dev == NULL)
1488                 goto out_unlock;
1489
1490         err = -ENETDOWN;
1491         if (!(dev->flags & IFF_UP))
1492                 goto out_unlock;
1493
1494         /*
1495          * You may not queue a frame bigger than the mtu. This is the lowest level
1496          * raw protocol and you must do your own fragmentation at this level.
1497          */
1498
1499         err = -EMSGSIZE;
1500         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN)
1501                 goto out_unlock;
1502
1503         if (!skb) {
1504                 size_t reserved = LL_RESERVED_SPACE(dev);
1505                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1506
1507                 rcu_read_unlock();
1508                 skb = sock_wmalloc(sk, len + reserved, 0, GFP_KERNEL);
1509                 if (skb == NULL)
1510                         return -ENOBUFS;
1511                 /* FIXME: Save some space for broken drivers that write a hard
1512                  * header at transmission time by themselves. PPP is the notable
1513                  * one here. This should really be fixed at the driver level.
1514                  */
1515                 skb_reserve(skb, reserved);
1516                 skb_reset_network_header(skb);
1517
1518                 /* Try to align data part correctly */
1519                 if (hhlen) {
1520                         skb->data -= hhlen;
1521                         skb->tail -= hhlen;
1522                         if (len < hhlen)
1523                                 skb_reset_network_header(skb);
1524                 }
1525                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1526                 if (err)
1527                         goto out_free;
1528                 goto retry;
1529         }
1530
1531         if (len > (dev->mtu + dev->hard_header_len)) {
1532                 /* Earlier code assumed this would be a VLAN pkt,
1533                  * double-check this now that we have the actual
1534                  * packet in hand.
1535                  */
1536                 struct ethhdr *ehdr;
1537                 skb_reset_mac_header(skb);
1538                 ehdr = eth_hdr(skb);
1539                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1540                         err = -EMSGSIZE;
1541                         goto out_unlock;
1542                 }
1543         }
1544
1545         skb->protocol = proto;
1546         skb->dev = dev;
1547         skb->priority = sk->sk_priority;
1548         skb->mark = sk->sk_mark;
1549         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1550         if (err < 0)
1551                 goto out_unlock;
1552
1553         dev_queue_xmit(skb);
1554         rcu_read_unlock();
1555         return len;
1556
1557 out_unlock:
1558         rcu_read_unlock();
1559 out_free:
1560         kfree_skb(skb);
1561         return err;
1562 }
1563
1564 static unsigned int run_filter(const struct sk_buff *skb,
1565                                       const struct sock *sk,
1566                                       unsigned int res)
1567 {
1568         struct sk_filter *filter;
1569
1570         rcu_read_lock();
1571         filter = rcu_dereference(sk->sk_filter);
1572         if (filter != NULL)
1573                 res = SK_RUN_FILTER(filter, skb);
1574         rcu_read_unlock();
1575
1576         return res;
1577 }
1578
1579 /*
1580  * This function makes lazy skb cloning in hope that most of packets
1581  * are discarded by BPF.
1582  *
1583  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1584  * and skb->cb are mangled. It works because (and until) packets
1585  * falling here are owned by current CPU. Output packets are cloned
1586  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1587  * sequencially, so that if we return skb to original state on exit,
1588  * we will not harm anyone.
1589  */
1590
1591 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1592                       struct packet_type *pt, struct net_device *orig_dev)
1593 {
1594         struct sock *sk;
1595         struct sockaddr_ll *sll;
1596         struct packet_sock *po;
1597         u8 *skb_head = skb->data;
1598         int skb_len = skb->len;
1599         unsigned int snaplen, res;
1600
1601         if (skb->pkt_type == PACKET_LOOPBACK)
1602                 goto drop;
1603
1604         sk = pt->af_packet_priv;
1605         po = pkt_sk(sk);
1606
1607         if (!net_eq(dev_net(dev), sock_net(sk)))
1608                 goto drop;
1609
1610         skb->dev = dev;
1611
1612         if (dev->header_ops) {
1613                 /* The device has an explicit notion of ll header,
1614                  * exported to higher levels.
1615                  *
1616                  * Otherwise, the device hides details of its frame
1617                  * structure, so that corresponding packet head is
1618                  * never delivered to user.
1619                  */
1620                 if (sk->sk_type != SOCK_DGRAM)
1621                         skb_push(skb, skb->data - skb_mac_header(skb));
1622                 else if (skb->pkt_type == PACKET_OUTGOING) {
1623                         /* Special case: outgoing packets have ll header at head */
1624                         skb_pull(skb, skb_network_offset(skb));
1625                 }
1626         }
1627
1628         snaplen = skb->len;
1629
1630         res = run_filter(skb, sk, snaplen);
1631         if (!res)
1632                 goto drop_n_restore;
1633         if (snaplen > res)
1634                 snaplen = res;
1635
1636         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1637                 goto drop_n_acct;
1638
1639         if (skb_shared(skb)) {
1640                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1641                 if (nskb == NULL)
1642                         goto drop_n_acct;
1643
1644                 if (skb_head != skb->data) {
1645                         skb->data = skb_head;
1646                         skb->len = skb_len;
1647                 }
1648                 kfree_skb(skb);
1649                 skb = nskb;
1650         }
1651
1652         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1653                      sizeof(skb->cb));
1654
1655         sll = &PACKET_SKB_CB(skb)->sa.ll;
1656         sll->sll_family = AF_PACKET;
1657         sll->sll_hatype = dev->type;
1658         sll->sll_protocol = skb->protocol;
1659         sll->sll_pkttype = skb->pkt_type;
1660         if (unlikely(po->origdev))
1661                 sll->sll_ifindex = orig_dev->ifindex;
1662         else
1663                 sll->sll_ifindex = dev->ifindex;
1664
1665         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1666
1667         PACKET_SKB_CB(skb)->origlen = skb->len;
1668
1669         if (pskb_trim(skb, snaplen))
1670                 goto drop_n_acct;
1671
1672         skb_set_owner_r(skb, sk);
1673         skb->dev = NULL;
1674         skb_dst_drop(skb);
1675
1676         /* drop conntrack reference */
1677         nf_reset(skb);
1678
1679         spin_lock(&sk->sk_receive_queue.lock);
1680         po->stats.tp_packets++;
1681         skb->dropcount = atomic_read(&sk->sk_drops);
1682         __skb_queue_tail(&sk->sk_receive_queue, skb);
1683         spin_unlock(&sk->sk_receive_queue.lock);
1684         sk->sk_data_ready(sk, skb->len);
1685         return 0;
1686
1687 drop_n_acct:
1688         spin_lock(&sk->sk_receive_queue.lock);
1689         po->stats.tp_drops++;
1690         atomic_inc(&sk->sk_drops);
1691         spin_unlock(&sk->sk_receive_queue.lock);
1692
1693 drop_n_restore:
1694         if (skb_head != skb->data && skb_shared(skb)) {
1695                 skb->data = skb_head;
1696                 skb->len = skb_len;
1697         }
1698 drop:
1699         consume_skb(skb);
1700         return 0;
1701 }
1702
1703 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1704                        struct packet_type *pt, struct net_device *orig_dev)
1705 {
1706         struct sock *sk;
1707         struct packet_sock *po;
1708         struct sockaddr_ll *sll;
1709         union {
1710                 struct tpacket_hdr *h1;
1711                 struct tpacket2_hdr *h2;
1712                 struct tpacket3_hdr *h3;
1713                 void *raw;
1714         } h;
1715         u8 *skb_head = skb->data;
1716         int skb_len = skb->len;
1717         unsigned int snaplen, res;
1718         unsigned long status = TP_STATUS_USER;
1719         unsigned short macoff, netoff, hdrlen;
1720         struct sk_buff *copy_skb = NULL;
1721         struct timeval tv;
1722         struct timespec ts;
1723         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1724
1725         if (skb->pkt_type == PACKET_LOOPBACK)
1726                 goto drop;
1727
1728         sk = pt->af_packet_priv;
1729         po = pkt_sk(sk);
1730
1731         if (!net_eq(dev_net(dev), sock_net(sk)))
1732                 goto drop;
1733
1734         if (dev->header_ops) {
1735                 if (sk->sk_type != SOCK_DGRAM)
1736                         skb_push(skb, skb->data - skb_mac_header(skb));
1737                 else if (skb->pkt_type == PACKET_OUTGOING) {
1738                         /* Special case: outgoing packets have ll header at head */
1739                         skb_pull(skb, skb_network_offset(skb));
1740                 }
1741         }
1742
1743         if (skb->ip_summed == CHECKSUM_PARTIAL)
1744                 status |= TP_STATUS_CSUMNOTREADY;
1745
1746         snaplen = skb->len;
1747
1748         res = run_filter(skb, sk, snaplen);
1749         if (!res)
1750                 goto drop_n_restore;
1751         if (snaplen > res)
1752                 snaplen = res;
1753
1754         if (sk->sk_type == SOCK_DGRAM) {
1755                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1756                                   po->tp_reserve;
1757         } else {
1758                 unsigned maclen = skb_network_offset(skb);
1759                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1760                                        (maclen < 16 ? 16 : maclen)) +
1761                         po->tp_reserve;
1762                 macoff = netoff - maclen;
1763         }
1764         if (po->tp_version <= TPACKET_V2) {
1765                 if (macoff + snaplen > po->rx_ring.frame_size) {
1766                         if (po->copy_thresh &&
1767                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1768                                 if (skb_shared(skb)) {
1769                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1770                                 } else {
1771                                         copy_skb = skb_get(skb);
1772                                         skb_head = skb->data;
1773                                 }
1774                                 if (copy_skb)
1775                                         skb_set_owner_r(copy_skb, sk);
1776                         }
1777                         snaplen = po->rx_ring.frame_size - macoff;
1778                         if ((int)snaplen < 0)
1779                                 snaplen = 0;
1780                 }
1781         }
1782         spin_lock(&sk->sk_receive_queue.lock);
1783         h.raw = packet_current_rx_frame(po, skb,
1784                                         TP_STATUS_KERNEL, (macoff+snaplen));
1785         if (!h.raw)
1786                 goto ring_is_full;
1787         if (po->tp_version <= TPACKET_V2) {
1788                 packet_increment_rx_head(po, &po->rx_ring);
1789         /*
1790          * LOSING will be reported till you read the stats,
1791          * because it's COR - Clear On Read.
