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 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1285 {
1286         struct packet_sock *po = pkt_sk(sk);
1287         struct packet_fanout *f, *match;
1288         u8 type = type_flags & 0xff;
1289         u8 defrag = (type_flags & PACKET_FANOUT_FLAG_DEFRAG) ? 1 : 0;
1290         int err;
1291
1292         switch (type) {
1293         case PACKET_FANOUT_HASH:
1294         case PACKET_FANOUT_LB:
1295         case PACKET_FANOUT_CPU:
1296                 break;
1297         default:
1298                 return -EINVAL;
1299         }
1300
1301         if (!po->running)
1302                 return -EINVAL;
1303
1304         if (po->fanout)
1305                 return -EALREADY;
1306
1307         mutex_lock(&fanout_mutex);
1308         match = NULL;
1309         list_for_each_entry(f, &fanout_list, list) {
1310                 if (f->id == id &&
1311                     read_pnet(&f->net) == sock_net(sk)) {
1312                         match = f;
1313                         break;
1314                 }
1315         }
1316         err = -EINVAL;
1317         if (match && match->defrag != defrag)
1318                 goto out;
1319         if (!match) {
1320                 err = -ENOMEM;
1321                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1322                 if (!match)
1323                         goto out;
1324                 write_pnet(&match->net, sock_net(sk));
1325                 match->id = id;
1326                 match->type = type;
1327                 match->defrag = defrag;
1328                 atomic_set(&match->rr_cur, 0);
1329                 INIT_LIST_HEAD(&match->list);
1330                 spin_lock_init(&match->lock);
1331                 atomic_set(&match->sk_ref, 0);
1332                 match->prot_hook.type = po->prot_hook.type;
1333                 match->prot_hook.dev = po->prot_hook.dev;
1334                 match->prot_hook.func = packet_rcv_fanout;
1335                 match->prot_hook.af_packet_priv = match;
1336                 dev_add_pack(&match->prot_hook);
1337                 list_add(&match->list, &fanout_list);
1338         }
1339         err = -EINVAL;
1340         if (match->type == type &&
1341             match->prot_hook.type == po->prot_hook.type &&
1342             match->prot_hook.dev == po->prot_hook.dev) {
1343                 err = -ENOSPC;
1344                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1345                         __dev_remove_pack(&po->prot_hook);
1346                         po->fanout = match;
1347                         atomic_inc(&match->sk_ref);
1348                         __fanout_link(sk, po);
1349                         err = 0;
1350                 }
1351         }
1352 out:
1353         mutex_unlock(&fanout_mutex);
1354         return err;
1355 }
1356
1357 static void fanout_release(struct sock *sk)
1358 {
1359         struct packet_sock *po = pkt_sk(sk);
1360         struct packet_fanout *f;
1361
1362         f = po->fanout;
1363         if (!f)
1364                 return;
1365
1366         po->fanout = NULL;
1367
1368         mutex_lock(&fanout_mutex);
1369         if (atomic_dec_and_test(&f->sk_ref)) {
1370                 list_del(&f->list);
1371                 dev_remove_pack(&f->prot_hook);
1372                 kfree(f);
1373         }
1374         mutex_unlock(&fanout_mutex);
1375 }
1376
1377 static const struct proto_ops packet_ops;
1378
1379 static const struct proto_ops packet_ops_spkt;
1380
1381 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1382                            struct packet_type *pt, struct net_device *orig_dev)
1383 {
1384         struct sock *sk;
1385         struct sockaddr_pkt *spkt;
1386
1387         /*
1388          *      When we registered the protocol we saved the socket in the data
1389          *      field for just this event.
1390          */
1391
1392         sk = pt->af_packet_priv;
1393
1394         /*
1395          *      Yank back the headers [hope the device set this
1396          *      right or kerboom...]
1397          *
1398          *      Incoming packets have ll header pulled,
1399          *      push it back.
1400          *
1401          *      For outgoing ones skb->data == skb_mac_header(skb)
1402          *      so that this procedure is noop.
1403          */
1404
1405         if (skb->pkt_type == PACKET_LOOPBACK)
1406                 goto out;
1407
1408         if (!net_eq(dev_net(dev), sock_net(sk)))
1409                 goto out;
1410
1411         skb = skb_share_check(skb, GFP_ATOMIC);
1412         if (skb == NULL)
1413                 goto oom;
1414
1415         /* drop any routing info */
1416         skb_dst_drop(skb);
1417
1418         /* drop conntrack reference */
1419         nf_reset(skb);
1420
1421         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1422
1423         skb_push(skb, skb->data - skb_mac_header(skb));
1424
1425         /*
1426          *      The SOCK_PACKET socket receives _all_ frames.
1427          */
1428
1429         spkt->spkt_family = dev->type;
1430         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1431         spkt->spkt_protocol = skb->protocol;
1432
1433         /*
1434          *      Charge the memory to the socket. This is done specifically
1435          *      to prevent sockets using all the memory up.
1436          */
1437
1438         if (sock_queue_rcv_skb(sk, skb) == 0)
1439                 return 0;
1440
1441 out:
1442         kfree_skb(skb);
1443 oom:
1444         return 0;
1445 }
1446
1447
1448 /*
1449  *      Output a raw packet to a device layer. This bypasses all the other
1450  *      protocol layers and you must therefore supply it with a complete frame
1451  */
1452
1453 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1454                                struct msghdr *msg, size_t len)
1455 {
1456         struct sock *sk = sock->sk;
1457         struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1458         struct sk_buff *skb = NULL;
1459         struct net_device *dev;
1460         __be16 proto = 0;
1461         int err;
1462
1463         /*
1464          *      Get and verify the address.
1465          */
1466
1467         if (saddr) {
1468                 if (msg->msg_namelen < sizeof(struct sockaddr))
1469                         return -EINVAL;
1470                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1471                         proto = saddr->spkt_protocol;
1472         } else
1473                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1474
1475         /*
1476          *      Find the device first to size check it
1477          */
1478
1479         saddr->spkt_device[13] = 0;
1480 retry:
1481         rcu_read_lock();
1482         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1483         err = -ENODEV;
1484         if (dev == NULL)
1485                 goto out_unlock;
1486
1487         err = -ENETDOWN;
1488         if (!(dev->flags & IFF_UP))
1489                 goto out_unlock;
1490
1491         /*
1492          * You may not queue a frame bigger than the mtu. This is the lowest level
1493          * raw protocol and you must do your own fragmentation at this level.
1494          */
1495
1496         err = -EMSGSIZE;
1497         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN)
1498                 goto out_unlock;
1499
1500         if (!skb) {
1501                 size_t reserved = LL_RESERVED_SPACE(dev);
1502                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1503
1504                 rcu_read_unlock();
1505                 skb = sock_wmalloc(sk, len + reserved, 0, GFP_KERNEL);
1506                 if (skb == NULL)
1507                         return -ENOBUFS;
1508                 /* FIXME: Save some space for broken drivers that write a hard
1509                  * header at transmission time by themselves. PPP is the notable
1510                  * one here. This should really be fixed at the driver level.
1511                  */
1512                 skb_reserve(skb, reserved);
1513                 skb_reset_network_header(skb);
1514
1515                 /* Try to align data part correctly */
1516                 if (hhlen) {
1517                         skb->data -= hhlen;
1518                         skb->tail -= hhlen;
1519                         if (len < hhlen)
1520                                 skb_reset_network_header(skb);
1521                 }
1522                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1523                 if (err)
1524                         goto out_free;
1525                 goto retry;
1526         }
1527
1528         if (len > (dev->mtu + dev->hard_header_len)) {
1529                 /* Earlier code assumed this would be a VLAN pkt,
1530                  * double-check this now that we have the actual
1531                  * packet in hand.
1532                  */
1533                 struct ethhdr *ehdr;
1534                 skb_reset_mac_header(skb);
1535                 ehdr = eth_hdr(skb);
1536                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1537                         err = -EMSGSIZE;
1538                         goto out_unlock;
1539                 }
1540         }
1541
1542         skb->protocol = proto;
1543         skb->dev = dev;
1544         skb->priority = sk->sk_priority;
1545         skb->mark = sk->sk_mark;
1546         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1547         if (err < 0)
1548                 goto out_unlock;
1549
1550         dev_queue_xmit(skb);
1551         rcu_read_unlock();
1552         return len;
1553
1554 out_unlock:
1555         rcu_read_unlock();
1556 out_free:
1557         kfree_skb(skb);
1558         return err;
1559 }
1560
1561 static unsigned int run_filter(const struct sk_buff *skb,
1562                                       const struct sock *sk,
1563                                       unsigned int res)
1564 {
1565         struct sk_filter *filter;
1566
1567         rcu_read_lock();
1568         filter = rcu_dereference(sk->sk_filter);
1569         if (filter != NULL)
1570                 res = SK_RUN_FILTER(filter, skb);
1571         rcu_read_unlock();
1572
1573         return res;
1574 }
1575
1576 /*
1577  * This function makes lazy skb cloning in hope that most of packets
1578  * are discarded by BPF.
1579  *
1580  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1581  * and skb->cb are mangled. It works because (and until) packets
1582  * falling here are owned by current CPU. Output packets are cloned
1583  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1584  * sequencially, so that if we return skb to original state on exit,
1585  * we will not harm anyone.
1586  */
1587
1588 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1589                       struct packet_type *pt, struct net_device *orig_dev)
1590 {
1591         struct sock *sk;
1592         struct sockaddr_ll *sll;
1593         struct packet_sock *po;
1594         u8 *skb_head = skb->data;
1595         int skb_len = skb->len;
1596         unsigned int snaplen, res;
1597
1598         if (skb->pkt_type == PACKET_LOOPBACK)
1599                 goto drop;
1600
1601         sk = pt->af_packet_priv;
1602         po = pkt_sk(sk);
1603
1604         if (!net_eq(dev_net(dev), sock_net(sk)))
1605                 goto drop;
1606
1607         skb->dev = dev;
1608
1609         if (dev->header_ops) {
1610                 /* The device has an explicit notion of ll header,
1611                  * exported to higher levels.
1612                  *
1613                  * Otherwise, the device hides details of its frame
1614                  * structure, so that corresponding packet head is
1615                  * never delivered to user.
