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/uaccess.h>
  77 #include <asm/ioctls.h>
  78 #include <asm/page.h>
  79 #include <asm/cacheflush.h>
  80 #include <asm/io.h>
  81 #include <linux/proc_fs.h>
  82 #include <linux/seq_file.h>
  83 #include <linux/poll.h>
  84 #include <linux/module.h>
  85 #include <linux/init.h>
  86 #include <linux/mutex.h>
  87 #include <linux/if_vlan.h>
  88 #include <linux/virtio_net.h>
  89 #include <linux/errqueue.h>
  90 #include <linux/net_tstamp.h>
  91
  92 #ifdef CONFIG_INET
  93 #include <net/inet_common.h>
  94 #endif
  95
  96 /*
  97    Assumptions:
  98    - if device has no dev->hard_header routine, it adds and removes ll header
  99      inside itself. In this case ll header is invisible outside of device,
 100      but higher levels still should reserve dev->hard_header_len.
 101      Some devices are enough clever to reallocate skb, when header
 102      will not fit to reserved space (tunnel), another ones are silly
 103      (PPP).
 104    - packet socket receives packets with pulled ll header,
 105      so that SOCK_RAW should push it back.
 106
 107 On receive:
 108 -----------
 109
 110 Incoming, dev->hard_header!=NULL
 111    mac_header -> ll header
 112    data       -> data
 113
 114 Outgoing, dev->hard_header!=NULL
 115    mac_header -> ll header
 116    data       -> ll header
 117
 118 Incoming, dev->hard_header==NULL
 119    mac_header -> UNKNOWN position. It is very likely, that it points to ll
 120                  header.  PPP makes it, that is wrong, because introduce
 121                  assymetry between rx and tx paths.
 122    data       -> data
 123
 124 Outgoing, dev->hard_header==NULL
 125    mac_header -> data. ll header is still not built!
 126    data       -> data
 127
 128 Resume
 129   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 130
 131
 132 On transmit:
 133 ------------
 134
 135 dev->hard_header != NULL
 136    mac_header -> ll header
 137    data       -> ll header
 138
 139 dev->hard_header == NULL (ll header is added by device, we cannot control it)
 140    mac_header -> data
 141    data       -> data
 142
 143    We should set nh.raw on output to correct posistion,
 144    packet classifier depends on it.
 145  */
 146
 147 /* Private packet socket structures. */
 148
 149 struct packet_mclist {
 150         struct packet_mclist    *next;
 151         int                     ifindex;
 152         int                     count;
 153         unsigned short          type;
 154         unsigned short          alen;
 155         unsigned char           addr[MAX_ADDR_LEN];
 156 };
 157 /* identical to struct packet_mreq except it has
 158  * a longer address field.
 159  */
 160 struct packet_mreq_max {
 161         int             mr_ifindex;
 162         unsigned short  mr_type;
 163         unsigned short  mr_alen;
 164         unsigned char   mr_address[MAX_ADDR_LEN];
 165 };
 166
 167 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 168                 int closing, int tx_ring);
 169
 170
 171 #define V3_ALIGNMENT    (8)
 172
 173 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 174
 175 #define BLK_PLUS_PRIV(sz_of_priv) \
 176         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 177
 178 /* kbdq - kernel block descriptor queue */
 179 struct tpacket_kbdq_core {
 180         struct pgv      *pkbdq;
 181         unsigned int    feature_req_word;
 182         unsigned int    hdrlen;
 183         unsigned char   reset_pending_on_curr_blk;
 184         unsigned char   delete_blk_timer;
 185         unsigned short  kactive_blk_num;
 186         unsigned short  blk_sizeof_priv;
 187
 188         /* last_kactive_blk_num:
 189          * trick to see if user-space has caught up
 190          * in order to avoid refreshing timer when every single pkt arrives.
 191          */
 192         unsigned short  last_kactive_blk_num;
 193
 194         char            *pkblk_start;
 195         char            *pkblk_end;
 196         int             kblk_size;
 197         unsigned int    knum_blocks;
 198         uint64_t        knxt_seq_num;
 199         char            *prev;
 200         char            *nxt_offset;
 201         struct sk_buff  *skb;
 202
 203         atomic_t        blk_fill_in_prog;
 204
 205         /* Default is set to 8ms */
 206 #define DEFAULT_PRB_RETIRE_TOV  (8)
 207
 208         unsigned short  retire_blk_tov;
 209         unsigned short  version;
 210         unsigned long   tov_in_jiffies;
 211
 212         /* timer to retire an outstanding block */
 213         struct timer_list retire_blk_timer;
 214 };
 215
 216 #define PGV_FROM_VMALLOC 1
 217 struct pgv {
 218         char *buffer;
 219 };
 220
 221 struct packet_ring_buffer {
 222         struct pgv              *pg_vec;
 223         unsigned int            head;
 224         unsigned int            frames_per_block;
 225         unsigned int            frame_size;
 226         unsigned int            frame_max;
 227
 228         unsigned int            pg_vec_order;
 229         unsigned int            pg_vec_pages;
 230         unsigned int            pg_vec_len;
 231
 232         struct tpacket_kbdq_core        prb_bdqc;
 233         atomic_t                pending;
 234 };
 235
 236 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
 237 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
 238 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
 239 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
 240 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
 241 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
 242 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
 243
 244 struct packet_sock;
 245 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
 246
 247 static void *packet_previous_frame(struct packet_sock *po,
 248                 struct packet_ring_buffer *rb,
 249                 int status);
 250 static void packet_increment_head(struct packet_ring_buffer *buff);
 251 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
 252                         struct tpacket_block_desc *);
 253 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 254                         struct packet_sock *);
 255 static void prb_retire_current_block(struct tpacket_kbdq_core *,
 256                 struct packet_sock *, unsigned int status);
 257 static int prb_queue_frozen(struct tpacket_kbdq_core *);
 258 static void prb_open_block(struct tpacket_kbdq_core *,
 259                 struct tpacket_block_desc *);
 260 static void prb_retire_rx_blk_timer_expired(unsigned long);
 261 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 262 static void prb_init_blk_timer(struct packet_sock *,
 263                 struct tpacket_kbdq_core *,
 264                 void (*func) (unsigned long));
 265 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 266 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 267                 struct tpacket3_hdr *);
 268 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 269                 struct tpacket3_hdr *);
 270 static void packet_flush_mclist(struct sock *sk);
 271
 272 struct packet_fanout;
 273 struct packet_sock {
 274         /* struct sock has to be the first member of packet_sock */
 275         struct sock             sk;
 276         struct packet_fanout    *fanout;
 277         struct tpacket_stats    stats;
 278         union  tpacket_stats_u  stats_u;
 279         struct packet_ring_buffer       rx_ring;
 280         struct packet_ring_buffer       tx_ring;
 281         int                     copy_thresh;
 282         spinlock_t              bind_lock;
 283         struct mutex            pg_vec_lock;
 284         unsigned int            running:1,      /* prot_hook is attached*/
 285                                 auxdata:1,
 286                                 origdev:1,
 287                                 has_vnet_hdr:1;
 288         int                     ifindex;        /* bound device         */
 289         __be16                  num;
 290         struct packet_mclist    *mclist;
 291         atomic_t                mapped;
 292         enum tpacket_versions   tp_version;
 293         unsigned int            tp_hdrlen;
 294         unsigned int            tp_reserve;
 295         unsigned int            tp_loss:1;
 296         unsigned int            tp_tstamp;
 297         struct packet_type      prot_hook ____cacheline_aligned_in_smp;
 298 };
 299
 300 #define PACKET_FANOUT_MAX       256
 301
 302 struct packet_fanout {
 303 #ifdef CONFIG_NET_NS
 304         struct net              *net;
 305 #endif
 306         unsigned int            num_members;
 307         u16                     id;
 308         u8                      type;
 309         u8                      defrag;
 310         atomic_t                rr_cur;
 311         struct list_head        list;
 312         struct sock             *arr[PACKET_FANOUT_MAX];
 313         spinlock_t              lock;
 314         atomic_t                sk_ref;
 315         struct packet_type      prot_hook ____cacheline_aligned_in_smp;
 316 };
 317
 318 struct packet_skb_cb {
 319         unsigned int origlen;
 320         union {
 321                 struct sockaddr_pkt pkt;
 322                 struct sockaddr_ll ll;
 323         } sa;
 324 };
 325
 326 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 327
 328 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 329 #define GET_PBLOCK_DESC(x, bid) \
 330         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 331 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 332         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 333 #define GET_NEXT_PRB_BLK_NUM(x) \
 334         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 335         ((x)->kactive_blk_num+1) : 0)
 336
 337 static struct packet_sock *pkt_sk(struct sock *sk)
 338 {
 339         return (struct packet_sock *)sk;
 340 }
 341
 342 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 343 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 344
 345 /* register_prot_hook must be invoked with the po->bind_lock held,
 346  * or from a context in which asynchronous accesses to the packet
 347  * socket is not possible (packet_create()).
 348  */
 349 static void register_prot_hook(struct sock *sk)
 350 {
 351         struct packet_sock *po = pkt_sk(sk);
 352         if (!po->running) {
 353                 if (po->fanout)
 354                         __fanout_link(sk, po);
 355                 else
 356                         dev_add_pack(&po->prot_hook);
 357                 sock_hold(sk);
 358                 po->running = 1;
 359         }
 360 }
 361
 362 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
 363  * held.   If the sync parameter is true, we will temporarily drop
 364  * the po->bind_lock and do a synchronize_net to make sure no
 365  * asynchronous packet processing paths still refer to the elements
 366  * of po->prot_hook.  If the sync parameter is false, it is the
 367  * callers responsibility to take care of this.
 368  */
 369 static void __unregister_prot_hook(struct sock *sk, bool sync)
 370 {
 371         struct packet_sock *po = pkt_sk(sk);
 372
 373         po->running = 0;
 374         if (po->fanout)
 375                 __fanout_unlink(sk, po);
 376         else
 377                 __dev_remove_pack(&po->prot_hook);
 378         __sock_put(sk);
 379
 380         if (sync) {
 381                 spin_unlock(&po->bind_lock);
 382                 synchronize_net();
 383                 spin_lock(&po->bind_lock);
 384         }
 385 }
 386
 387 static void unregister_prot_hook(struct sock *sk, bool sync)
 388 {
 389         struct packet_sock *po = pkt_sk(sk);
 390
 391         if (po->running)
 392                 __unregister_prot_hook(sk, sync);
 393 }
 394
 395 static inline __pure struct page *pgv_to_page(void *addr)
 396 {
 397         if (is_vmalloc_addr(addr))
 398                 return vmalloc_to_page(addr);
 399         return virt_to_page(addr);
 400 }
 401
 402 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 403 {
 404         union {
 405                 struct tpacket_hdr *h1;
 406                 struct tpacket2_hdr *h2;
 407                 void *raw;
 408         } h;
 409
 410         h.raw = frame;
 411         switch (po->tp_version) {
 412         case TPACKET_V1:
 413                 h.h1->tp_status = status;
 414                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 415                 break;
 416         case TPACKET_V2:
 417                 h.h2->tp_status = status;
 418                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 419                 break;
 420         case TPACKET_V3:
 421         default:
 422                 WARN(1, "TPACKET version not supported.\n");
 423                 BUG();
 424         }
 425
 426         smp_wmb();
 427 }
 428
 429 static int __packet_get_status(struct packet_sock *po, void *frame)
 430 {
 431         union {
 432                 struct tpacket_hdr *h1;
 433                 struct tpacket2_hdr *h2;
 434                 void *raw;
 435         } h;
 436
 437         smp_rmb();
 438
 439         h.raw = frame;
 440         switch (po->tp_version) {
 441         case TPACKET_V1:
 442                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 443                 return h.h1->tp_status;
 444         case TPACKET_V2:
 445                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 446                 return h.h2->tp_status;
 447         case TPACKET_V3:
 448         default:
 449                 WARN(1, "TPACKET version not supported.\n");
 450                 BUG();
 451                 return 0;
 452         }
 453 }
 454
 455 static void *packet_lookup_frame(struct packet_sock *po,
 456                 struct packet_ring_buffer *rb,
 457                 unsigned int position,
 458                 int status)
 459 {
 460         unsigned int pg_vec_pos, frame_offset;
 461         union {
 462                 struct tpacket_hdr *h1;
 463                 struct tpacket2_hdr *h2;
 464                 void *raw;
 465         } h;
 466
 467         pg_vec_pos = position / rb->frames_per_block;
 468         frame_offset = position % rb->frames_per_block;
 469
 470         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 471                 (frame_offset * rb->frame_size);
 472
 473         if (status != __packet_get_status(po, h.raw))
 474                 return NULL;
 475
 476         return h.raw;
 477 }
 478
 479 static void *packet_current_frame(struct packet_sock *po,
 480                 struct packet_ring_buffer *rb,
 481                 int status)
 482 {
 483         return packet_lookup_frame(po, rb, rb->head, status);
 484 }
 485
 486 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 487 {
 488         del_timer_sync(&pkc->retire_blk_timer);
 489 }
 490
 491 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 492                 int tx_ring,
 493                 struct sk_buff_head *rb_queue)
 494 {
 495         struct tpacket_kbdq_core *pkc;
 496
 497         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 498
 499         spin_lock(&rb_queue->lock);
 500         pkc->delete_blk_timer = 1;
 501         spin_unlock(&rb_queue->lock);
 502
 503         prb_del_retire_blk_timer(pkc);
 504 }
 505
 506 static void prb_init_blk_timer(struct packet_sock *po,
 507                 struct tpacket_kbdq_core *pkc,
 508                 void (*func) (unsigned long))
 509 {
 510         init_timer(&pkc->retire_blk_timer);
 511         pkc->retire_blk_timer.data = (long)po;
 512         pkc->retire_blk_timer.function = func;
 513         pkc->retire_blk_timer.expires = jiffies;
 514 }
 515
 516 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
 517 {
 518         struct tpacket_kbdq_core *pkc;
 519
 520         if (tx_ring)
 521                 BUG();
 522
 523         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 524         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
 525 }
 526
 527 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 528                                 int blk_size_in_bytes)
 529 {
 530         struct net_device *dev;
 531         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 532         struct ethtool_cmd ecmd;
 533         int err;
 534         u32 speed;
 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         speed = ethtool_cmd_speed(&ecmd);
 544         rtnl_unlock();
 545         if (!err) {
 546                 /*
 547                  * If the link speed is so slow you don't really
 548                  * need to worry about perf anyways
 549                  */
 550                 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
 551                         return DEFAULT_PRB_RETIRE_TOV;
 552                 } else {
 553                         msec = 1;
 554                         div = speed / 1000;
 555                 }
 556         }
 557
 558         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 559
 560         if (div)
 561                 mbits /= div;
 562
 563         tmo = mbits * msec;
 564
 565         if (div)
 566                 return tmo+1;
 567         return tmo;
 568 }
 569
 570 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 571                         union tpacket_req_u *req_u)
 572 {
 573         p1->feature_req_word = req_u->req3.tp_feature_req_word;
 574 }
 575
 576 static void init_prb_bdqc(struct packet_sock *po,
 577                         struct packet_ring_buffer *rb,
 578                         struct pgv *pg_vec,
 579                         union tpacket_req_u *req_u, int tx_ring)
 580 {
 581         struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
 582         struct tpacket_block_desc *pbd;
 583
 584         memset(p1, 0x0, sizeof(*p1));
 585
 586         p1->knxt_seq_num = 1;
 587         p1->pkbdq = pg_vec;
 588         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 589         p1->pkblk_start = pg_vec[0].buffer;
 590         p1->kblk_size = req_u->req3.tp_block_size;
 591         p1->knum_blocks = req_u->req3.tp_block_nr;
 592         p1->hdrlen = po->tp_hdrlen;
 593         p1->version = po->tp_version;
 594         p1->last_kactive_blk_num = 0;
 595         po->stats_u.stats3.tp_freeze_q_cnt = 0;
 596         if (req_u->req3.tp_retire_blk_tov)
 597                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 598         else
 599                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 600                                                 req_u->req3.tp_block_size);
 601         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 602         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 603
 604         prb_init_ft_ops(p1, req_u);
 605         prb_setup_retire_blk_timer(po, tx_ring);
 606         prb_open_block(p1, pbd);
 607 }
 608
 609 /*  Do NOT update the last_blk_num first.
