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 NULL;
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 bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1277 {
1278         if (ptype->af_packet_priv == (void*)((struct packet_sock *)sk)->fanout)
1279                 return true;
1280
1281         return false;
1282 }
1283
1284 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1285 {
1286         struct packet_sock *po = pkt_sk(sk);
1287         struct packet_fanout *f, *match;
1288         u8 type = type_flags & 0xff;
1289         u8 defrag = (type_flags & PACKET_FANOUT_FLAG_DEFRAG) ? 1 : 0;
1290         int err;
1291
1292         switch (type) {
1293         case PACKET_FANOUT_HASH:
1294         case PACKET_FANOUT_LB:
1295         case PACKET_FANOUT_CPU:
1296                 break;
1297         default:
1298                 return -EINVAL;
1299         }
1300
1301         if (!po->running)
1302                 return -EINVAL;
1303
1304         if (po->fanout)
1305                 return -EALREADY;
1306
1307         mutex_lock(&fanout_mutex);
1308         match = NULL;
1309         list_for_each_entry(f, &fanout_list, list) {
1310                 if (f->id == id &&
1311                     read_pnet(&f->net) == sock_net(sk)) {
1312                         match = f;
1313                         break;
1314                 }
1315         }
1316         err = -EINVAL;
1317         if (match && match->defrag != defrag)
1318                 goto out;
1319         if (!match) {
1320                 err = -ENOMEM;
1321                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1322                 if (!match)
1323                         goto out;
1324                 write_pnet(&match->net, sock_net(sk));
1325                 match->id = id;
1326                 match->type = type;
1327                 match->defrag = defrag;
1328                 atomic_set(&match->rr_cur, 0);
1329                 INIT_LIST_HEAD(&match->list);
1330                 spin_lock_init(&match->lock);
1331                 atomic_set(&match->sk_ref, 0);
1332                 match->prot_hook.type = po->prot_hook.type;
1333                 match->prot_hook.dev = po->prot_hook.dev;
1334                 match->prot_hook.func = packet_rcv_fanout;
1335                 match->prot_hook.af_packet_priv = match;
1336                 match->prot_hook.id_match = match_fanout_group;
1337                 dev_add_pack(&match->prot_hook);
1338                 list_add(&match->list, &fanout_list);
1339         }
1340         err = -EINVAL;
1341         if (match->type == type &&
1342             match->prot_hook.type == po->prot_hook.type &&
1343             match->prot_hook.dev == po->prot_hook.dev) {
1344                 err = -ENOSPC;
1345                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1346                         __dev_remove_pack(&po->prot_hook);
1347                         po->fanout = match;
1348                         atomic_inc(&match->sk_ref);
1349                         __fanout_link(sk, po);
1350                         err = 0;
1351                 }
1352         }
1353 out:
1354         mutex_unlock(&fanout_mutex);
1355         return err;
1356 }
1357
1358 static void fanout_release(struct sock *sk)
1359 {
1360         struct packet_sock *po = pkt_sk(sk);
1361         struct packet_fanout *f;
1362
1363         f = po->fanout;
1364         if (!f)
1365                 return;
1366
1367         po->fanout = NULL;
1368
1369         mutex_lock(&fanout_mutex);
1370         if (atomic_dec_and_test(&f->sk_ref)) {
1371                 list_del(&f->list);
1372                 dev_remove_pack(&f->prot_hook);
1373                 kfree(f);
1374         }
1375         mutex_unlock(&fanout_mutex);
1376 }
1377
1378 static const struct proto_ops packet_ops;
1379
1380 static const struct proto_ops packet_ops_spkt;
1381
1382 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1383                            struct packet_type *pt, struct net_device *orig_dev)
1384 {
1385         struct sock *sk;
1386         struct sockaddr_pkt *spkt;
1387
1388         /*
1389          *      When we registered the protocol we saved the socket in the data
1390          *      field for just this event.
1391          */
1392
1393         sk = pt->af_packet_priv;
1394
1395         /*
1396          *      Yank back the headers [hope the device set this
1397          *      right or kerboom...]
1398          *
1399          *      Incoming packets have ll header pulled,
1400          *      push it back.
1401          *
1402          *      For outgoing ones skb->data == skb_mac_header(skb)
1403          *      so that this procedure is noop.
1404          */
1405
1406         if (skb->pkt_type == PACKET_LOOPBACK)
1407                 goto out;
1408
1409         if (!net_eq(dev_net(dev), sock_net(sk)))
1410                 goto out;
1411
1412         skb = skb_share_check(skb, GFP_ATOMIC);
1413         if (skb == NULL)
1414                 goto oom;
1415
1416         /* drop any routing info */
1417         skb_dst_drop(skb);
1418
1419         /* drop conntrack reference */
1420         nf_reset(skb);
1421
1422         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1423
1424         skb_push(skb, skb->data - skb_mac_header(skb));
1425
1426         /*
1427          *      The SOCK_PACKET socket receives _all_ frames.
1428          */
1429
1430         spkt->spkt_family = dev->type;
1431         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1432         spkt->spkt_protocol = skb->protocol;
1433
1434         /*
1435          *      Charge the memory to the socket. This is done specifically
1436          *      to prevent sockets using all the memory up.
