net/packet/af_packet.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              PACKET - implements raw packet sockets.
   7  *
   8  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  11  *
  12  * Fixes:
  13  *              Alan Cox        :       verify_area() now used correctly
  14  *              Alan Cox        :       new skbuff lists, look ma no backlogs!
  15  *              Alan Cox        :       tidied skbuff lists.
  16  *              Alan Cox        :       Now uses generic datagram routines I
  17  *                                      added. Also fixed the peek/read crash
  18  *                                      from all old Linux datagram code.
  19  *              Alan Cox        :       Uses the improved datagram code.
  20  *              Alan Cox        :       Added NULL's for socket options.
  21  *              Alan Cox        :       Re-commented the code.
  22  *              Alan Cox        :       Use new kernel side addressing
  23  *              Rob Janssen     :       Correct MTU usage.
  24  *              Dave Platt      :       Counter leaks caused by incorrect
  25  *                                      interrupt locking and some slightly
  26  *                                      dubious gcc output. Can you read
  27  *                                      compiler: it said _VOLATILE_
  28  *      Richard Kooijman        :       Timestamp fixes.
  29  *              Alan Cox        :       New buffers. Use sk->mac.raw.
  30  *              Alan Cox        :       sendmsg/recvmsg support.
  31  *              Alan Cox        :       Protocol setting support
  32  *      Alexey Kuznetsov        :       Untied from IPv4 stack.
  33  *      Cyrus Durgin            :       Fixed kerneld for kmod.
  34  *      Michal Ostrowski        :       Module initialization cleanup.
  35  *         Ulises Alonso        :       Frame number limit removal and
  36  *                                      packet_set_ring memory leak.
  37  *              Eric Biederman  :       Allow for > 8 byte hardware addresses.
  38  *                                      The convention is that longer addresses
  39  *                                      will simply extend the hardware address
  40  *                                      byte arrays at the end of sockaddr_ll
  41  *                                      and packet_mreq.
  42  *              Johann Baudy    :       Added TX RING.
  43  *              Chetan Loke     :       Implemented TPACKET_V3 block abstraction
  44  *                                      layer.
  45  *                                      Copyright (C) 2011, <lokec@ccs.neu.edu>
  46  *
  47  *
  48  *              This program is free software; you can redistribute it and/or
  49  *              modify it under the terms of the GNU General Public License
  50  *              as published by the Free Software Foundation; either version
  51  *              2 of the License, or (at your option) any later version.
  52  *
  53  */
  54
  55 #include <linux/types.h>
  56 #include <linux/mm.h>
  57 #include <linux/capability.h>
  58 #include <linux/fcntl.h>
  59 #include <linux/socket.h>
  60 #include <linux/in.h>
  61 #include <linux/inet.h>
  62 #include <linux/netdevice.h>
  63 #include <linux/if_packet.h>
  64 #include <linux/wireless.h>
  65 #include <linux/kernel.h>
  66 #include <linux/kmod.h>
  67 #include <linux/slab.h>
  68 #include <linux/vmalloc.h>
  69 #include <net/net_namespace.h>
  70 #include <net/ip.h>
  71 #include <net/protocol.h>
  72 #include <linux/skbuff.h>
  73 #include <net/sock.h>
  74 #include <linux/errno.h>
  75 #include <linux/timer.h>
  76 #include <asm/system.h>
  77 #include <asm/uaccess.h>
  78 #include <asm/ioctls.h>
  79 #include <asm/page.h>
  80 #include <asm/cacheflush.h>
  81 #include <asm/io.h>
  82 #include <linux/proc_fs.h>
  83 #include <linux/seq_file.h>
  84 #include <linux/poll.h>
  85 #include <linux/module.h>
  86 #include <linux/init.h>
  87 #include <linux/mutex.h>
  88 #include <linux/if_vlan.h>
  89 #include <linux/virtio_net.h>
  90 #include <linux/errqueue.h>
  91 #include <linux/net_tstamp.h>
  92
  93 #ifdef CONFIG_INET
  94 #include <net/inet_common.h>
  95 #endif
  96
  97 /*
  98    Assumptions:
  99    - if device has no dev->hard_header routine, it adds and removes ll header
 100      inside itself. In this case ll header is invisible outside of device,
 101      but higher levels still should reserve dev->hard_header_len.
 102      Some devices are enough clever to reallocate skb, when header
 103      will not fit to reserved space (tunnel), another ones are silly
 104      (PPP).
 105    - packet socket receives packets with pulled ll header,
 106      so that SOCK_RAW should push it back.
 107
 108 On receive:
 109 -----------
 110
 111 Incoming, dev->hard_header!=NULL
 112    mac_header -> ll header
 113    data       -> data
 114
 115 Outgoing, dev->hard_header!=NULL
 116    mac_header -> ll header
 117    data       -> ll header
 118
 119 Incoming, dev->hard_header==NULL
 120    mac_header -> UNKNOWN position. It is very likely, that it points to ll
 121                  header.  PPP makes it, that is wrong, because introduce
 122                  assymetry between rx and tx paths.
 123    data       -> data
 124
 125 Outgoing, dev->hard_header==NULL
 126    mac_header -> data. ll header is still not built!
 127    data       -> data
 128
 129 Resume
 130   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 131
 132
 133 On transmit:
 134 ------------
 135
 136 dev->hard_header != NULL
 137    mac_header -> ll header
 138    data       -> ll header
 139
 140 dev->hard_header == NULL (ll header is added by device, we cannot control it)
 141    mac_header -> data
 142    data       -> data
 143
 144    We should set nh.raw on output to correct posistion,
 145    packet classifier depends on it.
 146  */
 147
 148 /* Private packet socket structures. */
 149
 150 struct packet_mclist {
 151         struct packet_mclist    *next;
 152         int                     ifindex;
 153         int                     count;
 154         unsigned short          type;
 155         unsigned short          alen;
 156         unsigned char           addr[MAX_ADDR_LEN];
 157 };
 158 /* identical to struct packet_mreq except it has
 159  * a longer address field.
 160  */
 161 struct packet_mreq_max {
 162         int             mr_ifindex;
 163         unsigned short  mr_type;
 164         unsigned short  mr_alen;
 165         unsigned char   mr_address[MAX_ADDR_LEN];
 166 };
 167
 168 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 169                 int closing, int tx_ring);
 170
 171
 172 #define V3_ALIGNMENT    (8)
 173
 174 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 175
 176 #define BLK_PLUS_PRIV(sz_of_priv) \
 177         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 178
 179 /* kbdq - kernel block descriptor queue */
 180 struct tpacket_kbdq_core {
 181         struct pgv      *pkbdq;
 182         unsigned int    feature_req_word;
 183         unsigned int    hdrlen;
 184         unsigned char   reset_pending_on_curr_blk;
 185         unsigned char   delete_blk_timer;
 186         unsigned short  kactive_blk_num;
 187         unsigned short  blk_sizeof_priv;
 188
 189         /* last_kactive_blk_num:
 190          * trick to see if user-space has caught up
 191          * in order to avoid refreshing timer when every single pkt arrives.
 192          */
 193         unsigned short  last_kactive_blk_num;
 194
 195         char            *pkblk_start;
 196         char            *pkblk_end;
 197         int             kblk_size;
 198         unsigned int    knum_blocks;
 199         uint64_t        knxt_seq_num;
 200         char            *prev;
 201         char            *nxt_offset;
 202         struct sk_buff  *skb;
 203
 204         atomic_t        blk_fill_in_prog;
 205
 206         /* Default is set to 8ms */
 207 #define DEFAULT_PRB_RETIRE_TOV  (8)
 208
 209         unsigned short  retire_blk_tov;
 210         unsigned short  version;
 211         unsigned long   tov_in_jiffies;
 212
 213         /* timer to retire an outstanding block */
 214         struct timer_list retire_blk_timer;
 215 };
 216
 217 #define PGV_FROM_VMALLOC 1
 218 struct pgv {
 219         char *buffer;
 220 };
 221
 222 struct packet_ring_buffer {
 223         struct pgv              *pg_vec;
 224         unsigned int            head;
 225         unsigned int            frames_per_block;
 226         unsigned int            frame_size;
 227         unsigned int            frame_max;
 228
 229         unsigned int            pg_vec_order;
 230         unsigned int            pg_vec_pages;
 231         unsigned int            pg_vec_len;
 232
 233         struct tpacket_kbdq_core        prb_bdqc;
 234         atomic_t                pending;
 235 };
 236
 237 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
 238 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
 239 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
 240 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
 241 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
 242 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
 243 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
 244
 245 struct packet_sock;
 246 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
 247
 248 static void *packet_previous_frame(struct packet_sock *po,
 249                 struct packet_ring_buffer *rb,
 250                 int status);
 251 static void packet_increment_head(struct packet_ring_buffer *buff);
 252 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
 253                         struct tpacket_block_desc *);
 254 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 255                         struct packet_sock *);
 256 static void prb_retire_current_block(struct tpacket_kbdq_core *,
 257                 struct packet_sock *, unsigned int status);
 258 static int prb_queue_frozen(struct tpacket_kbdq_core *);
 259 static void prb_open_block(struct tpacket_kbdq_core *,
 260                 struct tpacket_block_desc *);
 261 static void prb_retire_rx_blk_timer_expired(unsigned long);
 262 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 263 static void prb_init_blk_timer(struct packet_sock *,
 264                 struct tpacket_kbdq_core *,
 265                 void (*func) (unsigned long));
 266 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 267 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 268                 struct tpacket3_hdr *);
 269 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 270                 struct tpacket3_hdr *);
 271 static void packet_flush_mclist(struct sock *sk);
 272
 273 struct packet_fanout;
 274 struct packet_sock {
 275         /* struct sock has to be the first member of packet_sock */
 276         struct sock             sk;
 277         struct packet_fanout    *fanout;
 278         struct tpacket_stats    stats;
 279         union  tpacket_stats_u  stats_u;
 280         struct packet_ring_buffer       rx_ring;
 281         struct packet_ring_buffer       tx_ring;
 282         int                     copy_thresh;
 283         spinlock_t              bind_lock;
 284         struct mutex            pg_vec_lock;
 285         unsigned int            running:1,      /* prot_hook is attached*/
 286                                 auxdata:1,
 287                                 origdev:1,
 288                                 has_vnet_hdr:1;
 289         int                     ifindex;        /* bound device         */
 290         __be16                  num;
 291         struct packet_mclist    *mclist;
 292         atomic_t                mapped;
 293         enum tpacket_versions   tp_version;
 294         unsigned int            tp_hdrlen;
 295         unsigned int            tp_reserve;
 296         unsigned int            tp_loss:1;
 297         unsigned int            tp_tstamp;
 298         struct packet_type      prot_hook ____cacheline_aligned_in_smp;
 299 };
 300
 301 #define PACKET_FANOUT_MAX       256
 302
 303 struct packet_fanout {
 304 #ifdef CONFIG_NET_NS
 305         struct net              *net;
 306 #endif
 307         unsigned int            num_members;
 308         u16                     id;
 309         u8                      type;
 310         u8                      defrag;
 311         atomic_t                rr_cur;
 312         struct list_head        list;
 313         struct sock             *arr[PACKET_FANOUT_MAX];
 314         spinlock_t              lock;
 315         atomic_t                sk_ref;
 316         struct packet_type      prot_hook ____cacheline_aligned_in_smp;
 317 };
 318
 319 struct packet_skb_cb {
 320         unsigned int origlen;
 321         union {
 322                 struct sockaddr_pkt pkt;
 323                 struct sockaddr_ll ll;
 324         } sa;
 325 };
 326
 327 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 328
 329 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 330 #define GET_PBLOCK_DESC(x, bid) \
 331         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 332 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 333         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 334 #define GET_NEXT_PRB_BLK_NUM(x) \
 335         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 336         ((x)->kactive_blk_num+1) : 0)
 337
 338 static struct packet_sock *pkt_sk(struct sock *sk)
 339 {
 340         return (struct packet_sock *)sk;
 341 }
 342
 343 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 344 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 345
 346 /* register_prot_hook must be invoked with the po->bind_lock held,
 347  * or from a context in which asynchronous accesses to the packet
 348  * socket is not possible (packet_create()).
 349  */
 350 static void register_prot_hook(struct sock *sk)
 351 {
 352         struct packet_sock *po = pkt_sk(sk);
 353         if (!po->running) {
 354                 if (po->fanout)
 355                         __fanout_link(sk, po);
 356                 else
 357                         dev_add_pack(&po->prot_hook);
 358                 sock_hold(sk);
 359                 po->running = 1;
 360         }
 361 }
 362
 363 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
 364  * held.   If the sync parameter is true, we will temporarily drop
 365  * the po->bind_lock and do a synchronize_net to make sure no
 366  * asynchronous packet processing paths still refer to the elements
 367  * of po->prot_hook.  If the sync parameter is false, it is the
 368  * callers responsibility to take care of this.