1792          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1793          * at packet level.
1794          */
1795                 if (po->stats.tp_drops)
1796                         status |= TP_STATUS_LOSING;
1797         }
1798         po->stats.tp_packets++;
1799         if (copy_skb) {
1800                 status |= TP_STATUS_COPY;
1801                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1802         }
1803         spin_unlock(&sk->sk_receive_queue.lock);
1804
1805         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1806
1807         switch (po->tp_version) {
1808         case TPACKET_V1:
1809                 h.h1->tp_len = skb->len;
1810                 h.h1->tp_snaplen = snaplen;
1811                 h.h1->tp_mac = macoff;
1812                 h.h1->tp_net = netoff;
1813                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1814                                 && shhwtstamps->syststamp.tv64)
1815                         tv = ktime_to_timeval(shhwtstamps->syststamp);
1816                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1817                                 && shhwtstamps->hwtstamp.tv64)
1818                         tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1819                 else if (skb->tstamp.tv64)
1820                         tv = ktime_to_timeval(skb->tstamp);
1821                 else
1822                         do_gettimeofday(&tv);
1823                 h.h1->tp_sec = tv.tv_sec;
1824                 h.h1->tp_usec = tv.tv_usec;
1825                 hdrlen = sizeof(*h.h1);
1826                 break;
1827         case TPACKET_V2:
1828                 h.h2->tp_len = skb->len;
1829                 h.h2->tp_snaplen = snaplen;
1830                 h.h2->tp_mac = macoff;
1831                 h.h2->tp_net = netoff;
1832                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1833                                 && shhwtstamps->syststamp.tv64)
1834                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1835                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1836                                 && shhwtstamps->hwtstamp.tv64)
1837                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1838                 else if (skb->tstamp.tv64)
1839                         ts = ktime_to_timespec(skb->tstamp);
1840                 else
1841                         getnstimeofday(&ts);
1842                 h.h2->tp_sec = ts.tv_sec;
1843                 h.h2->tp_nsec = ts.tv_nsec;
1844                 if (vlan_tx_tag_present(skb)) {
1845                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1846                         status |= TP_STATUS_VLAN_VALID;
1847                 } else {
1848                         h.h2->tp_vlan_tci = 0;
1849                 }
1850                 h.h2->tp_padding = 0;
1851                 hdrlen = sizeof(*h.h2);
1852                 break;
1853         case TPACKET_V3:
1854                 /* tp_nxt_offset,vlan are already populated above.
1855                  * So DONT clear those fields here
1856                  */
1857                 h.h3->tp_status |= status;
1858                 h.h3->tp_len = skb->len;
1859                 h.h3->tp_snaplen = snaplen;
1860                 h.h3->tp_mac = macoff;
1861                 h.h3->tp_net = netoff;
1862                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1863                                 && shhwtstamps->syststamp.tv64)
1864                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1865                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1866                                 && shhwtstamps->hwtstamp.tv64)
1867                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1868                 else if (skb->tstamp.tv64)
1869                         ts = ktime_to_timespec(skb->tstamp);
1870                 else
1871                         getnstimeofday(&ts);
1872                 h.h3->tp_sec  = ts.tv_sec;
1873                 h.h3->tp_nsec = ts.tv_nsec;
1874                 hdrlen = sizeof(*h.h3);
1875                 break;
1876         default:
1877                 BUG();
1878         }
1879
1880         sll = h.raw + TPACKET_ALIGN(hdrlen);
1881         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1882         sll->sll_family = AF_PACKET;
1883         sll->sll_hatype = dev->type;
1884         sll->sll_protocol = skb->protocol;
1885         sll->sll_pkttype = skb->pkt_type;
1886         if (unlikely(po->origdev))
1887                 sll->sll_ifindex = orig_dev->ifindex;
1888         else
1889                 sll->sll_ifindex = dev->ifindex;
1890
1891         smp_mb();
1892 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1893         {
1894                 u8 *start, *end;
1895
1896                 if (po->tp_version <= TPACKET_V2) {
1897                         end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1898                                 + macoff + snaplen);
1899                         for (start = h.raw; start < end; start += PAGE_SIZE)
1900                                 flush_dcache_page(pgv_to_page(start));
1901                 }
1902                 smp_wmb();
1903         }
1904 #endif
1905         if (po->tp_version <= TPACKET_V2)
1906                 __packet_set_status(po, h.raw, status);
1907         else
1908                 prb_clear_blk_fill_status(&po->rx_ring);
1909
1910         sk->sk_data_ready(sk, 0);
1911
1912 drop_n_restore:
1913         if (skb_head != skb->data && skb_shared(skb)) {
1914                 skb->data = skb_head;
1915                 skb->len = skb_len;
1916         }
1917 drop:
1918         kfree_skb(skb);
1919         return 0;
1920
1921 ring_is_full:
1922         po->stats.tp_drops++;
1923         spin_unlock(&sk->sk_receive_queue.lock);
1924
1925         sk->sk_data_ready(sk, 0);
1926         kfree_skb(copy_skb);
1927         goto drop_n_restore;
1928 }
1929
1930 static void tpacket_destruct_skb(struct sk_buff *skb)
1931 {
1932         struct packet_sock *po = pkt_sk(skb->sk);
1933         void *ph;
1934
1935         if (likely(po->tx_ring.pg_vec)) {
1936                 ph = skb_shinfo(skb)->destructor_arg;
1937                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1938                 atomic_dec(&po->tx_ring.pending);
1939                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1940         }
1941
1942         sock_wfree(skb);
1943 }
1944
1945 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1946                 void *frame, struct net_device *dev, int size_max,
1947                 __be16 proto, unsigned char *addr)
1948 {
1949         union {
1950                 struct tpacket_hdr *h1;
1951                 struct tpacket2_hdr *h2;
1952                 void *raw;
1953         } ph;
1954         int to_write, offset, len, tp_len, nr_frags, len_max;
1955         struct socket *sock = po->sk.sk_socket;
1956         struct page *page;
1957         void *data;
1958         int err;
1959
1960         ph.raw = frame;
1961
1962         skb->protocol = proto;
1963         skb->dev = dev;
1964         skb->priority = po->sk.sk_priority;
1965         skb->mark = po->sk.sk_mark;
1966         skb_shinfo(skb)->destructor_arg = ph.raw;
1967
1968         switch (po->tp_version) {
1969         case TPACKET_V2:
1970                 tp_len = ph.h2->tp_len;
1971                 break;
1972         default:
1973                 tp_len = ph.h1->tp_len;
1974                 break;
1975         }
1976         if (unlikely(tp_len > size_max)) {
1977                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1978                 return -EMSGSIZE;
1979         }
1980
1981         skb_reserve(skb, LL_RESERVED_SPACE(dev));
1982         skb_reset_network_header(skb);
1983
1984         data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1985         to_write = tp_len;
1986
1987         if (sock->type == SOCK_DGRAM) {
1988                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
1989                                 NULL, tp_len);
1990                 if (unlikely(err < 0))
1991                         return -EINVAL;
1992         } else if (dev->hard_header_len) {
1993                 /* net device doesn't like empty head */
1994                 if (unlikely(tp_len <= dev->hard_header_len)) {
1995                         pr_err("packet size is too short (%d < %d)\n",
1996                                tp_len, dev->hard_header_len);
1997                         return -EINVAL;
1998                 }
1999
2000                 skb_push(skb, dev->hard_header_len);
2001                 err = skb_store_bits(skb, 0, data,
2002                                 dev->hard_header_len);
2003                 if (unlikely(err))
2004                         return err;
2005
2006                 data += dev->hard_header_len;
2007                 to_write -= dev->hard_header_len;
2008         }
2009
2010         err = -EFAULT;
2011         offset = offset_in_page(data);
2012         len_max = PAGE_SIZE - offset;
2013         len = ((to_write > len_max) ? len_max : to_write);
2014
2015         skb->data_len = to_write;
2016         skb->len += to_write;
2017         skb->truesize += to_write;
2018         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2019
2020         while (likely(to_write)) {
2021                 nr_frags = skb_shinfo(skb)->nr_frags;
2022
2023                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2024                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2025                                MAX_SKB_FRAGS);
2026                         return -EFAULT;
2027                 }
2028
2029                 page = pgv_to_page(data);
2030                 data += len;
2031                 flush_dcache_page(page);
2032                 get_page(page);
2033                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2034                 to_write -= len;
2035                 offset = 0;
2036                 len_max = PAGE_SIZE;
2037                 len = ((to_write > len_max) ? len_max : to_write);
2038         }
2039
2040         return tp_len;
2041 }
2042
2043 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
2044 {
2045         struct net_device *dev;
2046
2047         rcu_read_lock();
2048         dev = rcu_dereference(po->cached_dev);
2049         if (dev)
2050                 dev_hold(dev);
2051         rcu_read_unlock();
2052
2053         return dev;
2054 }
2055
2056 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2057 {
2058         struct sk_buff *skb;
2059         struct net_device *dev;
2060         __be16 proto;
2061         int err, reserve = 0;
2062         void *ph;
2063         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2064         int tp_len, size_max;
2065         unsigned char *addr;
2066         int len_sum = 0;
2067         int status = 0;
2068
2069         mutex_lock(&po->pg_vec_lock);
2070
2071         err = -EBUSY;
2072         if (saddr == NULL) {
2073                 dev     = packet_cached_dev_get(po);
2074                 proto   = po->num;
2075                 addr    = NULL;
2076         } else {
2077                 err = -EINVAL;
2078                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2079                         goto out;
2080                 if (msg->msg_namelen < (saddr->sll_halen
2081                                         + offsetof(struct sockaddr_ll,
2082                                                 sll_addr)))
2083                         goto out;
2084                 proto   = saddr->sll_protocol;
2085                 addr    = saddr->sll_addr;
2086                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2087         }