 610  *  Assumes sk_buff_head lock is held.
 611  */
 612 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 613 {
 614         mod_timer(&pkc->retire_blk_timer,
 615                         jiffies + pkc->tov_in_jiffies);
 616         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 617 }
 618
 619 /*
 620  * Timer logic:
 621  * 1) We refresh the timer only when we open a block.
 622  *    By doing this we don't waste cycles refreshing the timer
 623  *        on packet-by-packet basis.
 624  *
 625  * With a 1MB block-size, on a 1Gbps line, it will take
 626  * i) ~8 ms to fill a block + ii) memcpy etc.
 627  * In this cut we are not accounting for the memcpy time.
 628  *
 629  * So, if the user sets the 'tmo' to 10ms then the timer
 630  * will never fire while the block is still getting filled
 631  * (which is what we want). However, the user could choose
 632  * to close a block early and that's fine.
 633  *
 634  * But when the timer does fire, we check whether or not to refresh it.
 635  * Since the tmo granularity is in msecs, it is not too expensive
 636  * to refresh the timer, lets say every '8' msecs.
 637  * Either the user can set the 'tmo' or we can derive it based on
 638  * a) line-speed and b) block-size.
 639  * prb_calc_retire_blk_tmo() calculates the tmo.
 640  *
 641  */
 642 static void prb_retire_rx_blk_timer_expired(unsigned long data)
 643 {
 644         struct packet_sock *po = (struct packet_sock *)data;
 645         struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
 646         unsigned int frozen;
 647         struct tpacket_block_desc *pbd;
 648
 649         spin_lock(&po->sk.sk_receive_queue.lock);
 650
 651         frozen = prb_queue_frozen(pkc);
 652         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 653
 654         if (unlikely(pkc->delete_blk_timer))
 655                 goto out;
 656
 657         /* We only need to plug the race when the block is partially filled.
 658          * tpacket_rcv:
 659          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 660          *              copy_bits() is in progress ...
 661          *              timer fires on other cpu:
 662          *              we can't retire the current block because copy_bits
 663          *              is in progress.
 664          *
 665          */
 666         if (BLOCK_NUM_PKTS(pbd)) {
 667                 while (atomic_read(&pkc->blk_fill_in_prog)) {
 668                         /* Waiting for skb_copy_bits to finish... */
 669                         cpu_relax();
 670                 }
 671         }
 672
 673         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 674                 if (!frozen) {
 675                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 676                         if (!prb_dispatch_next_block(pkc, po))
 677                                 goto refresh_timer;
 678                         else
 679                                 goto out;
 680                 } else {
 681                         /* Case 1. Queue was frozen because user-space was
 682                          *         lagging behind.
 683                          */
 684                         if (prb_curr_blk_in_use(pkc, pbd)) {
 685                                 /*
 686                                  * Ok, user-space is still behind.
 687                                  * So just refresh the timer.
 688                                  */
 689                                 goto refresh_timer;
 690                         } else {
 691                                /* Case 2. queue was frozen,user-space caught up,
 692                                 * now the link went idle && the timer fired.
 693                                 * We don't have a block to close.So we open this
 694                                 * block and restart the timer.
 695                                 * opening a block thaws the queue,restarts timer
 696                                 * Thawing/timer-refresh is a side effect.
 697                                 */
 698                                 prb_open_block(pkc, pbd);
 699                                 goto out;
 700                         }
 701                 }
 702         }
 703
 704 refresh_timer:
 705         _prb_refresh_rx_retire_blk_timer(pkc);
 706
 707 out:
 708         spin_unlock(&po->sk.sk_receive_queue.lock);
 709 }
 710
 711 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 712                 struct tpacket_block_desc *pbd1, __u32 status)
 713 {
 714         /* Flush everything minus the block header */
 715
 716 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 717         u8 *start, *end;
 718
 719         start = (u8 *)pbd1;
 720
 721         /* Skip the block header(we know header WILL fit in 4K) */
 722         start += PAGE_SIZE;
 723
 724         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 725         for (; start < end; start += PAGE_SIZE)
 726                 flush_dcache_page(pgv_to_page(start));
 727
 728         smp_wmb();
 729 #endif
 730
 731         /* Now update the block status. */
 732
 733         BLOCK_STATUS(pbd1) = status;
 734
 735         /* Flush the block header */
 736
 737 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 738         start = (u8 *)pbd1;
 739         flush_dcache_page(pgv_to_page(start));
 740
 741         smp_wmb();
 742 #endif
 743 }
 744
 745 /*
 746  * Side effect:
 747  *
 748  * 1) flush the block
 749  * 2) Increment active_blk_num
 750  *
 751  * Note:We DONT refresh the timer on purpose.
 752  *      Because almost always the next block will be opened.
 753  */
 754 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 755                 struct tpacket_block_desc *pbd1,
 756                 struct packet_sock *po, unsigned int stat)
 757 {
 758         __u32 status = TP_STATUS_USER | stat;
 759
 760         struct tpacket3_hdr *last_pkt;
 761         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 762
 763         if (po->stats.tp_drops)
 764                 status |= TP_STATUS_LOSING;
 765
 766         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 767         last_pkt->tp_next_offset = 0;
 768
 769         /* Get the ts of the last pkt */
 770         if (BLOCK_NUM_PKTS(pbd1)) {
 771                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 772                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 773         } else {
 774                 /* Ok, we tmo'd - so get the current time */
 775                 struct timespec ts;
 776                 getnstimeofday(&ts);
 777                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 778                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 779         }
 780
 781         smp_wmb();
 782
 783         /* Flush the block */
 784         prb_flush_block(pkc1, pbd1, status);
 785
 786         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 787 }
 788
 789 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 790 {
 791         pkc->reset_pending_on_curr_blk = 0;
 792 }
 793
 794 /*
 795  * Side effect of opening a block:
 796  *
 797  * 1) prb_queue is thawed.
 798  * 2) retire_blk_timer is refreshed.
 799  *
 800  */
 801 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 802         struct tpacket_block_desc *pbd1)
 803 {
 804         struct timespec ts;
 805         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 806
 807         smp_rmb();
 808
 809         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd1))) {
 810
 811                 /* We could have just memset this but we will lose the
 812                  * flexibility of making the priv area sticky
 813                  */
 814                 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 815                 BLOCK_NUM_PKTS(pbd1) = 0;
 816                 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 817                 getnstimeofday(&ts);
 818                 h1->ts_first_pkt.ts_sec = ts.tv_sec;
 819                 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 820                 pkc1->pkblk_start = (char *)pbd1;
 821                 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 822                 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 823                 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 824                 pbd1->version = pkc1->version;
 825                 pkc1->prev = pkc1->nxt_offset;
 826                 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 827                 prb_thaw_queue(pkc1);
 828                 _prb_refresh_rx_retire_blk_timer(pkc1);
 829
 830                 smp_wmb();
 831
 832                 return;
 833         }
 834
 835         WARN(1, "ERROR block:%p is NOT FREE status:%d kactive_blk_num:%d\n",
 836                 pbd1, BLOCK_STATUS(pbd1), pkc1->kactive_blk_num);
 837         dump_stack();
 838         BUG();
 839 }
 840
 841 /*
 842  * Queue freeze logic:
 843  * 1) Assume tp_block_nr = 8 blocks.
 844  * 2) At time 't0', user opens Rx ring.
 845  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 846  * 4) user-space is either sleeping or processing block '0'.
 847  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 848  *    it will close block-7,loop around and try to fill block '0'.
 849  *    call-flow:
 850  *    __packet_lookup_frame_in_block
 851  *      prb_retire_current_block()
 852  *      prb_dispatch_next_block()
 853  *        |->(BLOCK_STATUS == USER) evaluates to true
 854  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 855  * 6) Now there are two cases:
 856  *    6.1) Link goes idle right after the queue is frozen.
 857  *         But remember, the last open_block() refreshed the timer.
 858  *         When this timer expires,it will refresh itself so that we can
 859  *         re-open block-0 in near future.
 860  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 861  *         case and __packet_lookup_frame_in_block will check if block-0
 862  *         is free and can now be re-used.
 863  */
 864 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 865                                   struct packet_sock *po)
 866 {
 867         pkc->reset_pending_on_curr_blk = 1;
 868         po->stats_u.stats3.tp_freeze_q_cnt++;
 869 }
 870
 871 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 872
 873 /*
 874  * If the next block is free then we will dispatch it
 875  * and return a good offset.
 876  * Else, we will freeze the queue.
 877  * So, caller must check the return value.
 878  */
 879 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 880                 struct packet_sock *po)
 881 {
 882         struct tpacket_block_desc *pbd;
 883
 884         smp_rmb();
 885
 886         /* 1. Get current block num */
 887         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 888
 889         /* 2. If this block is currently in_use then freeze the queue */
 890         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 891                 prb_freeze_queue(pkc, po);
 892                 return NULL;
 893         }
 894
 895         /*
 896          * 3.
 897          * open this block and return the offset where the first packet
 898          * needs to get stored.
 899          */
 900         prb_open_block(pkc, pbd);
 901         return (void *)pkc->nxt_offset;
 902 }
 903
 904 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 905                 struct packet_sock *po, unsigned int status)
 906 {
 907         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 908
 909         /* retire/close the current block */
 910         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 911                 /*
 912                  * Plug the case where copy_bits() is in progress on
 913                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 914                  * have space to copy the pkt in the current block and
 915                  * called prb_retire_current_block()
 916                  *
 917                  * We don't need to worry about the TMO case because
 918                  * the timer-handler already handled this case.
 919                  */
 920                 if (!(status & TP_STATUS_BLK_TMO)) {
 921                         while (atomic_read(&pkc->blk_fill_in_prog)) {
 922                                 /* Waiting for skb_copy_bits to finish... */
 923                                 cpu_relax();
 924                         }
 925                 }
 926                 prb_close_block(pkc, pbd, po, status);
 927                 return;
 928         }
 929
 930         WARN(1, "ERROR-pbd[%d]:%p\n", pkc->kactive_blk_num, pbd);
 931         dump_stack();
 932         BUG();
 933 }
 934
 935 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
 936                                       struct tpacket_block_desc *pbd)
 937 {
 938         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 939 }
 940
 941 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 942 {
 943         return pkc->reset_pending_on_curr_blk;
 944 }
 945
 946 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 947 {
 948         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 949         atomic_dec(&pkc->blk_fill_in_prog);
 950 }
 951
 952 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 953                         struct tpacket3_hdr *ppd)
 954 {
 955         ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
 956 }
 957
 958 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 959                         struct tpacket3_hdr *ppd)
 960 {
 961         ppd->hv1.tp_rxhash = 0;
 962 }
 963
 964 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
 965                         struct tpacket3_hdr *ppd)
 966 {
 967         if (vlan_tx_tag_present(pkc->skb)) {
 968                 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
 969                 ppd->tp_status = TP_STATUS_VLAN_VALID;
 970         } else {
 971                 ppd->hv1.tp_vlan_tci = ppd->tp_status = 0;
 972         }
 973 }
 974
 975 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
 976                         struct tpacket3_hdr *ppd)
 977 {
 978         prb_fill_vlan_info(pkc, ppd);
 979
 980         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
 981                 prb_fill_rxhash(pkc, ppd);
 982         else
 983                 prb_clear_rxhash(pkc, ppd);
 984 }
 985
 986 static void prb_fill_curr_block(char *curr,
 987                                 struct tpacket_kbdq_core *pkc,
 988                                 struct tpacket_block_desc *pbd,
 989                                 unsigned int len)
 990 {
 991         struct tpacket3_hdr *ppd;
 992
 993         ppd  = (struct tpacket3_hdr *)curr;
 994         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
 995         pkc->prev = curr;
 996         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
 997         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
 998         BLOCK_NUM_PKTS(pbd) += 1;
 999         atomic_inc(&pkc->blk_fill_in_prog);
1000         prb_run_all_ft_ops(pkc, ppd);
1001 }
1002
1003 /* Assumes caller has the sk->rx_queue.lock */
1004 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1005                                             struct sk_buff *skb,
1006                                                 int status,
1007                                             unsigned int len
1008                                             )
1009 {
1010         struct tpacket_kbdq_core *pkc;
1011         struct tpacket_block_desc *pbd;
1012         char *curr, *end;
1013
1014         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1015         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1016
1017         /* Queue is frozen when user space is lagging behind */
1018         if (prb_queue_frozen(pkc)) {
1019                 /*
1020                  * Check if that last block which caused the queue to freeze,
1021                  * is still in_use by user-space.
1022                  */
1023                 if (prb_curr_blk_in_use(pkc, pbd)) {
1024                         /* Can't record this packet */
1025                         return NULL;
1026                 } else {
1027                         /*
1028                          * Ok, the block was released by user-space.
1029                          * Now let's open that block.
1030                          * opening a block also thaws the queue.
1031                          * Thawing is a side effect.