1616                  */
1617                 if (sk->sk_type != SOCK_DGRAM)
1618                         skb_push(skb, skb->data - skb_mac_header(skb));
1619                 else if (skb->pkt_type == PACKET_OUTGOING) {
1620                         /* Special case: outgoing packets have ll header at head */
1621                         skb_pull(skb, skb_network_offset(skb));
1622                 }
1623         }
1624
1625         snaplen = skb->len;
1626
1627         res = run_filter(skb, sk, snaplen);
1628         if (!res)
1629                 goto drop_n_restore;
1630         if (snaplen > res)
1631                 snaplen = res;
1632
1633         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1634                 goto drop_n_acct;
1635
1636         if (skb_shared(skb)) {
1637                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1638                 if (nskb == NULL)
1639                         goto drop_n_acct;
1640
1641                 if (skb_head != skb->data) {
1642                         skb->data = skb_head;
1643                         skb->len = skb_len;
1644                 }
1645                 kfree_skb(skb);
1646                 skb = nskb;
1647         }
1648
1649         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1650                      sizeof(skb->cb));
1651
1652         sll = &PACKET_SKB_CB(skb)->sa.ll;
1653         sll->sll_family = AF_PACKET;
1654         sll->sll_hatype = dev->type;
1655         sll->sll_protocol = skb->protocol;
1656         sll->sll_pkttype = skb->pkt_type;
1657         if (unlikely(po->origdev))
1658                 sll->sll_ifindex = orig_dev->ifindex;
1659         else
1660                 sll->sll_ifindex = dev->ifindex;
1661
1662         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1663
1664         PACKET_SKB_CB(skb)->origlen = skb->len;
1665
1666         if (pskb_trim(skb, snaplen))
1667                 goto drop_n_acct;
1668
1669         skb_set_owner_r(skb, sk);
1670         skb->dev = NULL;
1671         skb_dst_drop(skb);
1672
1673         /* drop conntrack reference */
1674         nf_reset(skb);
1675
1676         spin_lock(&sk->sk_receive_queue.lock);
1677         po->stats.tp_packets++;
1678         skb->dropcount = atomic_read(&sk->sk_drops);
1679         __skb_queue_tail(&sk->sk_receive_queue, skb);
1680         spin_unlock(&sk->sk_receive_queue.lock);
1681         sk->sk_data_ready(sk, skb->len);
1682         return 0;
1683
1684 drop_n_acct:
1685         spin_lock(&sk->sk_receive_queue.lock);
1686         po->stats.tp_drops++;
1687         atomic_inc(&sk->sk_drops);
1688         spin_unlock(&sk->sk_receive_queue.lock);
1689
1690 drop_n_restore:
1691         if (skb_head != skb->data && skb_shared(skb)) {
1692                 skb->data = skb_head;
1693                 skb->len = skb_len;
1694         }
1695 drop:
1696         consume_skb(skb);
1697         return 0;
1698 }
1699
1700 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1701                        struct packet_type *pt, struct net_device *orig_dev)
1702 {
1703         struct sock *sk;
1704         struct packet_sock *po;
1705         struct sockaddr_ll *sll;
1706         union {
1707                 struct tpacket_hdr *h1;
1708                 struct tpacket2_hdr *h2;
1709                 struct tpacket3_hdr *h3;
1710                 void *raw;
1711         } h;
1712         u8 *skb_head = skb->data;
1713         int skb_len = skb->len;
1714         unsigned int snaplen, res;
1715         unsigned long status = TP_STATUS_USER;
1716         unsigned short macoff, netoff, hdrlen;
1717         struct sk_buff *copy_skb = NULL;
1718         struct timeval tv;
1719         struct timespec ts;
1720         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1721
1722         if (skb->pkt_type == PACKET_LOOPBACK)
1723                 goto drop;
1724
1725         sk = pt->af_packet_priv;
1726         po = pkt_sk(sk);
1727
1728         if (!net_eq(dev_net(dev), sock_net(sk)))
1729                 goto drop;
1730
1731         if (dev->header_ops) {
1732                 if (sk->sk_type != SOCK_DGRAM)
1733                         skb_push(skb, skb->data - skb_mac_header(skb));
1734                 else if (skb->pkt_type == PACKET_OUTGOING) {
1735                         /* Special case: outgoing packets have ll header at head */
1736                         skb_pull(skb, skb_network_offset(skb));
1737                 }
1738         }
1739
1740         if (skb->ip_summed == CHECKSUM_PARTIAL)
1741                 status |= TP_STATUS_CSUMNOTREADY;
1742
1743         snaplen = skb->len;
1744
1745         res = run_filter(skb, sk, snaplen);
1746         if (!res)
1747                 goto drop_n_restore;
1748         if (snaplen > res)
1749                 snaplen = res;
1750
1751         if (sk->sk_type == SOCK_DGRAM) {
1752                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1753                                   po->tp_reserve;
1754         } else {
1755                 unsigned maclen = skb_network_offset(skb);
1756                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1757                                        (maclen < 16 ? 16 : maclen)) +
1758                         po->tp_reserve;
1759                 macoff = netoff - maclen;
1760         }
1761         if (po->tp_version <= TPACKET_V2) {
1762                 if (macoff + snaplen > po->rx_ring.frame_size) {
1763                         if (po->copy_thresh &&
1764                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1765                                 if (skb_shared(skb)) {
1766                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1767                                 } else {
1768                                         copy_skb = skb_get(skb);
1769                                         skb_head = skb->data;
1770                                 }
1771                                 if (copy_skb)
1772                                         skb_set_owner_r(copy_skb, sk);
1773                         }
1774                         snaplen = po->rx_ring.frame_size - macoff;
1775                         if ((int)snaplen < 0)
1776                                 snaplen = 0;
1777                 }
1778         }
1779         spin_lock(&sk->sk_receive_queue.lock);
1780         h.raw = packet_current_rx_frame(po, skb,
1781                                         TP_STATUS_KERNEL, (macoff+snaplen));
1782         if (!h.raw)
1783                 goto ring_is_full;
1784         if (po->tp_version <= TPACKET_V2) {
1785                 packet_increment_rx_head(po, &po->rx_ring);
1786         /*
1787          * LOSING will be reported till you read the stats,
1788          * because it's COR - Clear On Read.
1789          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1790          * at packet level.
1791          */
1792                 if (po->stats.tp_drops)
1793                         status |= TP_STATUS_LOSING;
1794         }
1795         po->stats.tp_packets++;
1796         if (copy_skb) {
1797                 status |= TP_STATUS_COPY;
1798                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1799         }
1800         spin_unlock(&sk->sk_receive_queue.lock);
1801
1802         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1803
1804         switch (po->tp_version) {
1805         case TPACKET_V1:
1806                 h.h1->tp_len = skb->len;
1807                 h.h1->tp_snaplen = snaplen;
1808                 h.h1->tp_mac = macoff;
1809                 h.h1->tp_net = netoff;
1810                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1811                                 && shhwtstamps->syststamp.tv64)
1812                         tv = ktime_to_timeval(shhwtstamps->syststamp);
1813                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1814                                 && shhwtstamps->hwtstamp.tv64)
1815                         tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1816                 else if (skb->tstamp.tv64)
1817                         tv = ktime_to_timeval(skb->tstamp);
1818                 else
1819                         do_gettimeofday(&tv);
1820                 h.h1->tp_sec = tv.tv_sec;
1821                 h.h1->tp_usec = tv.tv_usec;
1822                 hdrlen = sizeof(*h.h1);
1823                 break;
1824         case TPACKET_V2:
1825                 h.h2->tp_len = skb->len;
1826                 h.h2->tp_snaplen = snaplen;
1827                 h.h2->tp_mac = macoff;
1828                 h.h2->tp_net = netoff;
1829                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1830                                 && shhwtstamps->syststamp.tv64)
1831                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1832                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1833                                 && shhwtstamps->hwtstamp.tv64)
1834                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1835                 else if (skb->tstamp.tv64)
1836                         ts = ktime_to_timespec(skb->tstamp);
1837                 else
1838                         getnstimeofday(&ts);
1839                 h.h2->tp_sec = ts.tv_sec;
1840                 h.h2->tp_nsec = ts.tv_nsec;
1841                 if (vlan_tx_tag_present(skb)) {
1842                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1843                         status |= TP_STATUS_VLAN_VALID;
1844                 } else {
1845                         h.h2->tp_vlan_tci = 0;
1846                 }
1847                 h.h2->tp_padding = 0;
1848                 hdrlen = sizeof(*h.h2);
1849                 break;
1850         case TPACKET_V3:
1851                 /* tp_nxt_offset,vlan are already populated above.