2088
2089         err = -ENXIO;
2090         if (unlikely(dev == NULL))
2091                 goto out;
2092         err = -ENETDOWN;
2093         if (unlikely(!(dev->flags & IFF_UP)))
2094                 goto out_put;
2095
2096         reserve = dev->hard_header_len;
2097
2098         size_max = po->tx_ring.frame_size
2099                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2100
2101         if (size_max > dev->mtu + reserve)
2102                 size_max = dev->mtu + reserve;
2103
2104         do {
2105                 ph = packet_current_frame(po, &po->tx_ring,
2106                                 TP_STATUS_SEND_REQUEST);
2107
2108                 if (unlikely(ph == NULL)) {
2109                         schedule();
2110                         continue;
2111                 }
2112
2113                 status = TP_STATUS_SEND_REQUEST;
2114                 skb = sock_alloc_send_skb(&po->sk,
2115                                 LL_ALLOCATED_SPACE(dev)
2116                                 + sizeof(struct sockaddr_ll),
2117                                 0, &err);
2118
2119                 if (unlikely(skb == NULL))
2120                         goto out_status;
2121
2122                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2123                                 addr);
2124
2125                 if (unlikely(tp_len < 0)) {
2126                         if (po->tp_loss) {
2127                                 __packet_set_status(po, ph,
2128                                                 TP_STATUS_AVAILABLE);
2129                                 packet_increment_head(&po->tx_ring);
2130                                 kfree_skb(skb);
2131                                 continue;
2132                         } else {
2133                                 status = TP_STATUS_WRONG_FORMAT;
2134                                 err = tp_len;
2135                                 goto out_status;
2136                         }
2137                 }
2138
2139                 skb->destructor = tpacket_destruct_skb;
2140                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2141                 atomic_inc(&po->tx_ring.pending);
2142
2143                 status = TP_STATUS_SEND_REQUEST;
2144                 err = dev_queue_xmit(skb);
2145                 if (unlikely(err > 0)) {
2146                         err = net_xmit_errno(err);
2147                         if (err && __packet_get_status(po, ph) ==
2148                                    TP_STATUS_AVAILABLE) {
2149                                 /* skb was destructed already */
2150                                 skb = NULL;
2151                                 goto out_status;
2152                         }
2153                         /*
2154                          * skb was dropped but not destructed yet;
2155                          * let's treat it like congestion or err < 0
2156                          */
2157                         err = 0;
2158                 }
2159                 packet_increment_head(&po->tx_ring);
2160                 len_sum += tp_len;
2161         } while (likely((ph != NULL) ||
2162                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2163                          (atomic_read(&po->tx_ring.pending))))
2164                 );
2165
2166         err = len_sum;
2167         goto out_put;
2168
2169 out_status:
2170         __packet_set_status(po, ph, status);
2171         kfree_skb(skb);
2172 out_put:
2173         dev_put(dev);
2174 out:
2175         mutex_unlock(&po->pg_vec_lock);
2176         return err;
2177 }
2178
2179 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2180                                         size_t reserve, size_t len,
2181                                         size_t linear, int noblock,
2182                                         int *err)
2183 {
2184         struct sk_buff *skb;
2185
2186         /* Under a page?  Don't bother with paged skb. */
2187         if (prepad + len < PAGE_SIZE || !linear)
2188                 linear = len;
2189
2190         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2191                                    err);
2192         if (!skb)
2193                 return NULL;
2194
2195         skb_reserve(skb, reserve);
2196         skb_put(skb, linear);
2197         skb->data_len = len - linear;
2198         skb->len += len - linear;
2199
2200         return skb;
2201 }
2202
2203 static int packet_snd(struct socket *sock,
2204                           struct msghdr *msg, size_t len)
2205 {
2206         struct sock *sk = sock->sk;
2207         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2208         struct sk_buff *skb;
2209         struct net_device *dev;
2210         __be16 proto;
2211         unsigned char *addr;
2212         int err, reserve = 0;
2213         struct virtio_net_hdr vnet_hdr = { 0 };
2214         int offset = 0;
2215         int vnet_hdr_len;
2216         struct packet_sock *po = pkt_sk(sk);
2217         unsigned short gso_type = 0;
2218
2219         /*
2220          *      Get and verify the address.
2221          */
2222
2223         if (saddr == NULL) {
2224                 dev     = packet_cached_dev_get(po);
2225                 proto   = po->num;
2226                 addr    = NULL;
2227         } else {
2228                 err = -EINVAL;
2229                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2230                         goto out;
2231                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2232                         goto out;
2233                 proto   = saddr->sll_protocol;
2234                 addr    = saddr->sll_addr;
2235                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2236         }
2237
2238         err = -ENXIO;
2239         if (unlikely(dev == NULL))
2240                 goto out_unlock;
2241         err = -ENETDOWN;
2242         if (unlikely(!(dev->flags & IFF_UP)))
2243                 goto out_unlock;
2244
2245         if (sock->type == SOCK_RAW)
2246                 reserve = dev->hard_header_len;
2247         if (po->has_vnet_hdr) {
2248                 vnet_hdr_len = sizeof(vnet_hdr);
2249
2250                 err = -EINVAL;
2251                 if (len < vnet_hdr_len)
2252                         goto out_unlock;
2253
2254                 len -= vnet_hdr_len;
2255
2256                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2257                                        vnet_hdr_len);
2258                 if (err < 0)
2259                         goto out_unlock;
2260
2261                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2262                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2263                       vnet_hdr.hdr_len))
2264                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
2265                                                  vnet_hdr.csum_offset + 2;
2266
2267                 err = -EINVAL;
2268                 if (vnet_hdr.hdr_len > len)
2269                         goto out_unlock;
2270
2271                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2272                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2273                         case VIRTIO_NET_HDR_GSO_TCPV4:
2274                                 gso_type = SKB_GSO_TCPV4;
2275                                 break;
2276                         case VIRTIO_NET_HDR_GSO_TCPV6:
2277                                 gso_type = SKB_GSO_TCPV6;
2278                                 break;
2279                         case VIRTIO_NET_HDR_GSO_UDP:
2280                                 gso_type = SKB_GSO_UDP;
2281                                 break;
2282                         default:
2283                                 goto out_unlock;
2284                         }
2285
2286                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2287                                 gso_type |= SKB_GSO_TCP_ECN;
2288
2289                         if (vnet_hdr.gso_size == 0)
2290                                 goto out_unlock;
2291
2292                 }
2293         }
2294
2295         err = -EMSGSIZE;
2296         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN))
2297                 goto out_unlock;
2298
2299         err = -ENOBUFS;
2300         skb = packet_alloc_skb(sk, LL_ALLOCATED_SPACE(dev),
2301                                LL_RESERVED_SPACE(dev), len, vnet_hdr.hdr_len,
2302                                msg->msg_flags & MSG_DONTWAIT, &err);
2303         if (skb == NULL)
2304                 goto out_unlock;
2305
2306         skb_set_network_header(skb, reserve);
2307
2308         err = -EINVAL;
2309         if (sock->type == SOCK_DGRAM &&
2310             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2311                 goto out_free;
2312
2313         /* Returns -EFAULT on error */
2314         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2315         if (err)
2316                 goto out_free;
2317         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2318         if (err < 0)
2319                 goto out_free;
2320
2321         if (!gso_type && (len > dev->mtu + reserve)) {
2322                 /* Earlier code assumed this would be a VLAN pkt,
2323                  * double-check this now that we have the actual
2324                  * packet in hand.
2325                  */
2326                 struct ethhdr *ehdr;
2327                 skb_reset_mac_header(skb);
2328                 ehdr = eth_hdr(skb);
2329                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2330                         err = -EMSGSIZE;
2331                         goto out_free;
2332                 }
2333         }
2334
2335         skb->protocol = proto;
2336         skb->dev = dev;
2337         skb->priority = sk->sk_priority;
2338         skb->mark = sk->sk_mark;
2339
2340         if (po->has_vnet_hdr) {
2341                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2342                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2343                                                   vnet_hdr.csum_offset)) {
2344                                 err = -EINVAL;
2345                                 goto out_free;
2346                         }
2347                 }
2348
2349                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2350                 skb_shinfo(skb)->gso_type = gso_type;
2351
2352                 /* Header must be checked, and gso_segs computed. */
2353                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2354                 skb_shinfo(skb)->gso_segs = 0;
2355
2356                 len += vnet_hdr_len;
2357         }
2358
2359         /*
2360          *      Now send it
2361          */
2362
2363         err = dev_queue_xmit(skb);
2364         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2365                 goto out_unlock;
2366
2367         dev_put(dev);
2368
2369         return len;
2370
2371 out_free:
2372         kfree_skb(skb);
2373 out_unlock:
2374         if (dev)
2375                 dev_put(dev);
2376 out:
2377         return err;
2378 }
2379
2380 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2381                 struct msghdr *msg, size_t len)
2382 {
2383         struct sock *sk = sock->sk;
2384         struct packet_sock *po = pkt_sk(sk);
2385         if (po->tx_ring.pg_vec)
2386                 return tpacket_snd(po, msg);
2387         else
2388                 return packet_snd(sock, msg, len);
2389 }
2390
2391 /*
2392  *      Close a PACKET socket. This is fairly simple. We immediately go
2393  *      to 'closed' state and remove our protocol entry in the device list.