1437          */
1438
1439         if (sock_queue_rcv_skb(sk, skb) == 0)
1440                 return 0;
1441
1442 out:
1443         kfree_skb(skb);
1444 oom:
1445         return 0;
1446 }
1447
1448
1449 /*
1450  *      Output a raw packet to a device layer. This bypasses all the other
1451  *      protocol layers and you must therefore supply it with a complete frame
1452  */
1453
1454 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1455                                struct msghdr *msg, size_t len)
1456 {
1457         struct sock *sk = sock->sk;
1458         struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1459         struct sk_buff *skb = NULL;
1460         struct net_device *dev;
1461         __be16 proto = 0;
1462         int err;
1463         int extra_len = 0;
1464
1465         /*
1466          *      Get and verify the address.
1467          */
1468
1469         if (saddr) {
1470                 if (msg->msg_namelen < sizeof(struct sockaddr))
1471                         return -EINVAL;
1472                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1473                         proto = saddr->spkt_protocol;
1474         } else
1475                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1476
1477         /*
1478          *      Find the device first to size check it
1479          */
1480
1481         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1482 retry:
1483         rcu_read_lock();
1484         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1485         err = -ENODEV;
1486         if (dev == NULL)
1487                 goto out_unlock;
1488
1489         err = -ENETDOWN;
1490         if (!(dev->flags & IFF_UP))
1491                 goto out_unlock;
1492
1493         /*
1494          * You may not queue a frame bigger than the mtu. This is the lowest level
1495          * raw protocol and you must do your own fragmentation at this level.
1496          */
1497
1498         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1499                 if (!netif_supports_nofcs(dev)) {
1500                         err = -EPROTONOSUPPORT;
1501                         goto out_unlock;
1502                 }
1503                 extra_len = 4; /* We're doing our own CRC */
1504         }
1505
1506         err = -EMSGSIZE;
1507         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1508                 goto out_unlock;
1509
1510         if (!skb) {
1511                 size_t reserved = LL_RESERVED_SPACE(dev);
1512                 int tlen = dev->needed_tailroom;
1513                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1514
1515                 rcu_read_unlock();
1516                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1517                 if (skb == NULL)
1518                         return -ENOBUFS;
1519                 /* FIXME: Save some space for broken drivers that write a hard
1520                  * header at transmission time by themselves. PPP is the notable
1521                  * one here. This should really be fixed at the driver level.
1522                  */
1523                 skb_reserve(skb, reserved);
1524                 skb_reset_network_header(skb);
1525
1526                 /* Try to align data part correctly */
1527                 if (hhlen) {
1528                         skb->data -= hhlen;
1529                         skb->tail -= hhlen;
1530                         if (len < hhlen)
1531                                 skb_reset_network_header(skb);
1532                 }
1533                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1534                 if (err)
1535                         goto out_free;
1536                 goto retry;
1537         }
1538
1539         if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1540                 /* Earlier code assumed this would be a VLAN pkt,
1541                  * double-check this now that we have the actual
1542                  * packet in hand.
1543                  */
1544                 struct ethhdr *ehdr;
1545                 skb_reset_mac_header(skb);
1546                 ehdr = eth_hdr(skb);
1547                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1548                         err = -EMSGSIZE;
1549                         goto out_unlock;
1550                 }
1551         }
1552
1553         skb->protocol = proto;
1554         skb->dev = dev;
1555         skb->priority = sk->sk_priority;
1556         skb->mark = sk->sk_mark;
1557         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1558         if (err < 0)
1559                 goto out_unlock;
1560
1561         if (unlikely(extra_len == 4))
1562                 skb->no_fcs = 1;
1563
1564         dev_queue_xmit(skb);
1565         rcu_read_unlock();
1566         return len;
1567
1568 out_unlock:
1569         rcu_read_unlock();
1570 out_free:
1571         kfree_skb(skb);
1572         return err;
1573 }
1574
1575 static unsigned int run_filter(const struct sk_buff *skb,
1576                                       const struct sock *sk,
1577                                       unsigned int res)
1578 {
1579         struct sk_filter *filter;
1580
1581         rcu_read_lock();
1582         filter = rcu_dereference(sk->sk_filter);
1583         if (filter != NULL)
1584                 res = SK_RUN_FILTER(filter, skb);
1585         rcu_read_unlock();
1586
1587         return res;
1588 }
1589
1590 /*
1591  * This function makes lazy skb cloning in hope that most of packets
1592  * are discarded by BPF.
1593  *
1594  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1595  * and skb->cb are mangled. It works because (and until) packets
1596  * falling here are owned by current CPU. Output packets are cloned
1597  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1598  * sequencially, so that if we return skb to original state on exit,
1599  * we will not harm anyone.
1600  */
1601
1602 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1603                       struct packet_type *pt, struct net_device *orig_dev)
1604 {
1605         struct sock *sk;
1606         struct sockaddr_ll *sll;
1607         struct packet_sock *po;
1608         u8 *skb_head = skb->data;
1609         int skb_len = skb->len;
1610         unsigned int snaplen, res;
1611
1612         if (skb->pkt_type == PACKET_LOOPBACK)
1613                 goto drop;
1614
1615         sk = pt->af_packet_priv;
1616         po = pkt_sk(sk);
1617
1618         if (!net_eq(dev_net(dev), sock_net(sk)))
1619                 goto drop;
1620
1621         skb->dev = dev;
1622
1623         if (dev->header_ops) {
1624                 /* The device has an explicit notion of ll header,
1625                  * exported to higher levels.