1852                  * So DONT clear those fields here
1853                  */
1854                 h.h3->tp_status |= status;
1855                 h.h3->tp_len = skb->len;
1856                 h.h3->tp_snaplen = snaplen;
1857                 h.h3->tp_mac = macoff;
1858                 h.h3->tp_net = netoff;
1859                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1860                                 && shhwtstamps->syststamp.tv64)
1861                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1862                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1863                                 && shhwtstamps->hwtstamp.tv64)
1864                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1865                 else if (skb->tstamp.tv64)
1866                         ts = ktime_to_timespec(skb->tstamp);
1867                 else
1868                         getnstimeofday(&ts);
1869                 h.h3->tp_sec  = ts.tv_sec;
1870                 h.h3->tp_nsec = ts.tv_nsec;
1871                 hdrlen = sizeof(*h.h3);
1872                 break;
1873         default:
1874                 BUG();
1875         }
1876
1877         sll = h.raw + TPACKET_ALIGN(hdrlen);
1878         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1879         sll->sll_family = AF_PACKET;
1880         sll->sll_hatype = dev->type;
1881         sll->sll_protocol = skb->protocol;
1882         sll->sll_pkttype = skb->pkt_type;
1883         if (unlikely(po->origdev))
1884                 sll->sll_ifindex = orig_dev->ifindex;
1885         else
1886                 sll->sll_ifindex = dev->ifindex;
1887
1888         smp_mb();
1889 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1890         {
1891                 u8 *start, *end;
1892
1893                 if (po->tp_version <= TPACKET_V2) {
1894                         end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1895                                 + macoff + snaplen);
1896                         for (start = h.raw; start < end; start += PAGE_SIZE)
1897                                 flush_dcache_page(pgv_to_page(start));
1898                 }
1899                 smp_wmb();
1900         }
1901 #endif
1902         if (po->tp_version <= TPACKET_V2)
1903                 __packet_set_status(po, h.raw, status);
1904         else
1905                 prb_clear_blk_fill_status(&po->rx_ring);
1906
1907         sk->sk_data_ready(sk, 0);
1908
1909 drop_n_restore:
1910         if (skb_head != skb->data && skb_shared(skb)) {
1911                 skb->data = skb_head;
1912                 skb->len = skb_len;
1913         }
1914 drop:
1915         kfree_skb(skb);
1916         return 0;
1917
1918 ring_is_full:
1919         po->stats.tp_drops++;
1920         spin_unlock(&sk->sk_receive_queue.lock);
1921
1922         sk->sk_data_ready(sk, 0);
1923         kfree_skb(copy_skb);
1924         goto drop_n_restore;
1925 }
1926
1927 static void tpacket_destruct_skb(struct sk_buff *skb)
1928 {
1929         struct packet_sock *po = pkt_sk(skb->sk);
1930         void *ph;
1931
1932         if (likely(po->tx_ring.pg_vec)) {
1933                 ph = skb_shinfo(skb)->destructor_arg;
1934                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1935                 atomic_dec(&po->tx_ring.pending);
1936                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1937         }
1938
1939         sock_wfree(skb);
1940 }
1941
1942 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1943                 void *frame, struct net_device *dev, int size_max,
1944                 __be16 proto, unsigned char *addr)
1945 {
1946         union {
1947                 struct tpacket_hdr *h1;
1948                 struct tpacket2_hdr *h2;
1949                 void *raw;
1950         } ph;
1951         int to_write, offset, len, tp_len, nr_frags, len_max;
1952         struct socket *sock = po->sk.sk_socket;
1953         struct page *page;
1954         void *data;
1955         int err;
1956
1957         ph.raw = frame;
1958
1959         skb->protocol = proto;
1960         skb->dev = dev;
1961         skb->priority = po->sk.sk_priority;
1962         skb->mark = po->sk.sk_mark;
1963         skb_shinfo(skb)->destructor_arg = ph.raw;
1964
1965         switch (po->tp_version) {
1966         case TPACKET_V2:
1967                 tp_len = ph.h2->tp_len;
1968                 break;
1969         default:
1970                 tp_len = ph.h1->tp_len;
1971                 break;
1972         }
1973         if (unlikely(tp_len > size_max)) {
1974                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1975                 return -EMSGSIZE;
1976         }
1977
1978         skb_reserve(skb, LL_RESERVED_SPACE(dev));
1979         skb_reset_network_header(skb);
1980
1981         data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1982         to_write = tp_len;
1983
1984         if (sock->type == SOCK_DGRAM) {
1985                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
1986                                 NULL, tp_len);
1987                 if (unlikely(err < 0))
1988                         return -EINVAL;
1989         } else if (dev->hard_header_len) {
1990                 /* net device doesn't like empty head */
1991                 if (unlikely(tp_len <= dev->hard_header_len)) {
1992                         pr_err("packet size is too short (%d < %d)\n",
1993                                tp_len, dev->hard_header_len);
1994                         return -EINVAL;
1995                 }
1996
1997                 skb_push(skb, dev->hard_header_len);
1998                 err = skb_store_bits(skb, 0, data,
1999                                 dev->hard_header_len);
2000                 if (unlikely(err))
2001                         return err;
2002
2003                 data += dev->hard_header_len;
2004                 to_write -= dev->hard_header_len;
2005         }
2006
2007         err = -EFAULT;
2008         offset = offset_in_page(data);
2009         len_max = PAGE_SIZE - offset;
2010         len = ((to_write > len_max) ? len_max : to_write);
2011
2012         skb->data_len = to_write;
2013         skb->len += to_write;
2014         skb->truesize += to_write;
2015         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2016
2017         while (likely(to_write)) {
2018                 nr_frags = skb_shinfo(skb)->nr_frags;
2019
2020                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2021                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2022                                MAX_SKB_FRAGS);
2023                         return -EFAULT;
2024                 }
2025
2026                 page = pgv_to_page(data);
2027                 data += len;
2028                 flush_dcache_page(page);
2029                 get_page(page);
2030                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2031                 to_write -= len;
2032                 offset = 0;
2033                 len_max = PAGE_SIZE;
2034                 len = ((to_write > len_max) ? len_max : to_write);
2035         }
2036
2037         return tp_len;
2038 }
2039
2040 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2041 {
2042         struct sk_buff *skb;
2043         struct net_device *dev;
2044         __be16 proto;
2045         bool need_rls_dev = false;
2046         int err, reserve = 0;
2047         void *ph;
2048         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2049         int tp_len, size_max;
2050         unsigned char *addr;
2051         int len_sum = 0;
2052         int status = 0;
2053
2054         mutex_lock(&po->pg_vec_lock);
2055
2056         err = -EBUSY;
2057         if (saddr == NULL) {
2058                 dev = po->prot_hook.dev;
2059                 proto   = po->num;
2060                 addr    = NULL;
2061         } else {
2062                 err = -EINVAL;
2063                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2064                         goto out;
2065                 if (msg->msg_namelen < (saddr->sll_halen
2066                                         + offsetof(struct sockaddr_ll,
2067                                                 sll_addr)))
2068                         goto out;
2069                 proto   = saddr->sll_protocol;
2070                 addr    = saddr->sll_addr;
2071                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2072                 need_rls_dev = true;
2073         }
2074
2075         err = -ENXIO;
2076         if (unlikely(dev == NULL))
2077                 goto out;
2078
2079         reserve = dev->hard_header_len;
2080
2081         err = -ENETDOWN;
2082         if (unlikely(!(dev->flags & IFF_UP)))
2083                 goto out_put;
2084
2085         size_max = po->tx_ring.frame_size
2086                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2087
2088         if (size_max > dev->mtu + reserve)
2089                 size_max = dev->mtu + reserve;
2090
2091         do {
2092                 ph = packet_current_frame(po, &po->tx_ring,
2093                                 TP_STATUS_SEND_REQUEST);
2094
2095                 if (unlikely(ph == NULL)) {
2096                         schedule();
2097                         continue;
2098                 }
2099
2100                 status = TP_STATUS_SEND_REQUEST;
2101                 skb = sock_alloc_send_skb(&po->sk,
2102                                 LL_ALLOCATED_SPACE(dev)
2103                                 + sizeof(struct sockaddr_ll),
2104                                 0, &err);
2105
2106                 if (unlikely(skb == NULL))
2107                         goto out_status;
2108
2109                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2110                                 addr);
2111
2112                 if (unlikely(tp_len < 0)) {
2113                         if (po->tp_loss) {
2114                                 __packet_set_status(po, ph,
2115                                                 TP_STATUS_AVAILABLE);
2116                                 packet_increment_head(&po->tx_ring);
2117                                 kfree_skb(skb);
2118                                 continue;
2119                         } else {
2120                                 status = TP_STATUS_WRONG_FORMAT;
2121                                 err = tp_len;
2122                                 goto out_status;
2123                         }
2124                 }
2125
2126                 skb->destructor = tpacket_destruct_skb;
2127                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2128                 atomic_inc(&po->tx_ring.pending);
2129
2130                 status = TP_STATUS_SEND_REQUEST;
2131                 err = dev_queue_xmit(skb);
2132                 if (unlikely(err > 0)) {
2133                         err = net_xmit_errno(err);
2134                         if (err && __packet_get_status(po, ph) ==
2135                                    TP_STATUS_AVAILABLE) {
2136                                 /* skb was destructed already */
2137                                 skb = NULL;
2138                                 goto out_status;
2139                         }
2140                         /*
2141                          * skb was dropped but not destructed yet;
2142                          * let's treat it like congestion or err < 0
2143                          */
2144                         err = 0;
2145                 }
2146                 packet_increment_head(&po->tx_ring);
2147                 len_sum += tp_len;
2148         } while (likely((ph != NULL) ||
2149                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2150                          (atomic_read(&po->tx_ring.pending))))
2151                 );
2152
2153         err = len_sum;
2154         goto out_put;
2155
2156 out_status:
2157         __packet_set_status(po, ph, status);
2158         kfree_skb(skb);
2159 out_put:
2160         if (need_rls_dev)
2161                 dev_put(dev);
2162 out:
2163         mutex_unlock(&po->pg_vec_lock);
2164         return err;
2165 }
2166
2167 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2168                                         size_t reserve, size_t len,
2169                                         size_t linear, int noblock,
2170                                         int *err)
2171 {
2172         struct sk_buff *skb;
2173
2174         /* Under a page?  Don't bother with paged skb. */
2175         if (prepad + len < PAGE_SIZE || !linear)
2176                 linear = len;
2177
2178         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2179                                    err);
2180         if (!skb)
2181                 return NULL;
2182
2183         skb_reserve(skb, reserve);
2184         skb_put(skb, linear);
2185         skb->data_len = len - linear;
2186         skb->len += len - linear;
2187
2188         return skb;
2189 }
2190
2191 static int packet_snd(struct socket *sock,
2192                           struct msghdr *msg, size_t len)
2193 {
2194         struct sock *sk = sock->sk;
2195         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2196         struct sk_buff *skb;
2197         struct net_device *dev;
2198         __be16 proto;
2199         bool need_rls_dev = false;
2200         unsigned char *addr;
2201         int err, reserve = 0;
2202         struct virtio_net_hdr vnet_hdr = { 0 };
2203         int offset = 0;
2204         int vnet_hdr_len;
2205         struct packet_sock *po = pkt_sk(sk);
2206         unsigned short gso_type = 0;
2207
2208         /*
2209          *      Get and verify the address.