2394  */
2395
2396 static int packet_release(struct socket *sock)
2397 {
2398         struct sock *sk = sock->sk;
2399         struct packet_sock *po;
2400         struct net *net;
2401         union tpacket_req_u req_u;
2402
2403         if (!sk)
2404                 return 0;
2405
2406         net = sock_net(sk);
2407         po = pkt_sk(sk);
2408
2409         spin_lock_bh(&net->packet.sklist_lock);
2410         sk_del_node_init_rcu(sk);
2411         sock_prot_inuse_add(net, sk->sk_prot, -1);
2412         spin_unlock_bh(&net->packet.sklist_lock);
2413
2414         spin_lock(&po->bind_lock);
2415         unregister_prot_hook(sk, false);
2416         if (po->prot_hook.dev) {
2417                 dev_put(po->prot_hook.dev);
2418                 po->prot_hook.dev = NULL;
2419         }
2420         spin_unlock(&po->bind_lock);
2421
2422         packet_flush_mclist(sk);
2423
2424         if (po->rx_ring.pg_vec) {
2425                 memset(&req_u, 0, sizeof(req_u));
2426                 packet_set_ring(sk, &req_u, 1, 0);
2427         }
2428
2429         if (po->tx_ring.pg_vec) {
2430                 memset(&req_u, 0, sizeof(req_u));
2431                 packet_set_ring(sk, &req_u, 1, 1);
2432         }
2433
2434         fanout_release(sk);
2435
2436         synchronize_net();
2437         /*
2438          *      Now the socket is dead. No more input will appear.
2439          */
2440         sock_orphan(sk);
2441         sock->sk = NULL;
2442
2443         /* Purge queues */
2444
2445         skb_queue_purge(&sk->sk_receive_queue);
2446         sk_refcnt_debug_release(sk);
2447
2448         sock_put(sk);
2449         return 0;
2450 }
2451
2452 /*
2453  *      Attach a packet hook.
2454  */
2455
2456 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2457 {
2458         struct packet_sock *po = pkt_sk(sk);
2459
2460         if (po->fanout) {
2461                 if (dev)
2462                         dev_put(dev);
2463
2464                 return -EINVAL;
2465         }
2466
2467         lock_sock(sk);
2468
2469         spin_lock(&po->bind_lock);
2470         unregister_prot_hook(sk, true);
2471         po->num = protocol;
2472         po->prot_hook.type = protocol;
2473         if (po->prot_hook.dev)
2474                 dev_put(po->prot_hook.dev);
2475         po->prot_hook.dev = dev;
2476
2477         po->ifindex = dev ? dev->ifindex : 0;
2478
2479         if (protocol == 0)
2480                 goto out_unlock;
2481
2482         if (!dev || (dev->flags & IFF_UP)) {
2483                 register_prot_hook(sk);
2484         } else {
2485                 sk->sk_err = ENETDOWN;
2486                 if (!sock_flag(sk, SOCK_DEAD))
2487                         sk->sk_error_report(sk);
2488         }
2489
2490 out_unlock:
2491         spin_unlock(&po->bind_lock);
2492         release_sock(sk);
2493         return 0;
2494 }
2495
2496 /*
2497  *      Bind a packet socket to a device
2498  */
2499
2500 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2501                             int addr_len)
2502 {
2503         struct sock *sk = sock->sk;
2504         char name[15];
2505         struct net_device *dev;
2506         int err = -ENODEV;
2507
2508         /*
2509          *      Check legality
2510          */
2511
2512         if (addr_len != sizeof(struct sockaddr))
2513                 return -EINVAL;
2514         strlcpy(name, uaddr->sa_data, sizeof(name));
2515
2516         dev = dev_get_by_name(sock_net(sk), name);
2517         if (dev)
2518                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2519         return err;
2520 }
2521
2522 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2523 {
2524         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2525         struct sock *sk = sock->sk;
2526         struct net_device *dev = NULL;
2527         int err;
2528
2529
2530         /*
2531          *      Check legality
2532          */
2533
2534         if (addr_len < sizeof(struct sockaddr_ll))
2535                 return -EINVAL;
2536         if (sll->sll_family != AF_PACKET)
2537                 return -EINVAL;
2538
2539         if (sll->sll_ifindex) {
2540                 err = -ENODEV;
2541                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2542                 if (dev == NULL)
2543                         goto out;
2544         }
2545         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2546
2547 out:
2548         return err;
2549 }
2550
2551 static struct proto packet_proto = {
2552         .name     = "PACKET",
2553         .owner    = THIS_MODULE,
2554         .obj_size = sizeof(struct packet_sock),
2555 };
2556
2557 /*
2558  *      Create a packet of type SOCK_PACKET.
2559  */
2560
2561 static int packet_create(struct net *net, struct socket *sock, int protocol,
2562                          int kern)
2563 {
2564         struct sock *sk;
2565         struct packet_sock *po;
2566         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2567         int err;
2568
2569         if (!capable(CAP_NET_RAW))
2570                 return -EPERM;
2571         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2572             sock->type != SOCK_PACKET)
2573                 return -ESOCKTNOSUPPORT;
2574
2575         sock->state = SS_UNCONNECTED;
2576
2577         err = -ENOBUFS;
2578         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2579         if (sk == NULL)
2580                 goto out;
2581
2582         sock->ops = &packet_ops;
2583         if (sock->type == SOCK_PACKET)
2584                 sock->ops = &packet_ops_spkt;
2585
2586         sock_init_data(sock, sk);
2587
2588         po = pkt_sk(sk);
2589         sk->sk_family = PF_PACKET;
2590         po->num = proto;
2591         RCU_INIT_POINTER(po->cached_dev, NULL);
2592
2593         sk->sk_destruct = packet_sock_destruct;
2594         sk_refcnt_debug_inc(sk);
2595
2596         /*
2597          *      Attach a protocol block
2598          */
2599
2600         spin_lock_init(&po->bind_lock);
2601         mutex_init(&po->pg_vec_lock);
2602         po->prot_hook.func = packet_rcv;
2603
2604         if (sock->type == SOCK_PACKET)
2605                 po->prot_hook.func = packet_rcv_spkt;
2606
2607         po->prot_hook.af_packet_priv = sk;
2608
2609         if (proto) {
2610                 po->prot_hook.type = proto;
2611                 register_prot_hook(sk);
2612         }
2613
2614         spin_lock_bh(&net->packet.sklist_lock);
2615         sk_add_node_rcu(sk, &net->packet.sklist);
2616         sock_prot_inuse_add(net, &packet_proto, 1);
2617         spin_unlock_bh(&net->packet.sklist_lock);
2618
2619         return 0;
2620 out:
2621         return err;
2622 }
2623
2624 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2625 {
2626         struct sock_exterr_skb *serr;
2627         struct sk_buff *skb, *skb2;
2628         int copied, err;
2629
2630         err = -EAGAIN;
2631         skb = skb_dequeue(&sk->sk_error_queue);
2632         if (skb == NULL)
2633                 goto out;
2634
2635         copied = skb->len;
2636         if (copied > len) {
2637                 msg->msg_flags |= MSG_TRUNC;
2638                 copied = len;
2639         }
2640         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2641         if (err)
2642                 goto out_free_skb;
2643
2644         sock_recv_timestamp(msg, sk, skb);
2645
2646         serr = SKB_EXT_ERR(skb);
2647         put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2648                  sizeof(serr->ee), &serr->ee);
2649
2650         msg->msg_flags |= MSG_ERRQUEUE;
2651         err = copied;
2652
2653         /* Reset and regenerate socket error */
2654         spin_lock_bh(&sk->sk_error_queue.lock);
2655         sk->sk_err = 0;
2656         if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2657                 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2658                 spin_unlock_bh(&sk->sk_error_queue.lock);
2659                 sk->sk_error_report(sk);
2660         } else
2661                 spin_unlock_bh(&sk->sk_error_queue.lock);
2662
2663 out_free_skb:
2664         kfree_skb(skb);
2665 out:
2666         return err;
2667 }
2668
2669 /*
2670  *      Pull a packet from our receive queue and hand it to the user.
2671  *      If necessary we block.
2672  */
2673
2674 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2675                           struct msghdr *msg, size_t len, int flags)
2676 {
2677         struct sock *sk = sock->sk;
2678         struct sk_buff *skb;
2679         int copied, err;
2680         int vnet_hdr_len = 0;
2681
2682         err = -EINVAL;
2683         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2684                 goto out;
2685
2686 #if 0
2687         /* What error should we return now? EUNATTACH? */
2688         if (pkt_sk(sk)->ifindex < 0)
2689                 return -ENODEV;
2690 #endif
2691
2692         if (flags & MSG_ERRQUEUE) {
2693                 err = packet_recv_error(sk, msg, len);
2694                 goto out;
2695         }
2696
2697         /*
2698          *      Call the generic datagram receiver. This handles all sorts
2699          *      of horrible races and re-entrancy so we can forget about it
2700          *      in the protocol layers.
