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         struct tpacket_kbdq_core *pkc;
 520
 521         if (tx_ring)
 522                 BUG();
 523
 524         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 525         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
 526 }
 527
 528 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 529                                 int blk_size_in_bytes)
 530 {
 531         struct net_device *dev;
 532         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 533         struct ethtool_cmd ecmd;
 534         int err;
 535
 536         rtnl_lock();
 537         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 538         if (unlikely(!dev)) {
 539                 rtnl_unlock();
 540                 return DEFAULT_PRB_RETIRE_TOV;
 541         }
 542         err = __ethtool_get_settings(dev, &ecmd);
 543         rtnl_unlock();
 544         if (!err) {
 545                 switch (ecmd.speed) {
 546                 case SPEED_10000:
 547                         msec = 1;
 548                         div = 10000/1000;
 549                         break;
 550                 case SPEED_1000:
 551                         msec = 1;
 552                         div = 1000/1000;
 553                         break;
 554                 /*
 555                  * If the link speed is so slow you don't really
 556                  * need to worry about perf anyways
 557                  */
 558                 case SPEED_100:
 559                 case SPEED_10:
 560                 default:
 561                         return DEFAULT_PRB_RETIRE_TOV;
 562                 }
 563         }
 564
 565         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 566
 567         if (div)
 568                 mbits /= div;
 569
 570         tmo = mbits * msec;
 571
 572         if (div)
 573                 return tmo+1;
 574         return tmo;
 575 }
 576
 577 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 578                         union tpacket_req_u *req_u)
 579 {
 580         p1->feature_req_word = req_u->req3.tp_feature_req_word;
 581 }
 582
 583 static void init_prb_bdqc(struct packet_sock *po,
 584                         struct packet_ring_buffer *rb,
 585                         struct pgv *pg_vec,
 586                         union tpacket_req_u *req_u, int tx_ring)
 587 {
 588         struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
 589         struct tpacket_block_desc *pbd;
 590
 591         memset(p1, 0x0, sizeof(*p1));
 592
 593         p1->knxt_seq_num = 1;
 594         p1->pkbdq = pg_vec;
 595         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 596         p1->pkblk_start = (char *)pg_vec[0].buffer;
 597         p1->kblk_size = req_u->req3.tp_block_size;
 598         p1->knum_blocks = req_u->req3.tp_block_nr;
 599         p1->hdrlen = po->tp_hdrlen;
 600         p1->version = po->tp_version;
 601         p1->last_kactive_blk_num = 0;
 602         po->stats_u.stats3.tp_freeze_q_cnt = 0;
 603         if (req_u->req3.tp_retire_blk_tov)
 604                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 605         else
 606                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 607                                                 req_u->req3.tp_block_size);
 608         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 609         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 610
 611         prb_init_ft_ops(p1, req_u);
 612         prb_setup_retire_blk_timer(po, tx_ring);
 613         prb_open_block(p1, pbd);
 614 }
 615
 616 /*  Do NOT update the last_blk_num first.
 617  *  Assumes sk_buff_head lock is held.
 618  */
 619 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 620 {
 621         mod_timer(&pkc->retire_blk_timer,
 622                         jiffies + pkc->tov_in_jiffies);
 623         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 624 }
 625
 626 /*
 627  * Timer logic:
 628  * 1) We refresh the timer only when we open a block.
 629  *    By doing this we don't waste cycles refreshing the timer
 630  *        on packet-by-packet basis.
 631  *
 632  * With a 1MB block-size, on a 1Gbps line, it will take
 633  * i) ~8 ms to fill a block + ii) memcpy etc.
 634  * In this cut we are not accounting for the memcpy time.
 635  *
 636  * So, if the user sets the 'tmo' to 10ms then the timer
 637  * will never fire while the block is still getting filled
 638  * (which is what we want). However, the user could choose
 639  * to close a block early and that's fine.
 640  *
 641  * But when the timer does fire, we check whether or not to refresh it.
 642  * Since the tmo granularity is in msecs, it is not too expensive
 643  * to refresh the timer, lets say every '8' msecs.
 644  * Either the user can set the 'tmo' or we can derive it based on
 645  * a) line-speed and b) block-size.
 646  * prb_calc_retire_blk_tmo() calculates the tmo.
 647  *
 648  */
 649 static void prb_retire_rx_blk_timer_expired(unsigned long data)
 650 {
 651         struct packet_sock *po = (struct packet_sock *)data;
 652         struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
 653         unsigned int frozen;
 654         struct tpacket_block_desc *pbd;
 655
 656         spin_lock(&po->sk.sk_receive_queue.lock);
 657
 658         frozen = prb_queue_frozen(pkc);
 659         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 660
 661         if (unlikely(pkc->delete_blk_timer))
 662                 goto out;
 663
 664         /* We only need to plug the race when the block is partially filled.
 665          * tpacket_rcv:
 666          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 667          *              copy_bits() is in progress ...
 668          *              timer fires on other cpu:
 669          *              we can't retire the current block because copy_bits
 670          *              is in progress.
 671          *
 672          */
 673         if (BLOCK_NUM_PKTS(pbd)) {
 674                 while (atomic_read(&pkc->blk_fill_in_prog)) {
 675                         /* Waiting for skb_copy_bits to finish... */
 676                         cpu_relax();
 677                 }
 678         }
 679
 680         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 681                 if (!frozen) {
 682                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 683                         if (!prb_dispatch_next_block(pkc, po))
 684                                 goto refresh_timer;
 685                         else
 686                                 goto out;
 687                 } else {
 688                         /* Case 1. Queue was frozen because user-space was
 689                          *         lagging behind.
 690                          */
 691                         if (prb_curr_blk_in_use(pkc, pbd)) {
 692                                 /*
 693                                  * Ok, user-space is still behind.
 694                                  * So just refresh the timer.
 695                                  */
 696                                 goto refresh_timer;
 697                         } else {
 698                                /* Case 2. queue was frozen,user-space caught up,
 699                                 * now the link went idle && the timer fired.
 700                                 * We don't have a block to close.So we open this
 701                                 * block and restart the timer.
 702                                 * opening a block thaws the queue,restarts timer
 703                                 * Thawing/timer-refresh is a side effect.
 704                                 */
 705                                 prb_open_block(pkc, pbd);
 706                                 goto out;
 707                         }
 708                 }
 709         }
 710
 711 refresh_timer:
 712         _prb_refresh_rx_retire_blk_timer(pkc);
 713
 714 out:
 715         spin_unlock(&po->sk.sk_receive_queue.lock);
 716 }
 717
 718 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 719                 struct tpacket_block_desc *pbd1, __u32 status)
 720 {
 721         /* Flush everything minus the block header */
 722
 723 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 724         u8 *start, *end;
 725
 726         start = (u8 *)pbd1;
 727
 728         /* Skip the block header(we know header WILL fit in 4K) */
 729         start += PAGE_SIZE;
 730
 731         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 732         for (; start < end; start += PAGE_SIZE)
 733                 flush_dcache_page(pgv_to_page(start));
 734
 735         smp_wmb();
 736 #endif
 737
 738         /* Now update the block status. */
 739
 740         BLOCK_STATUS(pbd1) = status;
 741
 742         /* Flush the block header */
 743
 744 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 745         start = (u8 *)pbd1;
 746         flush_dcache_page(pgv_to_page(start));
 747
 748         smp_wmb();
 749 #endif
 750 }
 751
 752 /*
 753  * Side effect:
 754  *
 755  * 1) flush the block
 756  * 2) Increment active_blk_num
 757  *
 758  * Note:We DONT refresh the timer on purpose.
 759  *      Because almost always the next block will be opened.
 760  */
 761 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 762                 struct tpacket_block_desc *pbd1,
 763                 struct packet_sock *po, unsigned int stat)
 764 {
 765         __u32 status = TP_STATUS_USER | stat;
 766
 767         struct tpacket3_hdr *last_pkt;
 768         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 769
 770         if (po->stats.tp_drops)
 771                 status |= TP_STATUS_LOSING;
 772
 773         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 774         last_pkt->tp_next_offset = 0;
 775
 776         /* Get the ts of the last pkt */
 777         if (BLOCK_NUM_PKTS(pbd1)) {
 778                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 779                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 780         } else {
 781                 /* Ok, we tmo'd - so get the current time */
 782                 struct timespec ts;
 783                 getnstimeofday(&ts);
 784                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 785                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 786         }
 787
 788         smp_wmb();
 789
 790         /* Flush the block */
 791         prb_flush_block(pkc1, pbd1, status);
 792
 793         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 794 }
 795
 796 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 797 {
 798         pkc->reset_pending_on_curr_blk = 0;
 799 }
 800
 801 /*
 802  * Side effect of opening a block:
 803  *
 804  * 1) prb_queue is thawed.
 805  * 2) retire_blk_timer is refreshed.
 806  *
 807  */
 808 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 809         struct tpacket_block_desc *pbd1)
 810 {
 811         struct timespec ts;
 812         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 813
 814         smp_rmb();
 815
 816         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd1))) {
 817
 818                 /* We could have just memset this but we will lose the
 819                  * flexibility of making the priv area sticky
 820                  */
 821                 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 822                 BLOCK_NUM_PKTS(pbd1) = 0;
 823                 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 824                 getnstimeofday(&ts);
 825                 h1->ts_first_pkt.ts_sec = ts.tv_sec;
 826                 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 827                 pkc1->pkblk_start = (char *)pbd1;
 828                 pkc1->nxt_offset = (char *)(pkc1->pkblk_start +
 829                 BLK_PLUS_PRIV(pkc1->blk_sizeof_priv));
 830                 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 831                 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 832                 pbd1->version = pkc1->version;
 833                 pkc1->prev = pkc1->nxt_offset;
 834                 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 835                 prb_thaw_queue(pkc1);
 836                 _prb_refresh_rx_retire_blk_timer(pkc1);
 837
 838                 smp_wmb();
 839
 840                 return;
 841         }
 842
 843         WARN(1, "ERROR block:%p is NOT FREE status:%d kactive_blk_num:%d\n",
 844                 pbd1, BLOCK_STATUS(pbd1), pkc1->kactive_blk_num);
 845         dump_stack();
 846         BUG();
 847 }
 848
 849 /*
 850  * Queue freeze logic:
 851  * 1) Assume tp_block_nr = 8 blocks.
 852  * 2) At time 't0', user opens Rx ring.
 853  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 854  * 4) user-space is either sleeping or processing block '0'.
 855  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 856  *    it will close block-7,loop around and try to fill block '0'.
 857  *    call-flow:
 858  *    __packet_lookup_frame_in_block
 859  *      prb_retire_current_block()
 860  *      prb_dispatch_next_block()
 861  *        |->(BLOCK_STATUS == USER) evaluates to true
 862  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 863  * 6) Now there are two cases:
 864  *    6.1) Link goes idle right after the queue is frozen.
 865  *         But remember, the last open_block() refreshed the timer.
 866  *         When this timer expires,it will refresh itself so that we can
 867  *         re-open block-0 in near future.
 868  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 869  *         case and __packet_lookup_frame_in_block will check if block-0
 870  *         is free and can now be re-used.
 871  */
 872 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 873                                   struct packet_sock *po)
 874 {
 875         pkc->reset_pending_on_curr_blk = 1;
 876         po->stats_u.stats3.tp_freeze_q_cnt++;
 877 }
 878
 879 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 880
 881 /*
 882  * If the next block is free then we will dispatch it
 883  * and return a good offset.
 884  * Else, we will freeze the queue.
 885  * So, caller must check the return value.
 886  */
 887 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 888                 struct packet_sock *po)
 889 {
 890         struct tpacket_block_desc *pbd;
 891
 892         smp_rmb();
 893
 894         /* 1. Get current block num */
 895         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 896
 897         /* 2. If this block is currently in_use then freeze the queue */
 898         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 899                 prb_freeze_queue(pkc, po);
 900                 return NULL;
 901         }
 902
 903         /*
 904          * 3.
 905          * open this block and return the offset where the first packet
 906          * needs to get stored.
 907          */
 908         prb_open_block(pkc, pbd);
 909         return (void *)pkc->nxt_offset;
 910 }
 911
 912 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 913                 struct packet_sock *po, unsigned int status)
 914 {
 915         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 916
 917         /* retire/close the current block */
 918         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 919                 /*
 920                  * Plug the case where copy_bits() is in progress on
 921                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 922                  * have space to copy the pkt in the current block and
 923                  * called prb_retire_current_block()
 924                  *
 925                  * We don't need to worry about the TMO case because
 926                  * the timer-handler already handled this case.