2210          */
2211
2212         if (saddr == NULL) {
2213                 dev = po->prot_hook.dev;
2214                 proto   = po->num;
2215                 addr    = NULL;
2216         } else {
2217                 err = -EINVAL;
2218                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2219                         goto out;
2220                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2221                         goto out;
2222                 proto   = saddr->sll_protocol;
2223                 addr    = saddr->sll_addr;
2224                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2225                 need_rls_dev = true;
2226         }
2227
2228         err = -ENXIO;
2229         if (dev == NULL)
2230                 goto out_unlock;
2231         if (sock->type == SOCK_RAW)
2232                 reserve = dev->hard_header_len;
2233
2234         err = -ENETDOWN;
2235         if (!(dev->flags & IFF_UP))
2236                 goto out_unlock;
2237
2238         if (po->has_vnet_hdr) {
2239                 vnet_hdr_len = sizeof(vnet_hdr);
2240
2241                 err = -EINVAL;
2242                 if (len < vnet_hdr_len)
2243                         goto out_unlock;
2244
2245                 len -= vnet_hdr_len;
2246
2247                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2248                                        vnet_hdr_len);
2249                 if (err < 0)
2250                         goto out_unlock;
2251
2252                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2253                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2254                       vnet_hdr.hdr_len))
2255                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
2256                                                  vnet_hdr.csum_offset + 2;
2257
2258                 err = -EINVAL;
2259                 if (vnet_hdr.hdr_len > len)
2260                         goto out_unlock;
2261
2262                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2263                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2264                         case VIRTIO_NET_HDR_GSO_TCPV4:
2265                                 gso_type = SKB_GSO_TCPV4;
2266                                 break;
2267                         case VIRTIO_NET_HDR_GSO_TCPV6:
2268                                 gso_type = SKB_GSO_TCPV6;
2269                                 break;
2270                         case VIRTIO_NET_HDR_GSO_UDP:
2271                                 gso_type = SKB_GSO_UDP;
2272                                 break;
2273                         default:
2274                                 goto out_unlock;
2275                         }
2276
2277                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2278                                 gso_type |= SKB_GSO_TCP_ECN;
2279
2280                         if (vnet_hdr.gso_size == 0)
2281                                 goto out_unlock;
2282
2283                 }
2284         }
2285
2286         err = -EMSGSIZE;
2287         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN))
2288                 goto out_unlock;
2289
2290         err = -ENOBUFS;
2291         skb = packet_alloc_skb(sk, LL_ALLOCATED_SPACE(dev),
2292                                LL_RESERVED_SPACE(dev), len, vnet_hdr.hdr_len,
2293                                msg->msg_flags & MSG_DONTWAIT, &err);
2294         if (skb == NULL)
2295                 goto out_unlock;
2296
2297         skb_set_network_header(skb, reserve);
2298
2299         err = -EINVAL;
2300         if (sock->type == SOCK_DGRAM &&
2301             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2302                 goto out_free;
2303
2304         /* Returns -EFAULT on error */
2305         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2306         if (err)
2307                 goto out_free;
2308         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2309         if (err < 0)
2310                 goto out_free;
2311
2312         if (!gso_type && (len > dev->mtu + reserve)) {
2313                 /* Earlier code assumed this would be a VLAN pkt,
2314                  * double-check this now that we have the actual
2315                  * packet in hand.
2316                  */
2317                 struct ethhdr *ehdr;
2318                 skb_reset_mac_header(skb);
2319                 ehdr = eth_hdr(skb);
2320                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2321                         err = -EMSGSIZE;
2322                         goto out_free;
2323                 }
2324         }
2325
2326         skb->protocol = proto;
2327         skb->dev = dev;
2328         skb->priority = sk->sk_priority;
2329         skb->mark = sk->sk_mark;
2330
2331         if (po->has_vnet_hdr) {
2332                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2333                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2334                                                   vnet_hdr.csum_offset)) {
2335                                 err = -EINVAL;
2336                                 goto out_free;
2337                         }
2338                 }
2339
2340                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2341                 skb_shinfo(skb)->gso_type = gso_type;
2342
2343                 /* Header must be checked, and gso_segs computed. */
2344                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2345                 skb_shinfo(skb)->gso_segs = 0;
2346
2347                 len += vnet_hdr_len;
2348         }
2349
2350         /*
2351          *      Now send it
2352          */
2353
2354         err = dev_queue_xmit(skb);
2355         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2356                 goto out_unlock;
2357
2358         if (need_rls_dev)
2359                 dev_put(dev);
2360
2361         return len;
2362
2363 out_free:
2364         kfree_skb(skb);
2365 out_unlock:
2366         if (dev && need_rls_dev)
2367                 dev_put(dev);
2368 out:
2369         return err;
2370 }
2371
2372 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2373                 struct msghdr *msg, size_t len)
2374 {
2375         struct sock *sk = sock->sk;
2376         struct packet_sock *po = pkt_sk(sk);
2377         if (po->tx_ring.pg_vec)
2378                 return tpacket_snd(po, msg);
2379         else
2380                 return packet_snd(sock, msg, len);
2381 }
2382
2383 /*
2384  *      Close a PACKET socket. This is fairly simple. We immediately go
2385  *      to 'closed' state and remove our protocol entry in the device list.
2386  */
2387
2388 static int packet_release(struct socket *sock)
2389 {
2390         struct sock *sk = sock->sk;
2391         struct packet_sock *po;
2392         struct net *net;
2393         union tpacket_req_u req_u;
2394
2395         if (!sk)
2396                 return 0;
2397
2398         net = sock_net(sk);
2399         po = pkt_sk(sk);
2400
2401         spin_lock_bh(&net->packet.sklist_lock);
2402         sk_del_node_init_rcu(sk);
2403         sock_prot_inuse_add(net, sk->sk_prot, -1);
2404         spin_unlock_bh(&net->packet.sklist_lock);
2405
2406         spin_lock(&po->bind_lock);
2407         unregister_prot_hook(sk, false);
2408         if (po->prot_hook.dev) {
2409                 dev_put(po->prot_hook.dev);
2410                 po->prot_hook.dev = NULL;
2411         }
2412         spin_unlock(&po->bind_lock);
2413
2414         packet_flush_mclist(sk);
2415
2416         memset(&req_u, 0, sizeof(req_u));
2417
2418         if (po->rx_ring.pg_vec)
2419                 packet_set_ring(sk, &req_u, 1, 0);
2420
2421         if (po->tx_ring.pg_vec)
2422                 packet_set_ring(sk, &req_u, 1, 1);
2423
2424         fanout_release(sk);
2425
2426         synchronize_net();
2427         /*
2428          *      Now the socket is dead. No more input will appear.
2429          */
2430         sock_orphan(sk);
2431         sock->sk = NULL;
2432
2433         /* Purge queues */
2434
2435         skb_queue_purge(&sk->sk_receive_queue);
2436         sk_refcnt_debug_release(sk);
2437
2438         sock_put(sk);
2439         return 0;
2440 }
2441
2442 /*
2443  *      Attach a packet hook.
2444  */
2445
2446 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2447 {
2448         struct packet_sock *po = pkt_sk(sk);
2449
2450         if (po->fanout) {
2451                 if (dev)
2452                         dev_put(dev);
2453
2454                 return -EINVAL;
2455         }
2456
2457         lock_sock(sk);
2458
2459         spin_lock(&po->bind_lock);
2460         unregister_prot_hook(sk, true);
2461         po->num = protocol;
2462         po->prot_hook.type = protocol;
2463         if (po->prot_hook.dev)
2464                 dev_put(po->prot_hook.dev);
2465         po->prot_hook.dev = dev;
2466
2467         po->ifindex = dev ? dev->ifindex : 0;
2468
2469         if (protocol == 0)
2470                 goto out_unlock;
2471
2472         if (!dev || (dev->flags & IFF_UP)) {
2473                 register_prot_hook(sk);
2474         } else {
2475                 sk->sk_err = ENETDOWN;
2476                 if (!sock_flag(sk, SOCK_DEAD))
2477                         sk->sk_error_report(sk);
2478         }
2479
2480 out_unlock:
2481         spin_unlock(&po->bind_lock);
2482         release_sock(sk);
2483         return 0;
2484 }
2485
2486 /*
2487  *      Bind a packet socket to a device
2488  */
2489
2490 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2491                             int addr_len)
2492 {
2493         struct sock *sk = sock->sk;
2494         char name[15];
2495         struct net_device *dev;
2496         int err = -ENODEV;
2497
2498         /*
2499          *      Check legality
2500          */
2501
2502         if (addr_len != sizeof(struct sockaddr))
2503                 return -EINVAL;
2504         strlcpy(name, uaddr->sa_data, sizeof(name));
2505
2506         dev = dev_get_by_name(sock_net(sk), name);
2507         if (dev)
2508                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2509         return err;
2510 }
2511
2512 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2513 {
2514         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2515         struct sock *sk = sock->sk;
2516         struct net_device *dev = NULL;
2517         int err;
2518
2519
2520         /*
2521          *      Check legality
2522          */
2523
2524         if (addr_len < sizeof(struct sockaddr_ll))
2525                 return -EINVAL;
2526         if (sll->sll_family != AF_PACKET)
2527                 return -EINVAL;
2528
2529         if (sll->sll_ifindex) {
2530                 err = -ENODEV;
2531                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2532                 if (dev == NULL)
2533                         goto out;
2534         }
2535         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2536
2537 out:
2538         return err;
2539 }
2540
2541 static struct proto packet_proto = {
2542         .name     = "PACKET",
2543         .owner    = THIS_MODULE,
2544         .obj_size = sizeof(struct packet_sock),
2545 };
2546
2547 /*
2548  *      Create a packet of type SOCK_PACKET.