1626                  *
1627                  * Otherwise, the device hides details of its frame
1628                  * structure, so that corresponding packet head is
1629                  * never delivered to user.
1630                  */
1631                 if (sk->sk_type != SOCK_DGRAM)
1632                         skb_push(skb, skb->data - skb_mac_header(skb));
1633                 else if (skb->pkt_type == PACKET_OUTGOING) {
1634                         /* Special case: outgoing packets have ll header at head */
1635                         skb_pull(skb, skb_network_offset(skb));
1636                 }
1637         }
1638
1639         snaplen = skb->len;
1640
1641         res = run_filter(skb, sk, snaplen);
1642         if (!res)
1643                 goto drop_n_restore;
1644         if (snaplen > res)
1645                 snaplen = res;
1646
1647         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1648                 goto drop_n_acct;
1649
1650         if (skb_shared(skb)) {
1651                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1652                 if (nskb == NULL)
1653                         goto drop_n_acct;
1654
1655                 if (skb_head != skb->data) {
1656                         skb->data = skb_head;
1657                         skb->len = skb_len;
1658                 }
1659                 consume_skb(skb);
1660                 skb = nskb;
1661         }
1662
1663         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1664                      sizeof(skb->cb));
1665
1666         sll = &PACKET_SKB_CB(skb)->sa.ll;
1667         sll->sll_family = AF_PACKET;
1668         sll->sll_hatype = dev->type;
1669         sll->sll_protocol = skb->protocol;
1670         sll->sll_pkttype = skb->pkt_type;
1671         if (unlikely(po->origdev))
1672                 sll->sll_ifindex = orig_dev->ifindex;
1673         else
1674                 sll->sll_ifindex = dev->ifindex;
1675
1676         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1677
1678         PACKET_SKB_CB(skb)->origlen = skb->len;
1679
1680         if (pskb_trim(skb, snaplen))
1681                 goto drop_n_acct;
1682
1683         skb_set_owner_r(skb, sk);
1684         skb->dev = NULL;
1685         skb_dst_drop(skb);
1686
1687         /* drop conntrack reference */
1688         nf_reset(skb);
1689
1690         spin_lock(&sk->sk_receive_queue.lock);
1691         po->stats.tp_packets++;
1692         skb->dropcount = atomic_read(&sk->sk_drops);
1693         __skb_queue_tail(&sk->sk_receive_queue, skb);
1694         spin_unlock(&sk->sk_receive_queue.lock);
1695         sk->sk_data_ready(sk, skb->len);
1696         return 0;
1697
1698 drop_n_acct:
1699         spin_lock(&sk->sk_receive_queue.lock);
1700         po->stats.tp_drops++;
1701         atomic_inc(&sk->sk_drops);
1702         spin_unlock(&sk->sk_receive_queue.lock);
1703
1704 drop_n_restore:
1705         if (skb_head != skb->data && skb_shared(skb)) {
1706                 skb->data = skb_head;
1707                 skb->len = skb_len;
1708         }
1709 drop:
1710         consume_skb(skb);
1711         return 0;
1712 }
1713
1714 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1715                        struct packet_type *pt, struct net_device *orig_dev)
1716 {
1717         struct sock *sk;
1718         struct packet_sock *po;
1719         struct sockaddr_ll *sll;
1720         union {
1721                 struct tpacket_hdr *h1;
1722                 struct tpacket2_hdr *h2;
1723                 struct tpacket3_hdr *h3;
1724                 void *raw;
1725         } h;
1726         u8 *skb_head = skb->data;
1727         int skb_len = skb->len;
1728         unsigned int snaplen, res;
1729         unsigned long status = TP_STATUS_USER;
1730         unsigned short macoff, netoff, hdrlen;
1731         struct sk_buff *copy_skb = NULL;
1732         struct timeval tv;
1733         struct timespec ts;
1734         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1735
1736         if (skb->pkt_type == PACKET_LOOPBACK)
1737                 goto drop;
1738
1739         sk = pt->af_packet_priv;
1740         po = pkt_sk(sk);
1741
1742         if (!net_eq(dev_net(dev), sock_net(sk)))
1743                 goto drop;
1744
1745         if (dev->header_ops) {
1746                 if (sk->sk_type != SOCK_DGRAM)
1747                         skb_push(skb, skb->data - skb_mac_header(skb));
1748                 else if (skb->pkt_type == PACKET_OUTGOING) {
1749                         /* Special case: outgoing packets have ll header at head */
1750                         skb_pull(skb, skb_network_offset(skb));
1751                 }
1752         }
1753
1754         if (skb->ip_summed == CHECKSUM_PARTIAL)
1755                 status |= TP_STATUS_CSUMNOTREADY;
1756
1757         snaplen = skb->len;
1758
1759         res = run_filter(skb, sk, snaplen);
1760         if (!res)
1761                 goto drop_n_restore;
1762         if (snaplen > res)
1763                 snaplen = res;
1764
1765         if (sk->sk_type == SOCK_DGRAM) {
1766                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1767                                   po->tp_reserve;
1768         } else {
1769                 unsigned int maclen = skb_network_offset(skb);
1770                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1771                                        (maclen < 16 ? 16 : maclen)) +
1772                         po->tp_reserve;
1773                 macoff = netoff - maclen;
1774         }
1775         if (po->tp_version <= TPACKET_V2) {
1776                 if (macoff + snaplen > po->rx_ring.frame_size) {
1777                         if (po->copy_thresh &&
1778                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1779                                 if (skb_shared(skb)) {
1780                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1781                                 } else {
1782                                         copy_skb = skb_get(skb);
1783                                         skb_head = skb->data;
1784                                 }
1785                                 if (copy_skb)
1786                                         skb_set_owner_r(copy_skb, sk);
1787                         }
1788                         snaplen = po->rx_ring.frame_size - macoff;
1789                         if ((int)snaplen < 0)
1790                                 snaplen = 0;
1791                 }
1792         }
1793         spin_lock(&sk->sk_receive_queue.lock);
1794         h.raw = packet_current_rx_frame(po, skb,
1795                                         TP_STATUS_KERNEL, (macoff+snaplen));
1796         if (!h.raw)
1797                 goto ring_is_full;
1798         if (po->tp_version <= TPACKET_V2) {
1799                 packet_increment_rx_head(po, &po->rx_ring);
1800         /*
1801          * LOSING will be reported till you read the stats,
1802          * because it's COR - Clear On Read.