 927                  */
 928                 if (!(status & TP_STATUS_BLK_TMO)) {
 929                         while (atomic_read(&pkc->blk_fill_in_prog)) {
 930                                 /* Waiting for skb_copy_bits to finish... */
 931                                 cpu_relax();
 932                         }
 933                 }
 934                 prb_close_block(pkc, pbd, po, status);
 935                 return;
 936         }
 937
 938         WARN(1, "ERROR-pbd[%d]:%p\n", pkc->kactive_blk_num, pbd);
 939         dump_stack();
 940         BUG();
 941 }
 942
 943 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
 944                                       struct tpacket_block_desc *pbd)
 945 {
 946         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 947 }
 948
 949 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 950 {
 951         return pkc->reset_pending_on_curr_blk;
 952 }
 953
 954 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 955 {
 956         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 957         atomic_dec(&pkc->blk_fill_in_prog);
 958 }
 959
 960 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 961                         struct tpacket3_hdr *ppd)
 962 {
 963         ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
 964 }
 965
 966 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 967                         struct tpacket3_hdr *ppd)
 968 {
 969         ppd->hv1.tp_rxhash = 0;
 970 }
 971
 972 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
 973                         struct tpacket3_hdr *ppd)
 974 {
 975         if (vlan_tx_tag_present(pkc->skb)) {
 976                 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
 977                 ppd->tp_status = TP_STATUS_VLAN_VALID;
 978         } else {
 979                 ppd->hv1.tp_vlan_tci = ppd->tp_status = 0;
 980         }
 981 }
 982
 983 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
 984                         struct tpacket3_hdr *ppd)
 985 {
 986         prb_fill_vlan_info(pkc, ppd);
 987
 988         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
 989                 prb_fill_rxhash(pkc, ppd);
 990         else
 991                 prb_clear_rxhash(pkc, ppd);
 992 }
 993
 994 static void prb_fill_curr_block(char *curr,
 995                                 struct tpacket_kbdq_core *pkc,
 996                                 struct tpacket_block_desc *pbd,
 997                                 unsigned int len)
 998 {
 999         struct tpacket3_hdr *ppd;
1000
1001         ppd  = (struct tpacket3_hdr *)curr;
1002         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1003         pkc->prev = curr;
1004         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1005         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1006         BLOCK_NUM_PKTS(pbd) += 1;
1007         atomic_inc(&pkc->blk_fill_in_prog);
1008         prb_run_all_ft_ops(pkc, ppd);
1009 }
1010
1011 /* Assumes caller has the sk->rx_queue.lock */
1012 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1013                                             struct sk_buff *skb,
1014                                                 int status,
1015                                             unsigned int len
1016                                             )
1017 {
1018         struct tpacket_kbdq_core *pkc;
1019         struct tpacket_block_desc *pbd;
1020         char *curr, *end;
1021
1022         pkc = GET_PBDQC_FROM_RB(((struct packet_ring_buffer *)&po->rx_ring));
1023         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1024
1025         /* Queue is frozen when user space is lagging behind */
1026         if (prb_queue_frozen(pkc)) {
1027                 /*
1028                  * Check if that last block which caused the queue to freeze,
1029                  * is still in_use by user-space.
1030                  */
1031                 if (prb_curr_blk_in_use(pkc, pbd)) {
1032                         /* Can't record this packet */
1033                         return NULL;
1034                 } else {
1035                         /*
1036                          * Ok, the block was released by user-space.
1037                          * Now let's open that block.
1038                          * opening a block also thaws the queue.
1039                          * Thawing is a side effect.
1040                          */
1041                         prb_open_block(pkc, pbd);
1042                 }
1043         }
1044
1045         smp_mb();
1046         curr = pkc->nxt_offset;
1047         pkc->skb = skb;
1048         end = (char *) ((char *)pbd + pkc->kblk_size);
1049
1050         /* first try the current block */
1051         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1052                 prb_fill_curr_block(curr, pkc, pbd, len);
1053                 return (void *)curr;
1054         }
1055
1056         /* Ok, close the current block */
1057         prb_retire_current_block(pkc, po, 0);
1058
1059         /* Now, try to dispatch the next block */
1060         curr = (char *)prb_dispatch_next_block(pkc, po);
1061         if (curr) {
1062                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1063                 prb_fill_curr_block(curr, pkc, pbd, len);
1064                 return (void *)curr;
1065         }
1066
1067         /*
1068          * No free blocks are available.user_space hasn't caught up yet.
1069          * Queue was just frozen and now this packet will get dropped.
1070          */
1071         return NULL;
1072 }
1073
1074 static void *packet_current_rx_frame(struct packet_sock *po,
1075                                             struct sk_buff *skb,
1076                                             int status, unsigned int len)
1077 {
1078         char *curr = NULL;
1079         switch (po->tp_version) {
1080         case TPACKET_V1:
1081         case TPACKET_V2:
1082                 curr = packet_lookup_frame(po, &po->rx_ring,
1083                                         po->rx_ring.head, status);
1084                 return curr;
1085         case TPACKET_V3:
1086                 return __packet_lookup_frame_in_block(po, skb, status, len);
1087         default:
1088                 WARN(1, "TPACKET version not supported\n");
1089                 BUG();
1090                 return 0;
1091         }
1092 }
1093
1094 static void *prb_lookup_block(struct packet_sock *po,
1095                                      struct packet_ring_buffer *rb,
1096                                      unsigned int previous,
1097                                      int status)
1098 {
1099         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1100         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, previous);
1101
1102         if (status != BLOCK_STATUS(pbd))
1103                 return NULL;
1104         return pbd;
1105 }
1106
1107 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1108 {
1109         unsigned int prev;
1110         if (rb->prb_bdqc.kactive_blk_num)
1111                 prev = rb->prb_bdqc.kactive_blk_num-1;
1112         else
1113                 prev = rb->prb_bdqc.knum_blocks-1;
1114         return prev;
1115 }
1116
1117 /* Assumes caller has held the rx_queue.lock */
1118 static void *__prb_previous_block(struct packet_sock *po,
1119                                          struct packet_ring_buffer *rb,
1120                                          int status)
1121 {
1122         unsigned int previous = prb_previous_blk_num(rb);
1123         return prb_lookup_block(po, rb, previous, status);
1124 }
1125
1126 static void *packet_previous_rx_frame(struct packet_sock *po,
1127                                              struct packet_ring_buffer *rb,
1128                                              int status)
1129 {
1130         if (po->tp_version <= TPACKET_V2)
1131                 return packet_previous_frame(po, rb, status);
1132
1133         return __prb_previous_block(po, rb, status);
1134 }
1135
1136 static void packet_increment_rx_head(struct packet_sock *po,
1137                                             struct packet_ring_buffer *rb)
1138 {
1139         switch (po->tp_version) {
1140         case TPACKET_V1:
1141         case TPACKET_V2:
1142                 return packet_increment_head(rb);
1143         case TPACKET_V3:
1144         default:
1145                 WARN(1, "TPACKET version not supported.\n");
1146                 BUG();
1147                 return;
1148         }
1149 }
1150
1151 static void *packet_previous_frame(struct packet_sock *po,
1152                 struct packet_ring_buffer *rb,
1153                 int status)
1154 {
1155         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1156         return packet_lookup_frame(po, rb, previous, status);
1157 }
1158
1159 static void packet_increment_head(struct packet_ring_buffer *buff)
1160 {
1161         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1162 }
1163
1164 static void packet_sock_destruct(struct sock *sk)
1165 {
1166         skb_queue_purge(&sk->sk_error_queue);
1167
1168         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1169         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1170
1171         if (!sock_flag(sk, SOCK_DEAD)) {
1172                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1173                 return;
1174         }
1175
1176         sk_refcnt_debug_dec(sk);
1177 }
1178
1179 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1180 {
1181         int x = atomic_read(&f->rr_cur) + 1;
1182
1183         if (x >= num)
1184                 x = 0;
1185
1186         return x;
1187 }
1188
1189 static struct sock *fanout_demux_hash(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1190 {
1191         u32 idx, hash = skb->rxhash;
1192
1193         idx = ((u64)hash * num) >> 32;
1194
1195         return f->arr[idx];
1196 }
1197
1198 static struct sock *fanout_demux_lb(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1199 {
1200         int cur, old;
1201
1202         cur = atomic_read(&f->rr_cur);
1203         while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1204                                      fanout_rr_next(f, num))) != cur)
1205                 cur = old;
1206         return f->arr[cur];
1207 }
1208
1209 static struct sock *fanout_demux_cpu(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1210 {
1211         unsigned int cpu = smp_processor_id();
1212
1213         return f->arr[cpu % num];
1214 }
1215
1216 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1217                              struct packet_type *pt, struct net_device *orig_dev)
1218 {
1219         struct packet_fanout *f = pt->af_packet_priv;
1220         unsigned int num = f->num_members;
1221         struct packet_sock *po;
1222         struct sock *sk;
1223
1224         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1225             !num) {
1226                 kfree_skb(skb);
1227                 return 0;
1228         }
1229
1230         switch (f->type) {
1231         case PACKET_FANOUT_HASH:
1232         default:
1233                 if (f->defrag) {
1234                         skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1235                         if (!skb)
1236                                 return 0;
1237                 }
1238                 skb_get_rxhash(skb);
1239                 sk = fanout_demux_hash(f, skb, num);
1240                 break;
1241         case PACKET_FANOUT_LB:
1242                 sk = fanout_demux_lb(f, skb, num);
1243                 break;
1244         case PACKET_FANOUT_CPU:
1245                 sk = fanout_demux_cpu(f, skb, num);
1246                 break;
1247         }
1248
1249         po = pkt_sk(sk);
1250
1251         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1252 }
1253
1254 static DEFINE_MUTEX(fanout_mutex);
1255 static LIST_HEAD(fanout_list);
1256
1257 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1258 {
1259         struct packet_fanout *f = po->fanout;
1260
1261         spin_lock(&f->lock);
1262         f->arr[f->num_members] = sk;
1263         smp_wmb();
1264         f->num_members++;
1265         spin_unlock(&f->lock);
1266 }
1267
1268 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1269 {
1270         struct packet_fanout *f = po->fanout;
1271         int i;
1272
1273         spin_lock(&f->lock);
1274         for (i = 0; i < f->num_members; i++) {
1275                 if (f->arr[i] == sk)
1276                         break;
1277         }
1278         BUG_ON(i >= f->num_members);
1279         f->arr[i] = f->arr[f->num_members - 1];
1280         f->num_members--;
1281         spin_unlock(&f->lock);
1282 }
1283
1284 bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1285 {
1286         if (ptype->af_packet_priv == (void*)((struct packet_sock *)sk)->fanout)
1287                 return true;
1288
1289         return false;
1290 }
1291
1292 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1293 {
1294         struct packet_sock *po = pkt_sk(sk);
1295         struct packet_fanout *f, *match;
1296         u8 type = type_flags & 0xff;
1297         u8 defrag = (type_flags & PACKET_FANOUT_FLAG_DEFRAG) ? 1 : 0;
1298         int err;
1299
1300         switch (type) {
1301         case PACKET_FANOUT_HASH:
1302         case PACKET_FANOUT_LB:
1303         case PACKET_FANOUT_CPU:
1304                 break;
1305         default:
1306                 return -EINVAL;
1307         }
1308
1309         if (!po->running)
1310                 return -EINVAL;
1311
1312         if (po->fanout)
1313                 return -EALREADY;
1314
1315         mutex_lock(&fanout_mutex);
1316         match = NULL;
1317         list_for_each_entry(f, &fanout_list, list) {
1318                 if (f->id == id &&
1319                     read_pnet(&f->net) == sock_net(sk)) {
1320                         match = f;
1321                         break;
1322                 }
1323         }
1324         err = -EINVAL;
1325         if (match && match->defrag != defrag)
1326                 goto out;
1327         if (!match) {
1328                 err = -ENOMEM;
1329                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1330                 if (!match)
1331                         goto out;
1332                 write_pnet(&match->net, sock_net(sk));
1333                 match->id = id;
1334                 match->type = type;
1335                 match->defrag = defrag;
1336                 atomic_set(&match->rr_cur, 0);
1337                 INIT_LIST_HEAD(&match->list);
1338                 spin_lock_init(&match->lock);
1339                 atomic_set(&match->sk_ref, 0);
1340                 match->prot_hook.type = po->prot_hook.type;
1341                 match->prot_hook.dev = po->prot_hook.dev;
1342                 match->prot_hook.func = packet_rcv_fanout;
1343                 match->prot_hook.af_packet_priv = match;
1344                 match->prot_hook.id_match = match_fanout_group;
1345                 dev_add_pack(&match->prot_hook);
1346                 list_add(&match->list, &fanout_list);
1347         }
1348         err = -EINVAL;
1349         if (match->type == type &&
1350             match->prot_hook.type == po->prot_hook.type &&
1351             match->prot_hook.dev == po->prot_hook.dev) {
1352                 err = -ENOSPC;
1353                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1354                         __dev_remove_pack(&po->prot_hook);
1355                         po->fanout = match;
1356                         atomic_inc(&match->sk_ref);
1357                         __fanout_link(sk, po);
1358                         err = 0;
1359                 }
1360         }
1361 out:
1362         mutex_unlock(&fanout_mutex);
1363         return err;
1364 }
1365
1366 static void fanout_release(struct sock *sk)
1367 {
1368         struct packet_sock *po = pkt_sk(sk);
1369         struct packet_fanout *f;
1370
1371         f = po->fanout;
1372         if (!f)
1373                 return;
1374
1375         po->fanout = NULL;
1376
1377         mutex_lock(&fanout_mutex);
1378         if (atomic_dec_and_test(&f->sk_ref)) {
1379                 list_del(&f->list);
1380                 dev_remove_pack(&f->prot_hook);
1381                 kfree(f);
1382         }
1383         mutex_unlock(&fanout_mutex);
1384 }
1385
1386 static const struct proto_ops packet_ops;
1387
1388 static const struct proto_ops packet_ops_spkt;
1389
1390 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1391                            struct packet_type *pt, struct net_device *orig_dev)
1392 {
1393         struct sock *sk;
1394         struct sockaddr_pkt *spkt;
1395
1396         /*
1397          *      When we registered the protocol we saved the socket in the data
1398          *      field for just this event.