 369  */
 370 static void __unregister_prot_hook(struct sock *sk, bool sync)
 371 {
 372         struct packet_sock *po = pkt_sk(sk);
 373
 374         po->running = 0;
 375         if (po->fanout)
 376                 __fanout_unlink(sk, po);
 377         else
 378                 __dev_remove_pack(&po->prot_hook);
 379         __sock_put(sk);
 380
 381         if (sync) {
 382                 spin_unlock(&po->bind_lock);
 383                 synchronize_net();
 384                 spin_lock(&po->bind_lock);
 385         }
 386 }
 387
 388 static void unregister_prot_hook(struct sock *sk, bool sync)
 389 {
 390         struct packet_sock *po = pkt_sk(sk);
 391
 392         if (po->running)
 393                 __unregister_prot_hook(sk, sync);
 394 }
 395
 396 static inline __pure struct page *pgv_to_page(void *addr)
 397 {
 398         if (is_vmalloc_addr(addr))
 399                 return vmalloc_to_page(addr);
 400         return virt_to_page(addr);
 401 }
 402
 403 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 404 {
 405         union {
 406                 struct tpacket_hdr *h1;
 407                 struct tpacket2_hdr *h2;
 408                 void *raw;
 409         } h;
 410
 411         h.raw = frame;
 412         switch (po->tp_version) {
 413         case TPACKET_V1:
 414                 h.h1->tp_status = status;
 415                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 416                 break;
 417         case TPACKET_V2:
 418                 h.h2->tp_status = status;
 419                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 420                 break;
 421         case TPACKET_V3:
 422         default:
 423                 WARN(1, "TPACKET version not supported.\n");
 424                 BUG();
 425         }
 426
 427         smp_wmb();
 428 }
 429
 430 static int __packet_get_status(struct packet_sock *po, void *frame)
 431 {
 432         union {
 433                 struct tpacket_hdr *h1;
 434                 struct tpacket2_hdr *h2;
 435                 void *raw;
 436         } h;
 437
 438         smp_rmb();
 439
 440         h.raw = frame;
 441         switch (po->tp_version) {
 442         case TPACKET_V1:
 443                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 444                 return h.h1->tp_status;
 445         case TPACKET_V2:
 446                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 447                 return h.h2->tp_status;
 448         case TPACKET_V3:
 449         default:
 450                 WARN(1, "TPACKET version not supported.\n");
 451                 BUG();
 452                 return 0;
 453         }
 454 }
 455
 456 static void *packet_lookup_frame(struct packet_sock *po,
 457                 struct packet_ring_buffer *rb,
 458                 unsigned int position,
 459                 int status)
 460 {
 461         unsigned int pg_vec_pos, frame_offset;
 462         union {
 463                 struct tpacket_hdr *h1;
 464                 struct tpacket2_hdr *h2;
 465                 void *raw;
 466         } h;
 467
 468         pg_vec_pos = position / rb->frames_per_block;
 469         frame_offset = position % rb->frames_per_block;
 470
 471         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 472                 (frame_offset * rb->frame_size);
 473
 474         if (status != __packet_get_status(po, h.raw))
 475                 return NULL;
 476
 477         return h.raw;
 478 }
 479
 480 static void *packet_current_frame(struct packet_sock *po,
 481                 struct packet_ring_buffer *rb,
 482                 int status)
 483 {
 484         return packet_lookup_frame(po, rb, rb->head, status);
 485 }
 486
 487 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 488 {
 489         del_timer_sync(&pkc->retire_blk_timer);
 490 }
 491
 492 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 493                 int tx_ring,
 494                 struct sk_buff_head *rb_queue)
 495 {
 496         struct tpacket_kbdq_core *pkc;
 497
 498         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 499
 500         spin_lock(&rb_queue->lock);
 501         pkc->delete_blk_timer = 1;
 502         spin_unlock(&rb_queue->lock);
 503
 504         prb_del_retire_blk_timer(pkc);
 505 }
 506
 507 static void prb_init_blk_timer(struct packet_sock *po,
 508                 struct tpacket_kbdq_core *pkc,
 509                 void (*func) (unsigned long))
 510 {
 511         init_timer(&pkc->retire_blk_timer);
 512         pkc->retire_blk_timer.data = (long)po;
 513         pkc->retire_blk_timer.function = func;
 514         pkc->retire_blk_timer.expires = jiffies;
 515 }
 516
 517 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
 518 {
 519         if (tx_ring)
 520                 BUG();
 521
 522         prb_init_blk_timer(po, &po->rx_ring.prb_bdqc,
 523                            prb_retire_rx_blk_timer_expired);
 524 }
 525
 526 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 527                                 int blk_size_in_bytes)
 528 {
 529         struct net_device *dev;
 530         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 531         struct ethtool_cmd ecmd;
 532         int err;
 533
 534         rtnl_lock();
 535         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 536         if (unlikely(!dev)) {
 537                 rtnl_unlock();
 538                 return DEFAULT_PRB_RETIRE_TOV;
 539         }
 540         err = __ethtool_get_settings(dev, &ecmd);
 541         rtnl_unlock();
 542         if (!err) {
 543                 switch (ecmd.speed) {
 544                 case SPEED_10000:
 545                         msec = 1;
 546                         div = 10000/1000;
 547                         break;
 548                 case SPEED_1000:
 549                         msec = 1;
 550                         div = 1000/1000;
 551                         break;
 552                 /*
 553                  * If the link speed is so slow you don't really
 554                  * need to worry about perf anyways
 555                  */
 556                 case SPEED_100:
 557                 case SPEED_10:
 558                 default:
 559                         return DEFAULT_PRB_RETIRE_TOV;
 560                 }
 561         }
 562
 563         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 564
 565         if (div)
 566                 mbits /= div;
 567
 568         tmo = mbits * msec;
 569
 570         if (div)
 571                 return tmo+1;
 572         return tmo;
 573 }
 574
 575 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 576                         union tpacket_req_u *req_u)
 577 {
 578         p1->feature_req_word = req_u->req3.tp_feature_req_word;
 579 }
 580
 581 static void init_prb_bdqc(struct packet_sock *po,
 582                         struct packet_ring_buffer *rb,
 583                         struct pgv *pg_vec,
 584                         union tpacket_req_u *req_u, int tx_ring)
 585 {
 586         struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
 587         struct tpacket_block_desc *pbd;
 588
 589         memset(p1, 0x0, sizeof(*p1));
 590
 591         p1->knxt_seq_num = 1;
 592         p1->pkbdq = pg_vec;
 593         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 594         p1->pkblk_start = (char *)pg_vec[0].buffer;
 595         p1->kblk_size = req_u->req3.tp_block_size;
 596         p1->knum_blocks = req_u->req3.tp_block_nr;
 597         p1->hdrlen = po->tp_hdrlen;
 598         p1->version = po->tp_version;
 599         p1->last_kactive_blk_num = 0;
 600         po->stats_u.stats3.tp_freeze_q_cnt = 0;
 601         if (req_u->req3.tp_retire_blk_tov)
 602                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 603         else
 604                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 605                                                 req_u->req3.tp_block_size);
 606         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 607         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 608
 609         prb_init_ft_ops(p1, req_u);
 610         prb_setup_retire_blk_timer(po, tx_ring);
 611         prb_open_block(p1, pbd);
 612 }
 613
 614 /*  Do NOT update the last_blk_num first.
 615  *  Assumes sk_buff_head lock is held.
 616  */
 617 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 618 {
 619         mod_timer(&pkc->retire_blk_timer,
 620                         jiffies + pkc->tov_in_jiffies);
 621         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 622 }
 623
 624 /*
 625  * Timer logic:
 626  * 1) We refresh the timer only when we open a block.
 627  *    By doing this we don't waste cycles refreshing the timer
 628  *        on packet-by-packet basis.
 629  *
 630  * With a 1MB block-size, on a 1Gbps line, it will take
 631  * i) ~8 ms to fill a block + ii) memcpy etc.
 632  * In this cut we are not accounting for the memcpy time.
 633  *
 634  * So, if the user sets the 'tmo' to 10ms then the timer
 635  * will never fire while the block is still getting filled
 636  * (which is what we want). However, the user could choose
 637  * to close a block early and that's fine.
 638  *
 639  * But when the timer does fire, we check whether or not to refresh it.
 640  * Since the tmo granularity is in msecs, it is not too expensive
 641  * to refresh the timer, lets say every '8' msecs.
 642  * Either the user can set the 'tmo' or we can derive it based on
 643  * a) line-speed and b) block-size.
 644  * prb_calc_retire_blk_tmo() calculates the tmo.
 645  *
 646  */
 647 static void prb_retire_rx_blk_timer_expired(unsigned long data)
 648 {
 649         struct packet_sock *po = (struct packet_sock *)data;
 650         struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
 651         unsigned int frozen;
 652         struct tpacket_block_desc *pbd;
 653
 654         spin_lock(&po->sk.sk_receive_queue.lock);
 655
 656         frozen = prb_queue_frozen(pkc);
 657         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 658
 659         if (unlikely(pkc->delete_blk_timer))
 660                 goto out;
 661
 662         /* We only need to plug the race when the block is partially filled.
 663          * tpacket_rcv:
 664          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 665          *              copy_bits() is in progress ...
 666          *              timer fires on other cpu:
 667          *              we can't retire the current block because copy_bits
 668          *              is in progress.
 669          *
 670          */
 671         if (BLOCK_NUM_PKTS(pbd)) {
 672                 while (atomic_read(&pkc->blk_fill_in_prog)) {
 673                         /* Waiting for skb_copy_bits to finish... */
 674                         cpu_relax();
 675                 }
 676         }
 677
 678         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 679                 if (!frozen) {
 680                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 681                         if (!prb_dispatch_next_block(pkc, po))
 682                                 goto refresh_timer;
 683                         else
 684                                 goto out;
 685                 } else {
 686                         /* Case 1. Queue was frozen because user-space was
 687                          *         lagging behind.
 688                          */
 689                         if (prb_curr_blk_in_use(pkc, pbd)) {
 690                                 /*
 691                                  * Ok, user-space is still behind.
 692                                  * So just refresh the timer.
 693                                  */
 694                                 goto refresh_timer;
 695                         } else {
 696                                /* Case 2. queue was frozen,user-space caught up,
 697                                 * now the link went idle && the timer fired.
 698                                 * We don't have a block to close.So we open this
 699                                 * block and restart the timer.
 700                                 * opening a block thaws the queue,restarts timer
 701                                 * Thawing/timer-refresh is a side effect.
 702                                 */
 703                                 prb_open_block(pkc, pbd);
 704                                 goto out;
 705                         }
 706                 }
 707         }
 708
 709 refresh_timer:
 710         _prb_refresh_rx_retire_blk_timer(pkc);
 711
 712 out:
 713         spin_unlock(&po->sk.sk_receive_queue.lock);
 714 }
 715
 716 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 717                 struct tpacket_block_desc *pbd1, __u32 status)
 718 {
 719         /* Flush everything minus the block header */
 720
 721 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 722         u8 *start, *end;
 723
 724         start = (u8 *)pbd1;
 725
 726         /* Skip the block header(we know header WILL fit in 4K) */
 727         start += PAGE_SIZE;
 728
 729         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 730         for (; start < end; start += PAGE_SIZE)
 731                 flush_dcache_page(pgv_to_page(start));
 732
 733         smp_wmb();
 734 #endif
 735
 736         /* Now update the block status. */
 737
 738         BLOCK_STATUS(pbd1) = status;
 739
 740         /* Flush the block header */
 741
 742 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 743         start = (u8 *)pbd1;
 744         flush_dcache_page(pgv_to_page(start));
 745
 746         smp_wmb();
 747 #endif
 748 }
 749
 750 /*
 751  * Side effect:
 752  *
 753  * 1) flush the block
 754  * 2) Increment active_blk_num
 755  *
 756  * Note:We DONT refresh the timer on purpose.
 757  *      Because almost always the next block will be opened.
 758  */
 759 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 760                 struct tpacket_block_desc *pbd1,
 761                 struct packet_sock *po, unsigned int stat)
 762 {
 763         __u32 status = TP_STATUS_USER | stat;
 764
 765         struct tpacket3_hdr *last_pkt;
 766         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 767
 768         if (po->stats.tp_drops)
 769                 status |= TP_STATUS_LOSING;
 770
 771         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 772         last_pkt->tp_next_offset = 0;
 773
 774         /* Get the ts of the last pkt */
 775         if (BLOCK_NUM_PKTS(pbd1)) {
 776                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 777                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 778         } else {
 779                 /* Ok, we tmo'd - so get the current time */
 780                 struct timespec ts;
 781                 getnstimeofday(&ts);
 782                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 783                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 784         }
 785
 786         smp_wmb();
 787
 788         /* Flush the block */
 789         prb_flush_block(pkc1, pbd1, status);
 790
 791         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 792 }
 793
 794 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 795 {
 796         pkc->reset_pending_on_curr_blk = 0;
 797 }
 798
 799 /*
 800  * Side effect of opening a block:
 801  *
 802  * 1) prb_queue is thawed.
 803  * 2) retire_blk_timer is refreshed.
 804  *
 805  */
 806 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 807         struct tpacket_block_desc *pbd1)
 808 {
 809         struct timespec ts;
 810         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 811
 812         smp_rmb();
 813
 814         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd1))) {
 815
 816                 /* We could have just memset this but we will lose the
 817                  * flexibility of making the priv area sticky
 818                  */
 819                 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 820                 BLOCK_NUM_PKTS(pbd1) = 0;
 821                 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 822                 getnstimeofday(&ts);
 823                 h1->ts_first_pkt.ts_sec = ts.tv_sec;
 824                 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 825                 pkc1->pkblk_start = (char *)pbd1;
 826                 pkc1->nxt_offset = (char *)(pkc1->pkblk_start +
 827                 BLK_PLUS_PRIV(pkc1->blk_sizeof_priv));
 828                 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 829                 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 830                 pbd1->version = pkc1->version;
 831                 pkc1->prev = pkc1->nxt_offset;
 832                 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 833                 prb_thaw_queue(pkc1);
 834                 _prb_refresh_rx_retire_blk_timer(pkc1);
 835
 836                 smp_wmb();
 837
 838                 return;
 839         }
 840
 841         WARN(1, "ERROR block:%p is NOT FREE status:%d kactive_blk_num:%d\n",
 842                 pbd1, BLOCK_STATUS(pbd1), pkc1->kactive_blk_num);
 843         dump_stack();
 844         BUG();
 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         WARN(1, "ERROR-pbd[%d]:%p\n", pkc->kactive_blk_num, pbd);
 937         dump_stack();
 938         BUG();
 939 }
 940
 941 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
 942                                       struct tpacket_block_desc *pbd)
 943 {
 944         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 945 }
 946
 947 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 948 {
 949         return pkc->reset_pending_on_curr_blk;
 950 }
 951
 952 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 953 {
 954         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 955         atomic_dec(&pkc->blk_fill_in_prog);
 956 }
 957
 958 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 959                         struct tpacket3_hdr *ppd)
 960 {
 961         ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
 962 }
 963
 964 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 965                         struct tpacket3_hdr *ppd)
 966 {
 967         ppd->hv1.tp_rxhash = 0;
 968 }
 969
 970 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
 971                         struct tpacket3_hdr *ppd)
 972 {
 973         if (vlan_tx_tag_present(pkc->skb)) {
 974                 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
 975                 ppd->tp_status = TP_STATUS_VLAN_VALID;
 976         } else {
 977                 ppd->hv1.tp_vlan_tci = ppd->tp_status = 0;
 978         }
 979 }
 980
 981 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
 982                         struct tpacket3_hdr *ppd)
 983 {
 984         prb_fill_vlan_info(pkc, ppd);
 985
 986         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
 987                 prb_fill_rxhash(pkc, ppd);
 988         else
 989                 prb_clear_rxhash(pkc, ppd);
 990 }
 991
 992 static void prb_fill_curr_block(char *curr,
 993                                 struct tpacket_kbdq_core *pkc,
 994                                 struct tpacket_block_desc *pbd,
 995                                 unsigned int len)
 996 {
 997         struct tpacket3_hdr *ppd;
 998
 999         ppd  = (struct tpacket3_hdr *)curr;
1000         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1001         pkc->prev = curr;
1002         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1003         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1004         BLOCK_NUM_PKTS(pbd) += 1;
1005         atomic_inc(&pkc->blk_fill_in_prog);
1006         prb_run_all_ft_ops(pkc, ppd);
1007 }
1008
1009 /* Assumes caller has the sk->rx_queue.lock */
1010 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1011                                             struct sk_buff *skb,
1012                                                 int status,
1013                                             unsigned int len
1014                                             )
1015 {
1016         struct tpacket_kbdq_core *pkc;
1017         struct tpacket_block_desc *pbd;
1018         char *curr, *end;
1019
1020         pkc = GET_PBDQC_FROM_RB(((struct packet_ring_buffer *)&po->rx_ring));
1021         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1022
1023         /* Queue is frozen when user space is lagging behind */
1024         if (prb_queue_frozen(pkc)) {
1025                 /*
1026                  * Check if that last block which caused the queue to freeze,
1027                  * is still in_use by user-space.