2549  */
2550
2551 static int packet_create(struct net *net, struct socket *sock, int protocol,
2552                          int kern)
2553 {
2554         struct sock *sk;
2555         struct packet_sock *po;
2556         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2557         int err;
2558
2559         if (!capable(CAP_NET_RAW))
2560                 return -EPERM;
2561         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2562             sock->type != SOCK_PACKET)
2563                 return -ESOCKTNOSUPPORT;
2564
2565         sock->state = SS_UNCONNECTED;
2566
2567         err = -ENOBUFS;
2568         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2569         if (sk == NULL)
2570                 goto out;
2571
2572         sock->ops = &packet_ops;
2573         if (sock->type == SOCK_PACKET)
2574                 sock->ops = &packet_ops_spkt;
2575
2576         sock_init_data(sock, sk);
2577
2578         po = pkt_sk(sk);
2579         sk->sk_family = PF_PACKET;
2580         po->num = proto;
2581
2582         sk->sk_destruct = packet_sock_destruct;
2583         sk_refcnt_debug_inc(sk);
2584
2585         /*
2586          *      Attach a protocol block
2587          */
2588
2589         spin_lock_init(&po->bind_lock);
2590         mutex_init(&po->pg_vec_lock);
2591         po->prot_hook.func = packet_rcv;
2592
2593         if (sock->type == SOCK_PACKET)
2594                 po->prot_hook.func = packet_rcv_spkt;
2595
2596         po->prot_hook.af_packet_priv = sk;
2597
2598         if (proto) {
2599                 po->prot_hook.type = proto;
2600                 register_prot_hook(sk);
2601         }
2602
2603         spin_lock_bh(&net->packet.sklist_lock);
2604         sk_add_node_rcu(sk, &net->packet.sklist);
2605         sock_prot_inuse_add(net, &packet_proto, 1);
2606         spin_unlock_bh(&net->packet.sklist_lock);
2607
2608         return 0;
2609 out:
2610         return err;
2611 }
2612
2613 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2614 {
2615         struct sock_exterr_skb *serr;
2616         struct sk_buff *skb, *skb2;
2617         int copied, err;
2618
2619         err = -EAGAIN;
2620         skb = skb_dequeue(&sk->sk_error_queue);
2621         if (skb == NULL)
2622                 goto out;
2623
2624         copied = skb->len;
2625         if (copied > len) {
2626                 msg->msg_flags |= MSG_TRUNC;
2627                 copied = len;
2628         }
2629         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2630         if (err)
2631                 goto out_free_skb;
2632
2633         sock_recv_timestamp(msg, sk, skb);
2634
2635         serr = SKB_EXT_ERR(skb);
2636         put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2637                  sizeof(serr->ee), &serr->ee);
2638
2639         msg->msg_flags |= MSG_ERRQUEUE;
2640         err = copied;
2641
2642         /* Reset and regenerate socket error */
2643         spin_lock_bh(&sk->sk_error_queue.lock);
2644         sk->sk_err = 0;
2645         if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2646                 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2647                 spin_unlock_bh(&sk->sk_error_queue.lock);
2648                 sk->sk_error_report(sk);
2649         } else
2650                 spin_unlock_bh(&sk->sk_error_queue.lock);
2651
2652 out_free_skb:
2653         kfree_skb(skb);
2654 out:
2655         return err;
2656 }
2657
2658 /*
2659  *      Pull a packet from our receive queue and hand it to the user.
2660  *      If necessary we block.
2661  */
2662
2663 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2664                           struct msghdr *msg, size_t len, int flags)
2665 {
2666         struct sock *sk = sock->sk;
2667         struct sk_buff *skb;
2668         int copied, err;
2669         struct sockaddr_ll *sll;
2670         int vnet_hdr_len = 0;
2671
2672         err = -EINVAL;
2673         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2674                 goto out;
2675
2676 #if 0
2677         /* What error should we return now? EUNATTACH? */
2678         if (pkt_sk(sk)->ifindex < 0)
2679                 return -ENODEV;
2680 #endif
2681
2682         if (flags & MSG_ERRQUEUE) {
2683                 err = packet_recv_error(sk, msg, len);
2684                 goto out;
2685         }
2686
2687         /*
2688          *      Call the generic datagram receiver. This handles all sorts
2689          *      of horrible races and re-entrancy so we can forget about it
2690          *      in the protocol layers.
2691          *
2692          *      Now it will return ENETDOWN, if device have just gone down,
2693          *      but then it will block.
2694          */
2695
2696         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2697
2698         /*
2699          *      An error occurred so return it. Because skb_recv_datagram()
2700          *      handles the blocking we don't see and worry about blocking
2701          *      retries.
2702          */
2703
2704         if (skb == NULL)
2705                 goto out;
2706
2707         if (pkt_sk(sk)->has_vnet_hdr) {
2708                 struct virtio_net_hdr vnet_hdr = { 0 };
2709
2710                 err = -EINVAL;
2711                 vnet_hdr_len = sizeof(vnet_hdr);
2712                 if (len < vnet_hdr_len)
2713                         goto out_free;
2714
2715                 len -= vnet_hdr_len;
2716
2717                 if (skb_is_gso(skb)) {
2718                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2719
2720                         /* This is a hint as to how much should be linear. */
2721                         vnet_hdr.hdr_len = skb_headlen(skb);
2722                         vnet_hdr.gso_size = sinfo->gso_size;
2723                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2724                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2725                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2726                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2727                         else if (sinfo->gso_type & SKB_GSO_UDP)
2728                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2729                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2730                                 goto out_free;
2731                         else
2732                                 BUG();
2733                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2734                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2735                 } else
2736                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2737
2738                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2739                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2740                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2741                         vnet_hdr.csum_offset = skb->csum_offset;
2742                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2743                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2744                 } /* else everything is zero */
2745
2746                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2747                                      vnet_hdr_len);
2748                 if (err < 0)
2749                         goto out_free;
2750         }
2751
2752         /*
2753          *      If the address length field is there to be filled in, we fill
2754          *      it in now.
2755          */
2756
2757         sll = &PACKET_SKB_CB(skb)->sa.ll;
2758         if (sock->type == SOCK_PACKET)
2759                 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2760         else
2761                 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2762
2763         /*
2764          *      You lose any data beyond the buffer you gave. If it worries a
2765          *      user program they can ask the device for its MTU anyway.
2766          */
2767
2768         copied = skb->len;
2769         if (copied > len) {
2770                 copied = len;
2771                 msg->msg_flags |= MSG_TRUNC;
2772         }
2773
2774         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2775         if (err)
2776                 goto out_free;
2777
2778         sock_recv_ts_and_drops(msg, sk, skb);
2779
2780         if (msg->msg_name)
2781                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2782                        msg->msg_namelen);
2783
2784         if (pkt_sk(sk)->auxdata) {
2785                 struct tpacket_auxdata aux;
2786
2787                 aux.tp_status = TP_STATUS_USER;
2788                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2789                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2790                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2791                 aux.tp_snaplen = skb->len;
2792                 aux.tp_mac = 0;
2793                 aux.tp_net = skb_network_offset(skb);
2794                 if (vlan_tx_tag_present(skb)) {
2795                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2796                         aux.tp_status |= TP_STATUS_VLAN_VALID;
2797                 } else {
2798                         aux.tp_vlan_tci = 0;
2799                 }
2800                 aux.tp_padding = 0;
2801                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2802         }
2803
2804         /*
2805          *      Free or return the buffer as appropriate. Again this
2806          *      hides all the races and re-entrancy issues from us.
2807          */
2808         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2809
2810 out_free:
2811         skb_free_datagram(sk, skb);
2812 out:
2813         return err;
2814 }
2815
2816 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2817                                int *uaddr_len, int peer)
2818 {
2819         struct net_device *dev;
2820         struct sock *sk = sock->sk;
2821
2822         if (peer)
2823                 return -EOPNOTSUPP;
2824
2825         uaddr->sa_family = AF_PACKET;
2826         rcu_read_lock();
2827         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2828         if (dev)
2829                 strncpy(uaddr->sa_data, dev->name, 14);
2830         else
2831                 memset(uaddr->sa_data, 0, 14);
2832         rcu_read_unlock();
2833         *uaddr_len = sizeof(*uaddr);
2834
2835         return 0;
2836 }
2837
2838 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2839                           int *uaddr_len, int peer)
2840 {
2841         struct net_device *dev;
2842         struct sock *sk = sock->sk;
2843         struct packet_sock *po = pkt_sk(sk);
2844         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2845
2846         if (peer)
2847                 return -EOPNOTSUPP;
2848
2849         sll->sll_family = AF_PACKET;
2850         sll->sll_ifindex = po->ifindex;
2851         sll->sll_protocol = po->num;
2852         sll->sll_pkttype = 0;
2853         rcu_read_lock();
2854         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2855         if (dev) {
2856                 sll->sll_hatype = dev->type;
2857                 sll->sll_halen = dev->addr_len;
2858                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2859         } else {
2860                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
2861                 sll->sll_halen = 0;
2862         }
2863         rcu_read_unlock();
2864         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2865
2866         return 0;
2867 }
2868
2869 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2870                          int what)
2871 {
2872         switch (i->type) {
2873         case PACKET_MR_MULTICAST:
2874                 if (i->alen != dev->addr_len)
2875                         return -EINVAL;
2876                 if (what > 0)
2877                         return dev_mc_add(dev, i->addr);
2878                 else
2879                         return dev_mc_del(dev, i->addr);
2880                 break;
2881         case PACKET_MR_PROMISC:
2882                 return dev_set_promiscuity(dev, what);
2883                 break;
2884         case PACKET_MR_ALLMULTI:
2885                 return dev_set_allmulti(dev, what);
2886                 break;
2887         case PACKET_MR_UNICAST:
2888                 if (i->alen != dev->addr_len)
2889                         return -EINVAL;
2890                 if (what > 0)
2891                         return dev_uc_add(dev, i->addr);
2892                 else
2893                         return dev_uc_del(dev, i->addr);
2894                 break;
2895         default:
2896                 break;
2897         }
2898         return 0;
2899 }
2900
2901 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2902 {
2903         for ( ; i; i = i->next) {
2904                 if (i->ifindex == dev->ifindex)
2905                         packet_dev_mc(dev, i, what);
2906         }
2907 }
2908
2909 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2910 {
2911         struct packet_sock *po = pkt_sk(sk);
2912         struct packet_mclist *ml, *i;
2913         struct net_device *dev;
2914         int err;
2915
2916         rtnl_lock();
2917
2918         err = -ENODEV;
2919         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2920         if (!dev)
2921                 goto done;
2922
2923         err = -EINVAL;
2924         if (mreq->mr_alen > dev->addr_len)
2925                 goto done;
2926
2927         err = -ENOBUFS;
2928         i = kmalloc(sizeof(*i), GFP_KERNEL);
2929         if (i == NULL)
2930                 goto done;
2931
2932         err = 0;
2933         for (ml = po->mclist; ml; ml = ml->next) {