1399          */
1400
1401         sk = pt->af_packet_priv;
1402
1403         /*
1404          *      Yank back the headers [hope the device set this
1405          *      right or kerboom...]
1406          *
1407          *      Incoming packets have ll header pulled,
1408          *      push it back.
1409          *
1410          *      For outgoing ones skb->data == skb_mac_header(skb)
1411          *      so that this procedure is noop.
1412          */
1413
1414         if (skb->pkt_type == PACKET_LOOPBACK)
1415                 goto out;
1416
1417         if (!net_eq(dev_net(dev), sock_net(sk)))
1418                 goto out;
1419
1420         skb = skb_share_check(skb, GFP_ATOMIC);
1421         if (skb == NULL)
1422                 goto oom;
1423
1424         /* drop any routing info */
1425         skb_dst_drop(skb);
1426
1427         /* drop conntrack reference */
1428         nf_reset(skb);
1429
1430         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1431
1432         skb_push(skb, skb->data - skb_mac_header(skb));
1433
1434         /*
1435          *      The SOCK_PACKET socket receives _all_ frames.
1436          */
1437
1438         spkt->spkt_family = dev->type;
1439         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1440         spkt->spkt_protocol = skb->protocol;
1441
1442         /*
1443          *      Charge the memory to the socket. This is done specifically
1444          *      to prevent sockets using all the memory up.
1445          */
1446
1447         if (sock_queue_rcv_skb(sk, skb) == 0)
1448                 return 0;
1449
1450 out:
1451         kfree_skb(skb);
1452 oom:
1453         return 0;
1454 }
1455
1456
1457 /*
1458  *      Output a raw packet to a device layer. This bypasses all the other
1459  *      protocol layers and you must therefore supply it with a complete frame
1460  */
1461
1462 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1463                                struct msghdr *msg, size_t len)
1464 {
1465         struct sock *sk = sock->sk;
1466         struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1467         struct sk_buff *skb = NULL;
1468         struct net_device *dev;
1469         __be16 proto = 0;
1470         int err;
1471
1472         /*
1473          *      Get and verify the address.
1474          */
1475
1476         if (saddr) {
1477                 if (msg->msg_namelen < sizeof(struct sockaddr))
1478                         return -EINVAL;
1479                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1480                         proto = saddr->spkt_protocol;
1481         } else
1482                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1483
1484         /*
1485          *      Find the device first to size check it
1486          */
1487
1488         saddr->spkt_device[13] = 0;
1489 retry:
1490         rcu_read_lock();
1491         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1492         err = -ENODEV;
1493         if (dev == NULL)
1494                 goto out_unlock;
1495
1496         err = -ENETDOWN;
1497         if (!(dev->flags & IFF_UP))
1498                 goto out_unlock;
1499
1500         /*
1501          * You may not queue a frame bigger than the mtu. This is the lowest level
1502          * raw protocol and you must do your own fragmentation at this level.
1503          */
1504
1505         err = -EMSGSIZE;
1506         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN)
1507                 goto out_unlock;
1508
1509         if (!skb) {
1510                 size_t reserved = LL_RESERVED_SPACE(dev);
1511                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1512
1513                 rcu_read_unlock();
1514                 skb = sock_wmalloc(sk, len + reserved, 0, GFP_KERNEL);
1515                 if (skb == NULL)
1516                         return -ENOBUFS;
1517                 /* FIXME: Save some space for broken drivers that write a hard
1518                  * header at transmission time by themselves. PPP is the notable
1519                  * one here. This should really be fixed at the driver level.
1520                  */
1521                 skb_reserve(skb, reserved);
1522                 skb_reset_network_header(skb);
1523
1524                 /* Try to align data part correctly */
1525                 if (hhlen) {
1526                         skb->data -= hhlen;
1527                         skb->tail -= hhlen;
1528                         if (len < hhlen)
1529                                 skb_reset_network_header(skb);
1530                 }
1531                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1532                 if (err)
1533                         goto out_free;
1534                 goto retry;
1535         }
1536
1537         if (len > (dev->mtu + dev->hard_header_len)) {
1538                 /* Earlier code assumed this would be a VLAN pkt,
1539                  * double-check this now that we have the actual
1540                  * packet in hand.
1541                  */
1542                 struct ethhdr *ehdr;
1543                 skb_reset_mac_header(skb);
1544                 ehdr = eth_hdr(skb);
1545                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1546                         err = -EMSGSIZE;
1547                         goto out_unlock;
1548                 }
1549         }
1550
1551         skb->protocol = proto;
1552         skb->dev = dev;
1553         skb->priority = sk->sk_priority;
1554         skb->mark = sk->sk_mark;
1555         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1556         if (err < 0)
1557                 goto out_unlock;
1558
1559         dev_queue_xmit(skb);
1560         rcu_read_unlock();
1561         return len;
1562
1563 out_unlock:
1564         rcu_read_unlock();
1565 out_free:
1566         kfree_skb(skb);
1567         return err;
1568 }
1569
1570 static unsigned int run_filter(const struct sk_buff *skb,
1571                                       const struct sock *sk,
1572                                       unsigned int res)
1573 {
1574         struct sk_filter *filter;
1575
1576         rcu_read_lock();
1577         filter = rcu_dereference(sk->sk_filter);
1578         if (filter != NULL)
1579                 res = SK_RUN_FILTER(filter, skb);
1580         rcu_read_unlock();
1581
1582         return res;
1583 }
1584
1585 /*
1586  * This function makes lazy skb cloning in hope that most of packets
1587  * are discarded by BPF.
1588  *
1589  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1590  * and skb->cb are mangled. It works because (and until) packets
1591  * falling here are owned by current CPU. Output packets are cloned
1592  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1593  * sequencially, so that if we return skb to original state on exit,
1594  * we will not harm anyone.
1595  */
1596
1597 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1598                       struct packet_type *pt, struct net_device *orig_dev)
1599 {
1600         struct sock *sk;
1601         struct sockaddr_ll *sll;
1602         struct packet_sock *po;
1603         u8 *skb_head = skb->data;
1604         int skb_len = skb->len;
1605         unsigned int snaplen, res;
1606
1607         if (skb->pkt_type == PACKET_LOOPBACK)
1608                 goto drop;
1609
1610         sk = pt->af_packet_priv;
1611         po = pkt_sk(sk);
1612
1613         if (!net_eq(dev_net(dev), sock_net(sk)))
1614                 goto drop;
1615
1616         skb->dev = dev;
1617
1618         if (dev->header_ops) {
1619                 /* The device has an explicit notion of ll header,
1620                  * exported to higher levels.
1621                  *
1622                  * Otherwise, the device hides details of its frame
1623                  * structure, so that corresponding packet head is
1624                  * never delivered to user.
1625                  */
1626                 if (sk->sk_type != SOCK_DGRAM)
1627                         skb_push(skb, skb->data - skb_mac_header(skb));
1628                 else if (skb->pkt_type == PACKET_OUTGOING) {
1629                         /* Special case: outgoing packets have ll header at head */
1630                         skb_pull(skb, skb_network_offset(skb));
1631                 }
1632         }
1633
1634         snaplen = skb->len;
1635
1636         res = run_filter(skb, sk, snaplen);
1637         if (!res)
1638                 goto drop_n_restore;
1639         if (snaplen > res)
1640                 snaplen = res;
1641
1642         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1643                 goto drop_n_acct;
1644
1645         if (skb_shared(skb)) {
1646                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1647                 if (nskb == NULL)
1648                         goto drop_n_acct;
1649
1650                 if (skb_head != skb->data) {
1651                         skb->data = skb_head;
1652                         skb->len = skb_len;
1653                 }
1654                 kfree_skb(skb);
1655                 skb = nskb;
1656         }
1657
1658         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1659                      sizeof(skb->cb));
1660
1661         sll = &PACKET_SKB_CB(skb)->sa.ll;
1662         sll->sll_family = AF_PACKET;
1663         sll->sll_hatype = dev->type;
1664         sll->sll_protocol = skb->protocol;
1665         sll->sll_pkttype = skb->pkt_type;
1666         if (unlikely(po->origdev))
1667                 sll->sll_ifindex = orig_dev->ifindex;
1668         else
1669                 sll->sll_ifindex = dev->ifindex;
1670
1671         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1672
1673         PACKET_SKB_CB(skb)->origlen = skb->len;
1674
1675         if (pskb_trim(skb, snaplen))
1676                 goto drop_n_acct;
1677
1678         skb_set_owner_r(skb, sk);
1679         skb->dev = NULL;
1680         skb_dst_drop(skb);
1681
1682         /* drop conntrack reference */
1683         nf_reset(skb);
1684
1685         spin_lock(&sk->sk_receive_queue.lock);
1686         po->stats.tp_packets++;
1687         skb->dropcount = atomic_read(&sk->sk_drops);
1688         __skb_queue_tail(&sk->sk_receive_queue, skb);
1689         spin_unlock(&sk->sk_receive_queue.lock);
1690         sk->sk_data_ready(sk, skb->len);
1691         return 0;
1692
1693 drop_n_acct:
1694         spin_lock(&sk->sk_receive_queue.lock);
1695         po->stats.tp_drops++;
1696         atomic_inc(&sk->sk_drops);
1697         spin_unlock(&sk->sk_receive_queue.lock);
1698
1699 drop_n_restore:
1700         if (skb_head != skb->data && skb_shared(skb)) {
1701                 skb->data = skb_head;
1702                 skb->len = skb_len;
1703         }
1704 drop:
1705         consume_skb(skb);
1706         return 0;
1707 }
1708
1709 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1710                        struct packet_type *pt, struct net_device *orig_dev)
1711 {
1712         struct sock *sk;
1713         struct packet_sock *po;
1714         struct sockaddr_ll *sll;
1715         union {
1716                 struct tpacket_hdr *h1;
1717                 struct tpacket2_hdr *h2;
1718                 struct tpacket3_hdr *h3;
1719                 void *raw;
1720         } h;
1721         u8 *skb_head = skb->data;
1722         int skb_len = skb->len;
1723         unsigned int snaplen, res;
1724         unsigned long status = TP_STATUS_USER;
1725         unsigned short macoff, netoff, hdrlen;
1726         struct sk_buff *copy_skb = NULL;
1727         struct timeval tv;
1728         struct timespec ts;
1729         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1730
1731         if (skb->pkt_type == PACKET_LOOPBACK)
1732                 goto drop;
1733
1734         sk = pt->af_packet_priv;
1735         po = pkt_sk(sk);
1736
1737         if (!net_eq(dev_net(dev), sock_net(sk)))
1738                 goto drop;
1739
1740         if (dev->header_ops) {
1741                 if (sk->sk_type != SOCK_DGRAM)
1742                         skb_push(skb, skb->data - skb_mac_header(skb));
1743                 else if (skb->pkt_type == PACKET_OUTGOING) {
1744                         /* Special case: outgoing packets have ll header at head */
1745                         skb_pull(skb, skb_network_offset(skb));
1746                 }
1747         }
1748
1749         if (skb->ip_summed == CHECKSUM_PARTIAL)
1750                 status |= TP_STATUS_CSUMNOTREADY;
1751
1752         snaplen = skb->len;
1753
1754         res = run_filter(skb, sk, snaplen);
1755         if (!res)
1756                 goto drop_n_restore;
1757         if (snaplen > res)
1758                 snaplen = res;
1759
1760         if (sk->sk_type == SOCK_DGRAM) {
1761                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1762                                   po->tp_reserve;
1763         } else {
1764                 unsigned maclen = skb_network_offset(skb);
1765                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1766                                        (maclen < 16 ? 16 : maclen)) +
1767                         po->tp_reserve;
1768                 macoff = netoff - maclen;
1769         }
1770         if (po->tp_version <= TPACKET_V2) {
1771                 if (macoff + snaplen > po->rx_ring.frame_size) {
1772                         if (po->copy_thresh &&
1773                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1774                                 if (skb_shared(skb)) {
1775                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1776                                 } else {
1777                                         copy_skb = skb_get(skb);
1778                                         skb_head = skb->data;
1779                                 }
1780                                 if (copy_skb)
1781                                         skb_set_owner_r(copy_skb, sk);
1782                         }
1783                         snaplen = po->rx_ring.frame_size - macoff;
1784                         if ((int)snaplen < 0)
1785                                 snaplen = 0;
1786                 }
1787         }
1788         spin_lock(&sk->sk_receive_queue.lock);
1789         h.raw = packet_current_rx_frame(po, skb,
1790                                         TP_STATUS_KERNEL, (macoff+snaplen));
1791         if (!h.raw)
1792                 goto ring_is_full;
1793         if (po->tp_version <= TPACKET_V2) {
1794                 packet_increment_rx_head(po, &po->rx_ring);
1795         /*
1796          * LOSING will be reported till you read the stats,
1797          * because it's COR - Clear On Read.