1032                          */
1033                         prb_open_block(pkc, pbd);
1034                 }
1035         }
1036
1037         smp_mb();
1038         curr = pkc->nxt_offset;
1039         pkc->skb = skb;
1040         end = (char *)pbd + pkc->kblk_size;
1041
1042         /* first try the current block */
1043         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1044                 prb_fill_curr_block(curr, pkc, pbd, len);
1045                 return (void *)curr;
1046         }
1047
1048         /* Ok, close the current block */
1049         prb_retire_current_block(pkc, po, 0);
1050
1051         /* Now, try to dispatch the next block */
1052         curr = (char *)prb_dispatch_next_block(pkc, po);
1053         if (curr) {
1054                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1055                 prb_fill_curr_block(curr, pkc, pbd, len);
1056                 return (void *)curr;
1057         }
1058
1059         /*
1060          * No free blocks are available.user_space hasn't caught up yet.
1061          * Queue was just frozen and now this packet will get dropped.
1062          */
1063         return NULL;
1064 }
1065
1066 static void *packet_current_rx_frame(struct packet_sock *po,
1067                                             struct sk_buff *skb,
1068                                             int status, unsigned int len)
1069 {
1070         char *curr = NULL;
1071         switch (po->tp_version) {
1072         case TPACKET_V1:
1073         case TPACKET_V2:
1074                 curr = packet_lookup_frame(po, &po->rx_ring,
1075                                         po->rx_ring.head, status);
1076                 return curr;
1077         case TPACKET_V3:
1078                 return __packet_lookup_frame_in_block(po, skb, status, len);
1079         default:
1080                 WARN(1, "TPACKET version not supported\n");
1081                 BUG();
1082                 return 0;
1083         }
1084 }
1085
1086 static void *prb_lookup_block(struct packet_sock *po,
1087                                      struct packet_ring_buffer *rb,
1088                                      unsigned int previous,
1089                                      int status)
1090 {
1091         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1092         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, previous);
1093
1094         if (status != BLOCK_STATUS(pbd))
1095                 return NULL;
1096         return pbd;
1097 }
1098
1099 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1100 {
1101         unsigned int prev;
1102         if (rb->prb_bdqc.kactive_blk_num)
1103                 prev = rb->prb_bdqc.kactive_blk_num-1;
1104         else
1105                 prev = rb->prb_bdqc.knum_blocks-1;
1106         return prev;
1107 }
1108
1109 /* Assumes caller has held the rx_queue.lock */
1110 static void *__prb_previous_block(struct packet_sock *po,
1111                                          struct packet_ring_buffer *rb,
1112                                          int status)
1113 {
1114         unsigned int previous = prb_previous_blk_num(rb);
1115         return prb_lookup_block(po, rb, previous, status);
1116 }
1117
1118 static void *packet_previous_rx_frame(struct packet_sock *po,
1119                                              struct packet_ring_buffer *rb,
1120                                              int status)
1121 {
1122         if (po->tp_version <= TPACKET_V2)
1123                 return packet_previous_frame(po, rb, status);
1124
1125         return __prb_previous_block(po, rb, status);
1126 }
1127
1128 static void packet_increment_rx_head(struct packet_sock *po,
1129                                             struct packet_ring_buffer *rb)
1130 {
1131         switch (po->tp_version) {
1132         case TPACKET_V1:
1133         case TPACKET_V2:
1134                 return packet_increment_head(rb);
1135         case TPACKET_V3:
1136         default:
1137                 WARN(1, "TPACKET version not supported.\n");
1138                 BUG();
1139                 return;
1140         }
1141 }
1142
1143 static void *packet_previous_frame(struct packet_sock *po,
1144                 struct packet_ring_buffer *rb,
1145                 int status)
1146 {
1147         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1148         return packet_lookup_frame(po, rb, previous, status);
1149 }
1150
1151 static void packet_increment_head(struct packet_ring_buffer *buff)
1152 {
1153         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1154 }
1155
1156 static void packet_sock_destruct(struct sock *sk)
1157 {
1158         skb_queue_purge(&sk->sk_error_queue);
1159
1160         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1161         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1162
1163         if (!sock_flag(sk, SOCK_DEAD)) {
1164                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1165                 return;
1166         }
1167
1168         sk_refcnt_debug_dec(sk);
1169 }
1170
1171 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1172 {
1173         int x = atomic_read(&f->rr_cur) + 1;
1174
1175         if (x >= num)
1176                 x = 0;
1177
1178         return x;
1179 }
1180
1181 static struct sock *fanout_demux_hash(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1182 {
1183         u32 idx, hash = skb->rxhash;
1184
1185         idx = ((u64)hash * num) >> 32;
1186
1187         return f->arr[idx];
1188 }
1189
1190 static struct sock *fanout_demux_lb(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1191 {
1192         int cur, old;
1193
1194         cur = atomic_read(&f->rr_cur);
1195         while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1196                                      fanout_rr_next(f, num))) != cur)
1197                 cur = old;
1198         return f->arr[cur];
1199 }
1200
1201 static struct sock *fanout_demux_cpu(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1202 {
1203         unsigned int cpu = smp_processor_id();
1204
1205         return f->arr[cpu % num];
1206 }
1207
1208 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1209                              struct packet_type *pt, struct net_device *orig_dev)
1210 {
1211         struct packet_fanout *f = pt->af_packet_priv;
1212         unsigned int num = f->num_members;
1213         struct packet_sock *po;
1214         struct sock *sk;
1215
1216         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1217             !num) {
1218                 kfree_skb(skb);
1219                 return 0;
1220         }
1221
1222         switch (f->type) {
1223         case PACKET_FANOUT_HASH:
1224         default:
1225                 if (f->defrag) {
1226                         skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1227                         if (!skb)
1228                                 return 0;
1229                 }
1230                 skb_get_rxhash(skb);
1231                 sk = fanout_demux_hash(f, skb, num);
1232                 break;
1233         case PACKET_FANOUT_LB:
1234                 sk = fanout_demux_lb(f, skb, num);
1235                 break;
1236         case PACKET_FANOUT_CPU:
1237                 sk = fanout_demux_cpu(f, skb, num);
1238                 break;
1239         }
1240
1241         po = pkt_sk(sk);
1242
1243         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1244 }
1245
1246 static DEFINE_MUTEX(fanout_mutex);
1247 static LIST_HEAD(fanout_list);
1248
1249 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1250 {
1251         struct packet_fanout *f = po->fanout;
1252
1253         spin_lock(&f->lock);
1254         f->arr[f->num_members] = sk;
1255         smp_wmb();
1256         f->num_members++;
1257         spin_unlock(&f->lock);
1258 }
1259
1260 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1261 {
1262         struct packet_fanout *f = po->fanout;
1263         int i;
1264
1265         spin_lock(&f->lock);
1266         for (i = 0; i < f->num_members; i++) {
1267                 if (f->arr[i] == sk)
1268                         break;
1269         }
1270         BUG_ON(i >= f->num_members);
1271         f->arr[i] = f->arr[f->num_members - 1];
1272         f->num_members--;
1273         spin_unlock(&f->lock);
1274 }
1275
1276 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1277 {
1278         struct packet_sock *po = pkt_sk(sk);
1279         struct packet_fanout *f, *match;
1280         u8 type = type_flags & 0xff;
1281         u8 defrag = (type_flags & PACKET_FANOUT_FLAG_DEFRAG) ? 1 : 0;
1282         int err;
1283
1284         switch (type) {
1285         case PACKET_FANOUT_HASH:
1286         case PACKET_FANOUT_LB:
1287         case PACKET_FANOUT_CPU:
1288                 break;
1289         default:
1290                 return -EINVAL;
1291         }
1292
1293         if (!po->running)
1294                 return -EINVAL;
1295
1296         if (po->fanout)
1297                 return -EALREADY;
1298
1299         mutex_lock(&fanout_mutex);
1300         match = NULL;
1301         list_for_each_entry(f, &fanout_list, list) {
1302                 if (f->id == id &&
1303                     read_pnet(&f->net) == sock_net(sk)) {
1304                         match = f;
1305                         break;
1306                 }
1307         }
1308         err = -EINVAL;
1309         if (match && match->defrag != defrag)
1310                 goto out;
1311         if (!match) {
1312                 err = -ENOMEM;
1313                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1314                 if (!match)
1315                         goto out;
1316                 write_pnet(&match->net, sock_net(sk));
1317                 match->id = id;
1318                 match->type = type;
1319                 match->defrag = defrag;
1320                 atomic_set(&match->rr_cur, 0);
1321                 INIT_LIST_HEAD(&match->list);
1322                 spin_lock_init(&match->lock);
1323                 atomic_set(&match->sk_ref, 0);
1324                 match->prot_hook.type = po->prot_hook.type;
1325                 match->prot_hook.dev = po->prot_hook.dev;
1326                 match->prot_hook.func = packet_rcv_fanout;
1327                 match->prot_hook.af_packet_priv = match;
1328                 dev_add_pack(&match->prot_hook);
1329                 list_add(&match->list, &fanout_list);
1330         }
1331         err = -EINVAL;
1332         if (match->type == type &&
1333             match->prot_hook.type == po->prot_hook.type &&
1334             match->prot_hook.dev == po->prot_hook.dev) {
1335                 err = -ENOSPC;
1336                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1337                         __dev_remove_pack(&po->prot_hook);
1338                         po->fanout = match;
1339                         atomic_inc(&match->sk_ref);
1340                         __fanout_link(sk, po);
1341                         err = 0;
1342                 }
1343         }
1344 out:
1345         mutex_unlock(&fanout_mutex);
1346         return err;
1347 }
1348
1349 static void fanout_release(struct sock *sk)
1350 {
1351         struct packet_sock *po = pkt_sk(sk);
1352         struct packet_fanout *f;
1353
1354         f = po->fanout;
1355         if (!f)
1356                 return;
1357
1358         po->fanout = NULL;
1359
1360         mutex_lock(&fanout_mutex);
1361         if (atomic_dec_and_test(&f->sk_ref)) {
1362                 list_del(&f->list);
1363                 dev_remove_pack(&f->prot_hook);
1364                 kfree(f);
1365         }
1366         mutex_unlock(&fanout_mutex);
1367 }
1368
1369 static const struct proto_ops packet_ops;
1370
1371 static const struct proto_ops packet_ops_spkt;
1372
1373 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1374                            struct packet_type *pt, struct net_device *orig_dev)
1375 {
1376         struct sock *sk;
1377         struct sockaddr_pkt *spkt;
1378
1379         /*
1380          *      When we registered the protocol we saved the socket in the data
1381          *      field for just this event.
1382          */
1383
1384         sk = pt->af_packet_priv;
1385
1386         /*
1387          *      Yank back the headers [hope the device set this
1388          *      right or kerboom...]
1389          *
1390          *      Incoming packets have ll header pulled,
1391          *      push it back.
1392          *
1393          *      For outgoing ones skb->data == skb_mac_header(skb)
1394          *      so that this procedure is noop.
1395          */
1396
1397         if (skb->pkt_type == PACKET_LOOPBACK)
1398                 goto out;
1399
1400         if (!net_eq(dev_net(dev), sock_net(sk)))
1401                 goto out;
1402
1403         skb = skb_share_check(skb, GFP_ATOMIC);
1404         if (skb == NULL)
1405                 goto oom;
1406
1407         /* drop any routing info */
1408         skb_dst_drop(skb);
1409
1410         /* drop conntrack reference */
1411         nf_reset(skb);
1412
1413         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1414
1415         skb_push(skb, skb->data - skb_mac_header(skb));
1416
1417         /*
1418          *      The SOCK_PACKET socket receives _all_ frames.
1419          */
1420
1421         spkt->spkt_family = dev->type;
1422         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1423         spkt->spkt_protocol = skb->protocol;
1424
1425         /*
1426          *      Charge the memory to the socket. This is done specifically
1427          *      to prevent sockets using all the memory up.
1428          */
1429
1430         if (sock_queue_rcv_skb(sk, skb) == 0)
1431                 return 0;
1432
1433 out:
1434         kfree_skb(skb);
1435 oom:
1436         return 0;
1437 }
1438
1439
1440 /*
1441  *      Output a raw packet to a device layer. This bypasses all the other
1442  *      protocol layers and you must therefore supply it with a complete frame
1443  */
1444
1445 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1446                                struct msghdr *msg, size_t len)
1447 {
1448         struct sock *sk = sock->sk;
1449         struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1450         struct sk_buff *skb = NULL;
1451         struct net_device *dev;
1452         __be16 proto = 0;
1453         int err;
1454         int extra_len = 0;
1455
1456         /*
1457          *      Get and verify the address.
1458          */
1459
1460         if (saddr) {
1461                 if (msg->msg_namelen < sizeof(struct sockaddr))
1462                         return -EINVAL;
1463                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1464                         proto = saddr->spkt_protocol;
1465         } else
1466                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1467
1468         /*
1469          *      Find the device first to size check it
1470          */
1471
1472         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1473 retry:
1474         rcu_read_lock();
1475         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1476         err = -ENODEV;
1477         if (dev == NULL)
1478                 goto out_unlock;
1479
1480         err = -ENETDOWN;
1481         if (!(dev->flags & IFF_UP))
1482                 goto out_unlock;
1483
1484         /*
1485          * You may not queue a frame bigger than the mtu. This is the lowest level
1486          * raw protocol and you must do your own fragmentation at this level.
1487          */
1488
1489         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1490                 if (!netif_supports_nofcs(dev)) {
1491                         err = -EPROTONOSUPPORT;
1492                         goto out_unlock;
1493                 }
1494                 extra_len = 4; /* We're doing our own CRC */
1495         }
1496
1497         err = -EMSGSIZE;
1498         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1499                 goto out_unlock;
1500
1501         if (!skb) {
1502                 size_t reserved = LL_RESERVED_SPACE(dev);
1503                 int tlen = dev->needed_tailroom;
1504                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1505
1506                 rcu_read_unlock();
1507                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1508                 if (skb == NULL)
1509                         return -ENOBUFS;
1510                 /* FIXME: Save some space for broken drivers that write a hard
1511                  * header at transmission time by themselves. PPP is the notable
1512                  * one here. This should really be fixed at the driver level.
1513                  */
1514                 skb_reserve(skb, reserved);
1515                 skb_reset_network_header(skb);
1516
1517                 /* Try to align data part correctly */
1518                 if (hhlen) {
1519                         skb->data -= hhlen;
1520                         skb->tail -= hhlen;
1521                         if (len < hhlen)
1522                                 skb_reset_network_header(skb);
1523                 }
1524                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1525                 if (err)
1526                         goto out_free;
1527                 goto retry;
1528         }
1529
1530         if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1531                 /* Earlier code assumed this would be a VLAN pkt,
1532                  * double-check this now that we have the actual
1533                  * packet in hand.