1028                  */
1029                 if (prb_curr_blk_in_use(pkc, pbd)) {
1030                         /* Can't record this packet */
1031                         return NULL;
1032                 } else {
1033                         /*
1034                          * Ok, the block was released by user-space.
1035                          * Now let's open that block.
1036                          * opening a block also thaws the queue.
1037                          * Thawing is a side effect.
1038                          */
1039                         prb_open_block(pkc, pbd);
1040                 }
1041         }
1042
1043         smp_mb();
1044         curr = pkc->nxt_offset;
1045         pkc->skb = skb;
1046         end = (char *) ((char *)pbd + pkc->kblk_size);
1047
1048         /* first try the current block */
1049         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1050                 prb_fill_curr_block(curr, pkc, pbd, len);
1051                 return (void *)curr;
1052         }
1053
1054         /* Ok, close the current block */
1055         prb_retire_current_block(pkc, po, 0);
1056
1057         /* Now, try to dispatch the next block */
1058         curr = (char *)prb_dispatch_next_block(pkc, po);
1059         if (curr) {
1060                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1061                 prb_fill_curr_block(curr, pkc, pbd, len);
1062                 return (void *)curr;
1063         }
1064
1065         /*
1066          * No free blocks are available.user_space hasn't caught up yet.
1067          * Queue was just frozen and now this packet will get dropped.
1068          */
1069         return NULL;
1070 }
1071
1072 static void *packet_current_rx_frame(struct packet_sock *po,
1073                                             struct sk_buff *skb,
1074                                             int status, unsigned int len)
1075 {
1076         char *curr = NULL;
1077         switch (po->tp_version) {
1078         case TPACKET_V1:
1079         case TPACKET_V2:
1080                 curr = packet_lookup_frame(po, &po->rx_ring,
1081                                         po->rx_ring.head, status);
1082                 return curr;
1083         case TPACKET_V3:
1084                 return __packet_lookup_frame_in_block(po, skb, status, len);
1085         default:
1086                 WARN(1, "TPACKET version not supported\n");
1087                 BUG();
1088                 return 0;
1089         }
1090 }
1091
1092 static void *prb_lookup_block(struct packet_sock *po,
1093                                      struct packet_ring_buffer *rb,
1094                                      unsigned int previous,
1095                                      int status)
1096 {
1097         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1098         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, previous);
1099
1100         if (status != BLOCK_STATUS(pbd))
1101                 return NULL;
1102         return pbd;
1103 }
1104
1105 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1106 {
1107         unsigned int prev;
1108         if (rb->prb_bdqc.kactive_blk_num)
1109                 prev = rb->prb_bdqc.kactive_blk_num-1;
1110         else
1111                 prev = rb->prb_bdqc.knum_blocks-1;
1112         return prev;
1113 }
1114
1115 /* Assumes caller has held the rx_queue.lock */
1116 static void *__prb_previous_block(struct packet_sock *po,
1117                                          struct packet_ring_buffer *rb,
1118                                          int status)
1119 {
1120         unsigned int previous = prb_previous_blk_num(rb);
1121         return prb_lookup_block(po, rb, previous, status);
1122 }
1123
1124 static void *packet_previous_rx_frame(struct packet_sock *po,
1125                                              struct packet_ring_buffer *rb,
1126                                              int status)
1127 {
1128         if (po->tp_version <= TPACKET_V2)
1129                 return packet_previous_frame(po, rb, status);
1130
1131         return __prb_previous_block(po, rb, status);
1132 }
1133
1134 static void packet_increment_rx_head(struct packet_sock *po,
1135                                             struct packet_ring_buffer *rb)
1136 {
1137         switch (po->tp_version) {
1138         case TPACKET_V1:
1139         case TPACKET_V2:
1140                 return packet_increment_head(rb);
1141         case TPACKET_V3:
1142         default:
1143                 WARN(1, "TPACKET version not supported.\n");
1144                 BUG();
1145                 return;
1146         }
1147 }
1148
1149 static void *packet_previous_frame(struct packet_sock *po,
1150                 struct packet_ring_buffer *rb,
1151                 int status)
1152 {
1153         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1154         return packet_lookup_frame(po, rb, previous, status);
1155 }
1156
1157 static void packet_increment_head(struct packet_ring_buffer *buff)
1158 {
1159         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1160 }
1161
1162 static void packet_sock_destruct(struct sock *sk)
1163 {
1164         skb_queue_purge(&sk->sk_error_queue);
1165
1166         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1167         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1168
1169         if (!sock_flag(sk, SOCK_DEAD)) {
1170                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1171                 return;
1172         }
1173
1174         sk_refcnt_debug_dec(sk);
1175 }
1176
1177 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1178 {
1179         int x = atomic_read(&f->rr_cur) + 1;
1180
1181         if (x >= num)
1182                 x = 0;
1183
1184         return x;
1185 }
1186
1187 static struct sock *fanout_demux_hash(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1188 {
1189         u32 idx, hash = skb->rxhash;
1190
1191         idx = ((u64)hash * num) >> 32;
1192
1193         return f->arr[idx];
1194 }
1195
1196 static struct sock *fanout_demux_lb(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1197 {
1198         int cur, old;
1199
1200         cur = atomic_read(&f->rr_cur);
1201         while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1202                                      fanout_rr_next(f, num))) != cur)
1203                 cur = old;
1204         return f->arr[cur];
1205 }
1206
1207 static struct sock *fanout_demux_cpu(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1208 {
1209         unsigned int cpu = smp_processor_id();
1210
1211         return f->arr[cpu % num];
1212 }
1213
1214 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1215                              struct packet_type *pt, struct net_device *orig_dev)
1216 {
1217         struct packet_fanout *f = pt->af_packet_priv;
1218         unsigned int num = f->num_members;
1219         struct packet_sock *po;
1220         struct sock *sk;
1221
1222         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1223             !num) {
1224                 kfree_skb(skb);
1225                 return 0;
1226         }
1227
1228         switch (f->type) {
1229         case PACKET_FANOUT_HASH:
1230         default:
1231                 if (f->defrag) {
1232                         skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1233                         if (!skb)
1234                                 return 0;
1235                 }
1236                 skb_get_rxhash(skb);
1237                 sk = fanout_demux_hash(f, skb, num);
1238                 break;
1239         case PACKET_FANOUT_LB:
1240                 sk = fanout_demux_lb(f, skb, num);
1241                 break;
1242         case PACKET_FANOUT_CPU:
1243                 sk = fanout_demux_cpu(f, skb, num);
1244                 break;
1245         }
1246
1247         po = pkt_sk(sk);
1248
1249         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1250 }
1251
1252 static DEFINE_MUTEX(fanout_mutex);
1253 static LIST_HEAD(fanout_list);
1254
1255 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1256 {
1257         struct packet_fanout *f = po->fanout;
1258
1259         spin_lock(&f->lock);
1260         f->arr[f->num_members] = sk;
1261         smp_wmb();
1262         f->num_members++;
1263         spin_unlock(&f->lock);
1264 }
1265
1266 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1267 {
1268         struct packet_fanout *f = po->fanout;
1269         int i;
1270
1271         spin_lock(&f->lock);
1272         for (i = 0; i < f->num_members; i++) {
1273                 if (f->arr[i] == sk)
1274                         break;
1275         }
1276         BUG_ON(i >= f->num_members);
1277         f->arr[i] = f->arr[f->num_members - 1];
1278         f->num_members--;
1279         spin_unlock(&f->lock);
1280 }
1281
1282 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1283 {
1284         struct packet_sock *po = pkt_sk(sk);
1285         struct packet_fanout *f, *match;
1286         u8 type = type_flags & 0xff;
1287         u8 defrag = (type_flags & PACKET_FANOUT_FLAG_DEFRAG) ? 1 : 0;
1288         int err;
1289
1290         switch (type) {
1291         case PACKET_FANOUT_HASH:
1292         case PACKET_FANOUT_LB:
1293         case PACKET_FANOUT_CPU:
1294                 break;
1295         default:
1296                 return -EINVAL;
1297         }
1298
1299         if (!po->running)
1300                 return -EINVAL;
1301
1302         if (po->fanout)
1303                 return -EALREADY;
1304
1305         mutex_lock(&fanout_mutex);
1306         match = NULL;
1307         list_for_each_entry(f, &fanout_list, list) {
1308                 if (f->id == id &&
1309                     read_pnet(&f->net) == sock_net(sk)) {
1310                         match = f;
1311                         break;
1312                 }
1313         }
1314         err = -EINVAL;
1315         if (match && match->defrag != defrag)
1316                 goto out;
1317         if (!match) {
1318                 err = -ENOMEM;
1319                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1320                 if (!match)
1321                         goto out;
1322                 write_pnet(&match->net, sock_net(sk));
1323                 match->id = id;
1324                 match->type = type;
1325                 match->defrag = defrag;
1326                 atomic_set(&match->rr_cur, 0);
1327                 INIT_LIST_HEAD(&match->list);
1328                 spin_lock_init(&match->lock);
1329                 atomic_set(&match->sk_ref, 0);
1330                 match->prot_hook.type = po->prot_hook.type;
1331                 match->prot_hook.dev = po->prot_hook.dev;
1332                 match->prot_hook.func = packet_rcv_fanout;
1333                 match->prot_hook.af_packet_priv = match;
1334                 dev_add_pack(&match->prot_hook);
1335                 list_add(&match->list, &fanout_list);
1336         }
1337         err = -EINVAL;
1338         if (match->type == type &&
1339             match->prot_hook.type == po->prot_hook.type &&
1340             match->prot_hook.dev == po->prot_hook.dev) {
1341                 err = -ENOSPC;
1342                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1343                         __dev_remove_pack(&po->prot_hook);
1344                         po->fanout = match;
1345                         atomic_inc(&match->sk_ref);
1346                         __fanout_link(sk, po);
1347                         err = 0;
1348                 }
1349         }
1350 out:
1351         mutex_unlock(&fanout_mutex);
1352         return err;
1353 }
1354
1355 static void fanout_release(struct sock *sk)
1356 {
1357         struct packet_sock *po = pkt_sk(sk);
1358         struct packet_fanout *f;
1359
1360         f = po->fanout;
1361         if (!f)
1362                 return;
1363
1364         po->fanout = NULL;
1365
1366         mutex_lock(&fanout_mutex);
1367         if (atomic_dec_and_test(&f->sk_ref)) {
1368                 list_del(&f->list);
1369                 dev_remove_pack(&f->prot_hook);
1370                 kfree(f);
1371         }
1372         mutex_unlock(&fanout_mutex);
1373 }
1374
1375 static const struct proto_ops packet_ops;
1376
1377 static const struct proto_ops packet_ops_spkt;
1378
1379 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1380                            struct packet_type *pt, struct net_device *orig_dev)
1381 {
1382         struct sock *sk;
1383         struct sockaddr_pkt *spkt;
1384
1385         /*
1386          *      When we registered the protocol we saved the socket in the data
1387          *      field for just this event.
1388          */
1389
1390         sk = pt->af_packet_priv;
1391
1392         /*
1393          *      Yank back the headers [hope the device set this
1394          *      right or kerboom...]
1395          *
1396          *      Incoming packets have ll header pulled,
1397          *      push it back.
1398          *
1399          *      For outgoing ones skb->data == skb_mac_header(skb)
1400          *      so that this procedure is noop.
1401          */
1402
1403         if (skb->pkt_type == PACKET_LOOPBACK)
1404                 goto out;
1405
1406         if (!net_eq(dev_net(dev), sock_net(sk)))
1407                 goto out;
1408
1409         skb = skb_share_check(skb, GFP_ATOMIC);
1410         if (skb == NULL)
1411                 goto oom;
1412
1413         /* drop any routing info */
1414         skb_dst_drop(skb);
1415
1416         /* drop conntrack reference */
1417         nf_reset(skb);
1418
1419         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1420
1421         skb_push(skb, skb->data - skb_mac_header(skb));
1422
1423         /*
1424          *      The SOCK_PACKET socket receives _all_ frames.
1425          */
1426
1427         spkt->spkt_family = dev->type;
1428         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1429         spkt->spkt_protocol = skb->protocol;
1430
1431         /*
1432          *      Charge the memory to the socket. This is done specifically
1433          *      to prevent sockets using all the memory up.
1434          */
1435
1436         if (sock_queue_rcv_skb(sk, skb) == 0)
1437                 return 0;
1438
1439 out:
1440         kfree_skb(skb);
1441 oom:
1442         return 0;
1443 }
1444
1445
1446 /*
1447  *      Output a raw packet to a device layer. This bypasses all the other
1448  *      protocol layers and you must therefore supply it with a complete frame
1449  */
1450
1451 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1452                                struct msghdr *msg, size_t len)
1453 {
1454         struct sock *sk = sock->sk;
1455         struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1456         struct sk_buff *skb = NULL;
1457         struct net_device *dev;
1458         __be16 proto = 0;
1459         int err;
1460
1461         /*
1462          *      Get and verify the address.
1463          */
1464
1465         if (saddr) {
1466                 if (msg->msg_namelen < sizeof(struct sockaddr))
1467                         return -EINVAL;
1468                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1469                         proto = saddr->spkt_protocol;
1470         } else
1471                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1472
1473         /*
1474          *      Find the device first to size check it
1475          */
1476
1477         saddr->spkt_device[13] = 0;
1478 retry:
1479         rcu_read_lock();
1480         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1481         err = -ENODEV;
1482         if (dev == NULL)
1483                 goto out_unlock;
1484
1485         err = -ENETDOWN;
1486         if (!(dev->flags & IFF_UP))
1487                 goto out_unlock;
1488
1489         /*
1490          * You may not queue a frame bigger than the mtu. This is the lowest level
1491          * raw protocol and you must do your own fragmentation at this level.
1492          */
1493
1494         err = -EMSGSIZE;
1495         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN)
1496                 goto out_unlock;
1497
1498         if (!skb) {
1499                 size_t reserved = LL_RESERVED_SPACE(dev);
1500                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1501
1502                 rcu_read_unlock();
1503                 skb = sock_wmalloc(sk, len + reserved, 0, GFP_KERNEL);
1504                 if (skb == NULL)
1505                         return -ENOBUFS;
1506                 /* FIXME: Save some space for broken drivers that write a hard
1507                  * header at transmission time by themselves. PPP is the notable
1508                  * one here. This should really be fixed at the driver level.
1509                  */
1510                 skb_reserve(skb, reserved);
1511                 skb_reset_network_header(skb);
1512
1513                 /* Try to align data part correctly */
1514                 if (hhlen) {
1515                         skb->data -= hhlen;
1516                         skb->tail -= hhlen;
1517                         if (len < hhlen)
1518                                 skb_reset_network_header(skb);
1519                 }
1520                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1521                 if (err)
1522                         goto out_free;
1523                 goto retry;
1524         }
1525
1526         if (len > (dev->mtu + dev->hard_header_len)) {
1527                 /* Earlier code assumed this would be a VLAN pkt,
1528                  * double-check this now that we have the actual
1529                  * packet in hand.