1798          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1799          * at packet level.
1800          */
1801                 if (po->stats.tp_drops)
1802                         status |= TP_STATUS_LOSING;
1803         }
1804         po->stats.tp_packets++;
1805         if (copy_skb) {
1806                 status |= TP_STATUS_COPY;
1807                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1808         }
1809         spin_unlock(&sk->sk_receive_queue.lock);
1810
1811         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1812
1813         switch (po->tp_version) {
1814         case TPACKET_V1:
1815                 h.h1->tp_len = skb->len;
1816                 h.h1->tp_snaplen = snaplen;
1817                 h.h1->tp_mac = macoff;
1818                 h.h1->tp_net = netoff;
1819                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1820                                 && shhwtstamps->syststamp.tv64)
1821                         tv = ktime_to_timeval(shhwtstamps->syststamp);
1822                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1823                                 && shhwtstamps->hwtstamp.tv64)
1824                         tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1825                 else if (skb->tstamp.tv64)
1826                         tv = ktime_to_timeval(skb->tstamp);
1827                 else
1828                         do_gettimeofday(&tv);
1829                 h.h1->tp_sec = tv.tv_sec;
1830                 h.h1->tp_usec = tv.tv_usec;
1831                 hdrlen = sizeof(*h.h1);
1832                 break;
1833         case TPACKET_V2:
1834                 h.h2->tp_len = skb->len;
1835                 h.h2->tp_snaplen = snaplen;
1836                 h.h2->tp_mac = macoff;
1837                 h.h2->tp_net = netoff;
1838                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1839                                 && shhwtstamps->syststamp.tv64)
1840                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1841                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1842                                 && shhwtstamps->hwtstamp.tv64)
1843                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1844                 else if (skb->tstamp.tv64)
1845                         ts = ktime_to_timespec(skb->tstamp);
1846                 else
1847                         getnstimeofday(&ts);
1848                 h.h2->tp_sec = ts.tv_sec;
1849                 h.h2->tp_nsec = ts.tv_nsec;
1850                 if (vlan_tx_tag_present(skb)) {
1851                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1852                         status |= TP_STATUS_VLAN_VALID;
1853                 } else {
1854                         h.h2->tp_vlan_tci = 0;
1855                 }
1856                 h.h2->tp_padding = 0;
1857                 hdrlen = sizeof(*h.h2);
1858                 break;
1859         case TPACKET_V3:
1860                 /* tp_nxt_offset,vlan are already populated above.
1861                  * So DONT clear those fields here
1862                  */
1863                 h.h3->tp_status |= status;
1864                 h.h3->tp_len = skb->len;
1865                 h.h3->tp_snaplen = snaplen;
1866                 h.h3->tp_mac = macoff;
1867                 h.h3->tp_net = netoff;
1868                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1869                                 && shhwtstamps->syststamp.tv64)
1870                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1871                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1872                                 && shhwtstamps->hwtstamp.tv64)
1873                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1874                 else if (skb->tstamp.tv64)
1875                         ts = ktime_to_timespec(skb->tstamp);
1876                 else
1877                         getnstimeofday(&ts);
1878                 h.h3->tp_sec  = ts.tv_sec;
1879                 h.h3->tp_nsec = ts.tv_nsec;
1880                 hdrlen = sizeof(*h.h3);
1881                 break;
1882         default:
1883                 BUG();
1884         }
1885
1886         sll = h.raw + TPACKET_ALIGN(hdrlen);
1887         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1888         sll->sll_family = AF_PACKET;
1889         sll->sll_hatype = dev->type;
1890         sll->sll_protocol = skb->protocol;
1891         sll->sll_pkttype = skb->pkt_type;
1892         if (unlikely(po->origdev))
1893                 sll->sll_ifindex = orig_dev->ifindex;
1894         else
1895                 sll->sll_ifindex = dev->ifindex;
1896
1897         smp_mb();
1898 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1899         {
1900                 u8 *start, *end;
1901
1902                 if (po->tp_version <= TPACKET_V2) {
1903                         end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1904                                 + macoff + snaplen);
1905                         for (start = h.raw; start < end; start += PAGE_SIZE)
1906                                 flush_dcache_page(pgv_to_page(start));
1907                 }
1908                 smp_wmb();
1909         }
1910 #endif
1911         if (po->tp_version <= TPACKET_V2)
1912                 __packet_set_status(po, h.raw, status);
1913         else
1914                 prb_clear_blk_fill_status(&po->rx_ring);
1915
1916         sk->sk_data_ready(sk, 0);
1917
1918 drop_n_restore:
1919         if (skb_head != skb->data && skb_shared(skb)) {
1920                 skb->data = skb_head;
1921                 skb->len = skb_len;
1922         }
1923 drop:
1924         kfree_skb(skb);
1925         return 0;
1926
1927 ring_is_full:
1928         po->stats.tp_drops++;
1929         spin_unlock(&sk->sk_receive_queue.lock);
1930
1931         sk->sk_data_ready(sk, 0);
1932         kfree_skb(copy_skb);
1933         goto drop_n_restore;
1934 }
1935
1936 static void tpacket_destruct_skb(struct sk_buff *skb)
1937 {
1938         struct packet_sock *po = pkt_sk(skb->sk);
1939         void *ph;
1940
1941         if (likely(po->tx_ring.pg_vec)) {
1942                 ph = skb_shinfo(skb)->destructor_arg;
1943                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1944                 atomic_dec(&po->tx_ring.pending);
1945                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1946         }
1947
1948         sock_wfree(skb);
1949 }
1950
1951 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1952                 void *frame, struct net_device *dev, int size_max,
1953                 __be16 proto, unsigned char *addr)
1954 {
1955         union {
1956                 struct tpacket_hdr *h1;
1957                 struct tpacket2_hdr *h2;
1958                 void *raw;
1959         } ph;
1960         int to_write, offset, len, tp_len, nr_frags, len_max;
1961         struct socket *sock = po->sk.sk_socket;
1962         struct page *page;
1963         void *data;
1964         int err;
1965
1966         ph.raw = frame;
1967
1968         skb->protocol = proto;
1969         skb->dev = dev;
1970         skb->priority = po->sk.sk_priority;
1971         skb->mark = po->sk.sk_mark;
1972         skb_shinfo(skb)->destructor_arg = ph.raw;
1973
1974         switch (po->tp_version) {
1975         case TPACKET_V2:
1976                 tp_len = ph.h2->tp_len;
1977                 break;
1978         default:
1979                 tp_len = ph.h1->tp_len;
1980                 break;
1981         }
1982         if (unlikely(tp_len > size_max)) {
1983                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1984                 return -EMSGSIZE;
1985         }
1986
1987         skb_reserve(skb, LL_RESERVED_SPACE(dev));
1988         skb_reset_network_header(skb);
1989
1990         data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1991         to_write = tp_len;
1992
1993         if (sock->type == SOCK_DGRAM) {
1994                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
1995                                 NULL, tp_len);
1996                 if (unlikely(err < 0))
1997                         return -EINVAL;
1998         } else if (dev->hard_header_len) {
1999                 /* net device doesn't like empty head */
2000                 if (unlikely(tp_len <= dev->hard_header_len)) {
2001                         pr_err("packet size is too short (%d < %d)\n",
2002                                tp_len, dev->hard_header_len);
2003                         return -EINVAL;
2004                 }
2005
2006                 skb_push(skb, dev->hard_header_len);
2007                 err = skb_store_bits(skb, 0, data,
2008                                 dev->hard_header_len);
2009                 if (unlikely(err))
2010                         return err;
2011
2012                 data += dev->hard_header_len;
2013                 to_write -= dev->hard_header_len;
2014         }
2015
2016         err = -EFAULT;
2017         offset = offset_in_page(data);
2018         len_max = PAGE_SIZE - offset;
2019         len = ((to_write > len_max) ? len_max : to_write);
2020
2021         skb->data_len = to_write;
2022         skb->len += to_write;
2023         skb->truesize += to_write;
2024         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2025
2026         while (likely(to_write)) {
2027                 nr_frags = skb_shinfo(skb)->nr_frags;
2028
2029                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2030                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2031                                MAX_SKB_FRAGS);
2032                         return -EFAULT;
2033                 }
2034
2035                 page = pgv_to_page(data);
2036                 data += len;
2037                 flush_dcache_page(page);
2038                 get_page(page);
2039                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2040                 to_write -= len;
2041                 offset = 0;
2042                 len_max = PAGE_SIZE;
2043                 len = ((to_write > len_max) ? len_max : to_write);
2044         }
2045
2046         return tp_len;
2047 }
2048
2049 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2050 {
2051         struct sk_buff *skb;
2052         struct net_device *dev;
2053         __be16 proto;
2054         bool need_rls_dev = false;
2055         int err, reserve = 0;
2056         void *ph;
2057         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2058         int tp_len, size_max;
2059         unsigned char *addr;
2060         int len_sum = 0;
2061         int status = 0;
2062
2063         mutex_lock(&po->pg_vec_lock);
2064
2065         err = -EBUSY;
2066         if (saddr == NULL) {
2067                 dev = po->prot_hook.dev;
2068                 proto   = po->num;
2069                 addr    = NULL;
2070         } else {
2071                 err = -EINVAL;
2072                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2073                         goto out;
2074                 if (msg->msg_namelen < (saddr->sll_halen
2075                                         + offsetof(struct sockaddr_ll,
2076                                                 sll_addr)))
2077                         goto out;
2078                 proto   = saddr->sll_protocol;
2079                 addr    = saddr->sll_addr;
2080                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2081                 need_rls_dev = true;
2082         }
2083
2084         err = -ENXIO;
2085         if (unlikely(dev == NULL))
2086                 goto out;
2087
2088         reserve = dev->hard_header_len;
2089
2090         err = -ENETDOWN;
2091         if (unlikely(!(dev->flags & IFF_UP)))
2092                 goto out_put;
2093
2094         size_max = po->tx_ring.frame_size
2095                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2096
2097         if (size_max > dev->mtu + reserve)
2098                 size_max = dev->mtu + reserve;
2099
2100         do {
2101                 ph = packet_current_frame(po, &po->tx_ring,
2102                                 TP_STATUS_SEND_REQUEST);
2103
2104                 if (unlikely(ph == NULL)) {
2105                         schedule();
2106                         continue;
2107                 }
2108
2109                 status = TP_STATUS_SEND_REQUEST;
2110                 skb = sock_alloc_send_skb(&po->sk,
2111                                 LL_ALLOCATED_SPACE(dev)
2112                                 + sizeof(struct sockaddr_ll),
2113                                 0, &err);
2114
2115                 if (unlikely(skb == NULL))
2116                         goto out_status;
2117
2118                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2119                                 addr);
2120
2121                 if (unlikely(tp_len < 0)) {
2122                         if (po->tp_loss) {
2123                                 __packet_set_status(po, ph,
2124                                                 TP_STATUS_AVAILABLE);
2125                                 packet_increment_head(&po->tx_ring);
2126                                 kfree_skb(skb);
2127                                 continue;
2128                         } else {
2129                                 status = TP_STATUS_WRONG_FORMAT;
2130                                 err = tp_len;
2131                                 goto out_status;
2132                         }
2133                 }
2134
2135                 skb->destructor = tpacket_destruct_skb;
2136                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2137                 atomic_inc(&po->tx_ring.pending);
2138
2139                 status = TP_STATUS_SEND_REQUEST;
2140                 err = dev_queue_xmit(skb);
2141                 if (unlikely(err > 0)) {
2142                         err = net_xmit_errno(err);
2143                         if (err && __packet_get_status(po, ph) ==
2144                                    TP_STATUS_AVAILABLE) {
2145                                 /* skb was destructed already */
2146                                 skb = NULL;
2147                                 goto out_status;
2148                         }
2149                         /*
2150                          * skb was dropped but not destructed yet;
2151                          * let's treat it like congestion or err < 0
2152                          */
2153                         err = 0;
2154                 }
2155                 packet_increment_head(&po->tx_ring);
2156                 len_sum += tp_len;
2157         } while (likely((ph != NULL) ||
2158                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2159                          (atomic_read(&po->tx_ring.pending))))
2160                 );
2161
2162         err = len_sum;
2163         goto out_put;
2164
2165 out_status:
2166         __packet_set_status(po, ph, status);
2167         kfree_skb(skb);
2168 out_put:
2169         if (need_rls_dev)
2170                 dev_put(dev);
2171 out:
2172         mutex_unlock(&po->pg_vec_lock);
2173         return err;
2174 }
2175
2176 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2177                                         size_t reserve, size_t len,
2178                                         size_t linear, int noblock,
2179                                         int *err)
2180 {
2181         struct sk_buff *skb;
2182
2183         /* Under a page?  Don't bother with paged skb. */
2184         if (prepad + len < PAGE_SIZE || !linear)
2185                 linear = len;
2186
2187         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2188                                    err);
2189         if (!skb)
2190                 return NULL;
2191
2192         skb_reserve(skb, reserve);
2193         skb_put(skb, linear);
2194         skb->data_len = len - linear;
2195         skb->len += len - linear;
2196
2197         return skb;
2198 }
2199
2200 static int packet_snd(struct socket *sock,
2201                           struct msghdr *msg, size_t len)
2202 {
2203         struct sock *sk = sock->sk;
2204         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2205         struct sk_buff *skb;
2206         struct net_device *dev;
2207         __be16 proto;
2208         bool need_rls_dev = false;
2209         unsigned char *addr;
2210         int err, reserve = 0;
2211         struct virtio_net_hdr vnet_hdr = { 0 };
2212         int offset = 0;
2213         int vnet_hdr_len;
2214         struct packet_sock *po = pkt_sk(sk);
2215         unsigned short gso_type = 0;
2216
2217         /*
2218          *      Get and verify the address.