1803          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1804          * at packet level.
1805          */
1806                 if (po->stats.tp_drops)
1807                         status |= TP_STATUS_LOSING;
1808         }
1809         po->stats.tp_packets++;
1810         if (copy_skb) {
1811                 status |= TP_STATUS_COPY;
1812                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1813         }
1814         spin_unlock(&sk->sk_receive_queue.lock);
1815
1816         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1817
1818         switch (po->tp_version) {
1819         case TPACKET_V1:
1820                 h.h1->tp_len = skb->len;
1821                 h.h1->tp_snaplen = snaplen;
1822                 h.h1->tp_mac = macoff;
1823                 h.h1->tp_net = netoff;
1824                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1825                                 && shhwtstamps->syststamp.tv64)
1826                         tv = ktime_to_timeval(shhwtstamps->syststamp);
1827                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1828                                 && shhwtstamps->hwtstamp.tv64)
1829                         tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1830                 else if (skb->tstamp.tv64)
1831                         tv = ktime_to_timeval(skb->tstamp);
1832                 else
1833                         do_gettimeofday(&tv);
1834                 h.h1->tp_sec = tv.tv_sec;
1835                 h.h1->tp_usec = tv.tv_usec;
1836                 hdrlen = sizeof(*h.h1);
1837                 break;
1838         case TPACKET_V2:
1839                 h.h2->tp_len = skb->len;
1840                 h.h2->tp_snaplen = snaplen;
1841                 h.h2->tp_mac = macoff;
1842                 h.h2->tp_net = netoff;
1843                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1844                                 && shhwtstamps->syststamp.tv64)
1845                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1846                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1847                                 && shhwtstamps->hwtstamp.tv64)
1848                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1849                 else if (skb->tstamp.tv64)
1850                         ts = ktime_to_timespec(skb->tstamp);
1851                 else
1852                         getnstimeofday(&ts);
1853                 h.h2->tp_sec = ts.tv_sec;
1854                 h.h2->tp_nsec = ts.tv_nsec;
1855                 if (vlan_tx_tag_present(skb)) {
1856                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1857                         status |= TP_STATUS_VLAN_VALID;
1858                 } else {
1859                         h.h2->tp_vlan_tci = 0;
1860                 }
1861                 h.h2->tp_padding = 0;
1862                 hdrlen = sizeof(*h.h2);
1863                 break;
1864         case TPACKET_V3:
1865                 /* tp_nxt_offset,vlan are already populated above.
1866                  * So DONT clear those fields here
1867                  */
1868                 h.h3->tp_status |= status;
1869                 h.h3->tp_len = skb->len;
1870                 h.h3->tp_snaplen = snaplen;
1871                 h.h3->tp_mac = macoff;
1872                 h.h3->tp_net = netoff;
1873                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1874                                 && shhwtstamps->syststamp.tv64)
1875                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1876                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1877                                 && shhwtstamps->hwtstamp.tv64)
1878                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1879                 else if (skb->tstamp.tv64)
1880                         ts = ktime_to_timespec(skb->tstamp);
1881                 else
1882                         getnstimeofday(&ts);
1883                 h.h3->tp_sec  = ts.tv_sec;
1884                 h.h3->tp_nsec = ts.tv_nsec;
1885                 hdrlen = sizeof(*h.h3);
1886                 break;
1887         default:
1888                 BUG();
1889         }
1890
1891         sll = h.raw + TPACKET_ALIGN(hdrlen);
1892         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1893         sll->sll_family = AF_PACKET;
1894         sll->sll_hatype = dev->type;
1895         sll->sll_protocol = skb->protocol;
1896         sll->sll_pkttype = skb->pkt_type;
1897         if (unlikely(po->origdev))
1898                 sll->sll_ifindex = orig_dev->ifindex;
1899         else
1900                 sll->sll_ifindex = dev->ifindex;
1901
1902         smp_mb();
1903 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1904         {
1905                 u8 *start, *end;
1906
1907                 if (po->tp_version <= TPACKET_V2) {
1908                         end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1909                                 + macoff + snaplen);
1910                         for (start = h.raw; start < end; start += PAGE_SIZE)
1911                                 flush_dcache_page(pgv_to_page(start));
1912                 }
1913                 smp_wmb();
1914         }
1915 #endif
1916         if (po->tp_version <= TPACKET_V2)
1917                 __packet_set_status(po, h.raw, status);
1918         else
1919                 prb_clear_blk_fill_status(&po->rx_ring);
1920
1921         sk->sk_data_ready(sk, 0);
1922
1923 drop_n_restore:
1924         if (skb_head != skb->data && skb_shared(skb)) {
1925                 skb->data = skb_head;