1534                  */
1535                 struct ethhdr *ehdr;
1536                 skb_reset_mac_header(skb);
1537                 ehdr = eth_hdr(skb);
1538                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1539                         err = -EMSGSIZE;
1540                         goto out_unlock;
1541                 }
1542         }
1543
1544         skb->protocol = proto;
1545         skb->dev = dev;
1546         skb->priority = sk->sk_priority;
1547         skb->mark = sk->sk_mark;
1548         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1549         if (err < 0)
1550                 goto out_unlock;
1551
1552         if (unlikely(extra_len == 4))
1553                 skb->no_fcs = 1;
1554
1555         dev_queue_xmit(skb);
1556         rcu_read_unlock();
1557         return len;
1558
1559 out_unlock:
1560         rcu_read_unlock();
1561 out_free:
1562         kfree_skb(skb);
1563         return err;
1564 }
1565
1566 static unsigned int run_filter(const struct sk_buff *skb,
1567                                       const struct sock *sk,
1568                                       unsigned int res)
1569 {
1570         struct sk_filter *filter;
1571
1572         rcu_read_lock();
1573         filter = rcu_dereference(sk->sk_filter);
1574         if (filter != NULL)
1575                 res = SK_RUN_FILTER(filter, skb);
1576         rcu_read_unlock();
1577
1578         return res;
1579 }
1580
1581 /*
1582  * This function makes lazy skb cloning in hope that most of packets
1583  * are discarded by BPF.
1584  *
1585  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1586  * and skb->cb are mangled. It works because (and until) packets
1587  * falling here are owned by current CPU. Output packets are cloned
1588  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1589  * sequencially, so that if we return skb to original state on exit,
1590  * we will not harm anyone.
1591  */
1592
1593 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1594                       struct packet_type *pt, struct net_device *orig_dev)
1595 {
1596         struct sock *sk;
1597         struct sockaddr_ll *sll;
1598         struct packet_sock *po;
1599         u8 *skb_head = skb->data;
1600         int skb_len = skb->len;
1601         unsigned int snaplen, res;
1602
1603         if (skb->pkt_type == PACKET_LOOPBACK)
1604                 goto drop;
1605
1606         sk = pt->af_packet_priv;
1607         po = pkt_sk(sk);
1608
1609         if (!net_eq(dev_net(dev), sock_net(sk)))
1610                 goto drop;
1611
1612         skb->dev = dev;
1613
1614         if (dev->header_ops) {
1615                 /* The device has an explicit notion of ll header,
1616                  * exported to higher levels.
1617                  *
1618                  * Otherwise, the device hides details of its frame
1619                  * structure, so that corresponding packet head is
1620                  * never delivered to user.
1621                  */
1622                 if (sk->sk_type != SOCK_DGRAM)
1623                         skb_push(skb, skb->data - skb_mac_header(skb));
1624                 else if (skb->pkt_type == PACKET_OUTGOING) {
1625                         /* Special case: outgoing packets have ll header at head */
1626                         skb_pull(skb, skb_network_offset(skb));
1627                 }
1628         }
1629
1630         snaplen = skb->len;
1631
1632         res = run_filter(skb, sk, snaplen);
1633         if (!res)
1634                 goto drop_n_restore;
1635         if (snaplen > res)
1636                 snaplen = res;
1637
1638         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1639                 goto drop_n_acct;
1640
1641         if (skb_shared(skb)) {
1642                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1643                 if (nskb == NULL)
1644                         goto drop_n_acct;
1645
1646                 if (skb_head != skb->data) {
1647                         skb->data = skb_head;
1648                         skb->len = skb_len;
1649                 }
1650                 consume_skb(skb);
1651                 skb = nskb;
1652         }
1653
1654         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1655                      sizeof(skb->cb));
1656
1657         sll = &PACKET_SKB_CB(skb)->sa.ll;
1658         sll->sll_family = AF_PACKET;
1659         sll->sll_hatype = dev->type;
1660         sll->sll_protocol = skb->protocol;
1661         sll->sll_pkttype = skb->pkt_type;
1662         if (unlikely(po->origdev))
1663                 sll->sll_ifindex = orig_dev->ifindex;
1664         else
1665                 sll->sll_ifindex = dev->ifindex;
1666
1667         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1668
1669         PACKET_SKB_CB(skb)->origlen = skb->len;
1670
1671         if (pskb_trim(skb, snaplen))
1672                 goto drop_n_acct;
1673
1674         skb_set_owner_r(skb, sk);
1675         skb->dev = NULL;
1676         skb_dst_drop(skb);
1677
1678         /* drop conntrack reference */
1679         nf_reset(skb);
1680
1681         spin_lock(&sk->sk_receive_queue.lock);
1682         po->stats.tp_packets++;
1683         skb->dropcount = atomic_read(&sk->sk_drops);
1684         __skb_queue_tail(&sk->sk_receive_queue, skb);
1685         spin_unlock(&sk->sk_receive_queue.lock);
1686         sk->sk_data_ready(sk, skb->len);
1687         return 0;
1688
1689 drop_n_acct:
1690         spin_lock(&sk->sk_receive_queue.lock);
1691         po->stats.tp_drops++;
1692         atomic_inc(&sk->sk_drops);
1693         spin_unlock(&sk->sk_receive_queue.lock);
1694
1695 drop_n_restore:
1696         if (skb_head != skb->data && skb_shared(skb)) {
1697                 skb->data = skb_head;
1698                 skb->len = skb_len;
1699         }
1700 drop:
1701         consume_skb(skb);
1702         return 0;
1703 }
1704
1705 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1706                        struct packet_type *pt, struct net_device *orig_dev)
1707 {
1708         struct sock *sk;
1709         struct packet_sock *po;
1710         struct sockaddr_ll *sll;
1711         union {
1712                 struct tpacket_hdr *h1;
1713                 struct tpacket2_hdr *h2;
1714                 struct tpacket3_hdr *h3;
1715                 void *raw;
1716         } h;
1717         u8 *skb_head = skb->data;
1718         int skb_len = skb->len;
1719         unsigned int snaplen, res;
1720         unsigned long status = TP_STATUS_USER;
1721         unsigned short macoff, netoff, hdrlen;
1722         struct sk_buff *copy_skb = NULL;
1723         struct timeval tv;
1724         struct timespec ts;
1725         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1726
1727         if (skb->pkt_type == PACKET_LOOPBACK)
1728                 goto drop;
1729
1730         sk = pt->af_packet_priv;
1731         po = pkt_sk(sk);
1732
1733         if (!net_eq(dev_net(dev), sock_net(sk)))
1734                 goto drop;
1735
1736         if (dev->header_ops) {
1737                 if (sk->sk_type != SOCK_DGRAM)
1738                         skb_push(skb, skb->data - skb_mac_header(skb));
1739                 else if (skb->pkt_type == PACKET_OUTGOING) {
1740                         /* Special case: outgoing packets have ll header at head */
1741                         skb_pull(skb, skb_network_offset(skb));
1742                 }
1743         }
1744
1745         if (skb->ip_summed == CHECKSUM_PARTIAL)
1746                 status |= TP_STATUS_CSUMNOTREADY;
1747
1748         snaplen = skb->len;
1749
1750         res = run_filter(skb, sk, snaplen);
1751         if (!res)
1752                 goto drop_n_restore;
1753         if (snaplen > res)
1754                 snaplen = res;
1755
1756         if (sk->sk_type == SOCK_DGRAM) {
1757                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1758                                   po->tp_reserve;
1759         } else {
1760                 unsigned int maclen = skb_network_offset(skb);
1761                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1762                                        (maclen < 16 ? 16 : maclen)) +
1763                         po->tp_reserve;
1764                 macoff = netoff - maclen;
1765         }
1766         if (po->tp_version <= TPACKET_V2) {
1767                 if (macoff + snaplen > po->rx_ring.frame_size) {
1768                         if (po->copy_thresh &&
1769                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1770                                 if (skb_shared(skb)) {
1771                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1772                                 } else {
1773                                         copy_skb = skb_get(skb);
1774                                         skb_head = skb->data;
1775                                 }
1776                                 if (copy_skb)
1777                                         skb_set_owner_r(copy_skb, sk);
1778                         }
1779                         snaplen = po->rx_ring.frame_size - macoff;
1780                         if ((int)snaplen < 0)
1781                                 snaplen = 0;
1782                 }
1783         }
1784         spin_lock(&sk->sk_receive_queue.lock);
1785         h.raw = packet_current_rx_frame(po, skb,
1786                                         TP_STATUS_KERNEL, (macoff+snaplen));
1787         if (!h.raw)
1788                 goto ring_is_full;
1789         if (po->tp_version <= TPACKET_V2) {
1790                 packet_increment_rx_head(po, &po->rx_ring);
1791         /*
1792          * LOSING will be reported till you read the stats,
1793          * because it's COR - Clear On Read.
1794          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1795          * at packet level.
1796          */
1797                 if (po->stats.tp_drops)
1798                         status |= TP_STATUS_LOSING;
1799         }
1800         po->stats.tp_packets++;
1801         if (copy_skb) {
1802                 status |= TP_STATUS_COPY;
1803                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1804         }
1805         spin_unlock(&sk->sk_receive_queue.lock);
1806
1807         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1808
1809         switch (po->tp_version) {
1810         case TPACKET_V1:
1811                 h.h1->tp_len = skb->len;
1812                 h.h1->tp_snaplen = snaplen;
1813                 h.h1->tp_mac = macoff;
1814                 h.h1->tp_net = netoff;
1815                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1816                                 && shhwtstamps->syststamp.tv64)
1817                         tv = ktime_to_timeval(shhwtstamps->syststamp);
1818                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1819                                 && shhwtstamps->hwtstamp.tv64)
1820                         tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1821                 else if (skb->tstamp.tv64)
1822                         tv = ktime_to_timeval(skb->tstamp);
1823                 else
1824                         do_gettimeofday(&tv);
1825                 h.h1->tp_sec = tv.tv_sec;
1826                 h.h1->tp_usec = tv.tv_usec;
1827                 hdrlen = sizeof(*h.h1);
1828                 break;
1829         case TPACKET_V2:
1830                 h.h2->tp_len = skb->len;
1831                 h.h2->tp_snaplen = snaplen;
1832                 h.h2->tp_mac = macoff;
1833                 h.h2->tp_net = netoff;
1834                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1835                                 && shhwtstamps->syststamp.tv64)
1836                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1837                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1838                                 && shhwtstamps->hwtstamp.tv64)
1839                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1840                 else if (skb->tstamp.tv64)
1841                         ts = ktime_to_timespec(skb->tstamp);
1842                 else
1843                         getnstimeofday(&ts);
1844                 h.h2->tp_sec = ts.tv_sec;
1845                 h.h2->tp_nsec = ts.tv_nsec;
1846                 if (vlan_tx_tag_present(skb)) {
1847                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1848                         status |= TP_STATUS_VLAN_VALID;
1849                 } else {
1850                         h.h2->tp_vlan_tci = 0;
1851                 }
1852                 h.h2->tp_padding = 0;
1853                 hdrlen = sizeof(*h.h2);
1854                 break;
1855         case TPACKET_V3:
1856                 /* tp_nxt_offset,vlan are already populated above.