1530                  */
1531                 struct ethhdr *ehdr;
1532                 skb_reset_mac_header(skb);
1533                 ehdr = eth_hdr(skb);
1534                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1535                         err = -EMSGSIZE;
1536                         goto out_unlock;
1537                 }
1538         }
1539
1540         skb->protocol = proto;
1541         skb->dev = dev;
1542         skb->priority = sk->sk_priority;
1543         skb->mark = sk->sk_mark;
1544         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1545         if (err < 0)
1546                 goto out_unlock;
1547
1548         dev_queue_xmit(skb);
1549         rcu_read_unlock();
1550         return len;
1551
1552 out_unlock:
1553         rcu_read_unlock();
1554 out_free:
1555         kfree_skb(skb);
1556         return err;
1557 }
1558
1559 static unsigned int run_filter(const struct sk_buff *skb,
1560                                       const struct sock *sk,
1561                                       unsigned int res)
1562 {
1563         struct sk_filter *filter;
1564
1565         rcu_read_lock();
1566         filter = rcu_dereference(sk->sk_filter);
1567         if (filter != NULL)
1568                 res = SK_RUN_FILTER(filter, skb);
1569         rcu_read_unlock();
1570
1571         return res;
1572 }
1573
1574 /*
1575  * This function makes lazy skb cloning in hope that most of packets
1576  * are discarded by BPF.
1577  *
1578  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1579  * and skb->cb are mangled. It works because (and until) packets
1580  * falling here are owned by current CPU. Output packets are cloned
1581  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1582  * sequencially, so that if we return skb to original state on exit,
1583  * we will not harm anyone.
1584  */
1585
1586 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1587                       struct packet_type *pt, struct net_device *orig_dev)
1588 {
1589         struct sock *sk;
1590         struct sockaddr_ll *sll;
1591         struct packet_sock *po;
1592         u8 *skb_head = skb->data;
1593         int skb_len = skb->len;
1594         unsigned int snaplen, res;
1595
1596         if (skb->pkt_type == PACKET_LOOPBACK)
1597                 goto drop;
1598
1599         sk = pt->af_packet_priv;
1600         po = pkt_sk(sk);
1601
1602         if (!net_eq(dev_net(dev), sock_net(sk)))
1603                 goto drop;
1604
1605         skb->dev = dev;
1606
1607         if (dev->header_ops) {
1608                 /* The device has an explicit notion of ll header,
1609                  * exported to higher levels.
1610                  *
1611                  * Otherwise, the device hides details of its frame
1612                  * structure, so that corresponding packet head is
1613                  * never delivered to user.
1614                  */
1615                 if (sk->sk_type != SOCK_DGRAM)
1616                         skb_push(skb, skb->data - skb_mac_header(skb));
1617                 else if (skb->pkt_type == PACKET_OUTGOING) {
1618                         /* Special case: outgoing packets have ll header at head */
1619                         skb_pull(skb, skb_network_offset(skb));
1620                 }
1621         }
1622
1623         snaplen = skb->len;
1624
1625         res = run_filter(skb, sk, snaplen);
1626         if (!res)
1627                 goto drop_n_restore;
1628         if (snaplen > res)
1629                 snaplen = res;
1630
1631         if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
1632             (unsigned)sk->sk_rcvbuf)
1633                 goto drop_n_acct;
1634
1635         if (skb_shared(skb)) {
1636                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1637                 if (nskb == NULL)
1638                         goto drop_n_acct;
1639
1640                 if (skb_head != skb->data) {
1641                         skb->data = skb_head;
1642                         skb->len = skb_len;
1643                 }
1644                 kfree_skb(skb);
1645                 skb = nskb;
1646         }
1647
1648         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1649                      sizeof(skb->cb));
1650
1651         sll = &PACKET_SKB_CB(skb)->sa.ll;
1652         sll->sll_family = AF_PACKET;
1653         sll->sll_hatype = dev->type;
1654         sll->sll_protocol = skb->protocol;
1655         sll->sll_pkttype = skb->pkt_type;
1656         if (unlikely(po->origdev))
1657                 sll->sll_ifindex = orig_dev->ifindex;
1658         else
1659                 sll->sll_ifindex = dev->ifindex;
1660
1661         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1662
1663         PACKET_SKB_CB(skb)->origlen = skb->len;
1664
1665         if (pskb_trim(skb, snaplen))
1666                 goto drop_n_acct;
1667
1668         skb_set_owner_r(skb, sk);
1669         skb->dev = NULL;
1670         skb_dst_drop(skb);
1671
1672         /* drop conntrack reference */
1673         nf_reset(skb);
1674
1675         spin_lock(&sk->sk_receive_queue.lock);
1676         po->stats.tp_packets++;
1677         skb->dropcount = atomic_read(&sk->sk_drops);
1678         __skb_queue_tail(&sk->sk_receive_queue, skb);
1679         spin_unlock(&sk->sk_receive_queue.lock);
1680         sk->sk_data_ready(sk, skb->len);
1681         return 0;
1682
1683 drop_n_acct:
1684         spin_lock(&sk->sk_receive_queue.lock);
1685         po->stats.tp_drops++;
1686         atomic_inc(&sk->sk_drops);
1687         spin_unlock(&sk->sk_receive_queue.lock);
1688
1689 drop_n_restore:
1690         if (skb_head != skb->data && skb_shared(skb)) {
1691                 skb->data = skb_head;
1692                 skb->len = skb_len;
1693         }
1694 drop:
1695         consume_skb(skb);
1696         return 0;
1697 }
1698
1699 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1700                        struct packet_type *pt, struct net_device *orig_dev)
1701 {
1702         struct sock *sk;
1703         struct packet_sock *po;
1704         struct sockaddr_ll *sll;
1705         union {
1706                 struct tpacket_hdr *h1;
1707                 struct tpacket2_hdr *h2;
1708                 struct tpacket3_hdr *h3;
1709                 void *raw;
1710         } h;
1711         u8 *skb_head = skb->data;
1712         int skb_len = skb->len;
1713         unsigned int snaplen, res;
1714         unsigned long status = TP_STATUS_USER;
1715         unsigned short macoff, netoff, hdrlen;
1716         struct sk_buff *copy_skb = NULL;
1717         struct timeval tv;
1718         struct timespec ts;
1719         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1720
1721         if (skb->pkt_type == PACKET_LOOPBACK)
1722                 goto drop;
1723
1724         sk = pt->af_packet_priv;
1725         po = pkt_sk(sk);
1726
1727         if (!net_eq(dev_net(dev), sock_net(sk)))
1728                 goto drop;
1729
1730         if (dev->header_ops) {
1731                 if (sk->sk_type != SOCK_DGRAM)
1732                         skb_push(skb, skb->data - skb_mac_header(skb));
1733                 else if (skb->pkt_type == PACKET_OUTGOING) {
1734                         /* Special case: outgoing packets have ll header at head */
1735                         skb_pull(skb, skb_network_offset(skb));
1736                 }
1737         }
1738
1739         if (skb->ip_summed == CHECKSUM_PARTIAL)
1740                 status |= TP_STATUS_CSUMNOTREADY;
1741
1742         snaplen = skb->len;
1743
1744         res = run_filter(skb, sk, snaplen);
1745         if (!res)
1746                 goto drop_n_restore;
1747         if (snaplen > res)
1748                 snaplen = res;
1749
1750         if (sk->sk_type == SOCK_DGRAM) {
1751                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1752                                   po->tp_reserve;
1753         } else {
1754                 unsigned maclen = skb_network_offset(skb);
1755                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1756                                        (maclen < 16 ? 16 : maclen)) +
1757                         po->tp_reserve;
1758                 macoff = netoff - maclen;
1759         }
1760         if (po->tp_version <= TPACKET_V2) {
1761                 if (macoff + snaplen > po->rx_ring.frame_size) {
1762                         if (po->copy_thresh &&
1763                                 atomic_read(&sk->sk_rmem_alloc) + skb->truesize
1764                                 < (unsigned)sk->sk_rcvbuf) {
1765                                 if (skb_shared(skb)) {
1766                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1767                                 } else {
1768                                         copy_skb = skb_get(skb);
1769                                         skb_head = skb->data;
1770                                 }
1771                                 if (copy_skb)
1772                                         skb_set_owner_r(copy_skb, sk);
1773                         }
1774                         snaplen = po->rx_ring.frame_size - macoff;
1775                         if ((int)snaplen < 0)
1776                                 snaplen = 0;
1777                 }
1778         }
1779         spin_lock(&sk->sk_receive_queue.lock);
1780         h.raw = packet_current_rx_frame(po, skb,
1781                                         TP_STATUS_KERNEL, (macoff+snaplen));
1782         if (!h.raw)
1783                 goto ring_is_full;
1784         if (po->tp_version <= TPACKET_V2) {
1785                 packet_increment_rx_head(po, &po->rx_ring);
1786         /*
1787          * LOSING will be reported till you read the stats,
1788          * because it's COR - Clear On Read.
1789          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1790          * at packet level.
1791          */
1792                 if (po->stats.tp_drops)
1793                         status |= TP_STATUS_LOSING;
1794         }
1795         po->stats.tp_packets++;
1796         if (copy_skb) {
1797                 status |= TP_STATUS_COPY;
1798                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1799         }
1800         spin_unlock(&sk->sk_receive_queue.lock);
1801
1802         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1803
1804         switch (po->tp_version) {
1805         case TPACKET_V1:
1806                 h.h1->tp_len = skb->len;
1807                 h.h1->tp_snaplen = snaplen;
1808                 h.h1->tp_mac = macoff;
1809                 h.h1->tp_net = netoff;
1810                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1811                                 && shhwtstamps->syststamp.tv64)
1812                         tv = ktime_to_timeval(shhwtstamps->syststamp);
1813                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1814                                 && shhwtstamps->hwtstamp.tv64)
1815                         tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1816                 else if (skb->tstamp.tv64)
1817                         tv = ktime_to_timeval(skb->tstamp);
1818                 else
1819                         do_gettimeofday(&tv);
1820                 h.h1->tp_sec = tv.tv_sec;
1821                 h.h1->tp_usec = tv.tv_usec;
1822                 hdrlen = sizeof(*h.h1);
1823                 break;
1824         case TPACKET_V2:
1825                 h.h2->tp_len = skb->len;
1826                 h.h2->tp_snaplen = snaplen;
1827                 h.h2->tp_mac = macoff;
1828                 h.h2->tp_net = netoff;
1829                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1830                                 && shhwtstamps->syststamp.tv64)
1831                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1832                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1833                                 && shhwtstamps->hwtstamp.tv64)
1834                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1835                 else if (skb->tstamp.tv64)
1836                         ts = ktime_to_timespec(skb->tstamp);
1837                 else
1838                         getnstimeofday(&ts);
1839                 h.h2->tp_sec = ts.tv_sec;
1840                 h.h2->tp_nsec = ts.tv_nsec;
1841                 if (vlan_tx_tag_present(skb)) {
1842                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1843                         status |= TP_STATUS_VLAN_VALID;
1844                 } else {
1845                         h.h2->tp_vlan_tci = 0;
1846                 }
1847                 h.h2->tp_padding = 0;
1848                 hdrlen = sizeof(*h.h2);
1849                 break;
1850         case TPACKET_V3:
1851                 /* tp_nxt_offset,vlan are already populated above.