1926                 skb->len = skb_len;
1927         }
1928 drop:
1929         kfree_skb(skb);
1930         return 0;
1931
1932 ring_is_full:
1933         po->stats.tp_drops++;
1934         spin_unlock(&sk->sk_receive_queue.lock);
1935
1936         sk->sk_data_ready(sk, 0);
1937         kfree_skb(copy_skb);
1938         goto drop_n_restore;
1939 }
1940
1941 static void tpacket_destruct_skb(struct sk_buff *skb)
1942 {
1943         struct packet_sock *po = pkt_sk(skb->sk);
1944         void *ph;
1945
1946         if (likely(po->tx_ring.pg_vec)) {
1947                 ph = skb_shinfo(skb)->destructor_arg;
1948                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1949                 atomic_dec(&po->tx_ring.pending);
1950                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1951         }
1952
1953         sock_wfree(skb);
1954 }
1955
1956 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1957                 void *frame, struct net_device *dev, int size_max,
1958                 __be16 proto, unsigned char *addr, int hlen)
1959 {
1960         union {
1961                 struct tpacket_hdr *h1;
1962                 struct tpacket2_hdr *h2;
1963                 void *raw;
1964         } ph;
1965         int to_write, offset, len, tp_len, nr_frags, len_max;
1966         struct socket *sock = po->sk.sk_socket;
1967         struct page *page;
1968         void *data;
1969         int err;
1970
1971         ph.raw = frame;
1972
1973         skb->protocol = proto;
1974         skb->dev = dev;
1975         skb->priority = po->sk.sk_priority;
1976         skb->mark = po->sk.sk_mark;
1977         skb_shinfo(skb)->destructor_arg = ph.raw;
1978
1979         switch (po->tp_version) {
1980         case TPACKET_V2:
1981                 tp_len = ph.h2->tp_len;
1982                 break;
1983         default:
1984                 tp_len = ph.h1->tp_len;
1985                 break;
1986         }
1987         if (unlikely(tp_len > size_max)) {
1988                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1989                 return -EMSGSIZE;
1990         }
1991
1992         skb_reserve(skb, hlen);
1993         skb_reset_network_header(skb);
1994
1995         data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1996         to_write = tp_len;
1997
1998         if (sock->type == SOCK_DGRAM) {
1999                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2000                                 NULL, tp_len);
2001                 if (unlikely(err < 0))
2002                         return -EINVAL;
2003         } else if (dev->hard_header_len) {
2004                 /* net device doesn't like empty head */
2005                 if (unlikely(tp_len <= dev->hard_header_len)) {
2006                         pr_err("packet size is too short (%d < %d)\n",
2007                                tp_len, dev->hard_header_len);
2008                         return -EINVAL;
2009                 }
2010
2011                 skb_push(skb, dev->hard_header_len);
2012                 err = skb_store_bits(skb, 0, data,
2013                                 dev->hard_header_len);
2014                 if (unlikely(err))
2015                         return err;
2016
2017                 data += dev->hard_header_len;
2018                 to_write -= dev->hard_header_len;
2019         }
2020
2021         err = -EFAULT;
2022         offset = offset_in_page(data);
2023         len_max = PAGE_SIZE - offset;
2024         len = ((to_write > len_max) ? len_max : to_write);
2025
2026         skb->data_len = to_write;
2027         skb->len += to_write;
2028         skb->truesize += to_write;
2029         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2030
2031         while (likely(to_write)) {
2032                 nr_frags = skb_shinfo(skb)->nr_frags;
2033
2034                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2035                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2036                                MAX_SKB_FRAGS);
2037                         return -EFAULT;
2038                 }
2039
2040                 page = pgv_to_page(data);
2041                 data += len;
2042                 flush_dcache_page(page);
2043                 get_page(page);
2044                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2045                 to_write -= len;
2046                 offset = 0;
2047                 len_max = PAGE_SIZE;
2048                 len = ((to_write > len_max) ? len_max : to_write);
2049         }
2050
2051         return tp_len;
2052 }
2053
2054 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2055 {
2056         struct sk_buff *skb;
2057         struct net_device *dev;
2058         __be16 proto;
2059         bool need_rls_dev = false;
2060         int err, reserve = 0;
2061         void *ph;
2062         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2063         int tp_len, size_max;
2064         unsigned char *addr;
2065         int len_sum = 0;
2066         int status = 0;
2067         int hlen, tlen;
2068
2069         mutex_lock(&po->pg_vec_lock);
2070
2071         err = -EBUSY;
2072         if (saddr == NULL) {
2073                 dev = po->prot_hook.dev;
2074                 proto   = po->num;
2075                 addr    = NULL;
2076         } else {
2077                 err = -EINVAL;
2078                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2079                         goto out;
2080                 if (msg->msg_namelen < (saddr->sll_halen
2081                                         + offsetof(struct sockaddr_ll,