2934                 if (ml->ifindex == mreq->mr_ifindex &&
2935                     ml->type == mreq->mr_type &&
2936                     ml->alen == mreq->mr_alen &&
2937                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2938                         ml->count++;
2939                         /* Free the new element ... */
2940                         kfree(i);
2941                         goto done;
2942                 }
2943         }
2944
2945         i->type = mreq->mr_type;
2946         i->ifindex = mreq->mr_ifindex;
2947         i->alen = mreq->mr_alen;
2948         memcpy(i->addr, mreq->mr_address, i->alen);
2949         i->count = 1;
2950         i->next = po->mclist;
2951         po->mclist = i;
2952         err = packet_dev_mc(dev, i, 1);
2953         if (err) {
2954                 po->mclist = i->next;
2955                 kfree(i);
2956         }
2957
2958 done:
2959         rtnl_unlock();
2960         return err;
2961 }
2962
2963 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2964 {
2965         struct packet_mclist *ml, **mlp;
2966
2967         rtnl_lock();
2968
2969         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
2970                 if (ml->ifindex == mreq->mr_ifindex &&
2971                     ml->type == mreq->mr_type &&
2972                     ml->alen == mreq->mr_alen &&
2973                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2974                         if (--ml->count == 0) {
2975                                 struct net_device *dev;
2976                                 *mlp = ml->next;
2977                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2978                                 if (dev)
2979                                         packet_dev_mc(dev, ml, -1);
2980                                 kfree(ml);
2981                         }
2982                         rtnl_unlock();
2983                         return 0;
2984                 }
2985         }
2986         rtnl_unlock();
2987         return -EADDRNOTAVAIL;
2988 }
2989
2990 static void packet_flush_mclist(struct sock *sk)
2991 {
2992         struct packet_sock *po = pkt_sk(sk);
2993         struct packet_mclist *ml;
2994
2995         if (!po->mclist)
2996                 return;
2997
2998         rtnl_lock();
2999         while ((ml = po->mclist) != NULL) {
3000                 struct net_device *dev;
3001
3002                 po->mclist = ml->next;
3003                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3004                 if (dev != NULL)
3005                         packet_dev_mc(dev, ml, -1);
3006                 kfree(ml);
3007         }
3008         rtnl_unlock();
3009 }
3010
3011 static int
3012 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3013 {
3014         struct sock *sk = sock->sk;
3015         struct packet_sock *po = pkt_sk(sk);
3016         int ret;
3017
3018         if (level != SOL_PACKET)
3019                 return -ENOPROTOOPT;
3020
3021         switch (optname) {
3022         case PACKET_ADD_MEMBERSHIP:
3023         case PACKET_DROP_MEMBERSHIP:
3024         {
3025                 struct packet_mreq_max mreq;
3026                 int len = optlen;
3027                 memset(&mreq, 0, sizeof(mreq));
3028                 if (len < sizeof(struct packet_mreq))
3029                         return -EINVAL;
3030                 if (len > sizeof(mreq))
3031                         len = sizeof(mreq);
3032                 if (copy_from_user(&mreq, optval, len))
3033                         return -EFAULT;
3034                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3035                         return -EINVAL;
3036                 if (optname == PACKET_ADD_MEMBERSHIP)
3037                         ret = packet_mc_add(sk, &mreq);
3038                 else
3039                         ret = packet_mc_drop(sk, &mreq);
3040                 return ret;
3041         }
3042
3043         case PACKET_RX_RING:
3044         case PACKET_TX_RING:
3045         {
3046                 union tpacket_req_u req_u;
3047                 int len;
3048
3049                 switch (po->tp_version) {
3050                 case TPACKET_V1:
3051                 case TPACKET_V2:
3052                         len = sizeof(req_u.req);
3053                         break;
3054                 case TPACKET_V3:
3055                 default:
3056                         len = sizeof(req_u.req3);
3057                         break;
3058                 }
3059                 if (optlen < len)
3060                         return -EINVAL;
3061                 if (pkt_sk(sk)->has_vnet_hdr)
3062                         return -EINVAL;
3063                 if (copy_from_user(&req_u.req, optval, len))
3064                         return -EFAULT;
3065                 return packet_set_ring(sk, &req_u, 0,
3066                         optname == PACKET_TX_RING);
3067         }
3068         case PACKET_COPY_THRESH:
3069         {
3070                 int val;
3071
3072                 if (optlen != sizeof(val))
3073                         return -EINVAL;
3074                 if (copy_from_user(&val, optval, sizeof(val)))
3075                         return -EFAULT;
3076
3077                 pkt_sk(sk)->copy_thresh = val;
3078                 return 0;
3079         }
3080         case PACKET_VERSION:
3081         {
3082                 int val;
3083
3084                 if (optlen != sizeof(val))
3085                         return -EINVAL;
3086                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3087                         return -EBUSY;
3088                 if (copy_from_user(&val, optval, sizeof(val)))
3089                         return -EFAULT;
3090                 switch (val) {
3091                 case TPACKET_V1:
3092                 case TPACKET_V2:
3093                 case TPACKET_V3:
3094                         po->tp_version = val;
3095                         return 0;
3096                 default:
3097                         return -EINVAL;
3098                 }
3099         }
3100         case PACKET_RESERVE:
3101         {
3102                 unsigned int val;
3103
3104                 if (optlen != sizeof(val))
3105                         return -EINVAL;
3106                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3107                         return -EBUSY;
3108                 if (copy_from_user(&val, optval, sizeof(val)))
3109                         return -EFAULT;
3110                 po->tp_reserve = val;
3111                 return 0;
3112         }
3113         case PACKET_LOSS:
3114         {
3115                 unsigned int val;
3116
3117                 if (optlen != sizeof(val))
3118                         return -EINVAL;
3119                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3120                         return -EBUSY;
3121                 if (copy_from_user(&val, optval, sizeof(val)))
3122                         return -EFAULT;
3123                 po->tp_loss = !!val;
3124                 return 0;
3125         }
3126         case PACKET_AUXDATA:
3127         {
3128                 int val;
3129
3130                 if (optlen < sizeof(val))
3131                         return -EINVAL;
3132                 if (copy_from_user(&val, optval, sizeof(val)))
3133                         return -EFAULT;
3134
3135                 po->auxdata = !!val;
3136                 return 0;
3137         }
3138         case PACKET_ORIGDEV:
3139         {
3140                 int val;
3141
3142                 if (optlen < sizeof(val))
3143                         return -EINVAL;
3144                 if (copy_from_user(&val, optval, sizeof(val)))
3145                         return -EFAULT;
3146
3147                 po->origdev = !!val;
3148                 return 0;
3149         }
3150         case PACKET_VNET_HDR:
3151         {
3152                 int val;
3153
3154                 if (sock->type != SOCK_RAW)
3155                         return -EINVAL;
3156                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3157                         return -EBUSY;
3158                 if (optlen < sizeof(val))
3159                         return -EINVAL;
3160                 if (copy_from_user(&val, optval, sizeof(val)))
3161                         return -EFAULT;
3162
3163                 po->has_vnet_hdr = !!val;
3164                 return 0;
3165         }
3166         case PACKET_TIMESTAMP:
3167         {
3168                 int val;
3169
3170                 if (optlen != sizeof(val))
3171                         return -EINVAL;
3172                 if (copy_from_user(&val, optval, sizeof(val)))
3173                         return -EFAULT;
3174
3175                 po->tp_tstamp = val;
3176                 return 0;
3177         }
3178         case PACKET_FANOUT:
3179         {
3180                 int val;
3181
3182                 if (optlen != sizeof(val))
3183                         return -EINVAL;
3184                 if (copy_from_user(&val, optval, sizeof(val)))
3185                         return -EFAULT;
3186
3187                 return fanout_add(sk, val & 0xffff, val >> 16);
3188         }
3189         default:
3190                 return -ENOPROTOOPT;
3191         }
3192 }
3193
3194 static int packet_getsockopt(struct socket *sock, int level, int optname,
3195                              char __user *optval, int __user *optlen)
3196 {
3197         int len;
3198         int val;
3199         struct sock *sk = sock->sk;
3200         struct packet_sock *po = pkt_sk(sk);
3201         void *data;
3202         struct tpacket_stats st;
3203         union tpacket_stats_u st_u;
3204
3205         if (level != SOL_PACKET)
3206                 return -ENOPROTOOPT;
3207
3208         if (get_user(len, optlen))
3209                 return -EFAULT;
3210
3211         if (len < 0)
3212                 return -EINVAL;
3213
3214         switch (optname) {
3215         case PACKET_STATISTICS:
3216                 if (po->tp_version == TPACKET_V3) {
3217                         len = sizeof(struct tpacket_stats_v3);
3218                 } else {
3219                         if (len > sizeof(struct tpacket_stats))
3220                                 len = sizeof(struct tpacket_stats);
3221                 }
3222                 spin_lock_bh(&sk->sk_receive_queue.lock);
3223                 if (po->tp_version == TPACKET_V3) {
3224                         memcpy(&st_u.stats3, &po->stats,
3225                         sizeof(struct tpacket_stats));
3226                         st_u.stats3.tp_freeze_q_cnt =
3227                         po->stats_u.stats3.tp_freeze_q_cnt;
3228                         st_u.stats3.tp_packets += po->stats.tp_drops;
3229                         data = &st_u.stats3;
3230                 } else {
3231                         st = po->stats;
3232                         st.tp_packets += st.tp_drops;
3233                         data = &st;
3234                 }
3235                 memset(&po->stats, 0, sizeof(st));
3236                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3237                 break;
3238         case PACKET_AUXDATA:
3239                 if (len > sizeof(int))
3240                         len = sizeof(int);
3241                 val = po->auxdata;
3242
3243                 data = &val;
3244                 break;
3245         case PACKET_ORIGDEV:
3246                 if (len > sizeof(int))
3247                         len = sizeof(int);
3248                 val = po->origdev;
3249
3250                 data = &val;
3251                 break;
3252         case PACKET_VNET_HDR:
3253                 if (len > sizeof(int))
3254                         len = sizeof(int);
3255                 val = po->has_vnet_hdr;
3256
3257                 data = &val;
3258                 break;
3259         case PACKET_VERSION:
3260                 if (len > sizeof(int))
3261                         len = sizeof(int);
3262                 val = po->tp_version;
3263                 data = &val;
3264                 break;
3265         case PACKET_HDRLEN:
3266                 if (len > sizeof(int))
3267                         len = sizeof(int);
3268                 if (copy_from_user(&val, optval, len))
3269                         return -EFAULT;
3270                 switch (val) {
3271                 case TPACKET_V1:
3272                         val = sizeof(struct tpacket_hdr);
3273                         break;
3274                 case TPACKET_V2:
3275                         val = sizeof(struct tpacket2_hdr);
3276                         break;
3277                 case TPACKET_V3:
3278                         val = sizeof(struct tpacket3_hdr);
3279                         break;
3280                 default:
3281                         return -EINVAL;
3282                 }
3283                 data = &val;
3284                 break;
3285         case PACKET_RESERVE:
3286                 if (len > sizeof(unsigned int))
3287                         len = sizeof(unsigned int);
3288                 val = po->tp_reserve;
3289                 data = &val;
3290                 break;
3291         case PACKET_LOSS:
3292                 if (len > sizeof(unsigned int))
3293                         len = sizeof(unsigned int);
3294                 val = po->tp_loss;
3295                 data = &val;
3296                 break;
3297         case PACKET_TIMESTAMP:
3298                 if (len > sizeof(int))
3299                         len = sizeof(int);
3300                 val = po->tp_tstamp;
3301                 data = &val;
3302                 break;
3303         case PACKET_FANOUT:
3304                 if (len > sizeof(int))
3305                         len = sizeof(int);
3306                 val = (po->fanout ?