2219          */
2220
2221         if (saddr == NULL) {
2222                 dev = po->prot_hook.dev;
2223                 proto   = po->num;
2224                 addr    = NULL;
2225         } else {
2226                 err = -EINVAL;
2227                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2228                         goto out;
2229                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2230                         goto out;
2231                 proto   = saddr->sll_protocol;
2232                 addr    = saddr->sll_addr;
2233                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2234                 need_rls_dev = true;
2235         }
2236
2237         err = -ENXIO;
2238         if (dev == NULL)
2239                 goto out_unlock;
2240         if (sock->type == SOCK_RAW)
2241                 reserve = dev->hard_header_len;
2242
2243         err = -ENETDOWN;
2244         if (!(dev->flags & IFF_UP))
2245                 goto out_unlock;
2246
2247         if (po->has_vnet_hdr) {
2248                 vnet_hdr_len = sizeof(vnet_hdr);
2249
2250                 err = -EINVAL;
2251                 if (len < vnet_hdr_len)
2252                         goto out_unlock;
2253
2254                 len -= vnet_hdr_len;
2255
2256                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2257                                        vnet_hdr_len);
2258                 if (err < 0)
2259                         goto out_unlock;
2260
2261                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2262                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2263                       vnet_hdr.hdr_len))
2264                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
2265                                                  vnet_hdr.csum_offset + 2;
2266
2267                 err = -EINVAL;
2268                 if (vnet_hdr.hdr_len > len)
2269                         goto out_unlock;
2270
2271                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2272                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2273                         case VIRTIO_NET_HDR_GSO_TCPV4:
2274                                 gso_type = SKB_GSO_TCPV4;
2275                                 break;
2276                         case VIRTIO_NET_HDR_GSO_TCPV6:
2277                                 gso_type = SKB_GSO_TCPV6;
2278                                 break;
2279                         case VIRTIO_NET_HDR_GSO_UDP:
2280                                 gso_type = SKB_GSO_UDP;
2281                                 break;
2282                         default:
2283                                 goto out_unlock;
2284                         }
2285
2286                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2287                                 gso_type |= SKB_GSO_TCP_ECN;
2288
2289                         if (vnet_hdr.gso_size == 0)
2290                                 goto out_unlock;
2291
2292                 }
2293         }
2294
2295         err = -EMSGSIZE;
2296         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN))
2297                 goto out_unlock;
2298
2299         err = -ENOBUFS;
2300         skb = packet_alloc_skb(sk, LL_ALLOCATED_SPACE(dev),
2301                                LL_RESERVED_SPACE(dev), len, vnet_hdr.hdr_len,
2302                                msg->msg_flags & MSG_DONTWAIT, &err);
2303         if (skb == NULL)
2304                 goto out_unlock;
2305
2306         skb_set_network_header(skb, reserve);
2307
2308         err = -EINVAL;
2309         if (sock->type == SOCK_DGRAM &&
2310             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2311                 goto out_free;
2312
2313         /* Returns -EFAULT on error */
2314         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2315         if (err)
2316                 goto out_free;
2317         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2318         if (err < 0)
2319                 goto out_free;
2320
2321         if (!gso_type && (len > dev->mtu + reserve)) {
2322                 /* Earlier code assumed this would be a VLAN pkt,
2323                  * double-check this now that we have the actual
2324                  * packet in hand.
2325                  */
2326                 struct ethhdr *ehdr;
2327                 skb_reset_mac_header(skb);
2328                 ehdr = eth_hdr(skb);
2329                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2330                         err = -EMSGSIZE;
2331                         goto out_free;
2332                 }
2333         }
2334
2335         skb->protocol = proto;
2336         skb->dev = dev;
2337         skb->priority = sk->sk_priority;
2338         skb->mark = sk->sk_mark;
2339
2340         if (po->has_vnet_hdr) {
2341                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2342                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2343                                                   vnet_hdr.csum_offset)) {
2344                                 err = -EINVAL;
2345                                 goto out_free;
2346                         }
2347                 }
2348
2349                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2350                 skb_shinfo(skb)->gso_type = gso_type;
2351
2352                 /* Header must be checked, and gso_segs computed. */
2353                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2354                 skb_shinfo(skb)->gso_segs = 0;
2355
2356                 len += vnet_hdr_len;
2357         }
2358
2359         /*
2360          *      Now send it
2361          */
2362
2363         err = dev_queue_xmit(skb);
2364         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2365                 goto out_unlock;
2366
2367         if (need_rls_dev)
2368                 dev_put(dev);
2369
2370         return len;
2371
2372 out_free:
2373         kfree_skb(skb);
2374 out_unlock:
2375         if (dev && need_rls_dev)
2376                 dev_put(dev);
2377 out:
2378         return err;
2379 }
2380
2381 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2382                 struct msghdr *msg, size_t len)
2383 {
2384         struct sock *sk = sock->sk;
2385         struct packet_sock *po = pkt_sk(sk);
2386         if (po->tx_ring.pg_vec)
2387                 return tpacket_snd(po, msg);
2388         else
2389                 return packet_snd(sock, msg, len);
2390 }
2391
2392 /*
2393  *      Close a PACKET socket. This is fairly simple. We immediately go
2394  *      to 'closed' state and remove our protocol entry in the device list.
2395  */
2396
2397 static int packet_release(struct socket *sock)
2398 {
2399         struct sock *sk = sock->sk;
2400         struct packet_sock *po;
2401         struct net *net;
2402         union tpacket_req_u req_u;
2403
2404         if (!sk)
2405                 return 0;
2406
2407         net = sock_net(sk);
2408         po = pkt_sk(sk);
2409
2410         spin_lock_bh(&net->packet.sklist_lock);
2411         sk_del_node_init_rcu(sk);
2412         sock_prot_inuse_add(net, sk->sk_prot, -1);
2413         spin_unlock_bh(&net->packet.sklist_lock);
2414
2415         spin_lock(&po->bind_lock);
2416         unregister_prot_hook(sk, false);
2417         if (po->prot_hook.dev) {
2418                 dev_put(po->prot_hook.dev);
2419                 po->prot_hook.dev = NULL;
2420         }
2421         spin_unlock(&po->bind_lock);
2422
2423         packet_flush_mclist(sk);
2424
2425         if (po->rx_ring.pg_vec) {
2426                 memset(&req_u, 0, sizeof(req_u));
2427                 packet_set_ring(sk, &req_u, 1, 0);
2428         }
2429
2430         if (po->tx_ring.pg_vec) {
2431                 memset(&req_u, 0, sizeof(req_u));
2432                 packet_set_ring(sk, &req_u, 1, 1);
2433         }
2434
2435         fanout_release(sk);
2436
2437         synchronize_net();
2438         /*
2439          *      Now the socket is dead. No more input will appear.
2440          */
2441         sock_orphan(sk);
2442         sock->sk = NULL;
2443
2444         /* Purge queues */
2445
2446         skb_queue_purge(&sk->sk_receive_queue);
2447         sk_refcnt_debug_release(sk);
2448
2449         sock_put(sk);
2450         return 0;
2451 }
2452
2453 /*
2454  *      Attach a packet hook.
2455  */
2456
2457 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2458 {
2459         struct packet_sock *po = pkt_sk(sk);
2460
2461         if (po->fanout) {
2462                 if (dev)
2463                         dev_put(dev);
2464
2465                 return -EINVAL;
2466         }
2467
2468         lock_sock(sk);
2469
2470         spin_lock(&po->bind_lock);
2471         unregister_prot_hook(sk, true);
2472         po->num = protocol;
2473         po->prot_hook.type = protocol;
2474         if (po->prot_hook.dev)
2475                 dev_put(po->prot_hook.dev);
2476         po->prot_hook.dev = dev;
2477
2478         po->ifindex = dev ? dev->ifindex : 0;
2479
2480         if (protocol == 0)
2481                 goto out_unlock;
2482
2483         if (!dev || (dev->flags & IFF_UP)) {
2484                 register_prot_hook(sk);
2485         } else {
2486                 sk->sk_err = ENETDOWN;
2487                 if (!sock_flag(sk, SOCK_DEAD))
2488                         sk->sk_error_report(sk);
2489         }
2490
2491 out_unlock:
2492         spin_unlock(&po->bind_lock);
2493         release_sock(sk);
2494         return 0;
2495 }
2496
2497 /*
2498  *      Bind a packet socket to a device
2499  */
2500
2501 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2502                             int addr_len)
2503 {
2504         struct sock *sk = sock->sk;
2505         char name[15];
2506         struct net_device *dev;
2507         int err = -ENODEV;
2508
2509         /*
2510          *      Check legality
2511          */
2512
2513         if (addr_len != sizeof(struct sockaddr))
2514                 return -EINVAL;
2515         strlcpy(name, uaddr->sa_data, sizeof(name));
2516
2517         dev = dev_get_by_name(sock_net(sk), name);
2518         if (dev)
2519                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2520         return err;
2521 }
2522
2523 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2524 {
2525         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2526         struct sock *sk = sock->sk;
2527         struct net_device *dev = NULL;
2528         int err;
2529
2530
2531         /*
2532          *      Check legality
2533          */
2534
2535         if (addr_len < sizeof(struct sockaddr_ll))
2536                 return -EINVAL;
2537         if (sll->sll_family != AF_PACKET)
2538                 return -EINVAL;
2539
2540         if (sll->sll_ifindex) {
2541                 err = -ENODEV;
2542                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2543                 if (dev == NULL)
2544                         goto out;
2545         }
2546         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2547
2548 out:
2549         return err;
2550 }
2551
2552 static struct proto packet_proto = {
2553         .name     = "PACKET",
2554         .owner    = THIS_MODULE,
2555         .obj_size = sizeof(struct packet_sock),
2556 };
2557
2558 /*
2559  *      Create a packet of type SOCK_PACKET.
2560  */
2561
2562 static int packet_create(struct net *net, struct socket *sock, int protocol,
2563                          int kern)
2564 {
2565         struct sock *sk;
2566         struct packet_sock *po;
2567         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2568         int err;
2569
2570         if (!capable(CAP_NET_RAW))
2571                 return -EPERM;
2572         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2573             sock->type != SOCK_PACKET)
2574                 return -ESOCKTNOSUPPORT;
2575
2576         sock->state = SS_UNCONNECTED;
2577
2578         err = -ENOBUFS;
2579         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2580         if (sk == NULL)
2581                 goto out;
2582
2583         sock->ops = &packet_ops;
2584         if (sock->type == SOCK_PACKET)
2585                 sock->ops = &packet_ops_spkt;
2586
2587         sock_init_data(sock, sk);
2588
2589         po = pkt_sk(sk);
2590         sk->sk_family = PF_PACKET;
2591         po->num = proto;
2592
2593         sk->sk_destruct = packet_sock_destruct;
2594         sk_refcnt_debug_inc(sk);
2595
2596         /*
2597          *      Attach a protocol block
2598          */
2599
2600         spin_lock_init(&po->bind_lock);
2601         mutex_init(&po->pg_vec_lock);
2602         po->prot_hook.func = packet_rcv;
2603
2604         if (sock->type == SOCK_PACKET)
2605                 po->prot_hook.func = packet_rcv_spkt;
2606
2607         po->prot_hook.af_packet_priv = sk;
2608
2609         if (proto) {
2610                 po->prot_hook.type = proto;
2611                 register_prot_hook(sk);
2612         }
2613
2614         spin_lock_bh(&net->packet.sklist_lock);
2615         sk_add_node_rcu(sk, &net->packet.sklist);
2616         sock_prot_inuse_add(net, &packet_proto, 1);
2617         spin_unlock_bh(&net->packet.sklist_lock);
2618
2619         return 0;
2620 out:
2621         return err;
2622 }
2623
2624 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2625 {
2626         struct sock_exterr_skb *serr;
2627         struct sk_buff *skb, *skb2;
2628         int copied, err;
2629
2630         err = -EAGAIN;
2631         skb = skb_dequeue(&sk->sk_error_queue);
2632         if (skb == NULL)
2633                 goto out;
2634
2635         copied = skb->len;
2636         if (copied > len) {
2637                 msg->msg_flags |= MSG_TRUNC;
2638                 copied = len;
2639         }
2640         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2641         if (err)
2642                 goto out_free_skb;
2643
2644         sock_recv_timestamp(msg, sk, skb);
2645
2646         serr = SKB_EXT_ERR(skb);
2647         put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2648                  sizeof(serr->ee), &serr->ee);
2649
2650         msg->msg_flags |= MSG_ERRQUEUE;
2651         err = copied;
2652
2653         /* Reset and regenerate socket error */
2654         spin_lock_bh(&sk->sk_error_queue.lock);
2655         sk->sk_err = 0;
2656         if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2657                 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2658                 spin_unlock_bh(&sk->sk_error_queue.lock);
2659                 sk->sk_error_report(sk);
2660         } else
2661                 spin_unlock_bh(&sk->sk_error_queue.lock);
2662
2663 out_free_skb:
2664         kfree_skb(skb);
2665 out:
2666         return err;
2667 }
2668
2669 /*
2670  *      Pull a packet from our receive queue and hand it to the user.