2701          *
2702          *      Now it will return ENETDOWN, if device have just gone down,
2703          *      but then it will block.
2704          */
2705
2706         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2707
2708         /*
2709          *      An error occurred so return it. Because skb_recv_datagram()
2710          *      handles the blocking we don't see and worry about blocking
2711          *      retries.
2712          */
2713
2714         if (skb == NULL)
2715                 goto out;
2716
2717         if (pkt_sk(sk)->has_vnet_hdr) {
2718                 struct virtio_net_hdr vnet_hdr = { 0 };
2719
2720                 err = -EINVAL;
2721                 vnet_hdr_len = sizeof(vnet_hdr);
2722                 if (len < vnet_hdr_len)
2723                         goto out_free;
2724
2725                 len -= vnet_hdr_len;
2726
2727                 if (skb_is_gso(skb)) {
2728                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2729
2730                         /* This is a hint as to how much should be linear. */
2731                         vnet_hdr.hdr_len = skb_headlen(skb);
2732                         vnet_hdr.gso_size = sinfo->gso_size;
2733                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2734                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2735                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2736                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2737                         else if (sinfo->gso_type & SKB_GSO_UDP)
2738                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2739                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2740                                 goto out_free;
2741                         else
2742                                 BUG();
2743                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2744                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2745                 } else
2746                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2747
2748                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2749                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2750                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2751                         vnet_hdr.csum_offset = skb->csum_offset;
2752                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2753                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2754                 } /* else everything is zero */
2755
2756                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2757                                      vnet_hdr_len);
2758                 if (err < 0)
2759                         goto out_free;
2760         }
2761
2762         /* You lose any data beyond the buffer you gave. If it worries
2763          * a user program they can ask the device for its MTU
2764          * anyway.
2765          */
2766         copied = skb->len;
2767         if (copied > len) {
2768                 copied = len;
2769                 msg->msg_flags |= MSG_TRUNC;
2770         }
2771
2772         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2773         if (err)
2774                 goto out_free;
2775
2776         sock_recv_ts_and_drops(msg, sk, skb);
2777
2778         if (msg->msg_name) {
2779                 /* If the address length field is there to be filled
2780                  * in, we fill it in now.
2781                  */
2782                 if (sock->type == SOCK_PACKET) {
2783                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
2784                 } else {
2785                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2786                         msg->msg_namelen = sll->sll_halen +
2787                                 offsetof(struct sockaddr_ll, sll_addr);
2788                 }
2789                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2790                        msg->msg_namelen);
2791         }
2792
2793         if (pkt_sk(sk)->auxdata) {
2794                 struct tpacket_auxdata aux;
2795
2796                 aux.tp_status = TP_STATUS_USER;
2797                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2798                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2799                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2800                 aux.tp_snaplen = skb->len;
2801                 aux.tp_mac = 0;
2802                 aux.tp_net = skb_network_offset(skb);
2803                 if (vlan_tx_tag_present(skb)) {
2804                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2805                         aux.tp_status |= TP_STATUS_VLAN_VALID;
2806                 } else {
2807                         aux.tp_vlan_tci = 0;
2808                 }
2809                 aux.tp_padding = 0;
2810                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2811         }
2812
2813         /*
2814          *      Free or return the buffer as appropriate. Again this
2815          *      hides all the races and re-entrancy issues from us.
2816          */
2817         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2818
2819 out_free:
2820         skb_free_datagram(sk, skb);
2821 out:
2822         return err;
2823 }
2824
2825 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2826                                int *uaddr_len, int peer)
2827 {
2828         struct net_device *dev;
2829         struct sock *sk = sock->sk;
2830
2831         if (peer)
2832                 return -EOPNOTSUPP;
2833
2834         uaddr->sa_family = AF_PACKET;
2835         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
2836         rcu_read_lock();
2837         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2838         if (dev)
2839                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
2840         rcu_read_unlock();
2841         *uaddr_len = sizeof(*uaddr);
2842
2843         return 0;
2844 }
2845
2846 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2847                           int *uaddr_len, int peer)
2848 {
2849         struct net_device *dev;
2850         struct sock *sk = sock->sk;
2851         struct packet_sock *po = pkt_sk(sk);
2852         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2853
2854         if (peer)
2855                 return -EOPNOTSUPP;
2856
2857         sll->sll_family = AF_PACKET;
2858         sll->sll_ifindex = po->ifindex;
2859         sll->sll_protocol = po->num;
2860         sll->sll_pkttype = 0;
2861         rcu_read_lock();
2862         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2863         if (dev) {
2864                 sll->sll_hatype = dev->type;
2865                 sll->sll_halen = dev->addr_len;
2866                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2867         } else {
2868                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
2869                 sll->sll_halen = 0;
2870         }
2871         rcu_read_unlock();
2872         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2873
2874         return 0;
2875 }
2876
2877 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2878                          int what)
2879 {
2880         switch (i->type) {
2881         case PACKET_MR_MULTICAST:
2882                 if (i->alen != dev->addr_len)
2883                         return -EINVAL;
2884                 if (what > 0)
2885                         return dev_mc_add(dev, i->addr);
2886                 else
2887                         return dev_mc_del(dev, i->addr);
2888                 break;
2889         case PACKET_MR_PROMISC:
2890                 return dev_set_promiscuity(dev, what);
2891                 break;
2892         case PACKET_MR_ALLMULTI:
2893                 return dev_set_allmulti(dev, what);
2894                 break;
2895         case PACKET_MR_UNICAST:
2896                 if (i->alen != dev->addr_len)
2897                         return -EINVAL;
2898                 if (what > 0)
2899                         return dev_uc_add(dev, i->addr);
2900                 else
2901                         return dev_uc_del(dev, i->addr);
2902                 break;
2903         default:
2904                 break;
2905         }
2906         return 0;
2907 }
2908
2909 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2910 {
2911         for ( ; i; i = i->next) {
2912                 if (i->ifindex == dev->ifindex)
2913                         packet_dev_mc(dev, i, what);
2914         }
2915 }
2916
2917 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2918 {
2919         struct packet_sock *po = pkt_sk(sk);
2920         struct packet_mclist *ml, *i;
2921         struct net_device *dev;
2922         int err;
2923
2924         rtnl_lock();
2925
2926         err = -ENODEV;
2927         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2928         if (!dev)
2929                 goto done;
2930
2931         err = -EINVAL;
2932         if (mreq->mr_alen > dev->addr_len)
2933                 goto done;
2934
2935         err = -ENOBUFS;
2936         i = kmalloc(sizeof(*i), GFP_KERNEL);
2937         if (i == NULL)
2938                 goto done;
2939
2940         err = 0;
2941         for (ml = po->mclist; ml; ml = ml->next) {
2942                 if (ml->ifindex == mreq->mr_ifindex &&
2943                     ml->type == mreq->mr_type &&
2944                     ml->alen == mreq->mr_alen &&
2945                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2946                         ml->count++;
2947                         /* Free the new element ... */
2948                         kfree(i);
2949                         goto done;
2950                 }
2951         }
2952
2953         i->type = mreq->mr_type;
2954         i->ifindex = mreq->mr_ifindex;
2955         i->alen = mreq->mr_alen;
2956         memcpy(i->addr, mreq->mr_address, i->alen);
2957         i->count = 1;
2958         i->next = po->mclist;
2959         po->mclist = i;
2960         err = packet_dev_mc(dev, i, 1);
2961         if (err) {
2962                 po->mclist = i->next;
2963                 kfree(i);
2964         }
2965
2966 done:
2967         rtnl_unlock();
2968         return err;
2969 }
2970
2971 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2972 {
2973         struct packet_mclist *ml, **mlp;
2974
2975         rtnl_lock();
2976
2977         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
2978                 if (ml->ifindex == mreq->mr_ifindex &&
2979                     ml->type == mreq->mr_type &&
2980                     ml->alen == mreq->mr_alen &&
2981                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2982                         if (--ml->count == 0) {
2983                                 struct net_device *dev;
2984                                 *mlp = ml->next;
2985                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2986                                 if (dev)
2987                                         packet_dev_mc(dev, ml, -1);
2988                                 kfree(ml);
2989                         }
2990                         rtnl_unlock();
2991                         return 0;
2992                 }
2993         }
2994         rtnl_unlock();
2995         return -EADDRNOTAVAIL;
2996 }
2997
2998 static void packet_flush_mclist(struct sock *sk)
2999 {
3000         struct packet_sock *po = pkt_sk(sk);
3001         struct packet_mclist *ml;
3002
3003         if (!po->mclist)