1852                  * So DONT clear those fields here
1853                  */
1854                 h.h3->tp_status |= status;
1855                 h.h3->tp_len = skb->len;
1856                 h.h3->tp_snaplen = snaplen;
1857                 h.h3->tp_mac = macoff;
1858                 h.h3->tp_net = netoff;
1859                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1860                                 && shhwtstamps->syststamp.tv64)
1861                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1862                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1863                                 && shhwtstamps->hwtstamp.tv64)
1864                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1865                 else if (skb->tstamp.tv64)
1866                         ts = ktime_to_timespec(skb->tstamp);
1867                 else
1868                         getnstimeofday(&ts);
1869                 h.h3->tp_sec  = ts.tv_sec;
1870                 h.h3->tp_nsec = ts.tv_nsec;
1871                 hdrlen = sizeof(*h.h3);
1872                 break;
1873         default:
1874                 BUG();
1875         }
1876
1877         sll = h.raw + TPACKET_ALIGN(hdrlen);
1878         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1879         sll->sll_family = AF_PACKET;
1880         sll->sll_hatype = dev->type;
1881         sll->sll_protocol = skb->protocol;
1882         sll->sll_pkttype = skb->pkt_type;
1883         if (unlikely(po->origdev))
1884                 sll->sll_ifindex = orig_dev->ifindex;
1885         else
1886                 sll->sll_ifindex = dev->ifindex;
1887
1888         smp_mb();
1889 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1890         {
1891                 u8 *start, *end;
1892
1893                 if (po->tp_version <= TPACKET_V2) {
1894                         end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1895                                 + macoff + snaplen);
1896                         for (start = h.raw; start < end; start += PAGE_SIZE)
1897                                 flush_dcache_page(pgv_to_page(start));
1898                 }
1899                 smp_wmb();
1900         }
1901 #endif
1902         if (po->tp_version <= TPACKET_V2)
1903                 __packet_set_status(po, h.raw, status);
1904         else
1905                 prb_clear_blk_fill_status(&po->rx_ring);
1906
1907         sk->sk_data_ready(sk, 0);
1908
1909 drop_n_restore:
1910         if (skb_head != skb->data && skb_shared(skb)) {
1911                 skb->data = skb_head;
1912                 skb->len = skb_len;
1913         }
1914 drop:
1915         kfree_skb(skb);
1916         return 0;
1917
1918 ring_is_full:
1919         po->stats.tp_drops++;
1920         spin_unlock(&sk->sk_receive_queue.lock);
1921
1922         sk->sk_data_ready(sk, 0);
1923         kfree_skb(copy_skb);
1924         goto drop_n_restore;
1925 }
1926
1927 static void tpacket_destruct_skb(struct sk_buff *skb)
1928 {
1929         struct packet_sock *po = pkt_sk(skb->sk);
1930         void *ph;
1931
1932         if (likely(po->tx_ring.pg_vec)) {
1933                 ph = skb_shinfo(skb)->destructor_arg;
1934                 BUG_ON(__packet_get_status(po, ph) != TP_STATUS_SENDING);
1935                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1936                 atomic_dec(&po->tx_ring.pending);
1937                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1938         }
1939
1940         sock_wfree(skb);
1941 }
1942
1943 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1944                 void *frame, struct net_device *dev, int size_max,
1945                 __be16 proto, unsigned char *addr)
1946 {
1947         union {
1948                 struct tpacket_hdr *h1;
1949                 struct tpacket2_hdr *h2;
1950                 void *raw;
1951         } ph;
1952         int to_write, offset, len, tp_len, nr_frags, len_max;
1953         struct socket *sock = po->sk.sk_socket;
1954         struct page *page;
1955         void *data;
1956         int err;
1957
1958         ph.raw = frame;
1959
1960         skb->protocol = proto;
1961         skb->dev = dev;
1962         skb->priority = po->sk.sk_priority;
1963         skb->mark = po->sk.sk_mark;
1964         skb_shinfo(skb)->destructor_arg = ph.raw;
1965
1966         switch (po->tp_version) {
1967         case TPACKET_V2:
1968                 tp_len = ph.h2->tp_len;
1969                 break;
1970         default:
1971                 tp_len = ph.h1->tp_len;
1972                 break;
1973         }
1974         if (unlikely(tp_len > size_max)) {
1975                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1976                 return -EMSGSIZE;
1977         }
1978
1979         skb_reserve(skb, LL_RESERVED_SPACE(dev));
1980         skb_reset_network_header(skb);
1981
1982         data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1983         to_write = tp_len;
1984
1985         if (sock->type == SOCK_DGRAM) {
1986                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
1987                                 NULL, tp_len);
1988                 if (unlikely(err < 0))
1989                         return -EINVAL;
1990         } else if (dev->hard_header_len) {
1991                 /* net device doesn't like empty head */
1992                 if (unlikely(tp_len <= dev->hard_header_len)) {
1993                         pr_err("packet size is too short (%d < %d)\n",
1994                                tp_len, dev->hard_header_len);
1995                         return -EINVAL;
1996                 }
1997
1998                 skb_push(skb, dev->hard_header_len);
1999                 err = skb_store_bits(skb, 0, data,
2000                                 dev->hard_header_len);
2001                 if (unlikely(err))
2002                         return err;
2003
2004                 data += dev->hard_header_len;
2005                 to_write -= dev->hard_header_len;
2006         }
2007
2008         err = -EFAULT;
2009         offset = offset_in_page(data);
2010         len_max = PAGE_SIZE - offset;
2011         len = ((to_write > len_max) ? len_max : to_write);
2012
2013         skb->data_len = to_write;
2014         skb->len += to_write;
2015         skb->truesize += to_write;
2016         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2017
2018         while (likely(to_write)) {
2019                 nr_frags = skb_shinfo(skb)->nr_frags;
2020
2021                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2022                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2023                                MAX_SKB_FRAGS);
2024                         return -EFAULT;
2025                 }
2026
2027                 page = pgv_to_page(data);
2028                 data += len;
2029                 flush_dcache_page(page);
2030                 get_page(page);
2031                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2032                 to_write -= len;
2033                 offset = 0;
2034                 len_max = PAGE_SIZE;
2035                 len = ((to_write > len_max) ? len_max : to_write);
2036         }
2037
2038         return tp_len;
2039 }
2040
2041 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2042 {
2043         struct sk_buff *skb;
2044         struct net_device *dev;
2045         __be16 proto;
2046         bool need_rls_dev = false;
2047         int err, reserve = 0;
2048         void *ph;
2049         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2050         int tp_len, size_max;
2051         unsigned char *addr;
2052         int len_sum = 0;
2053         int status = 0;
2054
2055         mutex_lock(&po->pg_vec_lock);
2056
2057         err = -EBUSY;
2058         if (saddr == NULL) {
2059                 dev = po->prot_hook.dev;
2060                 proto   = po->num;
2061                 addr    = NULL;
2062         } else {
2063                 err = -EINVAL;
2064                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2065                         goto out;
2066                 if (msg->msg_namelen < (saddr->sll_halen
2067                                         + offsetof(struct sockaddr_ll,
2068                                                 sll_addr)))
2069                         goto out;
2070                 proto   = saddr->sll_protocol;
2071                 addr    = saddr->sll_addr;
2072                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2073                 need_rls_dev = true;
2074         }
2075
2076         err = -ENXIO;
2077         if (unlikely(dev == NULL))
2078                 goto out;
2079
2080         reserve = dev->hard_header_len;
2081
2082         err = -ENETDOWN;
2083         if (unlikely(!(dev->flags & IFF_UP)))
2084                 goto out_put;
2085
2086         size_max = po->tx_ring.frame_size
2087                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2088
2089         if (size_max > dev->mtu + reserve)
2090                 size_max = dev->mtu + reserve;
2091
2092         do {
2093                 ph = packet_current_frame(po, &po->tx_ring,
2094                                 TP_STATUS_SEND_REQUEST);
2095
2096                 if (unlikely(ph == NULL)) {
2097                         schedule();
2098                         continue;
2099                 }
2100
2101                 status = TP_STATUS_SEND_REQUEST;
2102                 skb = sock_alloc_send_skb(&po->sk,
2103                                 LL_ALLOCATED_SPACE(dev)
2104                                 + sizeof(struct sockaddr_ll),
2105                                 0, &err);
2106
2107                 if (unlikely(skb == NULL))
2108                         goto out_status;
2109
2110                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2111                                 addr);
2112
2113                 if (unlikely(tp_len < 0)) {
2114                         if (po->tp_loss) {
2115                                 __packet_set_status(po, ph,
2116                                                 TP_STATUS_AVAILABLE);
2117                                 packet_increment_head(&po->tx_ring);
2118                                 kfree_skb(skb);
2119                                 continue;
2120                         } else {
2121                                 status = TP_STATUS_WRONG_FORMAT;
2122                                 err = tp_len;
2123                                 goto out_status;
2124                         }
2125                 }
2126
2127                 skb->destructor = tpacket_destruct_skb;
2128                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2129                 atomic_inc(&po->tx_ring.pending);
2130
2131                 status = TP_STATUS_SEND_REQUEST;
2132                 err = dev_queue_xmit(skb);
2133                 if (unlikely(err > 0)) {
2134                         err = net_xmit_errno(err);
2135                         if (err && __packet_get_status(po, ph) ==
2136                                    TP_STATUS_AVAILABLE) {
2137                                 /* skb was destructed already */
2138                                 skb = NULL;
2139                                 goto out_status;
2140                         }
2141                         /*
2142                          * skb was dropped but not destructed yet;
2143                          * let's treat it like congestion or err < 0
2144                          */
2145                         err = 0;
2146                 }
2147                 packet_increment_head(&po->tx_ring);
2148                 len_sum += tp_len;
2149         } while (likely((ph != NULL) ||
2150                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2151                          (atomic_read(&po->tx_ring.pending))))
2152                 );
2153
2154         err = len_sum;
2155         goto out_put;
2156
2157 out_status:
2158         __packet_set_status(po, ph, status);
2159         kfree_skb(skb);
2160 out_put:
2161         if (need_rls_dev)
2162                 dev_put(dev);
2163 out:
2164         mutex_unlock(&po->pg_vec_lock);
2165         return err;
2166 }
2167
2168 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2169                                         size_t reserve, size_t len,
2170                                         size_t linear, int noblock,
2171                                         int *err)
2172 {
2173         struct sk_buff *skb;
2174
2175         /* Under a page?  Don't bother with paged skb. */
2176         if (prepad + len < PAGE_SIZE || !linear)
2177                 linear = len;
2178
2179         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2180                                    err);
2181         if (!skb)
2182                 return NULL;
2183
2184         skb_reserve(skb, reserve);
2185         skb_put(skb, linear);
2186         skb->data_len = len - linear;
2187         skb->len += len - linear;
2188
2189         return skb;
2190 }
2191
2192 static int packet_snd(struct socket *sock,
2193                           struct msghdr *msg, size_t len)
2194 {
2195         struct sock *sk = sock->sk;
2196         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2197         struct sk_buff *skb;
2198         struct net_device *dev;
2199         __be16 proto;
2200         bool need_rls_dev = false;
2201         unsigned char *addr;
2202         int err, reserve = 0;
2203         struct virtio_net_hdr vnet_hdr = { 0 };
2204         int offset = 0;
2205         int vnet_hdr_len;
2206         struct packet_sock *po = pkt_sk(sk);
2207         unsigned short gso_type = 0;
2208
2209         /*
2210          *      Get and verify the address.
2211          */
2212
2213         if (saddr == NULL) {
2214                 dev = po->prot_hook.dev;
2215                 proto   = po->num;
2216                 addr    = NULL;
2217         } else {
2218                 err = -EINVAL;
2219                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2220                         goto out;
2221                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2222                         goto out;
2223                 proto   = saddr->sll_protocol;
2224                 addr    = saddr->sll_addr;
2225                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2226                 need_rls_dev = true;
2227         }
2228
2229         err = -ENXIO;
2230         if (dev == NULL)
2231                 goto out_unlock;
2232         if (sock->type == SOCK_RAW)
2233                 reserve = dev->hard_header_len;
2234
2235         err = -ENETDOWN;
2236         if (!(dev->flags & IFF_UP))
2237                 goto out_unlock;
2238
2239         if (po->has_vnet_hdr) {
2240                 vnet_hdr_len = sizeof(vnet_hdr);
2241
2242                 err = -EINVAL;
2243                 if (len < vnet_hdr_len)
2244                         goto out_unlock;
2245
2246                 len -= vnet_hdr_len;
2247
2248                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2249                                        vnet_hdr_len);
2250                 if (err < 0)
2251                         goto out_unlock;
2252
2253                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2254                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2255                       vnet_hdr.hdr_len))
2256                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
2257                                                  vnet_hdr.csum_offset + 2;
2258
2259                 err = -EINVAL;
2260                 if (vnet_hdr.hdr_len > len)
2261                         goto out_unlock;
2262
2263                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2264                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2265                         case VIRTIO_NET_HDR_GSO_TCPV4:
2266                                 gso_type = SKB_GSO_TCPV4;
2267                                 break;
2268                         case VIRTIO_NET_HDR_GSO_TCPV6:
2269                                 gso_type = SKB_GSO_TCPV6;
2270                                 break;
2271                         case VIRTIO_NET_HDR_GSO_UDP:
2272                                 gso_type = SKB_GSO_UDP;
2273                                 break;
2274                         default:
2275                                 goto out_unlock;
2276                         }
2277
2278                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2279                                 gso_type |= SKB_GSO_TCP_ECN;
2280
2281                         if (vnet_hdr.gso_size == 0)
2282                                 goto out_unlock;
2283
2284                 }
2285         }
2286
2287         err = -EMSGSIZE;
2288         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN))
2289                 goto out_unlock;
2290
2291         err = -ENOBUFS;
2292         skb = packet_alloc_skb(sk, LL_ALLOCATED_SPACE(dev),
2293                                LL_RESERVED_SPACE(dev), len, vnet_hdr.hdr_len,
2294                                msg->msg_flags & MSG_DONTWAIT, &err);
2295         if (skb == NULL)
2296                 goto out_unlock;
2297
2298         skb_set_network_header(skb, reserve);
2299
2300         err = -EINVAL;
2301         if (sock->type == SOCK_DGRAM &&
2302             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2303                 goto out_free;
2304
2305         /* Returns -EFAULT on error */
2306         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2307         if (err)
2308                 goto out_free;
2309         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2310         if (err < 0)
2311                 goto out_free;
2312
2313         if (!gso_type && (len > dev->mtu + reserve)) {
2314                 /* Earlier code assumed this would be a VLAN pkt,
2315                  * double-check this now that we have the actual
2316                  * packet in hand.
2317                  */
2318                 struct ethhdr *ehdr;
2319                 skb_reset_mac_header(skb);
2320                 ehdr = eth_hdr(skb);
2321                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2322                         err = -EMSGSIZE;
2323                         goto out_free;
2324                 }
2325         }
2326
2327         skb->protocol = proto;
2328         skb->dev = dev;
2329         skb->priority = sk->sk_priority;
2330         skb->mark = sk->sk_mark;
2331
2332         if (po->has_vnet_hdr) {
2333                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2334                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2335                                                   vnet_hdr.csum_offset)) {
2336                                 err = -EINVAL;
2337                                 goto out_free;
2338                         }
2339                 }
2340
2341                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2342                 skb_shinfo(skb)->gso_type = gso_type;
2343
2344                 /* Header must be checked, and gso_segs computed. */
2345                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2346                 skb_shinfo(skb)->gso_segs = 0;
2347
2348                 len += vnet_hdr_len;
2349         }
2350
2351         /*
2352          *      Now send it
2353          */
2354
2355         err = dev_queue_xmit(skb);
2356         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2357                 goto out_unlock;
2358
2359         if (need_rls_dev)
2360                 dev_put(dev);
2361
2362         return len;
2363
2364 out_free:
2365         kfree_skb(skb);
2366 out_unlock:
2367         if (dev && need_rls_dev)
2368                 dev_put(dev);
2369 out:
2370         return err;
2371 }
2372
2373 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2374                 struct msghdr *msg, size_t len)
2375 {
2376         struct sock *sk = sock->sk;
2377         struct packet_sock *po = pkt_sk(sk);
2378         if (po->tx_ring.pg_vec)
2379                 return tpacket_snd(po, msg);
2380         else
2381                 return packet_snd(sock, msg, len);
2382 }
2383
2384 /*
2385  *      Close a PACKET socket. This is fairly simple. We immediately go
2386  *      to 'closed' state and remove our protocol entry in the device list.
2387  */
2388
2389 static int packet_release(struct socket *sock)
2390 {
2391         struct sock *sk = sock->sk;
2392         struct packet_sock *po;
2393         struct net *net;
2394         union tpacket_req_u req_u;
2395
2396         if (!sk)
2397                 return 0;
2398
2399         net = sock_net(sk);
2400         po = pkt_sk(sk);
2401
2402         spin_lock_bh(&net->packet.sklist_lock);
2403         sk_del_node_init_rcu(sk);
2404         sock_prot_inuse_add(net, sk->sk_prot, -1);
2405         spin_unlock_bh(&net->packet.sklist_lock);
2406
2407         spin_lock(&po->bind_lock);
2408         unregister_prot_hook(sk, false);
2409         if (po->prot_hook.dev) {
2410                 dev_put(po->prot_hook.dev);
2411                 po->prot_hook.dev = NULL;
2412         }
2413         spin_unlock(&po->bind_lock);
2414
2415         packet_flush_mclist(sk);
2416
2417         memset(&req_u, 0, sizeof(req_u));
2418
2419         if (po->rx_ring.pg_vec)
2420                 packet_set_ring(sk, &req_u, 1, 0);
2421
2422         if (po->tx_ring.pg_vec)
2423                 packet_set_ring(sk, &req_u, 1, 1);
2424
2425         fanout_release(sk);
2426
2427         synchronize_net();
2428         /*
2429          *      Now the socket is dead. No more input will appear.
2430          */
2431         sock_orphan(sk);
2432         sock->sk = NULL;
2433
2434         /* Purge queues */
2435
2436         skb_queue_purge(&sk->sk_receive_queue);
2437         sk_refcnt_debug_release(sk);
2438
2439         sock_put(sk);
2440         return 0;
2441 }
2442
2443 /*
2444  *      Attach a packet hook.