1857                  * So DONT clear those fields here
1858                  */
1859                 h.h3->tp_status |= status;
1860                 h.h3->tp_len = skb->len;
1861                 h.h3->tp_snaplen = snaplen;
1862                 h.h3->tp_mac = macoff;
1863                 h.h3->tp_net = netoff;
1864                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1865                                 && shhwtstamps->syststamp.tv64)
1866                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1867                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1868                                 && shhwtstamps->hwtstamp.tv64)
1869                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1870                 else if (skb->tstamp.tv64)
1871                         ts = ktime_to_timespec(skb->tstamp);
1872                 else
1873                         getnstimeofday(&ts);
1874                 h.h3->tp_sec  = ts.tv_sec;
1875                 h.h3->tp_nsec = ts.tv_nsec;
1876                 hdrlen = sizeof(*h.h3);
1877                 break;
1878         default:
1879                 BUG();
1880         }
1881
1882         sll = h.raw + TPACKET_ALIGN(hdrlen);
1883         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1884         sll->sll_family = AF_PACKET;
1885         sll->sll_hatype = dev->type;
1886         sll->sll_protocol = skb->protocol;
1887         sll->sll_pkttype = skb->pkt_type;
1888         if (unlikely(po->origdev))
1889                 sll->sll_ifindex = orig_dev->ifindex;
1890         else
1891                 sll->sll_ifindex = dev->ifindex;
1892
1893         smp_mb();
1894 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1895         {
1896                 u8 *start, *end;
1897
1898                 if (po->tp_version <= TPACKET_V2) {
1899                         end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1900                                 + macoff + snaplen);
1901                         for (start = h.raw; start < end; start += PAGE_SIZE)
1902                                 flush_dcache_page(pgv_to_page(start));
1903                 }
1904                 smp_wmb();
1905         }
1906 #endif
1907         if (po->tp_version <= TPACKET_V2)
1908                 __packet_set_status(po, h.raw, status);
1909         else
1910                 prb_clear_blk_fill_status(&po->rx_ring);
1911
1912         sk->sk_data_ready(sk, 0);
1913
1914 drop_n_restore:
1915         if (skb_head != skb->data && skb_shared(skb)) {
1916                 skb->data = skb_head;
1917                 skb->len = skb_len;
1918         }
1919 drop:
1920         kfree_skb(skb);
1921         return 0;
1922
1923 ring_is_full:
1924         po->stats.tp_drops++;
1925         spin_unlock(&sk->sk_receive_queue.lock);
1926
1927         sk->sk_data_ready(sk, 0);
1928         kfree_skb(copy_skb);
1929         goto drop_n_restore;
1930 }
1931
1932 static void tpacket_destruct_skb(struct sk_buff *skb)
1933 {
1934         struct packet_sock *po = pkt_sk(skb->sk);
1935         void *ph;
1936
1937         if (likely(po->tx_ring.pg_vec)) {
1938                 ph = skb_shinfo(skb)->destructor_arg;
1939                 BUG_ON(__packet_get_status(po, ph) != TP_STATUS_SENDING);
1940                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1941                 atomic_dec(&po->tx_ring.pending);
1942                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1943         }
1944
1945         sock_wfree(skb);
1946 }
1947
1948 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1949                 void *frame, struct net_device *dev, int size_max,
1950                 __be16 proto, unsigned char *addr, int hlen)
1951 {
1952         union {
1953                 struct tpacket_hdr *h1;
1954                 struct tpacket2_hdr *h2;
1955                 void *raw;
1956         } ph;
1957         int to_write, offset, len, tp_len, nr_frags, len_max;
1958         struct socket *sock = po->sk.sk_socket;
1959         struct page *page;
1960         void *data;
1961         int err;
1962
1963         ph.raw = frame;
1964
1965         skb->protocol = proto;
1966         skb->dev = dev;
1967         skb->priority = po->sk.sk_priority;
1968         skb->mark = po->sk.sk_mark;
1969         skb_shinfo(skb)->destructor_arg = ph.raw;
1970
1971         switch (po->tp_version) {
1972         case TPACKET_V2:
1973                 tp_len = ph.h2->tp_len;
1974                 break;
1975         default:
1976                 tp_len = ph.h1->tp_len;
1977                 break;
1978         }
1979         if (unlikely(tp_len > size_max)) {
1980                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1981                 return -EMSGSIZE;
1982         }
1983
1984         skb_reserve(skb, hlen);
1985         skb_reset_network_header(skb);
1986
1987         data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1988         to_write = tp_len;
1989
1990         if (sock->type == SOCK_DGRAM) {
1991                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
1992                                 NULL, tp_len);
1993                 if (unlikely(err < 0))
1994                         return -EINVAL;
1995         } else if (dev->hard_header_len) {
1996                 /* net device doesn't like empty head */
1997                 if (unlikely(tp_len <= dev->hard_header_len)) {
1998                         pr_err("packet size is too short (%d < %d)\n",
1999                                tp_len, dev->hard_header_len);
2000                         return -EINVAL;
2001                 }
2002
2003                 skb_push(skb, dev->hard_header_len);
2004                 err = skb_store_bits(skb, 0, data,
2005                                 dev->hard_header_len);
2006                 if (unlikely(err))
2007                         return err;
2008
2009                 data += dev->hard_header_len;
2010                 to_write -= dev->hard_header_len;
2011         }
2012
2013         err = -EFAULT;
2014         offset = offset_in_page(data);
2015         len_max = PAGE_SIZE - offset;
2016         len = ((to_write > len_max) ? len_max : to_write);
2017
2018         skb->data_len = to_write;
2019         skb->len += to_write;
2020         skb->truesize += to_write;
2021         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2022
2023         while (likely(to_write)) {
2024                 nr_frags = skb_shinfo(skb)->nr_frags;
2025
2026                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2027                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2028                                MAX_SKB_FRAGS);
2029                         return -EFAULT;
2030                 }
2031
2032                 page = pgv_to_page(data);
2033                 data += len;
2034                 flush_dcache_page(page);
2035                 get_page(page);
2036                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2037                 to_write -= len;
2038                 offset = 0;
2039                 len_max = PAGE_SIZE;
2040                 len = ((to_write > len_max) ? len_max : to_write);
2041         }
2042
2043         return tp_len;
2044 }
2045
2046 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2047 {
2048         struct sk_buff *skb;
2049         struct net_device *dev;
2050         __be16 proto;
2051         bool need_rls_dev = false;
2052         int err, reserve = 0;
2053         void *ph;
2054         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2055         int tp_len, size_max;
2056         unsigned char *addr;
2057         int len_sum = 0;
2058         int status = 0;
2059         int hlen, tlen;
2060
2061         mutex_lock(&po->pg_vec_lock);
2062
2063         err = -EBUSY;
2064         if (saddr == NULL) {
2065                 dev = po->prot_hook.dev;
2066                 proto   = po->num;
2067                 addr    = NULL;
2068         } else {
2069                 err = -EINVAL;
2070                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2071                         goto out;
2072                 if (msg->msg_namelen < (saddr->sll_halen
2073                                         + offsetof(struct sockaddr_ll,
2074                                                 sll_addr)))
2075                         goto out;
2076                 proto   = saddr->sll_protocol;
2077                 addr    = saddr->sll_addr;
2078                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2079                 need_rls_dev = true;
2080         }
2081
2082         err = -ENXIO;
2083         if (unlikely(dev == NULL))
2084                 goto out;
2085
2086         reserve = dev->hard_header_len;
2087
2088         err = -ENETDOWN;
2089         if (unlikely(!(dev->flags & IFF_UP)))
2090                 goto out_put;
2091
2092         size_max = po->tx_ring.frame_size
2093                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2094
2095         if (size_max > dev->mtu + reserve)
2096                 size_max = dev->mtu + reserve;
2097
2098         do {
2099                 ph = packet_current_frame(po, &po->tx_ring,
2100                                 TP_STATUS_SEND_REQUEST);
2101
2102                 if (unlikely(ph == NULL)) {
2103                         schedule();
2104                         continue;
2105                 }
2106
2107                 status = TP_STATUS_SEND_REQUEST;
2108                 hlen = LL_RESERVED_SPACE(dev);
2109                 tlen = dev->needed_tailroom;
2110                 skb = sock_alloc_send_skb(&po->sk,
2111                                 hlen + tlen + sizeof(struct sockaddr_ll),
2112                                 0, &err);
2113
2114                 if (unlikely(skb == NULL))
2115                         goto out_status;
2116
2117                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2118                                 addr, hlen);
2119
2120                 if (unlikely(tp_len < 0)) {
2121                         if (po->tp_loss) {
2122                                 __packet_set_status(po, ph,
2123                                                 TP_STATUS_AVAILABLE);
2124                                 packet_increment_head(&po->tx_ring);
2125                                 kfree_skb(skb);
2126                                 continue;
2127                         } else {
2128                                 status = TP_STATUS_WRONG_FORMAT;
2129                                 err = tp_len;
2130                                 goto out_status;
2131                         }
2132                 }
2133
2134                 skb->destructor = tpacket_destruct_skb;
2135                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2136                 atomic_inc(&po->tx_ring.pending);
2137
2138                 status = TP_STATUS_SEND_REQUEST;
2139                 err = dev_queue_xmit(skb);
2140                 if (unlikely(err > 0)) {
2141                         err = net_xmit_errno(err);
2142                         if (err && __packet_get_status(po, ph) ==
2143                                    TP_STATUS_AVAILABLE) {
2144                                 /* skb was destructed already */
2145                                 skb = NULL;
2146                                 goto out_status;
2147                         }
2148                         /*
2149                          * skb was dropped but not destructed yet;
2150                          * let's treat it like congestion or err < 0
2151                          */
2152                         err = 0;
2153                 }
2154                 packet_increment_head(&po->tx_ring);
2155                 len_sum += tp_len;
2156         } while (likely((ph != NULL) ||
2157                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2158                          (atomic_read(&po->tx_ring.pending))))
2159                 );
2160
2161         err = len_sum;
2162         goto out_put;
2163
2164 out_status:
2165         __packet_set_status(po, ph, status);
2166         kfree_skb(skb);
2167 out_put:
2168         if (need_rls_dev)
2169                 dev_put(dev);
2170 out:
2171         mutex_unlock(&po->pg_vec_lock);
2172         return err;
2173 }
2174
2175 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2176                                         size_t reserve, size_t len,
2177                                         size_t linear, int noblock,
2178                                         int *err)
2179 {
2180         struct sk_buff *skb;
2181
2182         /* Under a page?  Don't bother with paged skb. */
2183         if (prepad + len < PAGE_SIZE || !linear)
2184                 linear = len;
2185
2186         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2187                                    err);
2188         if (!skb)
2189                 return NULL;
2190
2191         skb_reserve(skb, reserve);
2192         skb_put(skb, linear);
2193         skb->data_len = len - linear;
2194         skb->len += len - linear;
2195
2196         return skb;
2197 }
2198
2199 static int packet_snd(struct socket *sock,
2200                           struct msghdr *msg, size_t len)
2201 {
2202         struct sock *sk = sock->sk;
2203         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2204         struct sk_buff *skb;
2205         struct net_device *dev;
2206         __be16 proto;
2207         bool need_rls_dev = false;
2208         unsigned char *addr;
2209         int err, reserve = 0;
2210         struct virtio_net_hdr vnet_hdr = { 0 };
2211         int offset = 0;
2212         int vnet_hdr_len;
2213         struct packet_sock *po = pkt_sk(sk);
2214         unsigned short gso_type = 0;
2215         int hlen, tlen;
2216         int extra_len = 0;
2217
2218         /*
2219          *      Get and verify the address.
2220          */
2221
2222         if (saddr == NULL) {
2223                 dev = po->prot_hook.dev;
2224                 proto   = po->num;
2225                 addr    = NULL;
2226         } else {
2227                 err = -EINVAL;
2228                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2229                         goto out;
2230                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2231                         goto out;
2232                 proto   = saddr->sll_protocol;
2233                 addr    = saddr->sll_addr;
2234                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2235                 need_rls_dev = true;
2236         }
2237
2238         err = -ENXIO;
2239         if (dev == NULL)
2240                 goto out_unlock;
2241         if (sock->type == SOCK_RAW)
2242                 reserve = dev->hard_header_len;
2243
2244         err = -ENETDOWN;
2245         if (!(dev->flags & IFF_UP))
2246                 goto out_unlock;
2247
2248         if (po->has_vnet_hdr) {
2249                 vnet_hdr_len = sizeof(vnet_hdr);
2250
2251                 err = -EINVAL;
2252                 if (len < vnet_hdr_len)
2253                         goto out_unlock;
2254
2255                 len -= vnet_hdr_len;
2256
2257                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2258                                        vnet_hdr_len);
2259                 if (err < 0)
2260                         goto out_unlock;
2261
2262                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2263                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2264                       vnet_hdr.hdr_len))
2265                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
2266                                                  vnet_hdr.csum_offset + 2;
2267
2268                 err = -EINVAL;
2269                 if (vnet_hdr.hdr_len > len)
2270                         goto out_unlock;
2271
2272                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2273                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2274                         case VIRTIO_NET_HDR_GSO_TCPV4:
2275                                 gso_type = SKB_GSO_TCPV4;
2276                                 break;
2277                         case VIRTIO_NET_HDR_GSO_TCPV6:
2278                                 gso_type = SKB_GSO_TCPV6;
2279                                 break;
2280                         case VIRTIO_NET_HDR_GSO_UDP:
2281                                 gso_type = SKB_GSO_UDP;
2282                                 break;
2283                         default:
2284                                 goto out_unlock;
2285                         }
2286
2287                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2288                                 gso_type |= SKB_GSO_TCP_ECN;
2289
2290                         if (vnet_hdr.gso_size == 0)
2291                                 goto out_unlock;
2292
2293                 }
2294         }
2295
2296         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2297                 if (!netif_supports_nofcs(dev)) {
2298                         err = -EPROTONOSUPPORT;
2299                         goto out_unlock;
2300                 }
2301                 extra_len = 4; /* We're doing our own CRC */
2302         }
2303
2304         err = -EMSGSIZE;
2305         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2306                 goto out_unlock;
2307
2308         err = -ENOBUFS;
2309         hlen = LL_RESERVED_SPACE(dev);
2310         tlen = dev->needed_tailroom;
2311         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2312                                msg->msg_flags & MSG_DONTWAIT, &err);
2313         if (skb == NULL)
2314                 goto out_unlock;
2315
2316         skb_set_network_header(skb, reserve);
2317
2318         err = -EINVAL;
2319         if (sock->type == SOCK_DGRAM &&
2320             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2321                 goto out_free;
2322
2323         /* Returns -EFAULT on error */
2324         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2325         if (err)
2326                 goto out_free;
2327         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2328         if (err < 0)
2329                 goto out_free;
2330
2331         if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2332                 /* Earlier code assumed this would be a VLAN pkt,
2333                  * double-check this now that we have the actual
2334                  * packet in hand.
2335                  */
2336                 struct ethhdr *ehdr;
2337                 skb_reset_mac_header(skb);
2338                 ehdr = eth_hdr(skb);
2339                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2340                         err = -EMSGSIZE;
2341                         goto out_free;
2342                 }
2343         }
2344
2345         skb->protocol = proto;
2346         skb->dev = dev;
2347         skb->priority = sk->sk_priority;
2348         skb->mark = sk->sk_mark;
2349
2350         if (po->has_vnet_hdr) {
2351                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2352                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2353                                                   vnet_hdr.csum_offset)) {
2354                                 err = -EINVAL;
2355                                 goto out_free;
2356                         }
2357                 }
2358
2359                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2360                 skb_shinfo(skb)->gso_type = gso_type;
2361
2362                 /* Header must be checked, and gso_segs computed. */
2363                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2364                 skb_shinfo(skb)->gso_segs = 0;
2365
2366                 len += vnet_hdr_len;
2367         }
2368
2369         if (unlikely(extra_len == 4))
2370                 skb->no_fcs = 1;
2371
2372         /*
2373          *      Now send it
2374          */
2375
2376         err = dev_queue_xmit(skb);
2377         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2378                 goto out_unlock;
2379
2380         if (need_rls_dev)
2381                 dev_put(dev);
2382
2383         return len;
2384
2385 out_free:
2386         kfree_skb(skb);
2387 out_unlock:
2388         if (dev && need_rls_dev)
2389                 dev_put(dev);
2390 out:
2391         return err;
2392 }
2393
2394 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2395                 struct msghdr *msg, size_t len)
2396 {
2397         struct sock *sk = sock->sk;
2398         struct packet_sock *po = pkt_sk(sk);
2399         if (po->tx_ring.pg_vec)
2400                 return tpacket_snd(po, msg);
2401         else
2402                 return packet_snd(sock, msg, len);
2403 }
2404
2405 /*
2406  *      Close a PACKET socket. This is fairly simple. We immediately go
2407  *      to 'closed' state and remove our protocol entry in the device list.
2408  */
2409
2410 static int packet_release(struct socket *sock)
2411 {
2412         struct sock *sk = sock->sk;
2413         struct packet_sock *po;
2414         struct net *net;
2415         union tpacket_req_u req_u;
2416
2417         if (!sk)
2418                 return 0;
2419
2420         net = sock_net(sk);
2421         po = pkt_sk(sk);
2422
2423         spin_lock_bh(&net->packet.sklist_lock);
2424         sk_del_node_init_rcu(sk);
2425         sock_prot_inuse_add(net, sk->sk_prot, -1);
2426         spin_unlock_bh(&net->packet.sklist_lock);
2427
2428         spin_lock(&po->bind_lock);
2429         unregister_prot_hook(sk, false);
2430         if (po->prot_hook.dev) {
2431                 dev_put(po->prot_hook.dev);
2432                 po->prot_hook.dev = NULL;
2433         }
2434         spin_unlock(&po->bind_lock);
2435
2436         packet_flush_mclist(sk);
2437
2438         memset(&req_u, 0, sizeof(req_u));
2439
2440         if (po->rx_ring.pg_vec)
2441                 packet_set_ring(sk, &req_u, 1, 0);
2442
2443         if (po->tx_ring.pg_vec)
2444                 packet_set_ring(sk, &req_u, 1, 1);
2445
2446         fanout_release(sk);
2447
2448         synchronize_net();
2449         /*
2450          *      Now the socket is dead. No more input will appear.