2082                                                 sll_addr)))
2083                         goto out;
2084                 proto   = saddr->sll_protocol;
2085                 addr    = saddr->sll_addr;
2086                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2087                 need_rls_dev = true;
2088         }
2089
2090         err = -ENXIO;
2091         if (unlikely(dev == NULL))
2092                 goto out;
2093
2094         reserve = dev->hard_header_len;
2095
2096         err = -ENETDOWN;
2097         if (unlikely(!(dev->flags & IFF_UP)))
2098                 goto out_put;
2099
2100         size_max = po->tx_ring.frame_size
2101                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2102
2103         if (size_max > dev->mtu + reserve)
2104                 size_max = dev->mtu + reserve;
2105
2106         do {
2107                 ph = packet_current_frame(po, &po->tx_ring,
2108                                 TP_STATUS_SEND_REQUEST);
2109
2110                 if (unlikely(ph == NULL)) {
2111                         schedule();
2112                         continue;
2113                 }
2114
2115                 status = TP_STATUS_SEND_REQUEST;
2116                 hlen = LL_RESERVED_SPACE(dev);
2117                 tlen = dev->needed_tailroom;
2118                 skb = sock_alloc_send_skb(&po->sk,
2119                                 hlen + tlen + sizeof(struct sockaddr_ll),
2120                                 0, &err);
2121
2122                 if (unlikely(skb == NULL))
2123                         goto out_status;
2124
2125                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2126                                 addr, hlen);
2127
2128                 if (unlikely(tp_len < 0)) {
2129                         if (po->tp_loss) {
2130                                 __packet_set_status(po, ph,
2131                                                 TP_STATUS_AVAILABLE);
2132                                 packet_increment_head(&po->tx_ring);
2133                                 kfree_skb(skb);
2134                                 continue;
2135                         } else {
2136                                 status = TP_STATUS_WRONG_FORMAT;
2137                                 err = tp_len;
2138                                 goto out_status;
2139                         }
2140                 }
2141
2142                 skb->destructor = tpacket_destruct_skb;
2143                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2144                 atomic_inc(&po->tx_ring.pending);
2145
2146                 status = TP_STATUS_SEND_REQUEST;
2147                 err = dev_queue_xmit(skb);
2148                 if (unlikely(err > 0)) {
2149                         err = net_xmit_errno(err);
2150                         if (err && __packet_get_status(po, ph) ==
2151                                    TP_STATUS_AVAILABLE) {
2152                                 /* skb was destructed already */
2153                                 skb = NULL;
2154                                 goto out_status;
2155                         }
2156                         /*
2157                          * skb was dropped but not destructed yet;
2158                          * let's treat it like congestion or err < 0
2159                          */
2160                         err = 0;
2161                 }
2162                 packet_increment_head(&po->tx_ring);
2163                 len_sum += tp_len;
2164         } while (likely((ph != NULL) ||
2165                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2166                          (atomic_read(&po->tx_ring.pending))))
2167                 );
2168
2169         err = len_sum;
2170         goto out_put;
2171
2172 out_status:
2173         __packet_set_status(po, ph, status);
2174         kfree_skb(skb);
2175 out_put:
2176         if (need_rls_dev)
2177                 dev_put(dev);
2178 out:
2179         mutex_unlock(&po->pg_vec_lock);
2180         return err;
2181 }
2182
2183 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2184                                         size_t reserve, size_t len,
2185                                         size_t linear, int noblock,
2186                                         int *err)
2187 {
2188         struct sk_buff *skb;
2189
2190         /* Under a page?  Don't bother with paged skb. */
2191         if (prepad + len < PAGE_SIZE || !linear)
2192                 linear = len;
2193
2194         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2195                                    err);
2196         if (!skb)
2197                 return NULL;
2198
2199         skb_reserve(skb, reserve);
2200         skb_put(skb, linear);
2201         skb->data_len = len - linear;
2202         skb->len += len - linear;
2203
2204         return skb;
2205 }
2206
2207 static int packet_snd(struct socket *sock,
2208                           struct msghdr *msg, size_t len)
2209 {
2210         struct sock *sk = sock->sk;
2211         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2212         struct sk_buff *skb;
2213         struct net_device *dev;
2214         __be16 proto;
2215         bool need_rls_dev = false;
2216         unsigned char *addr;
2217         int err, reserve = 0;
2218         struct virtio_net_hdr vnet_hdr = { 0 };
2219         int offset = 0;
2220         int vnet_hdr_len;
2221         struct packet_sock *po = pkt_sk(sk);
2222         unsigned short gso_type = 0;
2223         int hlen, tlen;
2224         int extra_len = 0;
2225
2226         /*
2227          *      Get and verify the address.