2671  *      If necessary we block.
2672  */
2673
2674 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2675                           struct msghdr *msg, size_t len, int flags)
2676 {
2677         struct sock *sk = sock->sk;
2678         struct sk_buff *skb;
2679         int copied, err;
2680         struct sockaddr_ll *sll;
2681         int vnet_hdr_len = 0;
2682
2683         err = -EINVAL;
2684         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2685                 goto out;
2686
2687 #if 0
2688         /* What error should we return now? EUNATTACH? */
2689         if (pkt_sk(sk)->ifindex < 0)
2690                 return -ENODEV;
2691 #endif
2692
2693         if (flags & MSG_ERRQUEUE) {
2694                 err = packet_recv_error(sk, msg, len);
2695                 goto out;
2696         }
2697
2698         /*
2699          *      Call the generic datagram receiver. This handles all sorts
2700          *      of horrible races and re-entrancy so we can forget about it
2701          *      in the protocol layers.
2702          *
2703          *      Now it will return ENETDOWN, if device have just gone down,
2704          *      but then it will block.
2705          */
2706
2707         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2708
2709         /*
2710          *      An error occurred so return it. Because skb_recv_datagram()
2711          *      handles the blocking we don't see and worry about blocking
2712          *      retries.
2713          */
2714
2715         if (skb == NULL)
2716                 goto out;
2717
2718         if (pkt_sk(sk)->has_vnet_hdr) {
2719                 struct virtio_net_hdr vnet_hdr = { 0 };
2720
2721                 err = -EINVAL;
2722                 vnet_hdr_len = sizeof(vnet_hdr);
2723                 if (len < vnet_hdr_len)
2724                         goto out_free;
2725
2726                 len -= vnet_hdr_len;
2727
2728                 if (skb_is_gso(skb)) {
2729                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2730
2731                         /* This is a hint as to how much should be linear. */
2732                         vnet_hdr.hdr_len = skb_headlen(skb);
2733                         vnet_hdr.gso_size = sinfo->gso_size;
2734                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2735                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2736                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2737                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2738                         else if (sinfo->gso_type & SKB_GSO_UDP)
2739                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2740                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2741                                 goto out_free;
2742                         else
2743                                 BUG();
2744                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2745                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2746                 } else
2747                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2748
2749                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2750                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2751                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2752                         vnet_hdr.csum_offset = skb->csum_offset;
2753                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2754                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2755                 } /* else everything is zero */
2756
2757                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2758                                      vnet_hdr_len);
2759                 if (err < 0)
2760                         goto out_free;
2761         }
2762
2763         /*
2764          *      If the address length field is there to be filled in, we fill
2765          *      it in now.
2766          */
2767
2768         sll = &PACKET_SKB_CB(skb)->sa.ll;
2769         if (sock->type == SOCK_PACKET)
2770                 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2771         else
2772                 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2773
2774         /*
2775          *      You lose any data beyond the buffer you gave. If it worries a
2776          *      user program they can ask the device for its MTU anyway.
2777          */
2778
2779         copied = skb->len;
2780         if (copied > len) {
2781                 copied = len;
2782                 msg->msg_flags |= MSG_TRUNC;
2783         }
2784
2785         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2786         if (err)
2787                 goto out_free;
2788
2789         sock_recv_ts_and_drops(msg, sk, skb);
2790
2791         if (msg->msg_name)
2792                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2793                        msg->msg_namelen);
2794
2795         if (pkt_sk(sk)->auxdata) {
2796                 struct tpacket_auxdata aux;
2797
2798                 aux.tp_status = TP_STATUS_USER;
2799                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2800                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2801                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2802                 aux.tp_snaplen = skb->len;
2803                 aux.tp_mac = 0;
2804                 aux.tp_net = skb_network_offset(skb);
2805                 if (vlan_tx_tag_present(skb)) {
2806                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2807                         aux.tp_status |= TP_STATUS_VLAN_VALID;
2808                 } else {
2809                         aux.tp_vlan_tci = 0;
2810                 }
2811                 aux.tp_padding = 0;
2812                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2813         }
2814
2815         /*
2816          *      Free or return the buffer as appropriate. Again this
2817          *      hides all the races and re-entrancy issues from us.
2818          */
2819         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2820
2821 out_free:
2822         skb_free_datagram(sk, skb);
2823 out:
2824         return err;
2825 }
2826
2827 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2828                                int *uaddr_len, int peer)
2829 {
2830         struct net_device *dev;
2831         struct sock *sk = sock->sk;
2832
2833         if (peer)
2834                 return -EOPNOTSUPP;
2835
2836         uaddr->sa_family = AF_PACKET;
2837         rcu_read_lock();
2838         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2839         if (dev)
2840                 strncpy(uaddr->sa_data, dev->name, 14);
2841         else
2842                 memset(uaddr->sa_data, 0, 14);
2843         rcu_read_unlock();
2844         *uaddr_len = sizeof(*uaddr);
2845
2846         return 0;
2847 }
2848
2849 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2850                           int *uaddr_len, int peer)
2851 {
2852         struct net_device *dev;
2853         struct sock *sk = sock->sk;
2854         struct packet_sock *po = pkt_sk(sk);
2855         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2856
2857         if (peer)
2858                 return -EOPNOTSUPP;
2859
2860         sll->sll_family = AF_PACKET;
2861         sll->sll_ifindex = po->ifindex;
2862         sll->sll_protocol = po->num;
2863         sll->sll_pkttype = 0;
2864         rcu_read_lock();
2865         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2866         if (dev) {
2867                 sll->sll_hatype = dev->type;
2868                 sll->sll_halen = dev->addr_len;
2869                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2870         } else {
2871                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
2872                 sll->sll_halen = 0;
2873         }
2874         rcu_read_unlock();
2875         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2876
2877         return 0;
2878 }
2879
2880 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2881                          int what)
2882 {
2883         switch (i->type) {
2884         case PACKET_MR_MULTICAST:
2885                 if (i->alen != dev->addr_len)
2886                         return -EINVAL;
2887                 if (what > 0)
2888                         return dev_mc_add(dev, i->addr);
2889                 else
2890                         return dev_mc_del(dev, i->addr);
2891                 break;
2892         case PACKET_MR_PROMISC:
2893                 return dev_set_promiscuity(dev, what);
2894                 break;
2895         case PACKET_MR_ALLMULTI:
2896                 return dev_set_allmulti(dev, what);
2897                 break;
2898         case PACKET_MR_UNICAST:
2899                 if (i->alen != dev->addr_len)
2900                         return -EINVAL;
2901                 if (what > 0)
2902                         return dev_uc_add(dev, i->addr);
2903                 else
2904                         return dev_uc_del(dev, i->addr);
2905                 break;
2906         default:
2907                 break;
2908         }
2909         return 0;
2910 }
2911
2912 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2913 {
2914         for ( ; i; i = i->next) {
2915                 if (i->ifindex == dev->ifindex)
2916                         packet_dev_mc(dev, i, what);
2917         }
2918 }
2919
2920 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2921 {
2922         struct packet_sock *po = pkt_sk(sk);
2923         struct packet_mclist *ml, *i;
2924         struct net_device *dev;
2925         int err;
2926
2927         rtnl_lock();
2928
2929         err = -ENODEV;
2930         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2931         if (!dev)
2932                 goto done;
2933
2934         err = -EINVAL;
2935         if (mreq->mr_alen > dev->addr_len)
2936                 goto done;
2937
2938         err = -ENOBUFS;
2939         i = kmalloc(sizeof(*i), GFP_KERNEL);
2940         if (i == NULL)
2941                 goto done;
2942
2943         err = 0;
2944         for (ml = po->mclist; ml; ml = ml->next) {
2945                 if (ml->ifindex == mreq->mr_ifindex &&
2946                     ml->type == mreq->mr_type &&
2947                     ml->alen == mreq->mr_alen &&
2948                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2949                         ml->count++;
2950                         /* Free the new element ... */
2951                         kfree(i);
2952                         goto done;
2953                 }
2954         }
2955
2956         i->type = mreq->mr_type;
2957         i->ifindex = mreq->mr_ifindex;
2958         i->alen = mreq->mr_alen;
2959         memcpy(i->addr, mreq->mr_address, i->alen);
2960         i->count = 1;
2961         i->next = po->mclist;
2962         po->mclist = i;
2963         err = packet_dev_mc(dev, i, 1);
2964         if (err) {
2965                 po->mclist = i->next;
2966                 kfree(i);
2967         }
2968
2969 done:
2970         rtnl_unlock();
2971         return err;
2972 }
2973
2974 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2975 {
2976         struct packet_mclist *ml, **mlp;
2977
2978         rtnl_lock();
2979
2980         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
2981                 if (ml->ifindex == mreq->mr_ifindex &&
2982                     ml->type == mreq->mr_type &&
2983                     ml->alen == mreq->mr_alen &&
2984                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2985                         if (--ml->count == 0) {
2986                                 struct net_device *dev;
2987                                 *mlp = ml->next;
2988                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2989                                 if (dev)
2990                                         packet_dev_mc(dev, ml, -1);
2991                                 kfree(ml);
2992                         }
2993                         rtnl_unlock();
2994                         return 0;
2995                 }
2996         }
2997         rtnl_unlock();
2998         return -EADDRNOTAVAIL;
2999 }
3000
3001 static void packet_flush_mclist(struct sock *sk)
3002 {
3003         struct packet_sock *po = pkt_sk(sk);
3004         struct packet_mclist *ml;
3005
3006         if (!po->mclist)
3007                 return;
3008
3009         rtnl_lock();
3010         while ((ml = po->mclist) != NULL) {
3011                 struct net_device *dev;
3012
3013                 po->mclist = ml->next;
3014                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3015                 if (dev != NULL)
3016                         packet_dev_mc(dev, ml, -1);
3017                 kfree(ml);
3018         }
3019         rtnl_unlock();
3020 }
3021
3022 static int
3023 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3024 {
3025         struct sock *sk = sock->sk;
3026         struct packet_sock *po = pkt_sk(sk);
3027         int ret;
3028
3029         if (level != SOL_PACKET)
3030                 return -ENOPROTOOPT;
3031
3032         switch (optname) {
3033         case PACKET_ADD_MEMBERSHIP:
3034         case PACKET_DROP_MEMBERSHIP:
3035         {
3036                 struct packet_mreq_max mreq;
3037                 int len = optlen;
3038                 memset(&mreq, 0, sizeof(mreq));
3039                 if (len < sizeof(struct packet_mreq))
3040                         return -EINVAL;
3041                 if (len > sizeof(mreq))
3042                         len = sizeof(mreq);
3043                 if (copy_from_user(&mreq, optval, len))
3044                         return -EFAULT;
3045                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3046                         return -EINVAL;
3047                 if (optname == PACKET_ADD_MEMBERSHIP)
3048                         ret = packet_mc_add(sk, &mreq);
3049                 else
3050                         ret = packet_mc_drop(sk, &mreq);
3051                 return ret;
3052         }
3053
3054         case PACKET_RX_RING:
3055         case PACKET_TX_RING:
3056         {
3057                 union tpacket_req_u req_u;
3058                 int len;
3059
3060                 switch (po->tp_version) {
3061                 case TPACKET_V1:
3062                 case TPACKET_V2:
3063                         len = sizeof(req_u.req);
3064                         break;
3065                 case TPACKET_V3:
3066                 default:
3067                         len = sizeof(req_u.req3);
3068                         break;
3069                 }
3070                 if (optlen < len)
3071                         return -EINVAL;