2451          */
2452         sock_orphan(sk);
2453         sock->sk = NULL;
2454
2455         /* Purge queues */
2456
2457         skb_queue_purge(&sk->sk_receive_queue);
2458         sk_refcnt_debug_release(sk);
2459
2460         sock_put(sk);
2461         return 0;
2462 }
2463
2464 /*
2465  *      Attach a packet hook.
2466  */
2467
2468 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2469 {
2470         struct packet_sock *po = pkt_sk(sk);
2471
2472         if (po->fanout) {
2473                 if (dev)
2474                         dev_put(dev);
2475
2476                 return -EINVAL;
2477         }
2478
2479         lock_sock(sk);
2480
2481         spin_lock(&po->bind_lock);
2482         unregister_prot_hook(sk, true);
2483         po->num = protocol;
2484         po->prot_hook.type = protocol;
2485         if (po->prot_hook.dev)
2486                 dev_put(po->prot_hook.dev);
2487         po->prot_hook.dev = dev;
2488
2489         po->ifindex = dev ? dev->ifindex : 0;
2490
2491         if (protocol == 0)
2492                 goto out_unlock;
2493
2494         if (!dev || (dev->flags & IFF_UP)) {
2495                 register_prot_hook(sk);
2496         } else {
2497                 sk->sk_err = ENETDOWN;
2498                 if (!sock_flag(sk, SOCK_DEAD))
2499                         sk->sk_error_report(sk);
2500         }
2501
2502 out_unlock:
2503         spin_unlock(&po->bind_lock);
2504         release_sock(sk);
2505         return 0;
2506 }
2507
2508 /*
2509  *      Bind a packet socket to a device
2510  */
2511
2512 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2513                             int addr_len)
2514 {
2515         struct sock *sk = sock->sk;
2516         char name[15];
2517         struct net_device *dev;
2518         int err = -ENODEV;
2519
2520         /*
2521          *      Check legality
2522          */
2523
2524         if (addr_len != sizeof(struct sockaddr))
2525                 return -EINVAL;
2526         strlcpy(name, uaddr->sa_data, sizeof(name));
2527
2528         dev = dev_get_by_name(sock_net(sk), name);
2529         if (dev)
2530                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2531         return err;
2532 }
2533
2534 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2535 {
2536         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2537         struct sock *sk = sock->sk;
2538         struct net_device *dev = NULL;
2539         int err;
2540
2541
2542         /*
2543          *      Check legality
2544          */
2545
2546         if (addr_len < sizeof(struct sockaddr_ll))
2547                 return -EINVAL;
2548         if (sll->sll_family != AF_PACKET)
2549                 return -EINVAL;
2550
2551         if (sll->sll_ifindex) {
2552                 err = -ENODEV;
2553                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2554                 if (dev == NULL)
2555                         goto out;
2556         }
2557         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2558
2559 out:
2560         return err;
2561 }
2562
2563 static struct proto packet_proto = {
2564         .name     = "PACKET",
2565         .owner    = THIS_MODULE,
2566         .obj_size = sizeof(struct packet_sock),
2567 };
2568
2569 /*
2570  *      Create a packet of type SOCK_PACKET.
2571  */
2572
2573 static int packet_create(struct net *net, struct socket *sock, int protocol,
2574                          int kern)
2575 {
2576         struct sock *sk;
2577         struct packet_sock *po;
2578         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2579         int err;
2580
2581         if (!capable(CAP_NET_RAW))
2582                 return -EPERM;
2583         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2584             sock->type != SOCK_PACKET)
2585                 return -ESOCKTNOSUPPORT;
2586
2587         sock->state = SS_UNCONNECTED;
2588
2589         err = -ENOBUFS;
2590         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2591         if (sk == NULL)
2592                 goto out;
2593
2594         sock->ops = &packet_ops;
2595         if (sock->type == SOCK_PACKET)
2596                 sock->ops = &packet_ops_spkt;
2597
2598         sock_init_data(sock, sk);
2599
2600         po = pkt_sk(sk);
2601         sk->sk_family = PF_PACKET;
2602         po->num = proto;
2603
2604         sk->sk_destruct = packet_sock_destruct;
2605         sk_refcnt_debug_inc(sk);
2606
2607         /*
2608          *      Attach a protocol block
2609          */
2610
2611         spin_lock_init(&po->bind_lock);
2612         mutex_init(&po->pg_vec_lock);
2613         po->prot_hook.func = packet_rcv;
2614
2615         if (sock->type == SOCK_PACKET)
2616                 po->prot_hook.func = packet_rcv_spkt;
2617
2618         po->prot_hook.af_packet_priv = sk;
2619
2620         if (proto) {
2621                 po->prot_hook.type = proto;
2622                 register_prot_hook(sk);
2623         }
2624
2625         spin_lock_bh(&net->packet.sklist_lock);
2626         sk_add_node_rcu(sk, &net->packet.sklist);
2627         sock_prot_inuse_add(net, &packet_proto, 1);
2628         spin_unlock_bh(&net->packet.sklist_lock);
2629
2630         return 0;
2631 out:
2632         return err;
2633 }
2634
2635 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2636 {
2637         struct sock_exterr_skb *serr;
2638         struct sk_buff *skb, *skb2;
2639         int copied, err;
2640
2641         err = -EAGAIN;
2642         skb = skb_dequeue(&sk->sk_error_queue);
2643         if (skb == NULL)
2644                 goto out;
2645
2646         copied = skb->len;
2647         if (copied > len) {
2648                 msg->msg_flags |= MSG_TRUNC;
2649                 copied = len;
2650         }
2651         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2652         if (err)
2653                 goto out_free_skb;
2654
2655         sock_recv_timestamp(msg, sk, skb);
2656
2657         serr = SKB_EXT_ERR(skb);
2658         put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2659                  sizeof(serr->ee), &serr->ee);
2660
2661         msg->msg_flags |= MSG_ERRQUEUE;
2662         err = copied;
2663
2664         /* Reset and regenerate socket error */
2665         spin_lock_bh(&sk->sk_error_queue.lock);
2666         sk->sk_err = 0;
2667         if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2668                 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2669                 spin_unlock_bh(&sk->sk_error_queue.lock);
2670                 sk->sk_error_report(sk);
2671         } else
2672                 spin_unlock_bh(&sk->sk_error_queue.lock);
2673
2674 out_free_skb:
2675         kfree_skb(skb);
2676 out:
2677         return err;
2678 }
2679
2680 /*
2681  *      Pull a packet from our receive queue and hand it to the user.
2682  *      If necessary we block.
2683  */
2684
2685 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2686                           struct msghdr *msg, size_t len, int flags)
2687 {
2688         struct sock *sk = sock->sk;
2689         struct sk_buff *skb;
2690         int copied, err;
2691         struct sockaddr_ll *sll;
2692         int vnet_hdr_len = 0;
2693
2694         err = -EINVAL;
2695         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2696                 goto out;
2697
2698 #if 0
2699         /* What error should we return now? EUNATTACH? */
2700         if (pkt_sk(sk)->ifindex < 0)
2701                 return -ENODEV;
2702 #endif
2703
2704         if (flags & MSG_ERRQUEUE) {
2705                 err = packet_recv_error(sk, msg, len);
2706                 goto out;
2707         }
2708
2709         /*
2710          *      Call the generic datagram receiver. This handles all sorts
2711          *      of horrible races and re-entrancy so we can forget about it
2712          *      in the protocol layers.
2713          *
2714          *      Now it will return ENETDOWN, if device have just gone down,
2715          *      but then it will block.
2716          */
2717
2718         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2719
2720         /*
2721          *      An error occurred so return it. Because skb_recv_datagram()
2722          *      handles the blocking we don't see and worry about blocking
2723          *      retries.
2724          */
2725
2726         if (skb == NULL)
2727                 goto out;
2728
2729         if (pkt_sk(sk)->has_vnet_hdr) {
2730                 struct virtio_net_hdr vnet_hdr = { 0 };
2731
2732                 err = -EINVAL;
2733                 vnet_hdr_len = sizeof(vnet_hdr);
2734                 if (len < vnet_hdr_len)
2735                         goto out_free;
2736
2737                 len -= vnet_hdr_len;
2738
2739                 if (skb_is_gso(skb)) {
2740                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2741
2742                         /* This is a hint as to how much should be linear. */
2743                         vnet_hdr.hdr_len = skb_headlen(skb);
2744                         vnet_hdr.gso_size = sinfo->gso_size;
2745                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2746                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2747                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2748                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2749                         else if (sinfo->gso_type & SKB_GSO_UDP)
2750                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2751                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2752                                 goto out_free;
2753                         else
2754                                 BUG();
2755                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2756                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2757                 } else
2758                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2759
2760                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2761                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2762                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2763                         vnet_hdr.csum_offset = skb->csum_offset;
2764                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2765                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2766                 } /* else everything is zero */
2767
2768                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2769                                      vnet_hdr_len);
2770                 if (err < 0)
2771                         goto out_free;
2772         }
2773
2774         /*
2775          *      If the address length field is there to be filled in, we fill
2776          *      it in now.
2777          */
2778
2779         sll = &PACKET_SKB_CB(skb)->sa.ll;
2780         if (sock->type == SOCK_PACKET)
2781                 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2782         else
2783                 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2784
2785         /*
2786          *      You lose any data beyond the buffer you gave. If it worries a
2787          *      user program they can ask the device for its MTU anyway.
2788          */
2789
2790         copied = skb->len;
2791         if (copied > len) {
2792                 copied = len;
2793                 msg->msg_flags |= MSG_TRUNC;
2794         }
2795
2796         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2797         if (err)
2798                 goto out_free;
2799
2800         sock_recv_ts_and_drops(msg, sk, skb);
2801
2802         if (msg->msg_name)
2803                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2804                        msg->msg_namelen);
2805
2806         if (pkt_sk(sk)->auxdata) {
2807                 struct tpacket_auxdata aux;
2808
2809                 aux.tp_status = TP_STATUS_USER;
2810                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2811                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2812                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2813                 aux.tp_snaplen = skb->len;
2814                 aux.tp_mac = 0;
2815                 aux.tp_net = skb_network_offset(skb);
2816                 if (vlan_tx_tag_present(skb)) {
2817                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2818                         aux.tp_status |= TP_STATUS_VLAN_VALID;
2819                 } else {
2820                         aux.tp_vlan_tci = 0;
2821                 }
2822                 aux.tp_padding = 0;
2823                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2824         }
2825
2826         /*
2827          *      Free or return the buffer as appropriate. Again this
2828          *      hides all the races and re-entrancy issues from us.