2228          */
2229
2230         if (saddr == NULL) {
2231                 dev = po->prot_hook.dev;
2232                 proto   = po->num;
2233                 addr    = NULL;
2234         } else {
2235                 err = -EINVAL;
2236                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2237                         goto out;
2238                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2239                         goto out;
2240                 proto   = saddr->sll_protocol;
2241                 addr    = saddr->sll_addr;
2242                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2243                 need_rls_dev = true;
2244         }
2245
2246         err = -ENXIO;
2247         if (dev == NULL)
2248                 goto out_unlock;
2249         if (sock->type == SOCK_RAW)
2250                 reserve = dev->hard_header_len;
2251
2252         err = -ENETDOWN;
2253         if (!(dev->flags & IFF_UP))
2254                 goto out_unlock;
2255
2256         if (po->has_vnet_hdr) {
2257                 vnet_hdr_len = sizeof(vnet_hdr);
2258
2259                 err = -EINVAL;
2260                 if (len < vnet_hdr_len)
2261                         goto out_unlock;
2262
2263                 len -= vnet_hdr_len;
2264
2265                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2266                                        vnet_hdr_len);
2267                 if (err < 0)
2268                         goto out_unlock;
2269
2270                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2271                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2272                       vnet_hdr.hdr_len))
2273                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
2274                                                  vnet_hdr.csum_offset + 2;
2275
2276                 err = -EINVAL;
2277                 if (vnet_hdr.hdr_len > len)
2278                         goto out_unlock;
2279
2280                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2281                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2282                         case VIRTIO_NET_HDR_GSO_TCPV4:
2283                                 gso_type = SKB_GSO_TCPV4;
2284                                 break;
2285                         case VIRTIO_NET_HDR_GSO_TCPV6:
2286                                 gso_type = SKB_GSO_TCPV6;
2287                                 break;
2288                         case VIRTIO_NET_HDR_GSO_UDP:
2289                                 gso_type = SKB_GSO_UDP;
2290                                 break;
2291                         default:
2292                                 goto out_unlock;
2293                         }
2294
2295                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2296                                 gso_type |= SKB_GSO_TCP_ECN;
2297
2298                         if (vnet_hdr.gso_size == 0)
2299                                 goto out_unlock;
2300
2301                 }
2302         }
2303
2304         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2305                 if (!netif_supports_nofcs(dev)) {
2306                         err = -EPROTONOSUPPORT;
2307                         goto out_unlock;
2308                 }
2309                 extra_len = 4; /* We're doing our own CRC */
2310         }
2311
2312         err = -EMSGSIZE;
2313         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2314                 goto out_unlock;
2315
2316         err = -ENOBUFS;
2317         hlen = LL_RESERVED_SPACE(dev);
2318         tlen = dev->needed_tailroom;
2319         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2320                                msg->msg_flags & MSG_DONTWAIT, &err);
2321         if (skb == NULL)
2322                 goto out_unlock;
2323
2324         skb_set_network_header(skb, reserve);
2325
2326         err = -EINVAL;
2327         if (sock->type == SOCK_DGRAM &&
2328             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2329                 goto out_free;
2330
2331         /* Returns -EFAULT on error */
2332         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2333         if (err)
2334                 goto out_free;
2335         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2336         if (err < 0)
2337                 goto out_free;
2338
2339         if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2340                 /* Earlier code assumed this would be a VLAN pkt,
2341                  * double-check this now that we have the actual
2342                  * packet in hand.
2343                  */
2344                 struct ethhdr *ehdr;
2345                 skb_reset_mac_header(skb);
2346                 ehdr = eth_hdr(skb);
2347                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2348                         err = -EMSGSIZE;
2349                         goto out_free;
2350                 }
2351         }
2352
2353         skb->protocol = proto;
2354         skb->dev = dev;
2355         skb->priority = sk->sk_priority;
2356         skb->mark = sk->sk_mark;
2357
2358         if (po->has_vnet_hdr) {
2359                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2360                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2361                                                   vnet_hdr.csum_offset)) {
2362                                 err = -EINVAL;
2363                                 goto out_free;
2364                         }
2365                 }
2366
2367                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2368                 skb_shinfo(skb)->gso_type = gso_type;
2369
2370                 /* Header must be checked, and gso_segs computed. */
2371                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2372                 skb_shinfo(skb)->gso_segs = 0;
2373
2374                 len += vnet_hdr_len;
2375         }
2376
2377         if (unlikely(extra_len == 4))
2378                 skb->no_fcs = 1;
2379
2380         /*
2381          *      Now send it
2382          */
2383
2384         err = dev_queue_xmit(skb);
2385         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2386                 goto out_unlock;
2387
2388         if (need_rls_dev)
2389                 dev_put(dev);
2390
2391         return len;
2392
2393 out_free:
2394         kfree_skb(skb);
2395 out_unlock:
2396         if (dev && need_rls_dev)
2397                 dev_put(dev);
2398 out:
2399         return err;
2400 }
2401
2402 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2403                 struct msghdr *msg, size_t len)
2404 {
2405         struct sock *sk = sock->sk;
2406         struct packet_sock *po = pkt_sk(sk);
2407         if (po->tx_ring.pg_vec)
2408                 return tpacket_snd(po, msg);
2409         else
2410                 return packet_snd(sock, msg, len);
2411 }
2412
2413 /*
2414  *      Close a PACKET socket. This is fairly simple. We immediately go
2415  *      to 'closed' state and remove our protocol entry in the device list.