3072                 if (pkt_sk(sk)->has_vnet_hdr)
3073                         return -EINVAL;
3074                 if (copy_from_user(&req_u.req, optval, len))
3075                         return -EFAULT;
3076                 return packet_set_ring(sk, &req_u, 0,
3077                         optname == PACKET_TX_RING);
3078         }
3079         case PACKET_COPY_THRESH:
3080         {
3081                 int val;
3082
3083                 if (optlen != sizeof(val))
3084                         return -EINVAL;
3085                 if (copy_from_user(&val, optval, sizeof(val)))
3086                         return -EFAULT;
3087
3088                 pkt_sk(sk)->copy_thresh = val;
3089                 return 0;
3090         }
3091         case PACKET_VERSION:
3092         {
3093                 int val;
3094
3095                 if (optlen != sizeof(val))
3096                         return -EINVAL;
3097                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3098                         return -EBUSY;
3099                 if (copy_from_user(&val, optval, sizeof(val)))
3100                         return -EFAULT;
3101                 switch (val) {
3102                 case TPACKET_V1:
3103                 case TPACKET_V2:
3104                 case TPACKET_V3:
3105                         po->tp_version = val;
3106                         return 0;
3107                 default:
3108                         return -EINVAL;
3109                 }
3110         }
3111         case PACKET_RESERVE:
3112         {
3113                 unsigned int val;
3114
3115                 if (optlen != sizeof(val))
3116                         return -EINVAL;
3117                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3118                         return -EBUSY;
3119                 if (copy_from_user(&val, optval, sizeof(val)))
3120                         return -EFAULT;
3121                 po->tp_reserve = val;
3122                 return 0;
3123         }
3124         case PACKET_LOSS:
3125         {
3126                 unsigned int val;
3127
3128                 if (optlen != sizeof(val))
3129                         return -EINVAL;
3130                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3131                         return -EBUSY;
3132                 if (copy_from_user(&val, optval, sizeof(val)))
3133                         return -EFAULT;
3134                 po->tp_loss = !!val;
3135                 return 0;
3136         }
3137         case PACKET_AUXDATA:
3138         {
3139                 int val;
3140
3141                 if (optlen < sizeof(val))
3142                         return -EINVAL;
3143                 if (copy_from_user(&val, optval, sizeof(val)))
3144                         return -EFAULT;
3145
3146                 po->auxdata = !!val;
3147                 return 0;
3148         }
3149         case PACKET_ORIGDEV:
3150         {
3151                 int val;
3152
3153                 if (optlen < sizeof(val))
3154                         return -EINVAL;
3155                 if (copy_from_user(&val, optval, sizeof(val)))
3156                         return -EFAULT;
3157
3158                 po->origdev = !!val;
3159                 return 0;
3160         }
3161         case PACKET_VNET_HDR:
3162         {
3163                 int val;
3164
3165                 if (sock->type != SOCK_RAW)
3166                         return -EINVAL;
3167                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3168                         return -EBUSY;
3169                 if (optlen < sizeof(val))
3170                         return -EINVAL;
3171                 if (copy_from_user(&val, optval, sizeof(val)))
3172                         return -EFAULT;
3173
3174                 po->has_vnet_hdr = !!val;
3175                 return 0;
3176         }
3177         case PACKET_TIMESTAMP:
3178         {
3179                 int val;
3180
3181                 if (optlen != sizeof(val))
3182                         return -EINVAL;
3183                 if (copy_from_user(&val, optval, sizeof(val)))
3184                         return -EFAULT;
3185
3186                 po->tp_tstamp = val;
3187                 return 0;
3188         }
3189         case PACKET_FANOUT:
3190         {
3191                 int val;
3192
3193                 if (optlen != sizeof(val))
3194                         return -EINVAL;
3195                 if (copy_from_user(&val, optval, sizeof(val)))
3196                         return -EFAULT;
3197
3198                 return fanout_add(sk, val & 0xffff, val >> 16);
3199         }
3200         default:
3201                 return -ENOPROTOOPT;
3202         }
3203 }
3204
3205 static int packet_getsockopt(struct socket *sock, int level, int optname,
3206                              char __user *optval, int __user *optlen)
3207 {
3208         int len;
3209         int val;
3210         struct sock *sk = sock->sk;
3211         struct packet_sock *po = pkt_sk(sk);
3212         void *data;
3213         struct tpacket_stats st;
3214         union tpacket_stats_u st_u;
3215
3216         if (level != SOL_PACKET)
3217                 return -ENOPROTOOPT;
3218
3219         if (get_user(len, optlen))
3220                 return -EFAULT;
3221
3222         if (len < 0)
3223                 return -EINVAL;
3224
3225         switch (optname) {
3226         case PACKET_STATISTICS:
3227                 if (po->tp_version == TPACKET_V3) {
3228                         len = sizeof(struct tpacket_stats_v3);
3229                 } else {
3230                         if (len > sizeof(struct tpacket_stats))
3231                                 len = sizeof(struct tpacket_stats);
3232                 }
3233                 spin_lock_bh(&sk->sk_receive_queue.lock);
3234                 if (po->tp_version == TPACKET_V3) {
3235                         memcpy(&st_u.stats3, &po->stats,
3236                         sizeof(struct tpacket_stats));
3237                         st_u.stats3.tp_freeze_q_cnt =
3238                         po->stats_u.stats3.tp_freeze_q_cnt;
3239                         st_u.stats3.tp_packets += po->stats.tp_drops;
3240                         data = &st_u.stats3;
3241                 } else {
3242                         st = po->stats;
3243                         st.tp_packets += st.tp_drops;
3244                         data = &st;
3245                 }
3246                 memset(&po->stats, 0, sizeof(st));
3247                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3248                 break;
3249         case PACKET_AUXDATA:
3250                 if (len > sizeof(int))
3251                         len = sizeof(int);
3252                 val = po->auxdata;
3253
3254                 data = &val;
3255                 break;
3256         case PACKET_ORIGDEV:
3257                 if (len > sizeof(int))
3258                         len = sizeof(int);
3259                 val = po->origdev;
3260
3261                 data = &val;
3262                 break;
3263         case PACKET_VNET_HDR:
3264                 if (len > sizeof(int))
3265                         len = sizeof(int);
3266                 val = po->has_vnet_hdr;
3267
3268                 data = &val;
3269                 break;
3270         case PACKET_VERSION:
3271                 if (len > sizeof(int))
3272                         len = sizeof(int);
3273                 val = po->tp_version;
3274                 data = &val;
3275                 break;
3276         case PACKET_HDRLEN:
3277                 if (len > sizeof(int))
3278                         len = sizeof(int);
3279                 if (copy_from_user(&val, optval, len))
3280                         return -EFAULT;
3281                 switch (val) {
3282                 case TPACKET_V1:
3283                         val = sizeof(struct tpacket_hdr);
3284                         break;
3285                 case TPACKET_V2:
3286                         val = sizeof(struct tpacket2_hdr);
3287                         break;
3288                 case TPACKET_V3:
3289                         val = sizeof(struct tpacket3_hdr);
3290                         break;
3291                 default:
3292                         return -EINVAL;
3293                 }
3294                 data = &val;
3295                 break;
3296         case PACKET_RESERVE:
3297                 if (len > sizeof(unsigned int))
3298                         len = sizeof(unsigned int);
3299                 val = po->tp_reserve;
3300                 data = &val;
3301                 break;
3302         case PACKET_LOSS:
3303                 if (len > sizeof(unsigned int))
3304                         len = sizeof(unsigned int);
3305                 val = po->tp_loss;
3306                 data = &val;
3307                 break;
3308         case PACKET_TIMESTAMP:
3309                 if (len > sizeof(int))
3310                         len = sizeof(int);
3311                 val = po->tp_tstamp;
3312                 data = &val;
3313                 break;
3314         case PACKET_FANOUT:
3315                 if (len > sizeof(int))
3316                         len = sizeof(int);
3317                 val = (po->fanout ?
3318                        ((u32)po->fanout->id |
3319                         ((u32)po->fanout->type << 16)) :
3320                        0);
3321                 data = &val;
3322                 break;
3323         default:
3324                 return -ENOPROTOOPT;
3325         }
3326
3327         if (put_user(len, optlen))
3328                 return -EFAULT;
3329         if (copy_to_user(optval, data, len))
3330                 return -EFAULT;
3331         return 0;
3332 }
3333
3334
3335 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3336 {
3337         struct sock *sk;
3338         struct hlist_node *node;
3339         struct net_device *dev = data;
3340         struct net *net = dev_net(dev);
3341
3342         rcu_read_lock();
3343         sk_for_each_rcu(sk, node, &net->packet.sklist) {
3344                 struct packet_sock *po = pkt_sk(sk);
3345
3346                 switch (msg) {
3347                 case NETDEV_UNREGISTER:
3348                         if (po->mclist)
3349                                 packet_dev_mclist(dev, po->mclist, -1);
3350                         /* fallthrough */
3351
3352                 case NETDEV_DOWN:
3353                         if (dev->ifindex == po->ifindex) {
3354                                 spin_lock(&po->bind_lock);
3355                                 if (po->running) {
3356                                         __unregister_prot_hook(sk, false);
3357                                         sk->sk_err = ENETDOWN;
3358                                         if (!sock_flag(sk, SOCK_DEAD))
3359                                                 sk->sk_error_report(sk);
3360                                 }
3361                                 if (msg == NETDEV_UNREGISTER) {
3362                                         po->ifindex = -1;
3363                                         if (po->prot_hook.dev)
3364                                                 dev_put(po->prot_hook.dev);
3365                                         po->prot_hook.dev = NULL;
3366                                 }
3367                                 spin_unlock(&po->bind_lock);
3368                         }
3369                         break;
3370                 case NETDEV_UP:
3371                         if (dev->ifindex == po->ifindex) {
3372                                 spin_lock(&po->bind_lock);
3373                                 if (po->num)
3374                                         register_prot_hook(sk);
3375                                 spin_unlock(&po->bind_lock);
3376                         }
3377                         break;
3378                 }
3379         }
3380         rcu_read_unlock();
3381         return NOTIFY_DONE;
3382 }
3383
3384
3385 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3386                         unsigned long arg)
3387 {
3388         struct sock *sk = sock->sk;
3389
3390         switch (cmd) {
3391         case SIOCOUTQ:
3392         {
3393                 int amount = sk_wmem_alloc_get(sk);
3394
3395                 return put_user(amount, (int __user *)arg);
3396         }
3397         case SIOCINQ:
3398         {
3399                 struct sk_buff *skb;
3400                 int amount = 0;
3401
3402                 spin_lock_bh(&sk->sk_receive_queue.lock);
3403                 skb = skb_peek(&sk->sk_receive_queue);
3404                 if (skb)
3405                         amount = skb->len;
3406                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3407                 return put_user(amount, (int __user *)arg);
3408         }
3409         case SIOCGSTAMP:
3410                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3411         case SIOCGSTAMPNS:
3412                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3413
3414 #ifdef CONFIG_INET
3415         case SIOCADDRT:
3416         case SIOCDELRT:
3417         case SIOCDARP:
3418         case SIOCGARP:
3419         case SIOCSARP:
3420         case SIOCGIFADDR:
3421         case SIOCSIFADDR:
3422         case SIOCGIFBRDADDR:
3423         case SIOCSIFBRDADDR:
3424         case SIOCGIFNETMASK:
3425         case SIOCSIFNETMASK:
3426         case SIOCGIFDSTADDR:
3427         case SIOCSIFDSTADDR:
3428         case SIOCSIFFLAGS:
3429                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3430 #endif
3431
3432         default:
3433                 return -ENOIOCTLCMD;
3434         }
3435         return 0;
3436 }
3437
3438 static unsigned int packet_poll(struct file *file, struct socket *sock,
3439                                 poll_table *wait)
3440 {
3441         struct sock *sk = sock->sk;
3442         struct packet_sock *po = pkt_sk(sk);
3443         unsigned int mask = datagram_poll(file, sock, wait);
3444
3445         spin_lock_bh(&sk->sk_receive_queue.lock);
3446         if (po->rx_ring.pg_vec) {
3447                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3448                         TP_STATUS_KERNEL))
3449                         mask |= POLLIN | POLLRDNORM;
3450         }
3451         spin_unlock_bh(&sk->sk_receive_queue.lock);
3452         spin_lock_bh(&sk->sk_write_queue.lock);
3453         if (po->tx_ring.pg_vec) {
3454                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3455                         mask |= POLLOUT | POLLWRNORM;
3456         }
3457         spin_unlock_bh(&sk->sk_write_queue.lock);
3458         return mask;
3459 }
3460
3461
3462 /* Dirty? Well, I still did not learn better way to account
3463  * for user mmaps.