3004                 return;
3005
3006         rtnl_lock();
3007         while ((ml = po->mclist) != NULL) {
3008                 struct net_device *dev;
3009
3010                 po->mclist = ml->next;
3011                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3012                 if (dev != NULL)
3013                         packet_dev_mc(dev, ml, -1);
3014                 kfree(ml);
3015         }
3016         rtnl_unlock();
3017 }
3018
3019 static int
3020 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3021 {
3022         struct sock *sk = sock->sk;
3023         struct packet_sock *po = pkt_sk(sk);
3024         int ret;
3025
3026         if (level != SOL_PACKET)
3027                 return -ENOPROTOOPT;
3028
3029         switch (optname) {
3030         case PACKET_ADD_MEMBERSHIP:
3031         case PACKET_DROP_MEMBERSHIP:
3032         {
3033                 struct packet_mreq_max mreq;
3034                 int len = optlen;
3035                 memset(&mreq, 0, sizeof(mreq));
3036                 if (len < sizeof(struct packet_mreq))
3037                         return -EINVAL;
3038                 if (len > sizeof(mreq))
3039                         len = sizeof(mreq);
3040                 if (copy_from_user(&mreq, optval, len))
3041                         return -EFAULT;
3042                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3043                         return -EINVAL;
3044                 if (optname == PACKET_ADD_MEMBERSHIP)
3045                         ret = packet_mc_add(sk, &mreq);
3046                 else
3047                         ret = packet_mc_drop(sk, &mreq);
3048                 return ret;
3049         }
3050
3051         case PACKET_RX_RING:
3052         case PACKET_TX_RING:
3053         {
3054                 union tpacket_req_u req_u;
3055                 int len;
3056
3057                 switch (po->tp_version) {
3058                 case TPACKET_V1:
3059                 case TPACKET_V2:
3060                         len = sizeof(req_u.req);
3061                         break;
3062                 case TPACKET_V3:
3063                 default:
3064                         len = sizeof(req_u.req3);
3065                         break;
3066                 }
3067                 if (optlen < len)
3068                         return -EINVAL;
3069                 if (pkt_sk(sk)->has_vnet_hdr)
3070                         return -EINVAL;
3071                 if (copy_from_user(&req_u.req, optval, len))
3072                         return -EFAULT;
3073                 return packet_set_ring(sk, &req_u, 0,
3074                         optname == PACKET_TX_RING);
3075         }
3076         case PACKET_COPY_THRESH:
3077         {
3078                 int val;
3079
3080                 if (optlen != sizeof(val))
3081                         return -EINVAL;
3082                 if (copy_from_user(&val, optval, sizeof(val)))
3083                         return -EFAULT;
3084
3085                 pkt_sk(sk)->copy_thresh = val;
3086                 return 0;
3087         }
3088         case PACKET_VERSION:
3089         {
3090                 int val;
3091
3092                 if (optlen != sizeof(val))
3093                         return -EINVAL;
3094                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3095                         return -EBUSY;
3096                 if (copy_from_user(&val, optval, sizeof(val)))
3097                         return -EFAULT;
3098                 switch (val) {
3099                 case TPACKET_V1:
3100                 case TPACKET_V2:
3101                 case TPACKET_V3:
3102                         po->tp_version = val;
3103                         return 0;
3104                 default:
3105                         return -EINVAL;
3106                 }
3107         }
3108         case PACKET_RESERVE:
3109         {
3110                 unsigned int val;
3111
3112                 if (optlen != sizeof(val))
3113                         return -EINVAL;
3114                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3115                         return -EBUSY;
3116                 if (copy_from_user(&val, optval, sizeof(val)))
3117                         return -EFAULT;
3118                 po->tp_reserve = val;
3119                 return 0;
3120         }
3121         case PACKET_LOSS:
3122         {
3123                 unsigned int val;
3124
3125                 if (optlen != sizeof(val))
3126                         return -EINVAL;
3127                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3128                         return -EBUSY;
3129                 if (copy_from_user(&val, optval, sizeof(val)))
3130                         return -EFAULT;
3131                 po->tp_loss = !!val;
3132                 return 0;
3133         }
3134         case PACKET_AUXDATA:
3135         {
3136                 int val;
3137
3138                 if (optlen < sizeof(val))
3139                         return -EINVAL;
3140                 if (copy_from_user(&val, optval, sizeof(val)))
3141                         return -EFAULT;
3142
3143                 po->auxdata = !!val;
3144                 return 0;
3145         }
3146         case PACKET_ORIGDEV:
3147         {
3148                 int val;
3149
3150                 if (optlen < sizeof(val))
3151                         return -EINVAL;
3152                 if (copy_from_user(&val, optval, sizeof(val)))
3153                         return -EFAULT;
3154
3155                 po->origdev = !!val;
3156                 return 0;
3157         }
3158         case PACKET_VNET_HDR:
3159         {
3160                 int val;
3161
3162                 if (sock->type != SOCK_RAW)
3163                         return -EINVAL;
3164                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3165                         return -EBUSY;
3166                 if (optlen < sizeof(val))
3167                         return -EINVAL;
3168                 if (copy_from_user(&val, optval, sizeof(val)))
3169                         return -EFAULT;
3170
3171                 po->has_vnet_hdr = !!val;
3172                 return 0;
3173         }
3174         case PACKET_TIMESTAMP:
3175         {
3176                 int val;
3177
3178                 if (optlen != sizeof(val))
3179                         return -EINVAL;
3180                 if (copy_from_user(&val, optval, sizeof(val)))
3181                         return -EFAULT;
3182
3183                 po->tp_tstamp = val;
3184                 return 0;
3185         }
3186         case PACKET_FANOUT:
3187         {
3188                 int val;
3189
3190                 if (optlen != sizeof(val))
3191                         return -EINVAL;
3192                 if (copy_from_user(&val, optval, sizeof(val)))
3193                         return -EFAULT;
3194
3195                 return fanout_add(sk, val & 0xffff, val >> 16);
3196         }
3197         default:
3198                 return -ENOPROTOOPT;
3199         }
3200 }
3201
3202 static int packet_getsockopt(struct socket *sock, int level, int optname,
3203                              char __user *optval, int __user *optlen)
3204 {
3205         int len;
3206         int val;
3207         struct sock *sk = sock->sk;
3208         struct packet_sock *po = pkt_sk(sk);
3209         void *data;
3210         struct tpacket_stats st;
3211         union tpacket_stats_u st_u;
3212
3213         if (level != SOL_PACKET)
3214                 return -ENOPROTOOPT;
3215
3216         if (get_user(len, optlen))
3217                 return -EFAULT;
3218
3219         if (len < 0)
3220                 return -EINVAL;
3221
3222         switch (optname) {
3223         case PACKET_STATISTICS:
3224                 if (po->tp_version == TPACKET_V3) {
3225                         len = sizeof(struct tpacket_stats_v3);
3226                 } else {
3227                         if (len > sizeof(struct tpacket_stats))
3228                                 len = sizeof(struct tpacket_stats);
3229                 }
3230                 spin_lock_bh(&sk->sk_receive_queue.lock);
3231                 if (po->tp_version == TPACKET_V3) {
3232                         memcpy(&st_u.stats3, &po->stats,
3233                         sizeof(struct tpacket_stats));
3234                         st_u.stats3.tp_freeze_q_cnt =
3235                         po->stats_u.stats3.tp_freeze_q_cnt;
3236                         st_u.stats3.tp_packets += po->stats.tp_drops;
3237                         data = &st_u.stats3;
3238                 } else {
3239                         st = po->stats;
3240                         st.tp_packets += st.tp_drops;
3241                         data = &st;
3242                 }
3243                 memset(&po->stats, 0, sizeof(st));
3244                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3245                 break;
3246         case PACKET_AUXDATA:
3247                 if (len > sizeof(int))
3248                         len = sizeof(int);
3249                 val = po->auxdata;
3250
3251                 data = &val;
3252                 break;
3253         case PACKET_ORIGDEV:
3254                 if (len > sizeof(int))
3255                         len = sizeof(int);
3256                 val = po->origdev;
3257
3258                 data = &val;
3259                 break;
3260         case PACKET_VNET_HDR:
3261                 if (len > sizeof(int))
3262                         len = sizeof(int);
3263                 val = po->has_vnet_hdr;
3264
3265                 data = &val;
3266                 break;
3267         case PACKET_VERSION:
3268                 if (len > sizeof(int))
3269                         len = sizeof(int);
3270                 val = po->tp_version;
3271                 data = &val;
3272                 break;
3273         case PACKET_HDRLEN:
3274                 if (len > sizeof(int))
3275                         len = sizeof(int);
3276                 if (copy_from_user(&val, optval, len))
3277                         return -EFAULT;
3278                 switch (val) {
3279                 case TPACKET_V1:
3280                         val = sizeof(struct tpacket_hdr);
3281                         break;
3282                 case TPACKET_V2:
3283                         val = sizeof(struct tpacket2_hdr);
3284                         break;
3285                 case TPACKET_V3:
3286                         val = sizeof(struct tpacket3_hdr);
3287                         break;
3288                 default:
3289                         return -EINVAL;
3290                 }
3291                 data = &val;
3292                 break;
3293         case PACKET_RESERVE:
3294                 if (len > sizeof(unsigned int))
3295                         len = sizeof(unsigned int);
3296                 val = po->tp_reserve;
3297                 data = &val;
3298                 break;
3299         case PACKET_LOSS:
3300                 if (len > sizeof(unsigned int))
3301                         len = sizeof(unsigned int);
3302                 val = po->tp_loss;
3303                 data = &val;
3304                 break;
3305         case PACKET_TIMESTAMP:
3306                 if (len > sizeof(int))
3307                         len = sizeof(int);
3308                 val = po->tp_tstamp;
3309                 data = &val;
3310                 break;
3311         case PACKET_FANOUT:
3312                 if (len > sizeof(int))
3313                         len = sizeof(int);
3314                 val = (po->fanout ?