3307                        ((u32)po->fanout->id |
3308                         ((u32)po->fanout->type << 16)) :
3309                        0);
3310                 data = &val;
3311                 break;
3312         default:
3313                 return -ENOPROTOOPT;
3314         }
3315
3316         if (put_user(len, optlen))
3317                 return -EFAULT;
3318         if (copy_to_user(optval, data, len))
3319                 return -EFAULT;
3320         return 0;
3321 }
3322
3323
3324 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3325 {
3326         struct sock *sk;
3327         struct hlist_node *node;
3328         struct net_device *dev = data;
3329         struct net *net = dev_net(dev);
3330
3331         rcu_read_lock();
3332         sk_for_each_rcu(sk, node, &net->packet.sklist) {
3333                 struct packet_sock *po = pkt_sk(sk);
3334
3335                 switch (msg) {
3336                 case NETDEV_UNREGISTER:
3337                         if (po->mclist)
3338                                 packet_dev_mclist(dev, po->mclist, -1);
3339                         /* fallthrough */
3340
3341                 case NETDEV_DOWN:
3342                         if (dev->ifindex == po->ifindex) {
3343                                 spin_lock(&po->bind_lock);
3344                                 if (po->running) {
3345                                         __unregister_prot_hook(sk, false);
3346                                         sk->sk_err = ENETDOWN;
3347                                         if (!sock_flag(sk, SOCK_DEAD))
3348                                                 sk->sk_error_report(sk);
3349                                 }
3350                                 if (msg == NETDEV_UNREGISTER) {
3351                                         po->ifindex = -1;
3352                                         if (po->prot_hook.dev)
3353                                                 dev_put(po->prot_hook.dev);
3354                                         po->prot_hook.dev = NULL;
3355                                 }
3356                                 spin_unlock(&po->bind_lock);
3357                         }
3358                         break;
3359                 case NETDEV_UP:
3360                         if (dev->ifindex == po->ifindex) {
3361                                 spin_lock(&po->bind_lock);
3362                                 if (po->num)
3363                                         register_prot_hook(sk);
3364                                 spin_unlock(&po->bind_lock);
3365                         }
3366                         break;
3367                 }
3368         }
3369         rcu_read_unlock();
3370         return NOTIFY_DONE;
3371 }
3372
3373
3374 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3375                         unsigned long arg)
3376 {
3377         struct sock *sk = sock->sk;
3378
3379         switch (cmd) {
3380         case SIOCOUTQ:
3381         {
3382                 int amount = sk_wmem_alloc_get(sk);
3383
3384                 return put_user(amount, (int __user *)arg);
3385         }
3386         case SIOCINQ:
3387         {
3388                 struct sk_buff *skb;
3389                 int amount = 0;
3390
3391                 spin_lock_bh(&sk->sk_receive_queue.lock);
3392                 skb = skb_peek(&sk->sk_receive_queue);
3393                 if (skb)
3394                         amount = skb->len;
3395                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3396                 return put_user(amount, (int __user *)arg);
3397         }
3398         case SIOCGSTAMP:
3399                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3400         case SIOCGSTAMPNS:
3401                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3402
3403 #ifdef CONFIG_INET
3404         case SIOCADDRT:
3405         case SIOCDELRT:
3406         case SIOCDARP:
3407         case SIOCGARP:
3408         case SIOCSARP:
3409         case SIOCGIFADDR:
3410         case SIOCSIFADDR:
3411         case SIOCGIFBRDADDR:
3412         case SIOCSIFBRDADDR:
3413         case SIOCGIFNETMASK:
3414         case SIOCSIFNETMASK:
3415         case SIOCGIFDSTADDR:
3416         case SIOCSIFDSTADDR:
3417         case SIOCSIFFLAGS:
3418                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3419 #endif
3420
3421         default:
3422                 return -ENOIOCTLCMD;
3423         }
3424         return 0;
3425 }
3426
3427 static unsigned int packet_poll(struct file *file, struct socket *sock,
3428                                 poll_table *wait)
3429 {
3430         struct sock *sk = sock->sk;
3431         struct packet_sock *po = pkt_sk(sk);
3432         unsigned int mask = datagram_poll(file, sock, wait);
3433
3434         spin_lock_bh(&sk->sk_receive_queue.lock);
3435         if (po->rx_ring.pg_vec) {
3436                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3437                         TP_STATUS_KERNEL))
3438                         mask |= POLLIN | POLLRDNORM;
3439         }
3440         spin_unlock_bh(&sk->sk_receive_queue.lock);
3441         spin_lock_bh(&sk->sk_write_queue.lock);
3442         if (po->tx_ring.pg_vec) {
3443                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3444                         mask |= POLLOUT | POLLWRNORM;
3445         }
3446         spin_unlock_bh(&sk->sk_write_queue.lock);
3447         return mask;
3448 }
3449
3450
3451 /* Dirty? Well, I still did not learn better way to account
3452  * for user mmaps.
3453  */
3454
3455 static void packet_mm_open(struct vm_area_struct *vma)
3456 {
3457         struct file *file = vma->vm_file;
3458         struct socket *sock = file->private_data;
3459         struct sock *sk = sock->sk;
3460
3461         if (sk)
3462                 atomic_inc(&pkt_sk(sk)->mapped);
3463 }
3464
3465 static void packet_mm_close(struct vm_area_struct *vma)
3466 {
3467         struct file *file = vma->vm_file;
3468         struct socket *sock = file->private_data;
3469         struct sock *sk = sock->sk;
3470
3471         if (sk)
3472                 atomic_dec(&pkt_sk(sk)->mapped);
3473 }
3474
3475 static const struct vm_operations_struct packet_mmap_ops = {
3476         .open   =       packet_mm_open,
3477         .close  =       packet_mm_close,
3478 };
3479
3480 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3481                         unsigned int len)
3482 {
3483         int i;
3484
3485         for (i = 0; i < len; i++) {
3486                 if (likely(pg_vec[i].buffer)) {
3487                         if (is_vmalloc_addr(pg_vec[i].buffer))
3488                                 vfree(pg_vec[i].buffer);
3489                         else
3490                                 free_pages((unsigned long)pg_vec[i].buffer,
3491                                            order);
3492                         pg_vec[i].buffer = NULL;
3493                 }
3494         }
3495         kfree(pg_vec);
3496 }
3497
3498 static char *alloc_one_pg_vec_page(unsigned long order)
3499 {
3500         char *buffer = NULL;
3501         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3502                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3503
3504         buffer = (char *) __get_free_pages(gfp_flags, order);
3505
3506         if (buffer)
3507                 return buffer;
3508
3509         /*
3510          * __get_free_pages failed, fall back to vmalloc
3511          */
3512         buffer = vzalloc((1 << order) * PAGE_SIZE);
3513
3514         if (buffer)
3515                 return buffer;
3516
3517         /*
3518          * vmalloc failed, lets dig into swap here
3519          */
3520         gfp_flags &= ~__GFP_NORETRY;
3521         buffer = (char *)__get_free_pages(gfp_flags, order);
3522         if (buffer)
3523                 return buffer;
3524
3525         /*
3526          * complete and utter failure
3527          */
3528         return NULL;
3529 }
3530
3531 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3532 {
3533         unsigned int block_nr = req->tp_block_nr;
3534         struct pgv *pg_vec;
3535         int i;
3536
3537         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3538         if (unlikely(!pg_vec))
3539                 goto out;
3540
3541         for (i = 0; i < block_nr; i++) {
3542                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3543                 if (unlikely(!pg_vec[i].buffer))
3544                         goto out_free_pgvec;
3545         }
3546
3547 out:
3548         return pg_vec;
3549
3550 out_free_pgvec:
3551         free_pg_vec(pg_vec, order, block_nr);
3552         pg_vec = NULL;
3553         goto out;
3554 }
3555
3556 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3557                 int closing, int tx_ring)
3558 {
3559         struct pgv *pg_vec = NULL;
3560         struct packet_sock *po = pkt_sk(sk);
3561         int was_running, order = 0;
3562         struct packet_ring_buffer *rb;
3563         struct sk_buff_head *rb_queue;
3564         __be16 num;
3565         int err = -EINVAL;
3566         /* Added to avoid minimal code churn */
3567         struct tpacket_req *req = &req_u->req;
3568
3569         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3570         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3571                 WARN(1, "Tx-ring is not supported.\n");
3572                 goto out;
3573         }
3574
3575         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3576         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3577
3578         err = -EBUSY;
3579         if (!closing) {
3580                 if (atomic_read(&po->mapped))
3581                         goto out;
3582                 if (atomic_read(&rb->pending))
3583                         goto out;
3584         }
3585
3586         if (req->tp_block_nr) {
3587                 /* Sanity tests and some calculations */
3588                 err = -EBUSY;
3589                 if (unlikely(rb->pg_vec))
3590                         goto out;
3591
3592                 switch (po->tp_version) {
3593                 case TPACKET_V1:
3594                         po->tp_hdrlen = TPACKET_HDRLEN;
3595                         break;
3596                 case TPACKET_V2:
3597                         po->tp_hdrlen = TPACKET2_HDRLEN;
3598                         break;
3599                 case TPACKET_V3:
3600                         po->tp_hdrlen = TPACKET3_HDRLEN;
3601                         break;
3602                 }
3603
3604                 err = -EINVAL;
3605                 if (unlikely((int)req->tp_block_size <= 0))
3606                         goto out;
3607                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3608                         goto out;
3609                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3610                                         po->tp_reserve))
3611                         goto out;
3612                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3613                         goto out;
3614
3615                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3616                 if (unlikely(rb->frames_per_block <= 0))
3617                         goto out;