3464  */
3465
3466 static void packet_mm_open(struct vm_area_struct *vma)
3467 {
3468         struct file *file = vma->vm_file;
3469         struct socket *sock = file->private_data;
3470         struct sock *sk = sock->sk;
3471
3472         if (sk)
3473                 atomic_inc(&pkt_sk(sk)->mapped);
3474 }
3475
3476 static void packet_mm_close(struct vm_area_struct *vma)
3477 {
3478         struct file *file = vma->vm_file;
3479         struct socket *sock = file->private_data;
3480         struct sock *sk = sock->sk;
3481
3482         if (sk)
3483                 atomic_dec(&pkt_sk(sk)->mapped);
3484 }
3485
3486 static const struct vm_operations_struct packet_mmap_ops = {
3487         .open   =       packet_mm_open,
3488         .close  =       packet_mm_close,
3489 };
3490
3491 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3492                         unsigned int len)
3493 {
3494         int i;
3495
3496         for (i = 0; i < len; i++) {
3497                 if (likely(pg_vec[i].buffer)) {
3498                         if (is_vmalloc_addr(pg_vec[i].buffer))
3499                                 vfree(pg_vec[i].buffer);
3500                         else
3501                                 free_pages((unsigned long)pg_vec[i].buffer,
3502                                            order);
3503                         pg_vec[i].buffer = NULL;
3504                 }
3505         }
3506         kfree(pg_vec);
3507 }
3508
3509 static char *alloc_one_pg_vec_page(unsigned long order)
3510 {
3511         char *buffer = NULL;
3512         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3513                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3514
3515         buffer = (char *) __get_free_pages(gfp_flags, order);
3516
3517         if (buffer)
3518                 return buffer;
3519
3520         /*
3521          * __get_free_pages failed, fall back to vmalloc
3522          */
3523         buffer = vzalloc((1 << order) * PAGE_SIZE);
3524
3525         if (buffer)
3526                 return buffer;
3527
3528         /*
3529          * vmalloc failed, lets dig into swap here
3530          */
3531         gfp_flags &= ~__GFP_NORETRY;
3532         buffer = (char *)__get_free_pages(gfp_flags, order);
3533         if (buffer)
3534                 return buffer;
3535
3536         /*
3537          * complete and utter failure
3538          */
3539         return NULL;
3540 }
3541
3542 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3543 {
3544         unsigned int block_nr = req->tp_block_nr;
3545         struct pgv *pg_vec;
3546         int i;
3547
3548         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3549         if (unlikely(!pg_vec))
3550                 goto out;
3551
3552         for (i = 0; i < block_nr; i++) {
3553                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3554                 if (unlikely(!pg_vec[i].buffer))
3555                         goto out_free_pgvec;
3556         }
3557
3558 out:
3559         return pg_vec;
3560
3561 out_free_pgvec:
3562         free_pg_vec(pg_vec, order, block_nr);
3563         pg_vec = NULL;
3564         goto out;
3565 }
3566
3567 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3568                 int closing, int tx_ring)
3569 {
3570         struct pgv *pg_vec = NULL;
3571         struct packet_sock *po = pkt_sk(sk);
3572         int was_running, order = 0;
3573         struct packet_ring_buffer *rb;
3574         struct sk_buff_head *rb_queue;
3575         __be16 num;
3576         int err = -EINVAL;
3577         /* Added to avoid minimal code churn */
3578         struct tpacket_req *req = &req_u->req;
3579
3580         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3581         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3582                 WARN(1, "Tx-ring is not supported.\n");
3583                 goto out;
3584         }
3585
3586         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3587         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3588
3589         err = -EBUSY;
3590         if (!closing) {
3591                 if (atomic_read(&po->mapped))
3592                         goto out;
3593                 if (atomic_read(&rb->pending))
3594                         goto out;
3595         }
3596
3597         if (req->tp_block_nr) {
3598                 /* Sanity tests and some calculations */
3599                 err = -EBUSY;
3600                 if (unlikely(rb->pg_vec))
3601                         goto out;
3602
3603                 switch (po->tp_version) {
3604                 case TPACKET_V1:
3605                         po->tp_hdrlen = TPACKET_HDRLEN;
3606                         break;
3607                 case TPACKET_V2:
3608                         po->tp_hdrlen = TPACKET2_HDRLEN;
3609                         break;
3610                 case TPACKET_V3:
3611                         po->tp_hdrlen = TPACKET3_HDRLEN;
3612                         break;
3613                 }
3614
3615                 err = -EINVAL;
3616                 if (unlikely((int)req->tp_block_size <= 0))
3617                         goto out;
3618                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3619                         goto out;
3620                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3621                                         po->tp_reserve))
3622                         goto out;
3623                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3624                         goto out;
3625
3626                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3627                 if (unlikely(rb->frames_per_block <= 0))
3628                         goto out;
3629                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3630                                         req->tp_frame_nr))
3631                         goto out;
3632
3633                 err = -ENOMEM;
3634                 order = get_order(req->tp_block_size);
3635                 pg_vec = alloc_pg_vec(req, order);
3636                 if (unlikely(!pg_vec))
3637                         goto out;
3638                 switch (po->tp_version) {
3639                 case TPACKET_V3:
3640                 /* Transmit path is not supported. We checked
3641                  * it above but just being paranoid
3642                  */
3643                         if (!tx_ring)
3644                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3645                                 break;
3646                 default:
3647                         break;
3648                 }
3649         }
3650         /* Done */
3651         else {
3652                 err = -EINVAL;
3653                 if (unlikely(req->tp_frame_nr))
3654                         goto out;
3655         }
3656
3657         lock_sock(sk);
3658
3659         /* Detach socket from network */
3660         spin_lock(&po->bind_lock);
3661         was_running = po->running;
3662         num = po->num;
3663         if (was_running) {
3664                 po->num = 0;
3665                 __unregister_prot_hook(sk, false);
3666         }
3667         spin_unlock(&po->bind_lock);
3668
3669         synchronize_net();
3670
3671         err = -EBUSY;
3672         mutex_lock(&po->pg_vec_lock);
3673         if (closing || atomic_read(&po->mapped) == 0) {
3674                 err = 0;
3675                 spin_lock_bh(&rb_queue->lock);
3676                 swap(rb->pg_vec, pg_vec);
3677                 rb->frame_max = (req->tp_frame_nr - 1);
3678                 rb->head = 0;
3679                 rb->frame_size = req->tp_frame_size;
3680                 spin_unlock_bh(&rb_queue->lock);
3681
3682                 swap(rb->pg_vec_order, order);
3683                 swap(rb->pg_vec_len, req->tp_block_nr);
3684
3685                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3686                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3687                                                 tpacket_rcv : packet_rcv;
3688                 skb_queue_purge(rb_queue);
3689                 if (atomic_read(&po->mapped))
3690                         pr_err("packet_mmap: vma is busy: %d\n",
3691                                atomic_read(&po->mapped));
3692         }
3693         mutex_unlock(&po->pg_vec_lock);
3694
3695         spin_lock(&po->bind_lock);
3696         if (was_running) {
3697                 po->num = num;
3698                 register_prot_hook(sk);
3699         }
3700         spin_unlock(&po->bind_lock);
3701         if (closing && (po->tp_version > TPACKET_V2)) {
3702                 /* Because we don't support block-based V3 on tx-ring */
3703                 if (!tx_ring)
3704                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3705         }
3706         release_sock(sk);
3707
3708         if (pg_vec)
3709                 free_pg_vec(pg_vec, order, req->tp_block_nr);
3710 out:
3711         return err;
3712 }
3713
3714 static int packet_mmap(struct file *file, struct socket *sock,
3715                 struct vm_area_struct *vma)
3716 {
3717         struct sock *sk = sock->sk;
3718         struct packet_sock *po = pkt_sk(sk);
3719         unsigned long size, expected_size;
3720         struct packet_ring_buffer *rb;
3721         unsigned long start;
3722         int err = -EINVAL;
3723         int i;
3724
3725         if (vma->vm_pgoff)
3726                 return -EINVAL;
3727
3728         mutex_lock(&po->pg_vec_lock);
3729
3730         expected_size = 0;
3731         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3732                 if (rb->pg_vec) {
3733                         expected_size += rb->pg_vec_len
3734                                                 * rb->pg_vec_pages
3735                                                 * PAGE_SIZE;
3736                 }
3737         }
3738
3739         if (expected_size == 0)
3740                 goto out;
3741
3742         size = vma->vm_end - vma->vm_start;
3743         if (size != expected_size)
3744                 goto out;
3745
3746         start = vma->vm_start;
3747         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3748                 if (rb->pg_vec == NULL)
3749                         continue;
3750
3751                 for (i = 0; i < rb->pg_vec_len; i++) {
3752                         struct page *page;
3753                         void *kaddr = rb->pg_vec[i].buffer;
3754                         int pg_num;
3755
3756                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3757                                 page = pgv_to_page(kaddr);
3758                                 err = vm_insert_page(vma, start, page);
3759                                 if (unlikely(err))
3760                                         goto out;
3761                                 start += PAGE_SIZE;
3762                                 kaddr += PAGE_SIZE;
3763                         }
3764                 }
3765         }
3766
3767         atomic_inc(&po->mapped);
3768         vma->vm_ops = &packet_mmap_ops;
3769         err = 0;
3770
3771 out:
3772         mutex_unlock(&po->pg_vec_lock);
3773         return err;
3774 }
3775
3776 static const struct proto_ops packet_ops_spkt = {
3777         .family =       PF_PACKET,
3778         .owner =        THIS_MODULE,
3779         .release =      packet_release,
3780         .bind =         packet_bind_spkt,
3781         .connect =      sock_no_connect,
3782         .socketpair =   sock_no_socketpair,
3783         .accept =       sock_no_accept,
3784         .getname =      packet_getname_spkt,
3785         .poll =         datagram_poll,
3786         .ioctl =        packet_ioctl,
3787         .listen =       sock_no_listen,
3788         .shutdown =     sock_no_shutdown,
3789         .setsockopt =   sock_no_setsockopt,
3790         .getsockopt =   sock_no_getsockopt,
3791         .sendmsg =      packet_sendmsg_spkt,
3792         .recvmsg =      packet_recvmsg,
3793         .mmap =         sock_no_mmap,
3794         .sendpage =     sock_no_sendpage,
3795 };
3796
3797 static const struct proto_ops packet_ops = {
3798         .family =       PF_PACKET,
3799         .owner =        THIS_MODULE,
3800         .release =      packet_release,
3801         .bind =         packet_bind,
3802         .connect =      sock_no_connect,
3803         .socketpair =   sock_no_socketpair,
3804         .accept =       sock_no_accept,
3805         .getname =      packet_getname,
3806         .poll =         packet_poll,
3807         .ioctl =        packet_ioctl,
3808         .listen =       sock_no_listen,
3809         .shutdown =     sock_no_shutdown,
3810         .setsockopt =   packet_setsockopt,
3811         .getsockopt =   packet_getsockopt,
3812         .sendmsg =      packet_sendmsg,
3813         .recvmsg =      packet_recvmsg,
3814         .mmap =         packet_mmap,
3815         .sendpage =     sock_no_sendpage,
3816 };
3817
3818 static const struct net_proto_family packet_family_ops = {
3819         .family =       PF_PACKET,
3820         .create =       packet_create,
3821         .owner  =       THIS_MODULE,
3822 };
3823
3824 static struct notifier_block packet_netdev_notifier = {
3825         .notifier_call =        packet_notifier,
3826 };
3827
3828 #ifdef CONFIG_PROC_FS
3829
3830 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3831         __acquires(RCU)
3832 {
3833         struct net *net = seq_file_net(seq);
3834
3835         rcu_read_lock();
3836         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3837 }
3838
3839 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3840 {
3841         struct net *net = seq_file_net(seq);
3842         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3843 }
3844
3845 static void packet_seq_stop(struct seq_file *seq, void *v)
3846         __releases(RCU)
3847 {
3848         rcu_read_unlock();
3849 }
3850
3851 static int packet_seq_show(struct seq_file *seq, void *v)
3852 {
3853         if (v == SEQ_START_TOKEN)
3854                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
3855         else {
3856                 struct sock *s = sk_entry(v);
3857                 const struct packet_sock *po = pkt_sk(s);
3858
3859                 seq_printf(seq,
3860                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
3861                            s,
3862                            atomic_read(&s->sk_refcnt),
3863                            s->sk_type,
3864                            ntohs(po->num),
3865                            po->ifindex,
3866                            po->running,
3867                            atomic_read(&s->sk_rmem_alloc),
3868                            sock_i_uid(s),
3869                            sock_i_ino(s));
3870         }
3871
3872         return 0;
3873 }
3874
3875 static const struct seq_operations packet_seq_ops = {
3876         .start  = packet_seq_start,
3877         .next   = packet_seq_next,
3878         .stop   = packet_seq_stop,
3879         .show   = packet_seq_show,
3880 };
3881
3882 static int packet_seq_open(struct inode *inode, struct file *file)
3883 {
3884         return seq_open_net(inode, file, &packet_seq_ops,
3885                             sizeof(struct seq_net_private));
3886 }
3887
3888 static const struct file_operations packet_seq_fops = {
3889         .owner          = THIS_MODULE,
3890         .open           = packet_seq_open,
3891         .read           = seq_read,
3892         .llseek         = seq_lseek,
3893         .release        = seq_release_net,
3894 };
3895
3896 #endif
3897
3898 static int __net_init packet_net_init(struct net *net)
3899 {
3900         spin_lock_init(&net->packet.sklist_lock);
3901         INIT_HLIST_HEAD(&net->packet.sklist);
3902
3903         if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
3904                 return -ENOMEM;
3905
3906         return 0;
3907 }
3908
3909 static void __net_exit packet_net_exit(struct net *net)
3910 {
3911         proc_net_remove(net, "packet");
3912 }
3913
3914 static struct pernet_operations packet_net_ops = {
3915         .init = packet_net_init,
3916         .exit = packet_net_exit,
3917 };
3918
3919
3920 static void __exit packet_exit(void)
3921 {
3922         unregister_netdevice_notifier(&packet_netdev_notifier);
3923         unregister_pernet_subsys(&packet_net_ops);
3924         sock_unregister(PF_PACKET);
3925         proto_unregister(&packet_proto);
3926 }
3927
3928 static int __init packet_init(void)
3929 {
3930         int rc = proto_register(&packet_proto, 0);
3931
3932         if (rc != 0)
3933                 goto out;
3934
3935         sock_register(&packet_family_ops);
3936         register_pernet_subsys(&packet_net_ops);
3937         register_netdevice_notifier(&packet_netdev_notifier);
3938 out:
3939         return rc;
3940 }
3941
3942 module_init(packet_init);
3943 module_exit(packet_exit);
3944 MODULE_LICENSE("GPL");
3945 MODULE_ALIAS_NETPROTO(PF_PACKET);