3315                        ((u32)po->fanout->id |
3316                         ((u32)po->fanout->type << 16)) :
3317                        0);
3318                 data = &val;
3319                 break;
3320         default:
3321                 return -ENOPROTOOPT;
3322         }
3323
3324         if (put_user(len, optlen))
3325                 return -EFAULT;
3326         if (copy_to_user(optval, data, len))
3327                 return -EFAULT;
3328         return 0;
3329 }
3330
3331
3332 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3333 {
3334         struct sock *sk;
3335         struct hlist_node *node;
3336         struct net_device *dev = data;
3337         struct net *net = dev_net(dev);
3338
3339         rcu_read_lock();
3340         sk_for_each_rcu(sk, node, &net->packet.sklist) {
3341                 struct packet_sock *po = pkt_sk(sk);
3342
3343                 switch (msg) {
3344                 case NETDEV_UNREGISTER:
3345                         if (po->mclist)
3346                                 packet_dev_mclist(dev, po->mclist, -1);
3347                         /* fallthrough */
3348
3349                 case NETDEV_DOWN:
3350                         if (dev->ifindex == po->ifindex) {
3351                                 spin_lock(&po->bind_lock);
3352                                 if (po->running) {
3353                                         __unregister_prot_hook(sk, false);
3354                                         sk->sk_err = ENETDOWN;
3355                                         if (!sock_flag(sk, SOCK_DEAD))
3356                                                 sk->sk_error_report(sk);
3357                                 }
3358                                 if (msg == NETDEV_UNREGISTER) {
3359                                         po->ifindex = -1;
3360                                         if (po->prot_hook.dev)
3361                                                 dev_put(po->prot_hook.dev);
3362                                         po->prot_hook.dev = NULL;
3363                                 }
3364                                 spin_unlock(&po->bind_lock);
3365                         }
3366                         break;
3367                 case NETDEV_UP:
3368                         if (dev->ifindex == po->ifindex) {
3369                                 spin_lock(&po->bind_lock);
3370                                 if (po->num)
3371                                         register_prot_hook(sk);
3372                                 spin_unlock(&po->bind_lock);
3373                         }
3374                         break;
3375                 }
3376         }
3377         rcu_read_unlock();
3378         return NOTIFY_DONE;
3379 }
3380
3381
3382 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3383                         unsigned long arg)
3384 {
3385         struct sock *sk = sock->sk;
3386
3387         switch (cmd) {
3388         case SIOCOUTQ:
3389         {
3390                 int amount = sk_wmem_alloc_get(sk);
3391
3392                 return put_user(amount, (int __user *)arg);
3393         }
3394         case SIOCINQ:
3395         {
3396                 struct sk_buff *skb;
3397                 int amount = 0;
3398
3399                 spin_lock_bh(&sk->sk_receive_queue.lock);
3400                 skb = skb_peek(&sk->sk_receive_queue);
3401                 if (skb)
3402                         amount = skb->len;
3403                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3404                 return put_user(amount, (int __user *)arg);
3405         }
3406         case SIOCGSTAMP:
3407                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3408         case SIOCGSTAMPNS:
3409                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3410
3411 #ifdef CONFIG_INET
3412         case SIOCADDRT:
3413         case SIOCDELRT:
3414         case SIOCDARP:
3415         case SIOCGARP:
3416         case SIOCSARP:
3417         case SIOCGIFADDR:
3418         case SIOCSIFADDR:
3419         case SIOCGIFBRDADDR:
3420         case SIOCSIFBRDADDR:
3421         case SIOCGIFNETMASK:
3422         case SIOCSIFNETMASK:
3423         case SIOCGIFDSTADDR:
3424         case SIOCSIFDSTADDR:
3425         case SIOCSIFFLAGS:
3426                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3427 #endif
3428
3429         default:
3430                 return -ENOIOCTLCMD;
3431         }
3432         return 0;
3433 }
3434
3435 static unsigned int packet_poll(struct file *file, struct socket *sock,
3436                                 poll_table *wait)
3437 {
3438         struct sock *sk = sock->sk;
3439         struct packet_sock *po = pkt_sk(sk);
3440         unsigned int mask = datagram_poll(file, sock, wait);
3441
3442         spin_lock_bh(&sk->sk_receive_queue.lock);
3443         if (po->rx_ring.pg_vec) {
3444                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3445                         TP_STATUS_KERNEL))
3446                         mask |= POLLIN | POLLRDNORM;
3447         }
3448         spin_unlock_bh(&sk->sk_receive_queue.lock);
3449         spin_lock_bh(&sk->sk_write_queue.lock);
3450         if (po->tx_ring.pg_vec) {
3451                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3452                         mask |= POLLOUT | POLLWRNORM;
3453         }
3454         spin_unlock_bh(&sk->sk_write_queue.lock);
3455         return mask;
3456 }
3457
3458
3459 /* Dirty? Well, I still did not learn better way to account
3460  * for user mmaps.
3461  */
3462
3463 static void packet_mm_open(struct vm_area_struct *vma)
3464 {
3465         struct file *file = vma->vm_file;
3466         struct socket *sock = file->private_data;
3467         struct sock *sk = sock->sk;
3468
3469         if (sk)
3470                 atomic_inc(&pkt_sk(sk)->mapped);
3471 }
3472
3473 static void packet_mm_close(struct vm_area_struct *vma)
3474 {
3475         struct file *file = vma->vm_file;
3476         struct socket *sock = file->private_data;
3477         struct sock *sk = sock->sk;
3478
3479         if (sk)
3480                 atomic_dec(&pkt_sk(sk)->mapped);
3481 }
3482
3483 static const struct vm_operations_struct packet_mmap_ops = {
3484         .open   =       packet_mm_open,
3485         .close  =       packet_mm_close,
3486 };
3487
3488 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3489                         unsigned int len)
3490 {
3491         int i;
3492
3493         for (i = 0; i < len; i++) {
3494                 if (likely(pg_vec[i].buffer)) {
3495                         if (is_vmalloc_addr(pg_vec[i].buffer))
3496                                 vfree(pg_vec[i].buffer);
3497                         else
3498                                 free_pages((unsigned long)pg_vec[i].buffer,
3499                                            order);
3500                         pg_vec[i].buffer = NULL;
3501                 }
3502         }
3503         kfree(pg_vec);
3504 }
3505
3506 static char *alloc_one_pg_vec_page(unsigned long order)
3507 {
3508         char *buffer = NULL;
3509         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3510                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3511
3512         buffer = (char *) __get_free_pages(gfp_flags, order);
3513
3514         if (buffer)
3515                 return buffer;
3516
3517         /*
3518          * __get_free_pages failed, fall back to vmalloc
3519          */
3520         buffer = vzalloc((1 << order) * PAGE_SIZE);
3521
3522         if (buffer)
3523                 return buffer;
3524
3525         /*
3526          * vmalloc failed, lets dig into swap here
3527          */
3528         gfp_flags &= ~__GFP_NORETRY;
3529         buffer = (char *)__get_free_pages(gfp_flags, order);
3530         if (buffer)
3531                 return buffer;
3532
3533         /*
3534          * complete and utter failure
3535          */
3536         return NULL;
3537 }
3538
3539 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3540 {
3541         unsigned int block_nr = req->tp_block_nr;
3542         struct pgv *pg_vec;
3543         int i;
3544
3545         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3546         if (unlikely(!pg_vec))
3547                 goto out;
3548
3549         for (i = 0; i < block_nr; i++) {
3550                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3551                 if (unlikely(!pg_vec[i].buffer))
3552                         goto out_free_pgvec;
3553         }
3554
3555 out:
3556         return pg_vec;
3557
3558 out_free_pgvec:
3559         free_pg_vec(pg_vec, order, block_nr);
3560         pg_vec = NULL;
3561         goto out;
3562 }
3563
3564 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3565                 int closing, int tx_ring)
3566 {
3567         struct pgv *pg_vec = NULL;
3568         struct packet_sock *po = pkt_sk(sk);
3569         int was_running, order = 0;
3570         struct packet_ring_buffer *rb;
3571         struct sk_buff_head *rb_queue;
3572         __be16 num;
3573         int err = -EINVAL;
3574         /* Added to avoid minimal code churn */
3575         struct tpacket_req *req = &req_u->req;
3576
3577         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3578         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3579                 WARN(1, "Tx-ring is not supported.\n");
3580                 goto out;
3581         }
3582
3583         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3584         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3585
3586         err = -EBUSY;
3587         if (!closing) {
3588                 if (atomic_read(&po->mapped))
3589                         goto out;
3590                 if (atomic_read(&rb->pending))
3591                         goto out;
3592         }
3593
3594         if (req->tp_block_nr) {
3595                 /* Sanity tests and some calculations */
3596                 err = -EBUSY;
3597                 if (unlikely(rb->pg_vec))
3598                         goto out;
3599
3600                 switch (po->tp_version) {
3601                 case TPACKET_V1:
3602                         po->tp_hdrlen = TPACKET_HDRLEN;
3603                         break;
3604                 case TPACKET_V2:
3605                         po->tp_hdrlen = TPACKET2_HDRLEN;
3606                         break;
3607                 case TPACKET_V3:
3608                         po->tp_hdrlen = TPACKET3_HDRLEN;
3609                         break;
3610                 }
3611
3612                 err = -EINVAL;
3613                 if (unlikely((int)req->tp_block_size <= 0))
3614                         goto out;
3615                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3616                         goto out;
3617                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3618                                         po->tp_reserve))
3619                         goto out;
3620                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3621                         goto out;
3622
3623                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3624                 if (unlikely(rb->frames_per_block <= 0))
3625                         goto out;