3618                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3619                                         req->tp_frame_nr))
3620                         goto out;
3621
3622                 err = -ENOMEM;
3623                 order = get_order(req->tp_block_size);
3624                 pg_vec = alloc_pg_vec(req, order);
3625                 if (unlikely(!pg_vec))
3626                         goto out;
3627                 switch (po->tp_version) {
3628                 case TPACKET_V3:
3629                 /* Transmit path is not supported. We checked
3630                  * it above but just being paranoid
3631                  */
3632                         if (!tx_ring)
3633                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3634                                 break;
3635                 default:
3636                         break;
3637                 }
3638         }
3639         /* Done */
3640         else {
3641                 err = -EINVAL;
3642                 if (unlikely(req->tp_frame_nr))
3643                         goto out;
3644         }
3645
3646         lock_sock(sk);
3647
3648         /* Detach socket from network */
3649         spin_lock(&po->bind_lock);
3650         was_running = po->running;
3651         num = po->num;
3652         if (was_running) {
3653                 po->num = 0;
3654                 __unregister_prot_hook(sk, false);
3655         }
3656         spin_unlock(&po->bind_lock);
3657
3658         synchronize_net();
3659
3660         err = -EBUSY;
3661         mutex_lock(&po->pg_vec_lock);
3662         if (closing || atomic_read(&po->mapped) == 0) {
3663                 err = 0;
3664                 spin_lock_bh(&rb_queue->lock);
3665                 swap(rb->pg_vec, pg_vec);
3666                 rb->frame_max = (req->tp_frame_nr - 1);
3667                 rb->head = 0;
3668                 rb->frame_size = req->tp_frame_size;
3669                 spin_unlock_bh(&rb_queue->lock);
3670
3671                 swap(rb->pg_vec_order, order);
3672                 swap(rb->pg_vec_len, req->tp_block_nr);
3673
3674                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3675                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3676                                                 tpacket_rcv : packet_rcv;
3677                 skb_queue_purge(rb_queue);
3678                 if (atomic_read(&po->mapped))
3679                         pr_err("packet_mmap: vma is busy: %d\n",
3680                                atomic_read(&po->mapped));
3681         }
3682         mutex_unlock(&po->pg_vec_lock);
3683
3684         spin_lock(&po->bind_lock);
3685         if (was_running) {
3686                 po->num = num;
3687                 register_prot_hook(sk);
3688         }
3689         spin_unlock(&po->bind_lock);
3690         if (closing && (po->tp_version > TPACKET_V2)) {
3691                 /* Because we don't support block-based V3 on tx-ring */
3692                 if (!tx_ring)
3693                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3694         }
3695         release_sock(sk);
3696
3697         if (pg_vec)
3698                 free_pg_vec(pg_vec, order, req->tp_block_nr);
3699 out:
3700         return err;
3701 }
3702
3703 static int packet_mmap(struct file *file, struct socket *sock,
3704                 struct vm_area_struct *vma)
3705 {
3706         struct sock *sk = sock->sk;
3707         struct packet_sock *po = pkt_sk(sk);
3708         unsigned long size, expected_size;
3709         struct packet_ring_buffer *rb;
3710         unsigned long start;
3711         int err = -EINVAL;
3712         int i;
3713
3714         if (vma->vm_pgoff)
3715                 return -EINVAL;
3716
3717         mutex_lock(&po->pg_vec_lock);
3718
3719         expected_size = 0;
3720         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3721                 if (rb->pg_vec) {
3722                         expected_size += rb->pg_vec_len
3723                                                 * rb->pg_vec_pages
3724                                                 * PAGE_SIZE;
3725                 }
3726         }
3727
3728         if (expected_size == 0)
3729                 goto out;
3730
3731         size = vma->vm_end - vma->vm_start;
3732         if (size != expected_size)
3733                 goto out;
3734
3735         start = vma->vm_start;
3736         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3737                 if (rb->pg_vec == NULL)
3738                         continue;
3739
3740                 for (i = 0; i < rb->pg_vec_len; i++) {
3741                         struct page *page;
3742                         void *kaddr = rb->pg_vec[i].buffer;
3743                         int pg_num;
3744
3745                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3746                                 page = pgv_to_page(kaddr);
3747                                 err = vm_insert_page(vma, start, page);
3748                                 if (unlikely(err))
3749                                         goto out;
3750                                 start += PAGE_SIZE;
3751                                 kaddr += PAGE_SIZE;
3752                         }
3753                 }
3754         }
3755
3756         atomic_inc(&po->mapped);
3757         vma->vm_ops = &packet_mmap_ops;
3758         err = 0;
3759
3760 out:
3761         mutex_unlock(&po->pg_vec_lock);
3762         return err;
3763 }
3764
3765 static const struct proto_ops packet_ops_spkt = {
3766         .family =       PF_PACKET,
3767         .owner =        THIS_MODULE,
3768         .release =      packet_release,
3769         .bind =         packet_bind_spkt,
3770         .connect =      sock_no_connect,
3771         .socketpair =   sock_no_socketpair,
3772         .accept =       sock_no_accept,
3773         .getname =      packet_getname_spkt,
3774         .poll =         datagram_poll,
3775         .ioctl =        packet_ioctl,
3776         .listen =       sock_no_listen,
3777         .shutdown =     sock_no_shutdown,
3778         .setsockopt =   sock_no_setsockopt,
3779         .getsockopt =   sock_no_getsockopt,
3780         .sendmsg =      packet_sendmsg_spkt,
3781         .recvmsg =      packet_recvmsg,
3782         .mmap =         sock_no_mmap,
3783         .sendpage =     sock_no_sendpage,
3784 };
3785
3786 static const struct proto_ops packet_ops = {
3787         .family =       PF_PACKET,
3788         .owner =        THIS_MODULE,
3789         .release =      packet_release,
3790         .bind =         packet_bind,
3791         .connect =      sock_no_connect,
3792         .socketpair =   sock_no_socketpair,
3793         .accept =       sock_no_accept,
3794         .getname =      packet_getname,
3795         .poll =         packet_poll,
3796         .ioctl =        packet_ioctl,
3797         .listen =       sock_no_listen,
3798         .shutdown =     sock_no_shutdown,
3799         .setsockopt =   packet_setsockopt,
3800         .getsockopt =   packet_getsockopt,
3801         .sendmsg =      packet_sendmsg,
3802         .recvmsg =      packet_recvmsg,
3803         .mmap =         packet_mmap,
3804         .sendpage =     sock_no_sendpage,
3805 };
3806
3807 static const struct net_proto_family packet_family_ops = {
3808         .family =       PF_PACKET,
3809         .create =       packet_create,
3810         .owner  =       THIS_MODULE,
3811 };
3812
3813 static struct notifier_block packet_netdev_notifier = {
3814         .notifier_call =        packet_notifier,
3815 };
3816
3817 #ifdef CONFIG_PROC_FS
3818
3819 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3820         __acquires(RCU)
3821 {
3822         struct net *net = seq_file_net(seq);
3823
3824         rcu_read_lock();
3825         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3826 }
3827
3828 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3829 {
3830         struct net *net = seq_file_net(seq);
3831         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3832 }
3833
3834 static void packet_seq_stop(struct seq_file *seq, void *v)
3835         __releases(RCU)
3836 {
3837         rcu_read_unlock();
3838 }
3839
3840 static int packet_seq_show(struct seq_file *seq, void *v)
3841 {
3842         if (v == SEQ_START_TOKEN)
3843                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
3844         else {
3845                 struct sock *s = sk_entry(v);
3846                 const struct packet_sock *po = pkt_sk(s);
3847
3848                 seq_printf(seq,
3849                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
3850                            s,
3851                            atomic_read(&s->sk_refcnt),
3852                            s->sk_type,
3853                            ntohs(po->num),
3854                            po->ifindex,
3855                            po->running,
3856                            atomic_read(&s->sk_rmem_alloc),
3857                            sock_i_uid(s),
3858                            sock_i_ino(s));
3859         }
3860
3861         return 0;
3862 }
3863
3864 static const struct seq_operations packet_seq_ops = {
3865         .start  = packet_seq_start,
3866         .next   = packet_seq_next,
3867         .stop   = packet_seq_stop,
3868         .show   = packet_seq_show,
3869 };
3870
3871 static int packet_seq_open(struct inode *inode, struct file *file)
3872 {
3873         return seq_open_net(inode, file, &packet_seq_ops,
3874                             sizeof(struct seq_net_private));
3875 }
3876
3877 static const struct file_operations packet_seq_fops = {
3878         .owner          = THIS_MODULE,
3879         .open           = packet_seq_open,
3880         .read           = seq_read,
3881         .llseek         = seq_lseek,
3882         .release        = seq_release_net,
3883 };
3884
3885 #endif
3886
3887 static int __net_init packet_net_init(struct net *net)
3888 {
3889         spin_lock_init(&net->packet.sklist_lock);
3890         INIT_HLIST_HEAD(&net->packet.sklist);
3891
3892         if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
3893                 return -ENOMEM;
3894
3895         return 0;
3896 }
3897
3898 static void __net_exit packet_net_exit(struct net *net)
3899 {
3900         proc_net_remove(net, "packet");
3901 }
3902
3903 static struct pernet_operations packet_net_ops = {
3904         .init = packet_net_init,
3905         .exit = packet_net_exit,
3906 };
3907
3908
3909 static void __exit packet_exit(void)
3910 {
3911         unregister_netdevice_notifier(&packet_netdev_notifier);
3912         unregister_pernet_subsys(&packet_net_ops);
3913         sock_unregister(PF_PACKET);
3914         proto_unregister(&packet_proto);
3915 }
3916
3917 static int __init packet_init(void)
3918 {
3919         int rc = proto_register(&packet_proto, 0);
3920
3921         if (rc != 0)
3922                 goto out;
3923
3924         sock_register(&packet_family_ops);
3925         register_pernet_subsys(&packet_net_ops);
3926         register_netdevice_notifier(&packet_netdev_notifier);
3927 out:
3928         return rc;
3929 }
3930
3931 module_init(packet_init);
3932 module_exit(packet_exit);
3933 MODULE_LICENSE("GPL");
3934 MODULE_ALIAS_NETPROTO(PF_PACKET);