2445  */
2446
2447 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2448 {
2449         struct packet_sock *po = pkt_sk(sk);
2450
2451         if (po->fanout)
2452                 return -EINVAL;
2453
2454         lock_sock(sk);
2455
2456         spin_lock(&po->bind_lock);
2457         unregister_prot_hook(sk, true);
2458         po->num = protocol;
2459         po->prot_hook.type = protocol;
2460         if (po->prot_hook.dev)
2461                 dev_put(po->prot_hook.dev);
2462         po->prot_hook.dev = dev;
2463
2464         po->ifindex = dev ? dev->ifindex : 0;
2465
2466         if (protocol == 0)
2467                 goto out_unlock;
2468
2469         if (!dev || (dev->flags & IFF_UP)) {
2470                 register_prot_hook(sk);
2471         } else {
2472                 sk->sk_err = ENETDOWN;
2473                 if (!sock_flag(sk, SOCK_DEAD))
2474                         sk->sk_error_report(sk);
2475         }
2476
2477 out_unlock:
2478         spin_unlock(&po->bind_lock);
2479         release_sock(sk);
2480         return 0;
2481 }
2482
2483 /*
2484  *      Bind a packet socket to a device
2485  */
2486
2487 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2488                             int addr_len)
2489 {
2490         struct sock *sk = sock->sk;
2491         char name[15];
2492         struct net_device *dev;
2493         int err = -ENODEV;
2494
2495         /*
2496          *      Check legality
2497          */
2498
2499         if (addr_len != sizeof(struct sockaddr))
2500                 return -EINVAL;
2501         strlcpy(name, uaddr->sa_data, sizeof(name));
2502
2503         dev = dev_get_by_name(sock_net(sk), name);
2504         if (dev)
2505                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2506         return err;
2507 }
2508
2509 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2510 {
2511         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2512         struct sock *sk = sock->sk;
2513         struct net_device *dev = NULL;
2514         int err;
2515
2516
2517         /*
2518          *      Check legality
2519          */
2520
2521         if (addr_len < sizeof(struct sockaddr_ll))
2522                 return -EINVAL;
2523         if (sll->sll_family != AF_PACKET)
2524                 return -EINVAL;
2525
2526         if (sll->sll_ifindex) {
2527                 err = -ENODEV;
2528                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2529                 if (dev == NULL)
2530                         goto out;
2531         }
2532         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2533
2534 out:
2535         return err;
2536 }
2537
2538 static struct proto packet_proto = {
2539         .name     = "PACKET",
2540         .owner    = THIS_MODULE,
2541         .obj_size = sizeof(struct packet_sock),
2542 };
2543
2544 /*
2545  *      Create a packet of type SOCK_PACKET.
2546  */
2547
2548 static int packet_create(struct net *net, struct socket *sock, int protocol,
2549                          int kern)
2550 {
2551         struct sock *sk;
2552         struct packet_sock *po;
2553         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2554         int err;
2555
2556         if (!capable(CAP_NET_RAW))
2557                 return -EPERM;
2558         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2559             sock->type != SOCK_PACKET)
2560                 return -ESOCKTNOSUPPORT;
2561
2562         sock->state = SS_UNCONNECTED;
2563
2564         err = -ENOBUFS;
2565         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2566         if (sk == NULL)
2567                 goto out;
2568
2569         sock->ops = &packet_ops;
2570         if (sock->type == SOCK_PACKET)
2571                 sock->ops = &packet_ops_spkt;
2572
2573         sock_init_data(sock, sk);
2574
2575         po = pkt_sk(sk);
2576         sk->sk_family = PF_PACKET;
2577         po->num = proto;
2578
2579         sk->sk_destruct = packet_sock_destruct;
2580         sk_refcnt_debug_inc(sk);
2581
2582         /*
2583          *      Attach a protocol block
2584          */
2585
2586         spin_lock_init(&po->bind_lock);
2587         mutex_init(&po->pg_vec_lock);
2588         po->prot_hook.func = packet_rcv;
2589
2590         if (sock->type == SOCK_PACKET)
2591                 po->prot_hook.func = packet_rcv_spkt;
2592
2593         po->prot_hook.af_packet_priv = sk;
2594
2595         if (proto) {
2596                 po->prot_hook.type = proto;
2597                 register_prot_hook(sk);
2598         }
2599
2600         spin_lock_bh(&net->packet.sklist_lock);
2601         sk_add_node_rcu(sk, &net->packet.sklist);
2602         sock_prot_inuse_add(net, &packet_proto, 1);
2603         spin_unlock_bh(&net->packet.sklist_lock);
2604
2605         return 0;
2606 out:
2607         return err;
2608 }
2609
2610 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2611 {
2612         struct sock_exterr_skb *serr;
2613         struct sk_buff *skb, *skb2;
2614         int copied, err;
2615
2616         err = -EAGAIN;
2617         skb = skb_dequeue(&sk->sk_error_queue);
2618         if (skb == NULL)
2619                 goto out;
2620
2621         copied = skb->len;
2622         if (copied > len) {
2623                 msg->msg_flags |= MSG_TRUNC;
2624                 copied = len;
2625         }
2626         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2627         if (err)
2628                 goto out_free_skb;
2629
2630         sock_recv_timestamp(msg, sk, skb);
2631
2632         serr = SKB_EXT_ERR(skb);
2633         put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2634                  sizeof(serr->ee), &serr->ee);
2635
2636         msg->msg_flags |= MSG_ERRQUEUE;
2637         err = copied;
2638
2639         /* Reset and regenerate socket error */
2640         spin_lock_bh(&sk->sk_error_queue.lock);
2641         sk->sk_err = 0;
2642         if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2643                 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2644                 spin_unlock_bh(&sk->sk_error_queue.lock);
2645                 sk->sk_error_report(sk);
2646         } else
2647                 spin_unlock_bh(&sk->sk_error_queue.lock);
2648
2649 out_free_skb:
2650         kfree_skb(skb);
2651 out:
2652         return err;
2653 }
2654
2655 /*
2656  *      Pull a packet from our receive queue and hand it to the user.
2657  *      If necessary we block.
2658  */
2659
2660 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2661                           struct msghdr *msg, size_t len, int flags)
2662 {
2663         struct sock *sk = sock->sk;
2664         struct sk_buff *skb;
2665         int copied, err;
2666         struct sockaddr_ll *sll;
2667         int vnet_hdr_len = 0;
2668
2669         err = -EINVAL;
2670         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2671                 goto out;
2672
2673 #if 0
2674         /* What error should we return now? EUNATTACH? */
2675         if (pkt_sk(sk)->ifindex < 0)
2676                 return -ENODEV;
2677 #endif
2678
2679         if (flags & MSG_ERRQUEUE) {
2680                 err = packet_recv_error(sk, msg, len);
2681                 goto out;
2682         }
2683
2684         /*
2685          *      Call the generic datagram receiver. This handles all sorts
2686          *      of horrible races and re-entrancy so we can forget about it
2687          *      in the protocol layers.
2688          *
2689          *      Now it will return ENETDOWN, if device have just gone down,
2690          *      but then it will block.
2691          */
2692
2693         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2694
2695         /*
2696          *      An error occurred so return it. Because skb_recv_datagram()
2697          *      handles the blocking we don't see and worry about blocking
2698          *      retries.
2699          */
2700
2701         if (skb == NULL)
2702                 goto out;
2703
2704         if (pkt_sk(sk)->has_vnet_hdr) {
2705                 struct virtio_net_hdr vnet_hdr = { 0 };
2706
2707                 err = -EINVAL;
2708                 vnet_hdr_len = sizeof(vnet_hdr);
2709                 if (len < vnet_hdr_len)
2710                         goto out_free;
2711
2712                 len -= vnet_hdr_len;
2713
2714                 if (skb_is_gso(skb)) {
2715                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2716
2717                         /* This is a hint as to how much should be linear. */
2718                         vnet_hdr.hdr_len = skb_headlen(skb);
2719                         vnet_hdr.gso_size = sinfo->gso_size;
2720                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2721                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2722                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2723                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2724                         else if (sinfo->gso_type & SKB_GSO_UDP)
2725                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2726                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2727                                 goto out_free;
2728                         else
2729                                 BUG();
2730                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2731                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2732                 } else
2733                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2734
2735                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2736                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2737                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2738                         vnet_hdr.csum_offset = skb->csum_offset;
2739                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2740                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2741                 } /* else everything is zero */
2742
2743                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2744                                      vnet_hdr_len);
2745                 if (err < 0)
2746                         goto out_free;
2747         }
2748
2749         /*
2750          *      If the address length field is there to be filled in, we fill
2751          *      it in now.
2752          */
2753
2754         sll = &PACKET_SKB_CB(skb)->sa.ll;
2755         if (sock->type == SOCK_PACKET)
2756                 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2757         else
2758                 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2759
2760         /*
2761          *      You lose any data beyond the buffer you gave. If it worries a
2762          *      user program they can ask the device for its MTU anyway.
2763          */
2764
2765         copied = skb->len;
2766         if (copied > len) {
2767                 copied = len;
2768                 msg->msg_flags |= MSG_TRUNC;
2769         }
2770
2771         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2772         if (err)
2773                 goto out_free;
2774
2775         sock_recv_ts_and_drops(msg, sk, skb);
2776
2777         if (msg->msg_name)
2778                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2779                        msg->msg_namelen);
2780
2781         if (pkt_sk(sk)->auxdata) {
2782                 struct tpacket_auxdata aux;
2783
2784                 aux.tp_status = TP_STATUS_USER;
2785                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2786                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2787                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2788                 aux.tp_snaplen = skb->len;
2789                 aux.tp_mac = 0;
2790                 aux.tp_net = skb_network_offset(skb);
2791                 if (vlan_tx_tag_present(skb)) {
2792                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2793                         aux.tp_status |= TP_STATUS_VLAN_VALID;
2794                 } else {
2795                         aux.tp_vlan_tci = 0;
2796                 }
2797                 aux.tp_padding = 0;
2798                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2799         }
2800
2801         /*
2802          *      Free or return the buffer as appropriate. Again this
2803          *      hides all the races and re-entrancy issues from us.
2804          */
2805         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2806
2807 out_free:
2808         skb_free_datagram(sk, skb);
2809 out:
2810         return err;
2811 }
2812
2813 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2814                                int *uaddr_len, int peer)
2815 {
2816         struct net_device *dev;
2817         struct sock *sk = sock->sk;
2818
2819         if (peer)
2820                 return -EOPNOTSUPP;
2821
2822         uaddr->sa_family = AF_PACKET;
2823         rcu_read_lock();
2824         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2825         if (dev)
2826                 strncpy(uaddr->sa_data, dev->name, 14);
2827         else
2828                 memset(uaddr->sa_data, 0, 14);
2829         rcu_read_unlock();
2830         *uaddr_len = sizeof(*uaddr);
2831
2832         return 0;
2833 }
2834
2835 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2836                           int *uaddr_len, int peer)
2837 {
2838         struct net_device *dev;
2839         struct sock *sk = sock->sk;
2840         struct packet_sock *po = pkt_sk(sk);
2841         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2842
2843         if (peer)
2844                 return -EOPNOTSUPP;
2845
2846         sll->sll_family = AF_PACKET;
2847         sll->sll_ifindex = po->ifindex;
2848         sll->sll_protocol = po->num;
2849         sll->sll_pkttype = 0;
2850         rcu_read_lock();
2851         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2852         if (dev) {
2853                 sll->sll_hatype = dev->type;
2854                 sll->sll_halen = dev->addr_len;
2855                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2856         } else {
2857                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
2858                 sll->sll_halen = 0;
2859         }
2860         rcu_read_unlock();
2861         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2862
2863         return 0;
2864 }
2865
2866 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2867                          int what)
2868 {
2869         switch (i->type) {
2870         case PACKET_MR_MULTICAST:
2871                 if (i->alen != dev->addr_len)
2872                         return -EINVAL;
2873                 if (what > 0)
2874                         return dev_mc_add(dev, i->addr);
2875                 else
2876                         return dev_mc_del(dev, i->addr);
2877                 break;
2878         case PACKET_MR_PROMISC:
2879                 return dev_set_promiscuity(dev, what);
2880                 break;
2881         case PACKET_MR_ALLMULTI:
2882                 return dev_set_allmulti(dev, what);
2883                 break;
2884         case PACKET_MR_UNICAST:
2885                 if (i->alen != dev->addr_len)
2886                         return -EINVAL;
2887                 if (what > 0)
2888                         return dev_uc_add(dev, i->addr);
2889                 else
2890                         return dev_uc_del(dev, i->addr);
2891                 break;
2892         default:
2893                 break;
2894         }
2895         return 0;
2896 }
2897
2898 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2899 {
2900         for ( ; i; i = i->next) {
2901                 if (i->ifindex == dev->ifindex)
2902                         packet_dev_mc(dev, i, what);
2903         }
2904 }
2905
2906 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2907 {
2908         struct packet_sock *po = pkt_sk(sk);
2909         struct packet_mclist *ml, *i;
2910         struct net_device *dev;
2911         int err;
2912
2913         rtnl_lock();
2914
2915         err = -ENODEV;
2916         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2917         if (!dev)
2918                 goto done;
2919
2920         err = -EINVAL;
2921         if (mreq->mr_alen > dev->addr_len)
2922                 goto done;
2923
2924         err = -ENOBUFS;
2925         i = kmalloc(sizeof(*i), GFP_KERNEL);
2926         if (i == NULL)
2927                 goto done;
2928
2929         err = 0;
2930         for (ml = po->mclist; ml; ml = ml->next) {
2931                 if (ml->ifindex == mreq->mr_ifindex &&
2932                     ml->type == mreq->mr_type &&
2933                     ml->alen == mreq->mr_alen &&
2934                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2935                         ml->count++;
2936                         /* Free the new element ... */
2937                         kfree(i);
2938                         goto done;
2939                 }
2940         }
2941
2942         i->type = mreq->mr_type;
2943         i->ifindex = mreq->mr_ifindex;
2944         i->alen = mreq->mr_alen;
2945         memcpy(i->addr, mreq->mr_address, i->alen);
2946         i->count = 1;
2947         i->next = po->mclist;
2948         po->mclist = i;
2949         err = packet_dev_mc(dev, i, 1);
2950         if (err) {
2951                 po->mclist = i->next;
2952                 kfree(i);
2953         }
2954
2955 done:
2956         rtnl_unlock();
2957         return err;
2958 }
2959
2960 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2961 {
2962         struct packet_mclist *ml, **mlp;
2963
2964         rtnl_lock();
2965
2966         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
2967                 if (ml->ifindex == mreq->mr_ifindex &&
2968                     ml->type == mreq->mr_type &&
2969                     ml->alen == mreq->mr_alen &&