3626                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3627                                         req->tp_frame_nr))
3628                         goto out;
3629
3630                 err = -ENOMEM;
3631                 order = get_order(req->tp_block_size);
3632                 pg_vec = alloc_pg_vec(req, order);
3633                 if (unlikely(!pg_vec))
3634                         goto out;
3635                 switch (po->tp_version) {
3636                 case TPACKET_V3:
3637                 /* Transmit path is not supported. We checked
3638                  * it above but just being paranoid
3639                  */
3640                         if (!tx_ring)
3641                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3642                                 break;
3643                 default:
3644                         break;
3645                 }
3646         }
3647         /* Done */
3648         else {
3649                 err = -EINVAL;
3650                 if (unlikely(req->tp_frame_nr))
3651                         goto out;
3652         }
3653
3654         lock_sock(sk);
3655
3656         /* Detach socket from network */
3657         spin_lock(&po->bind_lock);
3658         was_running = po->running;
3659         num = po->num;
3660         if (was_running) {
3661                 po->num = 0;
3662                 __unregister_prot_hook(sk, false);
3663         }
3664         spin_unlock(&po->bind_lock);
3665
3666         synchronize_net();
3667
3668         err = -EBUSY;
3669         mutex_lock(&po->pg_vec_lock);
3670         if (closing || atomic_read(&po->mapped) == 0) {
3671                 err = 0;
3672                 spin_lock_bh(&rb_queue->lock);
3673                 swap(rb->pg_vec, pg_vec);
3674                 rb->frame_max = (req->tp_frame_nr - 1);
3675                 rb->head = 0;
3676                 rb->frame_size = req->tp_frame_size;
3677                 spin_unlock_bh(&rb_queue->lock);
3678
3679                 swap(rb->pg_vec_order, order);
3680                 swap(rb->pg_vec_len, req->tp_block_nr);
3681
3682                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3683                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3684                                                 tpacket_rcv : packet_rcv;
3685                 skb_queue_purge(rb_queue);
3686                 if (atomic_read(&po->mapped))
3687                         pr_err("packet_mmap: vma is busy: %d\n",
3688                                atomic_read(&po->mapped));
3689         }
3690         mutex_unlock(&po->pg_vec_lock);
3691
3692         spin_lock(&po->bind_lock);
3693         if (was_running) {
3694                 po->num = num;
3695                 register_prot_hook(sk);
3696         }
3697         spin_unlock(&po->bind_lock);
3698         if (closing && (po->tp_version > TPACKET_V2)) {
3699                 /* Because we don't support block-based V3 on tx-ring */
3700                 if (!tx_ring)
3701                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3702         }
3703         release_sock(sk);
3704
3705         if (pg_vec)
3706                 free_pg_vec(pg_vec, order, req->tp_block_nr);
3707 out:
3708         return err;
3709 }
3710
3711 static int packet_mmap(struct file *file, struct socket *sock,
3712                 struct vm_area_struct *vma)
3713 {
3714         struct sock *sk = sock->sk;
3715         struct packet_sock *po = pkt_sk(sk);
3716         unsigned long size, expected_size;
3717         struct packet_ring_buffer *rb;
3718         unsigned long start;
3719         int err = -EINVAL;
3720         int i;
3721
3722         if (vma->vm_pgoff)
3723                 return -EINVAL;
3724
3725         mutex_lock(&po->pg_vec_lock);
3726
3727         expected_size = 0;
3728         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3729                 if (rb->pg_vec) {
3730                         expected_size += rb->pg_vec_len
3731                                                 * rb->pg_vec_pages
3732                                                 * PAGE_SIZE;
3733                 }
3734         }
3735
3736         if (expected_size == 0)
3737                 goto out;
3738
3739         size = vma->vm_end - vma->vm_start;
3740         if (size != expected_size)
3741                 goto out;
3742
3743         start = vma->vm_start;
3744         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3745                 if (rb->pg_vec == NULL)
3746                         continue;
3747
3748                 for (i = 0; i < rb->pg_vec_len; i++) {
3749                         struct page *page;
3750                         void *kaddr = rb->pg_vec[i].buffer;
3751                         int pg_num;
3752
3753                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3754                                 page = pgv_to_page(kaddr);
3755                                 err = vm_insert_page(vma, start, page);
3756                                 if (unlikely(err))
3757                                         goto out;
3758                                 start += PAGE_SIZE;
3759                                 kaddr += PAGE_SIZE;
3760                         }
3761                 }
3762         }
3763
3764         atomic_inc(&po->mapped);
3765         vma->vm_ops = &packet_mmap_ops;
3766         err = 0;
3767
3768 out:
3769         mutex_unlock(&po->pg_vec_lock);
3770         return err;
3771 }
3772
3773 static const struct proto_ops packet_ops_spkt = {
3774         .family =       PF_PACKET,
3775         .owner =        THIS_MODULE,
3776         .release =      packet_release,
3777         .bind =         packet_bind_spkt,
3778         .connect =      sock_no_connect,
3779         .socketpair =   sock_no_socketpair,
3780         .accept =       sock_no_accept,
3781         .getname =      packet_getname_spkt,
3782         .poll =         datagram_poll,
3783         .ioctl =        packet_ioctl,
3784         .listen =       sock_no_listen,
3785         .shutdown =     sock_no_shutdown,
3786         .setsockopt =   sock_no_setsockopt,
3787         .getsockopt =   sock_no_getsockopt,
3788         .sendmsg =      packet_sendmsg_spkt,
3789         .recvmsg =      packet_recvmsg,
3790         .mmap =         sock_no_mmap,
3791         .sendpage =     sock_no_sendpage,
3792 };
3793
3794 static const struct proto_ops packet_ops = {
3795         .family =       PF_PACKET,
3796         .owner =        THIS_MODULE,
3797         .release =      packet_release,
3798         .bind =         packet_bind,
3799         .connect =      sock_no_connect,
3800         .socketpair =   sock_no_socketpair,
3801         .accept =       sock_no_accept,
3802         .getname =      packet_getname,
3803         .poll =         packet_poll,
3804         .ioctl =        packet_ioctl,
3805         .listen =       sock_no_listen,
3806         .shutdown =     sock_no_shutdown,
3807         .setsockopt =   packet_setsockopt,
3808         .getsockopt =   packet_getsockopt,
3809         .sendmsg =      packet_sendmsg,
3810         .recvmsg =      packet_recvmsg,
3811         .mmap =         packet_mmap,
3812         .sendpage =     sock_no_sendpage,
3813 };
3814
3815 static const struct net_proto_family packet_family_ops = {
3816         .family =       PF_PACKET,
3817         .create =       packet_create,
3818         .owner  =       THIS_MODULE,
3819 };
3820
3821 static struct notifier_block packet_netdev_notifier = {
3822         .notifier_call =        packet_notifier,
3823 };
3824
3825 #ifdef CONFIG_PROC_FS
3826
3827 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3828         __acquires(RCU)
3829 {
3830         struct net *net = seq_file_net(seq);
3831
3832         rcu_read_lock();
3833         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3834 }
3835
3836 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3837 {
3838         struct net *net = seq_file_net(seq);
3839         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3840 }
3841
3842 static void packet_seq_stop(struct seq_file *seq, void *v)
3843         __releases(RCU)
3844 {
3845         rcu_read_unlock();
3846 }
3847
3848 static int packet_seq_show(struct seq_file *seq, void *v)
3849 {
3850         if (v == SEQ_START_TOKEN)
3851                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
3852         else {
3853                 struct sock *s = sk_entry(v);
3854                 const struct packet_sock *po = pkt_sk(s);
3855
3856                 seq_printf(seq,
3857                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
3858                            s,
3859                            atomic_read(&s->sk_refcnt),
3860                            s->sk_type,
3861                            ntohs(po->num),
3862                            po->ifindex,
3863                            po->running,
3864                            atomic_read(&s->sk_rmem_alloc),
3865                            sock_i_uid(s),
3866                            sock_i_ino(s));
3867         }
3868
3869         return 0;
3870 }
3871
3872 static const struct seq_operations packet_seq_ops = {
3873         .start  = packet_seq_start,
3874         .next   = packet_seq_next,
3875         .stop   = packet_seq_stop,
3876         .show   = packet_seq_show,
3877 };
3878
3879 static int packet_seq_open(struct inode *inode, struct file *file)
3880 {
3881         return seq_open_net(inode, file, &packet_seq_ops,
3882                             sizeof(struct seq_net_private));
3883 }
3884
3885 static const struct file_operations packet_seq_fops = {
3886         .owner          = THIS_MODULE,
3887         .open           = packet_seq_open,
3888         .read           = seq_read,
3889         .llseek         = seq_lseek,
3890         .release        = seq_release_net,
3891 };
3892
3893 #endif
3894
3895 static int __net_init packet_net_init(struct net *net)
3896 {
3897         spin_lock_init(&net->packet.sklist_lock);
3898         INIT_HLIST_HEAD(&net->packet.sklist);
3899
3900         if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
3901                 return -ENOMEM;
3902
3903         return 0;
3904 }
3905
3906 static void __net_exit packet_net_exit(struct net *net)
3907 {
3908         proc_net_remove(net, "packet");
3909 }
3910
3911 static struct pernet_operations packet_net_ops = {
3912         .init = packet_net_init,
3913         .exit = packet_net_exit,
3914 };
3915
3916
3917 static void __exit packet_exit(void)
3918 {
3919         unregister_netdevice_notifier(&packet_netdev_notifier);
3920         unregister_pernet_subsys(&packet_net_ops);
3921         sock_unregister(PF_PACKET);
3922         proto_unregister(&packet_proto);
3923 }
3924
3925 static int __init packet_init(void)
3926 {
3927         int rc = proto_register(&packet_proto, 0);
3928
3929         if (rc != 0)
3930                 goto out;
3931
3932         sock_register(&packet_family_ops);
3933         register_pernet_subsys(&packet_net_ops);
3934         register_netdevice_notifier(&packet_netdev_notifier);
3935 out:
3936         return rc;
3937 }
3938
3939 module_init(packet_init);
3940 module_exit(packet_exit);
3941 MODULE_LICENSE("GPL");
3942 MODULE_ALIAS_NETPROTO(PF_PACKET);