2829          */
2830         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2831
2832 out_free:
2833         skb_free_datagram(sk, skb);
2834 out:
2835         return err;
2836 }
2837
2838 static int packet_getname_spkt(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
2844         if (peer)
2845                 return -EOPNOTSUPP;
2846
2847         uaddr->sa_family = AF_PACKET;
2848         rcu_read_lock();
2849         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2850         if (dev)
2851                 strncpy(uaddr->sa_data, dev->name, 14);
2852         else
2853                 memset(uaddr->sa_data, 0, 14);
2854         rcu_read_unlock();
2855         *uaddr_len = sizeof(*uaddr);
2856
2857         return 0;
2858 }
2859
2860 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2861                           int *uaddr_len, int peer)
2862 {
2863         struct net_device *dev;
2864         struct sock *sk = sock->sk;
2865         struct packet_sock *po = pkt_sk(sk);
2866         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2867
2868         if (peer)
2869                 return -EOPNOTSUPP;
2870
2871         sll->sll_family = AF_PACKET;
2872         sll->sll_ifindex = po->ifindex;
2873         sll->sll_protocol = po->num;
2874         sll->sll_pkttype = 0;
2875         rcu_read_lock();
2876         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2877         if (dev) {
2878                 sll->sll_hatype = dev->type;
2879                 sll->sll_halen = dev->addr_len;
2880                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2881         } else {
2882                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
2883                 sll->sll_halen = 0;
2884         }
2885         rcu_read_unlock();
2886         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2887
2888         return 0;
2889 }
2890
2891 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2892                          int what)
2893 {
2894         switch (i->type) {
2895         case PACKET_MR_MULTICAST:
2896                 if (i->alen != dev->addr_len)
2897                         return -EINVAL;
2898                 if (what > 0)
2899                         return dev_mc_add(dev, i->addr);
2900                 else
2901                         return dev_mc_del(dev, i->addr);
2902                 break;
2903         case PACKET_MR_PROMISC:
2904                 return dev_set_promiscuity(dev, what);
2905                 break;
2906         case PACKET_MR_ALLMULTI:
2907                 return dev_set_allmulti(dev, what);
2908                 break;
2909         case PACKET_MR_UNICAST:
2910                 if (i->alen != dev->addr_len)
2911                         return -EINVAL;
2912                 if (what > 0)
2913                         return dev_uc_add(dev, i->addr);
2914                 else
2915                         return dev_uc_del(dev, i->addr);
2916                 break;
2917         default:
2918                 break;
2919         }
2920         return 0;
2921 }
2922
2923 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2924 {
2925         for ( ; i; i = i->next) {
2926                 if (i->ifindex == dev->ifindex)
2927                         packet_dev_mc(dev, i, what);
2928         }
2929 }
2930
2931 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2932 {
2933         struct packet_sock *po = pkt_sk(sk);
2934         struct packet_mclist *ml, *i;
2935         struct net_device *dev;
2936         int err;
2937
2938         rtnl_lock();
2939
2940         err = -ENODEV;
2941         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2942         if (!dev)
2943                 goto done;
2944
2945         err = -EINVAL;
2946         if (mreq->mr_alen > dev->addr_len)
2947                 goto done;
2948
2949         err = -ENOBUFS;
2950         i = kmalloc(sizeof(*i), GFP_KERNEL);
2951         if (i == NULL)
2952                 goto done;
2953
2954         err = 0;
2955         for (ml = po->mclist; ml; ml = ml->next) {
2956                 if (ml->ifindex == mreq->mr_ifindex &&
2957                     ml->type == mreq->mr_type &&
2958                     ml->alen == mreq->mr_alen &&
2959                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2960                         ml->count++;
2961                         /* Free the new element ... */
2962                         kfree(i);
2963                         goto done;
2964                 }
2965         }
2966
2967         i->type = mreq->mr_type;
2968         i->ifindex = mreq->mr_ifindex;
2969         i->alen = mreq->mr_alen;
2970         memcpy(i->addr, mreq->mr_address, i->alen);
2971         i->count = 1;
2972         i->next = po->mclist;
2973         po->mclist = i;
2974         err = packet_dev_mc(dev, i, 1);
2975         if (err) {
2976                 po->mclist = i->next;
2977                 kfree(i);
2978         }
2979
2980 done:
2981         rtnl_unlock();
2982         return err;
2983 }
2984
2985 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2986 {
2987         struct packet_mclist *ml, **mlp;
2988
2989         rtnl_lock();
2990
2991         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
2992                 if (ml->ifindex == mreq->mr_ifindex &&
2993                     ml->type == mreq->mr_type &&
2994                     ml->alen == mreq->mr_alen &&
2995                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2996                         if (--ml->count == 0) {
2997                                 struct net_device *dev;
2998                                 *mlp = ml->next;
2999                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3000                                 if (dev)
3001                                         packet_dev_mc(dev, ml, -1);
3002                                 kfree(ml);
3003                         }
3004                         rtnl_unlock();
3005                         return 0;
3006                 }
3007         }
3008         rtnl_unlock();
3009         return -EADDRNOTAVAIL;
3010 }
3011
3012 static void packet_flush_mclist(struct sock *sk)
3013 {
3014         struct packet_sock *po = pkt_sk(sk);
3015         struct packet_mclist *ml;
3016
3017         if (!po->mclist)
3018                 return;
3019
3020         rtnl_lock();
3021         while ((ml = po->mclist) != NULL) {
3022                 struct net_device *dev;
3023
3024                 po->mclist = ml->next;
3025                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3026                 if (dev != NULL)
3027                         packet_dev_mc(dev, ml, -1);
3028                 kfree(ml);
3029         }
3030         rtnl_unlock();
3031 }
3032
3033 static int
3034 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3035 {
3036         struct sock *sk = sock->sk;
3037         struct packet_sock *po = pkt_sk(sk);
3038         int ret;
3039
3040         if (level != SOL_PACKET)
3041                 return -ENOPROTOOPT;
3042
3043         switch (optname) {
3044         case PACKET_ADD_MEMBERSHIP:
3045         case PACKET_DROP_MEMBERSHIP:
3046         {
3047                 struct packet_mreq_max mreq;
3048                 int len = optlen;
3049                 memset(&mreq, 0, sizeof(mreq));
3050                 if (len < sizeof(struct packet_mreq))
3051                         return -EINVAL;
3052                 if (len > sizeof(mreq))
3053                         len = sizeof(mreq);
3054                 if (copy_from_user(&mreq, optval, len))
3055                         return -EFAULT;
3056                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3057                         return -EINVAL;
3058                 if (optname == PACKET_ADD_MEMBERSHIP)
3059                         ret = packet_mc_add(sk, &mreq);
3060                 else
3061                         ret = packet_mc_drop(sk, &mreq);
3062                 return ret;
3063         }
3064
3065         case PACKET_RX_RING:
3066         case PACKET_TX_RING:
3067         {
3068                 union tpacket_req_u req_u;
3069                 int len;
3070
3071                 switch (po->tp_version) {
3072                 case TPACKET_V1:
3073                 case TPACKET_V2:
3074                         len = sizeof(req_u.req);
3075                         break;
3076                 case TPACKET_V3:
3077                 default:
3078                         len = sizeof(req_u.req3);
3079                         break;
3080                 }
3081                 if (optlen < len)
3082                         return -EINVAL;
3083                 if (pkt_sk(sk)->has_vnet_hdr)
3084                         return -EINVAL;
3085                 if (copy_from_user(&req_u.req, optval, len))
3086                         return -EFAULT;
3087                 return packet_set_ring(sk, &req_u, 0,
3088                         optname == PACKET_TX_RING);
3089         }
3090         case PACKET_COPY_THRESH:
3091         {
3092                 int val;
3093
3094                 if (optlen != sizeof(val))
3095                         return -EINVAL;
3096                 if (copy_from_user(&val, optval, sizeof(val)))
3097                         return -EFAULT;
3098
3099                 pkt_sk(sk)->copy_thresh = val;
3100                 return 0;
3101         }
3102         case PACKET_VERSION:
3103         {
3104                 int val;
3105
3106                 if (optlen != sizeof(val))
3107                         return -EINVAL;
3108                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3109                         return -EBUSY;
3110                 if (copy_from_user(&val, optval, sizeof(val)))
3111                         return -EFAULT;
3112                 switch (val) {
3113                 case TPACKET_V1:
3114                 case TPACKET_V2:
3115                 case TPACKET_V3:
3116                         po->tp_version = val;
3117                         return 0;
3118                 default:
3119                         return -EINVAL;
3120                 }
3121         }
3122         case PACKET_RESERVE:
3123         {
3124                 unsigned int val;
3125
3126                 if (optlen != sizeof(val))
3127                         return -EINVAL;
3128                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3129                         return -EBUSY;
3130                 if (copy_from_user(&val, optval, sizeof(val)))
3131                         return -EFAULT;
3132                 po->tp_reserve = val;
3133                 return 0;
3134         }
3135         case PACKET_LOSS:
3136         {
3137                 unsigned int val;
3138
3139                 if (optlen != sizeof(val))
3140                         return -EINVAL;
3141                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3142                         return -EBUSY;
3143                 if (copy_from_user(&val, optval, sizeof(val)))
3144                         return -EFAULT;
3145                 po->tp_loss = !!val;
3146                 return 0;
3147         }
3148         case PACKET_AUXDATA:
3149         {
3150                 int val;
3151
3152                 if (optlen < sizeof(val))
3153                         return -EINVAL;
3154                 if (copy_from_user(&val, optval, sizeof(val)))
3155                         return -EFAULT;
3156
3157                 po->auxdata = !!val;
3158                 return 0;
3159         }
3160         case PACKET_ORIGDEV:
3161         {
3162                 int val;
3163
3164                 if (optlen < sizeof(val))
3165                         return -EINVAL;
3166                 if (copy_from_user(&val, optval, sizeof(val)))
3167                         return -EFAULT;
3168
3169                 po->origdev = !!val;
3170                 return 0;
3171         }
3172         case PACKET_VNET_HDR:
3173         {
3174                 int val;
3175
3176                 if (sock->type != SOCK_RAW)
3177                         return -EINVAL;
3178                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3179                         return -EBUSY;
3180                 if (optlen < sizeof(val))
3181                         return -EINVAL;
3182                 if (copy_from_user(&val, optval, sizeof(val)))
3183                         return -EFAULT;
3184
3185                 po->has_vnet_hdr = !!val;
3186                 return 0;
3187         }
3188         case PACKET_TIMESTAMP:
3189         {
3190                 int val;
3191
3192                 if (optlen != sizeof(val))
3193                         return -EINVAL;
3194                 if (copy_from_user(&val, optval, sizeof(val)))
3195                         return -EFAULT;
3196
3197                 po->tp_tstamp = val;
3198                 return 0;
3199         }
3200         case PACKET_FANOUT:
3201         {
3202                 int val;
3203
3204                 if (optlen != sizeof(val))
3205                         return -EINVAL;
3206                 if (copy_from_user(&val, optval, sizeof(val)))
3207                         return -EFAULT;
3208
3209                 return fanout_add(sk, val & 0xffff, val >> 16);
3210         }
3211         default:
3212                 return -ENOPROTOOPT;
3213         }
3214 }
3215
3216 static int packet_getsockopt(struct socket *sock, int level, int optname,
3217                              char __user *optval, int __user *optlen)
3218 {
3219         int len;
3220         int val, lv = sizeof(val);
3221         struct sock *sk = sock->sk;
3222         struct packet_sock *po = pkt_sk(sk);
3223         void *data = &val;
3224         struct tpacket_stats st;
3225         union tpacket_stats_u st_u;
3226
3227         if (level != SOL_PACKET)
3228                 return -ENOPROTOOPT;
3229
3230         if (get_user(len, optlen))
3231                 return -EFAULT;
3232
3233         if (len < 0)
3234                 return -EINVAL;
3235
3236         switch (optname) {
3237         case PACKET_STATISTICS:
3238                 spin_lock_bh(&sk->sk_receive_queue.lock);
3239                 if (po->tp_version == TPACKET_V3) {
3240                         lv = sizeof(struct tpacket_stats_v3);
3241                         memcpy(&st_u.stats3, &po->stats,
3242                                sizeof(struct tpacket_stats));
3243                         st_u.stats3.tp_freeze_q_cnt =
3244                                         po->stats_u.stats3.tp_freeze_q_cnt;
3245                         st_u.stats3.tp_packets += po->stats.tp_drops;
3246                         data = &st_u.stats3;
3247                 } else {
3248                         lv = sizeof(struct tpacket_stats);
3249                         st = po->stats;
3250                         st.tp_packets += st.tp_drops;
3251                         data = &st;
3252                 }
3253                 memset(&po->stats, 0, sizeof(st));
3254                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3255                 break;
3256         case PACKET_AUXDATA:
3257                 val = po->auxdata;
3258                 break;
3259         case PACKET_ORIGDEV:
3260                 val = po->origdev;
3261                 break;
3262         case PACKET_VNET_HDR:
3263                 val = po->has_vnet_hdr;
3264                 break;
3265         case PACKET_VERSION:
3266                 val = po->tp_version;
3267                 break;
3268         case PACKET_HDRLEN:
3269                 if (len > sizeof(int))
3270                         len = sizeof(int);
3271                 if (copy_from_user(&val, optval, len))
3272                         return -EFAULT;
3273                 switch (val) {
3274                 case TPACKET_V1:
3275                         val = sizeof(struct tpacket_hdr);
3276                         break;
3277                 case TPACKET_V2:
3278                         val = sizeof(struct tpacket2_hdr);
3279                         break;
3280                 case TPACKET_V3:
3281                         val = sizeof(struct tpacket3_hdr);
3282                         break;
3283                 default:
3284                         return -EINVAL;
3285                 }
3286                 break;
3287         case PACKET_RESERVE:
3288                 val = po->tp_reserve;
3289                 break;
3290         case PACKET_LOSS:
3291                 val = po->tp_loss;
3292                 break;
3293         case PACKET_TIMESTAMP:
3294                 val = po->tp_tstamp;
3295                 break;
3296         case PACKET_FANOUT:
3297                 val = (po->fanout ?
3298                        ((u32)po->fanout->id |
3299                         ((u32)po->fanout->type << 16)) :
3300                        0);
3301                 break;
3302         default:
3303                 return -ENOPROTOOPT;
3304         }
3305
3306         if (len > lv)
3307                 len = lv;
3308         if (put_user(len, optlen))
3309                 return -EFAULT;
3310         if (copy_to_user(optval, data, len))
3311                 return -EFAULT;
3312         return 0;
3313 }
3314
3315
3316 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3317 {
3318         struct sock *sk;
3319         struct hlist_node *node;
3320         struct net_device *dev = data;
3321         struct net *net = dev_net(dev);
3322
3323         rcu_read_lock();
3324         sk_for_each_rcu(sk, node, &net->packet.sklist) {
3325                 struct packet_sock *po = pkt_sk(sk);
3326
3327                 switch (msg) {
3328                 case NETDEV_UNREGISTER:
3329                         if (po->mclist)
3330                                 packet_dev_mclist(dev, po->mclist, -1);
3331                         /* fallthrough */
3332
3333                 case NETDEV_DOWN:
3334                         if (dev->ifindex == po->ifindex) {
3335                                 spin_lock(&po->bind_lock);
3336                                 if (po->running) {
3337                                         __unregister_prot_hook(sk, false);
3338                                         sk->sk_err = ENETDOWN;
3339                                         if (!sock_flag(sk, SOCK_DEAD))
3340                                                 sk->sk_error_report(sk);
3341                                 }
3342                                 if (msg == NETDEV_UNREGISTER) {
3343                                         po->ifindex = -1;
3344                                         if (po->prot_hook.dev)
3345                                                 dev_put(po->prot_hook.dev);
3346                                         po->prot_hook.dev = NULL;
3347                                 }
3348                                 spin_unlock(&po->bind_lock);
3349                         }
3350                         break;
3351                 case NETDEV_UP:
3352                         if (dev->ifindex == po->ifindex) {
3353                                 spin_lock(&po->bind_lock);
3354                                 if (po->num)
3355                                         register_prot_hook(sk);
3356                                 spin_unlock(&po->bind_lock);
3357                         }
3358                         break;
3359                 }
3360         }
3361         rcu_read_unlock();
3362         return NOTIFY_DONE;
3363 }
3364
3365
3366 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3367                         unsigned long arg)
3368 {
3369         struct sock *sk = sock->sk;
3370
3371         switch (cmd) {
3372         case SIOCOUTQ:
3373         {
3374                 int amount = sk_wmem_alloc_get(sk);
3375
3376                 return put_user(amount, (int __user *)arg);
3377         }
3378         case SIOCINQ:
3379         {
3380                 struct sk_buff *skb;
3381                 int amount = 0;
3382
3383                 spin_lock_bh(&sk->sk_receive_queue.lock);
3384                 skb = skb_peek(&sk->sk_receive_queue);
3385                 if (skb)
3386                         amount = skb->len;
3387                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3388                 return put_user(amount, (int __user *)arg);
3389         }
3390         case SIOCGSTAMP:
3391                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3392         case SIOCGSTAMPNS:
3393                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3394
3395 #ifdef CONFIG_INET
3396         case SIOCADDRT:
3397         case SIOCDELRT:
3398         case SIOCDARP:
3399         case SIOCGARP:
3400         case SIOCSARP:
3401         case SIOCGIFADDR:
3402         case SIOCSIFADDR:
3403         case SIOCGIFBRDADDR:
3404         case SIOCSIFBRDADDR:
3405         case SIOCGIFNETMASK:
3406         case SIOCSIFNETMASK:
3407         case SIOCGIFDSTADDR:
3408         case SIOCSIFDSTADDR:
3409         case SIOCSIFFLAGS:
3410                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3411 #endif
3412
3413         default:
3414                 return -ENOIOCTLCMD;
3415         }
3416         return 0;
3417 }
3418
3419 static unsigned int packet_poll(struct file *file, struct socket *sock,
3420                                 poll_table *wait)
3421 {
3422         struct sock *sk = sock->sk;
3423         struct packet_sock *po = pkt_sk(sk);
3424         unsigned int mask = datagram_poll(file, sock, wait);
3425
3426         spin_lock_bh(&sk->sk_receive_queue.lock);
3427         if (po->rx_ring.pg_vec) {
3428                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3429                         TP_STATUS_KERNEL))
3430                         mask |= POLLIN | POLLRDNORM;
3431         }
3432         spin_unlock_bh(&sk->sk_receive_queue.lock);
3433         spin_lock_bh(&sk->sk_write_queue.lock);
3434         if (po->tx_ring.pg_vec) {
3435                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3436                         mask |= POLLOUT | POLLWRNORM;
3437         }
3438         spin_unlock_bh(&sk->sk_write_queue.lock);
3439         return mask;
3440 }
3441
3442
3443 /* Dirty? Well, I still did not learn better way to account
3444  * for user mmaps.