2970                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2971                         if (--ml->count == 0) {
2972                                 struct net_device *dev;
2973                                 *mlp = ml->next;
2974                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2975                                 if (dev)
2976                                         packet_dev_mc(dev, ml, -1);
2977                                 kfree(ml);
2978                         }
2979                         rtnl_unlock();
2980                         return 0;
2981                 }
2982         }
2983         rtnl_unlock();
2984         return -EADDRNOTAVAIL;
2985 }
2986
2987 static void packet_flush_mclist(struct sock *sk)
2988 {
2989         struct packet_sock *po = pkt_sk(sk);
2990         struct packet_mclist *ml;
2991
2992         if (!po->mclist)
2993                 return;
2994
2995         rtnl_lock();
2996         while ((ml = po->mclist) != NULL) {
2997                 struct net_device *dev;
2998
2999                 po->mclist = ml->next;
3000                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3001                 if (dev != NULL)
3002                         packet_dev_mc(dev, ml, -1);
3003                 kfree(ml);
3004         }
3005         rtnl_unlock();
3006 }
3007
3008 static int
3009 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3010 {
3011         struct sock *sk = sock->sk;
3012         struct packet_sock *po = pkt_sk(sk);
3013         int ret;
3014
3015         if (level != SOL_PACKET)
3016                 return -ENOPROTOOPT;
3017
3018         switch (optname) {
3019         case PACKET_ADD_MEMBERSHIP:
3020         case PACKET_DROP_MEMBERSHIP:
3021         {
3022                 struct packet_mreq_max mreq;
3023                 int len = optlen;
3024                 memset(&mreq, 0, sizeof(mreq));
3025                 if (len < sizeof(struct packet_mreq))
3026                         return -EINVAL;
3027                 if (len > sizeof(mreq))
3028                         len = sizeof(mreq);
3029                 if (copy_from_user(&mreq, optval, len))
3030                         return -EFAULT;
3031                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3032                         return -EINVAL;
3033                 if (optname == PACKET_ADD_MEMBERSHIP)
3034                         ret = packet_mc_add(sk, &mreq);
3035                 else
3036                         ret = packet_mc_drop(sk, &mreq);
3037                 return ret;
3038         }
3039
3040         case PACKET_RX_RING:
3041         case PACKET_TX_RING:
3042         {
3043                 union tpacket_req_u req_u;
3044                 int len;
3045
3046                 switch (po->tp_version) {
3047                 case TPACKET_V1:
3048                 case TPACKET_V2:
3049                         len = sizeof(req_u.req);
3050                         break;
3051                 case TPACKET_V3:
3052                 default:
3053                         len = sizeof(req_u.req3);
3054                         break;
3055                 }
3056                 if (optlen < len)
3057                         return -EINVAL;
3058                 if (pkt_sk(sk)->has_vnet_hdr)
3059                         return -EINVAL;
3060                 if (copy_from_user(&req_u.req, optval, len))
3061                         return -EFAULT;
3062                 return packet_set_ring(sk, &req_u, 0,
3063                         optname == PACKET_TX_RING);
3064         }
3065         case PACKET_COPY_THRESH:
3066         {
3067                 int val;
3068
3069                 if (optlen != sizeof(val))
3070                         return -EINVAL;
3071                 if (copy_from_user(&val, optval, sizeof(val)))
3072                         return -EFAULT;
3073
3074                 pkt_sk(sk)->copy_thresh = val;
3075                 return 0;
3076         }
3077         case PACKET_VERSION:
3078         {
3079                 int val;
3080
3081                 if (optlen != sizeof(val))
3082                         return -EINVAL;
3083                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3084                         return -EBUSY;
3085                 if (copy_from_user(&val, optval, sizeof(val)))
3086                         return -EFAULT;
3087                 switch (val) {
3088                 case TPACKET_V1:
3089                 case TPACKET_V2:
3090                 case TPACKET_V3:
3091                         po->tp_version = val;
3092                         return 0;
3093                 default:
3094                         return -EINVAL;
3095                 }
3096         }
3097         case PACKET_RESERVE:
3098         {
3099                 unsigned int val;
3100
3101                 if (optlen != sizeof(val))
3102                         return -EINVAL;
3103                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3104                         return -EBUSY;
3105                 if (copy_from_user(&val, optval, sizeof(val)))
3106                         return -EFAULT;
3107                 po->tp_reserve = val;
3108                 return 0;
3109         }
3110         case PACKET_LOSS:
3111         {
3112                 unsigned int val;
3113
3114                 if (optlen != sizeof(val))
3115                         return -EINVAL;
3116                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3117                         return -EBUSY;
3118                 if (copy_from_user(&val, optval, sizeof(val)))
3119                         return -EFAULT;
3120                 po->tp_loss = !!val;
3121                 return 0;
3122         }
3123         case PACKET_AUXDATA:
3124         {
3125                 int val;
3126
3127                 if (optlen < sizeof(val))
3128                         return -EINVAL;
3129                 if (copy_from_user(&val, optval, sizeof(val)))
3130                         return -EFAULT;
3131
3132                 po->auxdata = !!val;
3133                 return 0;
3134         }
3135         case PACKET_ORIGDEV:
3136         {
3137                 int val;
3138
3139                 if (optlen < sizeof(val))
3140                         return -EINVAL;
3141                 if (copy_from_user(&val, optval, sizeof(val)))
3142                         return -EFAULT;
3143
3144                 po->origdev = !!val;
3145                 return 0;
3146         }
3147         case PACKET_VNET_HDR:
3148         {
3149                 int val;
3150
3151                 if (sock->type != SOCK_RAW)
3152                         return -EINVAL;
3153                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3154                         return -EBUSY;
3155                 if (optlen < sizeof(val))
3156                         return -EINVAL;
3157                 if (copy_from_user(&val, optval, sizeof(val)))
3158                         return -EFAULT;
3159
3160                 po->has_vnet_hdr = !!val;
3161                 return 0;
3162         }
3163         case PACKET_TIMESTAMP:
3164         {
3165                 int val;
3166
3167                 if (optlen != sizeof(val))
3168                         return -EINVAL;
3169                 if (copy_from_user(&val, optval, sizeof(val)))
3170                         return -EFAULT;
3171
3172                 po->tp_tstamp = val;
3173                 return 0;
3174         }
3175         case PACKET_FANOUT:
3176         {
3177                 int val;
3178
3179                 if (optlen != sizeof(val))
3180                         return -EINVAL;
3181                 if (copy_from_user(&val, optval, sizeof(val)))
3182                         return -EFAULT;
3183
3184                 return fanout_add(sk, val & 0xffff, val >> 16);
3185         }
3186         default:
3187                 return -ENOPROTOOPT;
3188         }
3189 }
3190
3191 static int packet_getsockopt(struct socket *sock, int level, int optname,
3192                              char __user *optval, int __user *optlen)
3193 {
3194         int len;
3195         int val;
3196         struct sock *sk = sock->sk;
3197         struct packet_sock *po = pkt_sk(sk);
3198         void *data;
3199         struct tpacket_stats st;
3200         union tpacket_stats_u st_u;
3201
3202         if (level != SOL_PACKET)
3203                 return -ENOPROTOOPT;
3204
3205         if (get_user(len, optlen))
3206                 return -EFAULT;
3207
3208         if (len < 0)
3209                 return -EINVAL;
3210
3211         switch (optname) {
3212         case PACKET_STATISTICS:
3213                 if (po->tp_version == TPACKET_V3) {
3214                         len = sizeof(struct tpacket_stats_v3);
3215                 } else {
3216                         if (len > sizeof(struct tpacket_stats))
3217                                 len = sizeof(struct tpacket_stats);
3218                 }
3219                 spin_lock_bh(&sk->sk_receive_queue.lock);
3220                 if (po->tp_version == TPACKET_V3) {
3221                         memcpy(&st_u.stats3, &po->stats,
3222                         sizeof(struct tpacket_stats));
3223                         st_u.stats3.tp_freeze_q_cnt =
3224                         po->stats_u.stats3.tp_freeze_q_cnt;
3225                         st_u.stats3.tp_packets += po->stats.tp_drops;
3226                         data = &st_u.stats3;
3227                 } else {
3228                         st = po->stats;
3229                         st.tp_packets += st.tp_drops;
3230                         data = &st;
3231                 }
3232                 memset(&po->stats, 0, sizeof(st));
3233                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3234                 break;
3235         case PACKET_AUXDATA:
3236                 if (len > sizeof(int))
3237                         len = sizeof(int);
3238                 val = po->auxdata;
3239
3240                 data = &val;
3241                 break;
3242         case PACKET_ORIGDEV:
3243                 if (len > sizeof(int))
3244                         len = sizeof(int);
3245                 val = po->origdev;
3246
3247                 data = &val;
3248                 break;
3249         case PACKET_VNET_HDR:
3250                 if (len > sizeof(int))
3251                         len = sizeof(int);
3252                 val = po->has_vnet_hdr;
3253
3254                 data = &val;
3255                 break;
3256         case PACKET_VERSION:
3257                 if (len > sizeof(int))
3258                         len = sizeof(int);
3259                 val = po->tp_version;
3260                 data = &val;
3261                 break;
3262         case PACKET_HDRLEN:
3263                 if (len > sizeof(int))
3264                         len = sizeof(int);
3265                 if (copy_from_user(&val, optval, len))
3266                         return -EFAULT;
3267                 switch (val) {
3268                 case TPACKET_V1:
3269                         val = sizeof(struct tpacket_hdr);
3270                         break;
3271                 case TPACKET_V2:
3272                         val = sizeof(struct tpacket2_hdr);
3273                         break;
3274                 case TPACKET_V3:
3275                         val = sizeof(struct tpacket3_hdr);
3276                         break;
3277                 default:
3278                         return -EINVAL;
3279                 }
3280                 data = &val;
3281                 break;
3282         case PACKET_RESERVE:
3283                 if (len > sizeof(unsigned int))
3284                         len = sizeof(unsigned int);
3285                 val = po->tp_reserve;
3286                 data = &val;
3287                 break;
3288         case PACKET_LOSS:
3289                 if (len > sizeof(unsigned int))
3290                         len = sizeof(unsigned int);
3291                 val = po->tp_loss;
3292                 data = &val;
3293                 break;
3294         case PACKET_TIMESTAMP:
3295                 if (len > sizeof(int))
3296                         len = sizeof(int);
3297                 val = po->tp_tstamp;
3298                 data = &val;
3299                 break;
3300         case PACKET_FANOUT:
3301                 if (len > sizeof(int))
3302                         len = sizeof(int);
3303                 val = (po->fanout ?
3304                        ((u32)po->fanout->id |
3305                         ((u32)po->fanout->type << 16)) :
3306                        0);
3307                 data = &val;
3308                 break;
3309         default:
3310                 return -ENOPROTOOPT;
3311         }
3312
3313         if (put_user(len, optlen))
3314                 return -EFAULT;
3315         if (copy_to_user(optval, data, len))
3316                 return -EFAULT;
3317         return 0;
3318 }
3319
3320
3321 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3322 {
3323         struct sock *sk;
3324         struct hlist_node *node;
3325         struct net_device *dev = data;
3326         struct net *net = dev_net(dev);
3327
3328         rcu_read_lock();
3329         sk_for_each_rcu(sk, node, &net->packet.sklist) {
3330                 struct packet_sock *po = pkt_sk(sk);
3331
3332                 switch (msg) {
3333                 case NETDEV_UNREGISTER:
3334                         if (po->mclist)
3335                                 packet_dev_mclist(dev, po->mclist, -1);
3336                         /* fallthrough */
3337
3338                 case NETDEV_DOWN:
3339                         if (dev->ifindex == po->ifindex) {
3340                                 spin_lock(&po->bind_lock);
3341                                 if (po->running) {
3342                                         __unregister_prot_hook(sk, false);
3343                                         sk->sk_err = ENETDOWN;
3344                                         if (!sock_flag(sk, SOCK_DEAD))
3345                                                 sk->sk_error_report(sk);
3346                                 }
3347                                 if (msg == NETDEV_UNREGISTER) {
3348                                         po->ifindex = -1;
3349                                         if (po->prot_hook.dev)
3350                                                 dev_put(po->prot_hook.dev);
3351                                         po->prot_hook.dev = NULL;
3352                                 }
3353                                 spin_unlock(&po->bind_lock);
3354                         }
3355                         break;
3356                 case NETDEV_UP:
3357                         if (dev->ifindex == po->ifindex) {
3358                                 spin_lock(&po->bind_lock);
3359                                 if (po->num)
3360                                         register_prot_hook(sk);
3361                                 spin_unlock(&po->bind_lock);
3362                         }
3363                         break;
3364                 }
3365         }
3366         rcu_read_unlock();
3367         return NOTIFY_DONE;
3368 }
3369
3370
3371 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3372                         unsigned long arg)
3373 {
3374         struct sock *sk = sock->sk;
3375
3376         switch (cmd) {
3377         case SIOCOUTQ:
3378         {
3379                 int amount = sk_wmem_alloc_get(sk);
3380
3381                 return put_user(amount, (int __user *)arg);
3382         }
3383         case SIOCINQ:
3384         {
3385                 struct sk_buff *skb;
3386                 int amount = 0;
3387
3388                 spin_lock_bh(&sk->sk_receive_queue.lock);
3389                 skb = skb_peek(&sk->sk_receive_queue);
3390                 if (skb)
3391                         amount = skb->len;
3392                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3393                 return put_user(amount, (int __user *)arg);
3394         }
3395         case SIOCGSTAMP:
3396                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3397         case SIOCGSTAMPNS:
3398                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3399
3400 #ifdef CONFIG_INET
3401         case SIOCADDRT:
3402         case SIOCDELRT:
3403         case SIOCDARP:
3404         case SIOCGARP:
3405         case SIOCSARP:
3406         case SIOCGIFADDR:
3407         case SIOCSIFADDR:
3408         case SIOCGIFBRDADDR:
3409         case SIOCSIFBRDADDR:
3410         case SIOCGIFNETMASK:
3411         case SIOCSIFNETMASK:
3412         case SIOCGIFDSTADDR:
3413         case SIOCSIFDSTADDR:
3414         case SIOCSIFFLAGS:
3415                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3416 #endif
3417
3418         default:
3419                 return -ENOIOCTLCMD;
3420         }
3421         return 0;
3422 }
3423
3424 static unsigned int packet_poll(struct file *file, struct socket *sock,
3425                                 poll_table *wait)
3426 {
3427         struct sock *sk = sock->sk;
3428         struct packet_sock *po = pkt_sk(sk);
3429         unsigned int mask = datagram_poll(file, sock, wait);
3430
3431         spin_lock_bh(&sk->sk_receive_queue.lock);
3432         if (po->rx_ring.pg_vec) {
3433                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3434                         TP_STATUS_KERNEL))
3435                         mask |= POLLIN | POLLRDNORM;
3436         }
3437         spin_unlock_bh(&sk->sk_receive_queue.lock);
3438         spin_lock_bh(&sk->sk_write_queue.lock);
3439         if (po->tx_ring.pg_vec) {
3440                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3441                         mask |= POLLOUT | POLLWRNORM;
3442         }
3443         spin_unlock_bh(&sk->sk_write_queue.lock);
3444         return mask;
3445 }
3446
3447
3448 /* Dirty? Well, I still did not learn better way to account
3449  * for user mmaps.
3450  */
3451
3452 static void packet_mm_open(struct vm_area_struct *vma)
3453 {
3454         struct file *file = vma->vm_file;
3455         struct socket *sock = file->private_data;
3456         struct sock *sk = sock->sk;
3457
3458         if (sk)
3459                 atomic_inc(&pkt_sk(sk)->mapped);
3460 }
3461
3462 static void packet_mm_close(struct vm_area_struct *vma)
3463 {
3464         struct file *file = vma->vm_file;
3465         struct socket *sock = file->private_data;
3466         struct sock *sk = sock->sk;
3467
3468         if (sk)
3469                 atomic_dec(&pkt_sk(sk)->mapped);
3470 }
3471
3472 static const struct vm_operations_struct packet_mmap_ops = {
3473         .open   =       packet_mm_open,
3474         .close  =       packet_mm_close,
3475 };
3476
3477 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3478                         unsigned int len)
3479 {
3480         int i;
3481
3482         for (i = 0; i < len; i++) {
3483                 if (likely(pg_vec[i].buffer)) {
3484                         if (is_vmalloc_addr(pg_vec[i].buffer))
3485                                 vfree(pg_vec[i].buffer);
3486                         else
3487                                 free_pages((unsigned long)pg_vec[i].buffer,
3488                                            order);
3489                         pg_vec[i].buffer = NULL;
3490                 }
3491         }
3492         kfree(pg_vec);
3493 }
3494
3495 static char *alloc_one_pg_vec_page(unsigned long order)
3496 {
3497         char *buffer = NULL;
3498         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3499                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3500
3501         buffer = (char *) __get_free_pages(gfp_flags, order);
3502
3503         if (buffer)
3504                 return buffer;
3505
3506         /*
3507          * __get_free_pages failed, fall back to vmalloc
3508          */
3509         buffer = vzalloc((1 << order) * PAGE_SIZE);
3510
3511         if (buffer)
3512                 return buffer;
3513
3514         /*
3515          * vmalloc failed, lets dig into swap here
3516          */
3517         gfp_flags &= ~__GFP_NORETRY;
3518         buffer = (char *)__get_free_pages(gfp_flags, order);
3519         if (buffer)
3520                 return buffer;
3521
3522         /*
3523          * complete and utter failure
3524          */
3525         return NULL;
3526 }
3527
3528 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3529 {
3530         unsigned int block_nr = req->tp_block_nr;
3531         struct pgv *pg_vec;
3532         int i;
3533
3534         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3535         if (unlikely(!pg_vec))
3536                 goto out;
3537
3538         for (i = 0; i < block_nr; i++) {
3539                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3540                 if (unlikely(!pg_vec[i].buffer))
3541                         goto out_free_pgvec;
3542         }
3543
3544 out:
3545         return pg_vec;
3546
3547 out_free_pgvec:
3548         free_pg_vec(pg_vec, order, block_nr);
3549         pg_vec = NULL;
3550         goto out;
3551 }
3552
3553 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3554                 int closing, int tx_ring)
3555 {
3556         struct pgv *pg_vec = NULL;
3557         struct packet_sock *po = pkt_sk(sk);
3558         int was_running, order = 0;
3559         struct packet_ring_buffer *rb;
3560         struct sk_buff_head *rb_queue;
3561         __be16 num;
3562         int err = -EINVAL;
3563         /* Added to avoid minimal code churn */
3564         struct tpacket_req *req = &req_u->req;
3565
3566         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3567         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3568                 WARN(1, "Tx-ring is not supported.\n");
3569                 goto out;
3570         }
3571
3572         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3573         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3574
3575         err = -EBUSY;
3576         if (!closing) {
3577                 if (atomic_read(&po->mapped))
3578                         goto out;
3579                 if (atomic_read(&rb->pending))
3580                         goto out;
3581         }
3582
3583         if (req->tp_block_nr) {
3584                 /* Sanity tests and some calculations */
3585                 err = -EBUSY;
3586                 if (unlikely(rb->pg_vec))
3587                         goto out;
3588
3589                 switch (po->tp_version) {
3590                 case TPACKET_V1:
3591                         po->tp_hdrlen = TPACKET_HDRLEN;
3592                         break;
3593                 case TPACKET_V2:
3594                         po->tp_hdrlen = TPACKET2_HDRLEN;
3595                         break;
3596                 case TPACKET_V3:
3597                         po->tp_hdrlen = TPACKET3_HDRLEN;
3598                         break;
3599                 }
3600
3601                 err = -EINVAL;
3602                 if (unlikely((int)req->tp_block_size <= 0))
3603                         goto out;
3604                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3605                         goto out;
3606                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3607                                         po->tp_reserve))
3608                         goto out;
3609                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3610                         goto out;
3611
3612                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3613                 if (unlikely(rb->frames_per_block <= 0))
3614                         goto out;
3615                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3616                                         req->tp_frame_nr))
3617                         goto out;
3618
3619                 err = -ENOMEM;
3620                 order = get_order(req->tp_block_size);
3621                 pg_vec = alloc_pg_vec(req, order);
3622                 if (unlikely(!pg_vec))
3623                         goto out;
3624                 switch (po->tp_version) {
3625                 case TPACKET_V3:
3626                 /* Transmit path is not supported. We checked
3627                  * it above but just being paranoid
3628                  */
3629                         if (!tx_ring)
3630                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3631                                 break;
3632                 default:
3633                         break;
3634                 }
3635         }
3636         /* Done */
3637         else {
3638                 err = -EINVAL;
3639                 if (unlikely(req->tp_frame_nr))
3640                         goto out;
3641         }
3642
3643         lock_sock(sk);
3644
3645         /* Detach socket from network */
3646         spin_lock(&po->bind_lock);
3647         was_running = po->running;
3648         num = po->num;
3649         if (was_running) {
3650                 po->num = 0;
3651                 __unregister_prot_hook(sk, false);
3652         }
3653         spin_unlock(&po->bind_lock);
3654
3655         synchronize_net();
3656
3657         err = -EBUSY;
3658         mutex_lock(&po->pg_vec_lock);
3659         if (closing || atomic_read(&po->mapped) == 0) {
3660                 err = 0;
3661                 spin_lock_bh(&rb_queue->lock);
3662                 swap(rb->pg_vec, pg_vec);
3663                 rb->frame_max = (req->tp_frame_nr - 1);
3664                 rb->head = 0;
3665                 rb->frame_size = req->tp_frame_size;
3666                 spin_unlock_bh(&rb_queue->lock);
3667
3668                 swap(rb->pg_vec_order, order);
3669                 swap(rb->pg_vec_len, req->tp_block_nr);
3670
3671                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3672                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3673                                                 tpacket_rcv : packet_rcv;
3674                 skb_queue_purge(rb_queue);
3675                 if (atomic_read(&po->mapped))
3676                         pr_err("packet_mmap: vma is busy: %d\n",
3677                                atomic_read(&po->mapped));
3678         }
3679         mutex_unlock(&po->pg_vec_lock);
3680
3681         spin_lock(&po->bind_lock);
3682         if (was_running) {
3683                 po->num = num;
3684                 register_prot_hook(sk);
3685         }
3686         spin_unlock(&po->bind_lock);
3687         if (closing && (po->tp_version > TPACKET_V2)) {
3688                 /* Because we don't support block-based V3 on tx-ring */
3689                 if (!tx_ring)
3690                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3691         }
3692         release_sock(sk);
3693
3694         if (pg_vec)
3695                 free_pg_vec(pg_vec, order, req->tp_block_nr);
3696 out:
3697         return err;
3698 }
3699
3700 static int packet_mmap(struct file *file, struct socket *sock,
3701                 struct vm_area_struct *vma)
3702 {
3703         struct sock *sk = sock->sk;
3704         struct packet_sock *po = pkt_sk(sk);
3705         unsigned long size, expected_size;
3706         struct packet_ring_buffer *rb;
3707         unsigned long start;
3708         int err = -EINVAL;
3709         int i;
3710
3711         if (vma->vm_pgoff)
3712                 return -EINVAL;
3713
3714         mutex_lock(&po->pg_vec_lock);
3715
3716         expected_size = 0;
3717         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3718                 if (rb->pg_vec) {
3719                         expected_size += rb->pg_vec_len
3720                                                 * rb->pg_vec_pages
3721                                                 * PAGE_SIZE;
3722                 }
3723         }
3724
3725         if (expected_size == 0)
3726                 goto out;
3727
3728         size = vma->vm_end - vma->vm_start;
3729         if (size != expected_size)
3730                 goto out;
3731
3732         start = vma->vm_start;
3733         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3734                 if (rb->pg_vec == NULL)
3735                         continue;
3736
3737                 for (i = 0; i < rb->pg_vec_len; i++) {
3738                         struct page *page;
3739                         void *kaddr = rb->pg_vec[i].buffer;
3740                         int pg_num;
3741
3742                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3743                                 page = pgv_to_page(kaddr);
3744                                 err = vm_insert_page(vma, start, page);
3745                                 if (unlikely(err))
3746                                         goto out;
3747                                 start += PAGE_SIZE;
3748                                 kaddr += PAGE_SIZE;
3749                         }
3750                 }
3751         }
3752
3753         atomic_inc(&po->mapped);
3754         vma->vm_ops = &packet_mmap_ops;
3755         err = 0;
3756
3757 out:
3758         mutex_unlock(&po->pg_vec_lock);
3759         return err;
3760 }
3761
3762 static const struct proto_ops packet_ops_spkt = {
3763         .family =       PF_PACKET,
3764         .owner =        THIS_MODULE,
3765         .release =      packet_release,
3766         .bind =         packet_bind_spkt,
3767         .connect =      sock_no_connect,
3768         .socketpair =   sock_no_socketpair,
3769         .accept =       sock_no_accept,
3770         .getname =      packet_getname_spkt,
3771         .poll =         datagram_poll,
3772         .ioctl =        packet_ioctl,
3773         .listen =       sock_no_listen,
3774         .shutdown =     sock_no_shutdown,
3775         .setsockopt =   sock_no_setsockopt,
3776         .getsockopt =   sock_no_getsockopt,
3777         .sendmsg =      packet_sendmsg_spkt,
3778         .recvmsg =      packet_recvmsg,
3779         .mmap =         sock_no_mmap,
3780         .sendpage =     sock_no_sendpage,
3781 };
3782
3783 static const struct proto_ops packet_ops = {
3784         .family =       PF_PACKET,
3785         .owner =        THIS_MODULE,
3786         .release =      packet_release,
3787         .bind =         packet_bind,
3788         .connect =      sock_no_connect,
3789         .socketpair =   sock_no_socketpair,
3790         .accept =       sock_no_accept,
3791         .getname =      packet_getname,
3792         .poll =         packet_poll,
3793         .ioctl =        packet_ioctl,
3794         .listen =       sock_no_listen,
3795         .shutdown =     sock_no_shutdown,
3796         .setsockopt =   packet_setsockopt,
3797         .getsockopt =   packet_getsockopt,
3798         .sendmsg =      packet_sendmsg,
3799         .recvmsg =      packet_recvmsg,
3800         .mmap =         packet_mmap,
3801         .sendpage =     sock_no_sendpage,
3802 };
3803
3804 static const struct net_proto_family packet_family_ops = {
3805         .family =       PF_PACKET,
3806         .create =       packet_create,
3807         .owner  =       THIS_MODULE,
3808 };
3809
3810 static struct notifier_block packet_netdev_notifier = {
3811         .notifier_call =        packet_notifier,
3812 };
3813
3814 #ifdef CONFIG_PROC_FS
3815
3816 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3817         __acquires(RCU)
3818 {
3819         struct net *net = seq_file_net(seq);
3820
3821         rcu_read_lock();
3822         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3823 }
3824
3825 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3826 {
3827         struct net *net = seq_file_net(seq);
3828         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3829 }
3830
3831 static void packet_seq_stop(struct seq_file *seq, void *v)
3832         __releases(RCU)
3833 {
3834         rcu_read_unlock();
3835 }
3836
3837 static int packet_seq_show(struct seq_file *seq, void *v)
3838 {
3839         if (v == SEQ_START_TOKEN)
3840                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
3841         else {
3842                 struct sock *s = sk_entry(v);
3843                 const struct packet_sock *po = pkt_sk(s);
3844
3845                 seq_printf(seq,
3846                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
3847                            s,
3848                            atomic_read(&s->sk_refcnt),
3849                            s->sk_type,
3850                            ntohs(po->num),
3851                            po->ifindex,
3852                            po->running,
3853                            atomic_read(&s->sk_rmem_alloc),
3854                            sock_i_uid(s),
3855                            sock_i_ino(s));
3856         }
3857
3858         return 0;
3859 }
3860
3861 static const struct seq_operations packet_seq_ops = {
3862         .start  = packet_seq_start,
3863         .next   = packet_seq_next,
3864         .stop   = packet_seq_stop,
3865         .show   = packet_seq_show,
3866 };
3867
3868 static int packet_seq_open(struct inode *inode, struct file *file)
3869 {
3870         return seq_open_net(inode, file, &packet_seq_ops,
3871                             sizeof(struct seq_net_private));
3872 }
3873
3874 static const struct file_operations packet_seq_fops = {
3875         .owner          = THIS_MODULE,
3876         .open           = packet_seq_open,
3877         .read           = seq_read,
3878         .llseek         = seq_lseek,
3879         .release        = seq_release_net,
3880 };
3881
3882 #endif
3883
3884 static int __net_init packet_net_init(struct net *net)
3885 {
3886         spin_lock_init(&net->packet.sklist_lock);
3887         INIT_HLIST_HEAD(&net->packet.sklist);
3888
3889         if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
3890                 return -ENOMEM;
3891
3892         return 0;
3893 }
3894
3895 static void __net_exit packet_net_exit(struct net *net)
3896 {
3897         proc_net_remove(net, "packet");
3898 }
3899
3900 static struct pernet_operations packet_net_ops = {
3901         .init = packet_net_init,
3902         .exit = packet_net_exit,
3903 };
3904
3905
3906 static void __exit packet_exit(void)
3907 {
3908         unregister_netdevice_notifier(&packet_netdev_notifier);
3909         unregister_pernet_subsys(&packet_net_ops);
3910         sock_unregister(PF_PACKET);
3911         proto_unregister(&packet_proto);
3912 }
3913
3914 static int __init packet_init(void)
3915 {
3916         int rc = proto_register(&packet_proto, 0);
3917
3918         if (rc != 0)
3919                 goto out;
3920
3921         sock_register(&packet_family_ops);
3922         register_pernet_subsys(&packet_net_ops);
3923         register_netdevice_notifier(&packet_netdev_notifier);
3924 out:
3925         return rc;
3926 }
3927
3928 module_init(packet_init);
3929 module_exit(packet_exit);
3930 MODULE_LICENSE("GPL");
3931 MODULE_ALIAS_NETPROTO(PF_PACKET);