3445  */
3446
3447 static void packet_mm_open(struct vm_area_struct *vma)
3448 {
3449         struct file *file = vma->vm_file;
3450         struct socket *sock = file->private_data;
3451         struct sock *sk = sock->sk;
3452
3453         if (sk)
3454                 atomic_inc(&pkt_sk(sk)->mapped);
3455 }
3456
3457 static void packet_mm_close(struct vm_area_struct *vma)
3458 {
3459         struct file *file = vma->vm_file;
3460         struct socket *sock = file->private_data;
3461         struct sock *sk = sock->sk;
3462
3463         if (sk)
3464                 atomic_dec(&pkt_sk(sk)->mapped);
3465 }
3466
3467 static const struct vm_operations_struct packet_mmap_ops = {
3468         .open   =       packet_mm_open,
3469         .close  =       packet_mm_close,
3470 };
3471
3472 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3473                         unsigned int len)
3474 {
3475         int i;
3476
3477         for (i = 0; i < len; i++) {
3478                 if (likely(pg_vec[i].buffer)) {
3479                         if (is_vmalloc_addr(pg_vec[i].buffer))
3480                                 vfree(pg_vec[i].buffer);
3481                         else
3482                                 free_pages((unsigned long)pg_vec[i].buffer,
3483                                            order);
3484                         pg_vec[i].buffer = NULL;
3485                 }
3486         }
3487         kfree(pg_vec);
3488 }
3489
3490 static char *alloc_one_pg_vec_page(unsigned long order)
3491 {
3492         char *buffer = NULL;
3493         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3494                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3495
3496         buffer = (char *) __get_free_pages(gfp_flags, order);
3497
3498         if (buffer)
3499                 return buffer;
3500
3501         /*
3502          * __get_free_pages failed, fall back to vmalloc
3503          */
3504         buffer = vzalloc((1 << order) * PAGE_SIZE);
3505
3506         if (buffer)
3507                 return buffer;
3508
3509         /*
3510          * vmalloc failed, lets dig into swap here
3511          */
3512         gfp_flags &= ~__GFP_NORETRY;
3513         buffer = (char *)__get_free_pages(gfp_flags, order);
3514         if (buffer)
3515                 return buffer;
3516
3517         /*
3518          * complete and utter failure
3519          */
3520         return NULL;
3521 }
3522
3523 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3524 {
3525         unsigned int block_nr = req->tp_block_nr;
3526         struct pgv *pg_vec;
3527         int i;
3528
3529         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3530         if (unlikely(!pg_vec))
3531                 goto out;
3532
3533         for (i = 0; i < block_nr; i++) {
3534                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3535                 if (unlikely(!pg_vec[i].buffer))
3536                         goto out_free_pgvec;
3537         }
3538
3539 out:
3540         return pg_vec;
3541
3542 out_free_pgvec:
3543         free_pg_vec(pg_vec, order, block_nr);
3544         pg_vec = NULL;
3545         goto out;
3546 }
3547
3548 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3549                 int closing, int tx_ring)
3550 {
3551         struct pgv *pg_vec = NULL;
3552         struct packet_sock *po = pkt_sk(sk);
3553         int was_running, order = 0;
3554         struct packet_ring_buffer *rb;
3555         struct sk_buff_head *rb_queue;
3556         __be16 num;
3557         int err = -EINVAL;
3558         /* Added to avoid minimal code churn */
3559         struct tpacket_req *req = &req_u->req;
3560
3561         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3562         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3563                 WARN(1, "Tx-ring is not supported.\n");
3564                 goto out;
3565         }
3566
3567         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3568         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3569
3570         err = -EBUSY;
3571         if (!closing) {
3572                 if (atomic_read(&po->mapped))
3573                         goto out;
3574                 if (atomic_read(&rb->pending))
3575                         goto out;
3576         }
3577
3578         if (req->tp_block_nr) {
3579                 /* Sanity tests and some calculations */
3580                 err = -EBUSY;
3581                 if (unlikely(rb->pg_vec))
3582                         goto out;
3583
3584                 switch (po->tp_version) {
3585                 case TPACKET_V1:
3586                         po->tp_hdrlen = TPACKET_HDRLEN;
3587                         break;
3588                 case TPACKET_V2:
3589                         po->tp_hdrlen = TPACKET2_HDRLEN;
3590                         break;
3591                 case TPACKET_V3:
3592                         po->tp_hdrlen = TPACKET3_HDRLEN;
3593                         break;
3594                 }
3595
3596                 err = -EINVAL;
3597                 if (unlikely((int)req->tp_block_size <= 0))
3598                         goto out;
3599                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3600                         goto out;
3601                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3602                                         po->tp_reserve))
3603                         goto out;
3604                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3605                         goto out;
3606
3607                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3608                 if (unlikely(rb->frames_per_block <= 0))
3609                         goto out;
3610                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3611                                         req->tp_frame_nr))
3612                         goto out;
3613
3614                 err = -ENOMEM;
3615                 order = get_order(req->tp_block_size);
3616                 pg_vec = alloc_pg_vec(req, order);
3617                 if (unlikely(!pg_vec))
3618                         goto out;
3619                 switch (po->tp_version) {
3620                 case TPACKET_V3:
3621                 /* Transmit path is not supported. We checked
3622                  * it above but just being paranoid
3623                  */
3624                         if (!tx_ring)
3625                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3626                                 break;
3627                 default:
3628                         break;
3629                 }
3630         }
3631         /* Done */
3632         else {
3633                 err = -EINVAL;
3634                 if (unlikely(req->tp_frame_nr))
3635                         goto out;
3636         }
3637
3638         lock_sock(sk);
3639
3640         /* Detach socket from network */
3641         spin_lock(&po->bind_lock);
3642         was_running = po->running;
3643         num = po->num;
3644         if (was_running) {
3645                 po->num = 0;
3646                 __unregister_prot_hook(sk, false);
3647         }
3648         spin_unlock(&po->bind_lock);
3649
3650         synchronize_net();
3651
3652         err = -EBUSY;
3653         mutex_lock(&po->pg_vec_lock);
3654         if (closing || atomic_read(&po->mapped) == 0) {
3655                 err = 0;
3656                 spin_lock_bh(&rb_queue->lock);
3657                 swap(rb->pg_vec, pg_vec);
3658                 rb->frame_max = (req->tp_frame_nr - 1);
3659                 rb->head = 0;
3660                 rb->frame_size = req->tp_frame_size;
3661                 spin_unlock_bh(&rb_queue->lock);
3662
3663                 swap(rb->pg_vec_order, order);
3664                 swap(rb->pg_vec_len, req->tp_block_nr);
3665
3666                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3667                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3668                                                 tpacket_rcv : packet_rcv;
3669                 skb_queue_purge(rb_queue);
3670                 if (atomic_read(&po->mapped))
3671                         pr_err("packet_mmap: vma is busy: %d\n",
3672                                atomic_read(&po->mapped));
3673         }
3674         mutex_unlock(&po->pg_vec_lock);
3675
3676         spin_lock(&po->bind_lock);
3677         if (was_running) {
3678                 po->num = num;
3679                 register_prot_hook(sk);
3680         }
3681         spin_unlock(&po->bind_lock);
3682         if (closing && (po->tp_version > TPACKET_V2)) {
3683                 /* Because we don't support block-based V3 on tx-ring */
3684                 if (!tx_ring)
3685                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3686         }
3687         release_sock(sk);
3688
3689         if (pg_vec)
3690                 free_pg_vec(pg_vec, order, req->tp_block_nr);
3691 out:
3692         return err;
3693 }
3694
3695 static int packet_mmap(struct file *file, struct socket *sock,
3696                 struct vm_area_struct *vma)
3697 {
3698         struct sock *sk = sock->sk;
3699         struct packet_sock *po = pkt_sk(sk);
3700         unsigned long size, expected_size;
3701         struct packet_ring_buffer *rb;
3702         unsigned long start;
3703         int err = -EINVAL;
3704         int i;
3705
3706         if (vma->vm_pgoff)
3707                 return -EINVAL;
3708
3709         mutex_lock(&po->pg_vec_lock);
3710
3711         expected_size = 0;
3712         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3713                 if (rb->pg_vec) {
3714                         expected_size += rb->pg_vec_len
3715                                                 * rb->pg_vec_pages
3716                                                 * PAGE_SIZE;
3717                 }
3718         }
3719
3720         if (expected_size == 0)
3721                 goto out;
3722
3723         size = vma->vm_end - vma->vm_start;
3724         if (size != expected_size)
3725                 goto out;
3726
3727         start = vma->vm_start;
3728         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3729                 if (rb->pg_vec == NULL)
3730                         continue;
3731
3732                 for (i = 0; i < rb->pg_vec_len; i++) {
3733                         struct page *page;
3734                         void *kaddr = rb->pg_vec[i].buffer;
3735                         int pg_num;
3736
3737                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3738                                 page = pgv_to_page(kaddr);
3739                                 err = vm_insert_page(vma, start, page);
3740                                 if (unlikely(err))
3741                                         goto out;
3742                                 start += PAGE_SIZE;
3743                                 kaddr += PAGE_SIZE;
3744                         }
3745                 }
3746         }
3747
3748         atomic_inc(&po->mapped);
3749         vma->vm_ops = &packet_mmap_ops;
3750         err = 0;
3751
3752 out:
3753         mutex_unlock(&po->pg_vec_lock);
3754         return err;
3755 }
3756
3757 static const struct proto_ops packet_ops_spkt = {
3758         .family =       PF_PACKET,
3759         .owner =        THIS_MODULE,
3760         .release =      packet_release,
3761         .bind =         packet_bind_spkt,
3762         .connect =      sock_no_connect,
3763         .socketpair =   sock_no_socketpair,
3764         .accept =       sock_no_accept,
3765         .getname =      packet_getname_spkt,
3766         .poll =         datagram_poll,
3767         .ioctl =        packet_ioctl,
3768         .listen =       sock_no_listen,
3769         .shutdown =     sock_no_shutdown,
3770         .setsockopt =   sock_no_setsockopt,
3771         .getsockopt =   sock_no_getsockopt,
3772         .sendmsg =      packet_sendmsg_spkt,
3773         .recvmsg =      packet_recvmsg,
3774         .mmap =         sock_no_mmap,
3775         .sendpage =     sock_no_sendpage,
3776 };
3777
3778 static const struct proto_ops packet_ops = {
3779         .family =       PF_PACKET,
3780         .owner =        THIS_MODULE,
3781         .release =      packet_release,
3782         .bind =         packet_bind,
3783         .connect =      sock_no_connect,
3784         .socketpair =   sock_no_socketpair,
3785         .accept =       sock_no_accept,
3786         .getname =      packet_getname,
3787         .poll =         packet_poll,
3788         .ioctl =        packet_ioctl,
3789         .listen =       sock_no_listen,
3790         .shutdown =     sock_no_shutdown,
3791         .setsockopt =   packet_setsockopt,
3792         .getsockopt =   packet_getsockopt,
3793         .sendmsg =      packet_sendmsg,
3794         .recvmsg =      packet_recvmsg,
3795         .mmap =         packet_mmap,
3796         .sendpage =     sock_no_sendpage,
3797 };
3798
3799 static const struct net_proto_family packet_family_ops = {
3800         .family =       PF_PACKET,
3801         .create =       packet_create,
3802         .owner  =       THIS_MODULE,
3803 };
3804
3805 static struct notifier_block packet_netdev_notifier = {
3806         .notifier_call =        packet_notifier,
3807 };
3808
3809 #ifdef CONFIG_PROC_FS
3810
3811 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3812         __acquires(RCU)
3813 {
3814         struct net *net = seq_file_net(seq);
3815
3816         rcu_read_lock();
3817         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3818 }
3819
3820 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3821 {
3822         struct net *net = seq_file_net(seq);
3823         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3824 }
3825
3826 static void packet_seq_stop(struct seq_file *seq, void *v)
3827         __releases(RCU)
3828 {
3829         rcu_read_unlock();
3830 }
3831
3832 static int packet_seq_show(struct seq_file *seq, void *v)
3833 {
3834         if (v == SEQ_START_TOKEN)
3835                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
3836         else {
3837                 struct sock *s = sk_entry(v);
3838                 const struct packet_sock *po = pkt_sk(s);
3839
3840                 seq_printf(seq,
3841                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
3842                            s,
3843                            atomic_read(&s->sk_refcnt),
3844                            s->sk_type,
3845                            ntohs(po->num),
3846                            po->ifindex,
3847                            po->running,
3848                            atomic_read(&s->sk_rmem_alloc),
3849                            sock_i_uid(s),
3850                            sock_i_ino(s));
3851         }
3852
3853         return 0;
3854 }
3855
3856 static const struct seq_operations packet_seq_ops = {
3857         .start  = packet_seq_start,
3858         .next   = packet_seq_next,
3859         .stop   = packet_seq_stop,
3860         .show   = packet_seq_show,
3861 };
3862
3863 static int packet_seq_open(struct inode *inode, struct file *file)
3864 {
3865         return seq_open_net(inode, file, &packet_seq_ops,
3866                             sizeof(struct seq_net_private));
3867 }
3868
3869 static const struct file_operations packet_seq_fops = {
3870         .owner          = THIS_MODULE,
3871         .open           = packet_seq_open,
3872         .read           = seq_read,
3873         .llseek         = seq_lseek,
3874         .release        = seq_release_net,
3875 };
3876
3877 #endif
3878
3879 static int __net_init packet_net_init(struct net *net)
3880 {
3881         spin_lock_init(&net->packet.sklist_lock);
3882         INIT_HLIST_HEAD(&net->packet.sklist);
3883
3884         if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
3885                 return -ENOMEM;
3886
3887         return 0;
3888 }
3889
3890 static void __net_exit packet_net_exit(struct net *net)
3891 {
3892         proc_net_remove(net, "packet");
3893 }
3894
3895 static struct pernet_operations packet_net_ops = {
3896         .init = packet_net_init,
3897         .exit = packet_net_exit,
3898 };
3899
3900
3901 static void __exit packet_exit(void)
3902 {
3903         unregister_netdevice_notifier(&packet_netdev_notifier);
3904         unregister_pernet_subsys(&packet_net_ops);
3905         sock_unregister(PF_PACKET);
3906         proto_unregister(&packet_proto);
3907 }
3908
3909 static int __init packet_init(void)
3910 {
3911         int rc = proto_register(&packet_proto, 0);
3912
3913         if (rc != 0)
3914                 goto out;
3915
3916         sock_register(&packet_family_ops);
3917         register_pernet_subsys(&packet_net_ops);
3918         register_netdevice_notifier(&packet_netdev_notifier);
3919 out:
3920         return rc;
3921 }
3922
3923 module_init(packet_init);
3924 module_exit(packet_exit);
3925 MODULE_LICENSE("GPL");
3926 MODULE_ALIAS_NETPROTO(PF_PACKET);