2416  */
2417
2418 static int packet_release(struct socket *sock)
2419 {
2420         struct sock *sk = sock->sk;
2421         struct packet_sock *po;
2422         struct net *net;
2423         union tpacket_req_u req_u;
2424
2425         if (!sk)
2426                 return 0;
2427
2428         net = sock_net(sk);
2429         po = pkt_sk(sk);
2430
2431         spin_lock_bh(&net->packet.sklist_lock);
2432         sk_del_node_init_rcu(sk);
2433         sock_prot_inuse_add(net, sk->sk_prot, -1);
2434         spin_unlock_bh(&net->packet.sklist_lock);
2435
2436         spin_lock(&po->bind_lock);
2437         unregister_prot_hook(sk, false);
2438         if (po->prot_hook.dev) {
2439                 dev_put(po->prot_hook.dev);
2440                 po->prot_hook.dev = NULL;
2441         }
2442         spin_unlock(&po->bind_lock);
2443
2444         packet_flush_mclist(sk);
2445
2446         memset(&req_u, 0, sizeof(req_u));
2447
2448         if (po->rx_ring.pg_vec)
2449                 packet_set_ring(sk, &req_u, 1, 0);
2450
2451         if (po->tx_ring.pg_vec)
2452                 packet_set_ring(sk, &req_u, 1, 1);
2453
2454         fanout_release(sk);
2455
2456         synchronize_net();
2457         /*
2458          *      Now the socket is dead. No more input will appear.
2459          */
2460         sock_orphan(sk);
2461         sock->sk = NULL;
2462
2463         /* Purge queues */
2464
2465         skb_queue_purge(&sk->sk_receive_queue);
2466         sk_refcnt_debug_release(sk);
2467
2468         sock_put(sk);
2469         return 0;
2470 }
2471
2472 /*
2473  *      Attach a packet hook.
2474  */
2475
2476 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2477 {
2478         struct packet_sock *po = pkt_sk(sk);
2479
2480         if (po->fanout) {
2481                 if (dev)
2482                         dev_put(dev);
2483
2484                 return -EINVAL;
2485         }
2486
2487         lock_sock(sk);
2488
2489         spin_lock(&po->bind_lock);
2490         unregister_prot_hook(sk, true);
2491         po->num = protocol;
2492         po->prot_hook.type = protocol;
2493         if (po->prot_hook.dev)
2494                 dev_put(po->prot_hook.dev);
2495         po->prot_hook.dev = dev;
2496
2497         po->ifindex = dev ? dev->ifindex : 0;
2498
2499         if (protocol == 0)
2500                 goto out_unlock;
2501
2502         if (!dev || (dev->flags & IFF_UP)) {
2503                 register_prot_hook(sk);
2504         } else {
2505                 sk->sk_err = ENETDOWN;
2506                 if (!sock_flag(sk, SOCK_DEAD))
2507                         sk->sk_error_report(sk);
2508         }
2509
2510 out_unlock:
2511         spin_unlock(&po->bind_lock);
2512         release_sock(sk);
2513         return 0;
2514 }
2515
2516 /*
2517  *      Bind a packet socket to a device
2518  */
2519
2520 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2521                             int addr_len)
2522 {
2523         struct sock *sk = sock->sk;
2524         char name[15];
2525         struct net_device *dev;
2526         int err = -ENODEV;
2527
2528         /*
2529          *      Check legality
2530          */
2531
2532         if (addr_len != sizeof(struct sockaddr))
2533                 return -EINVAL;
2534         strlcpy(name, uaddr->sa_data, sizeof(name));
2535
2536         dev = dev_get_by_name(sock_net(sk), name);
2537         if (dev)
2538                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2539         return err;
2540 }
2541
2542 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2543 {
2544         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2545         struct sock *sk = sock->sk;
2546         struct net_device *dev = NULL;
2547         int err;
2548
2549
2550         /*
2551          *      Check legality
2552          */
2553
2554         if (addr_len < sizeof(struct sockaddr_ll))
2555                 return -EINVAL;
2556         if (sll->sll_family != AF_PACKET)
2557                 return -EINVAL;
2558
2559         if (sll->sll_ifindex) {
2560                 err = -ENODEV;
2561                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2562                 if (dev == NULL)
2563                         goto out;
2564         }
2565         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2566
2567 out:
2568         return err;
2569 }
2570
2571 static struct proto packet_proto = {
2572         .name     = "PACKET",
2573         .owner    = THIS_MODULE,
2574         .obj_size = sizeof(struct packet_sock),
2575 };
2576
2577 /*
2578  *      Create a packet of type SOCK_PACKET.
2579  */
2580
2581 static int packet_create(struct net *net, struct socket *sock, int protocol,
2582                          int kern)
2583 {
2584         struct sock *sk;
2585         struct packet_sock *po;
2586         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2587         int err;
2588
2589         if (!capable(CAP_NET_RAW))
2590                 return -EPERM;
2591         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2592             sock->type != SOCK_PACKET)
2593                 return -ESOCKTNOSUPPORT;
2594
2595         sock->state = SS_UNCONNECTED;
2596
2597         err = -ENOBUFS;
2598         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2599         if (sk == NULL)
2600                 goto out;
2601
2602         sock->ops = &packet_ops;
2603         if (sock->type == SOCK_PACKET)
2604                 sock->ops = &packet_ops_spkt;
2605
2606         sock_init_data(sock, sk);
2607
2608         po = pkt_sk(sk);
2609         sk->sk_family = PF_PACKET;
2610         po->num = proto;
2611
2612         sk->sk_destruct = packet_sock_destruct;
2613         sk_refcnt_debug_inc(sk);
2614
2615         /*
2616          *      Attach a protocol block
2617          */
2618
2619         spin_lock_init(&po->bind_lock);
2620         mutex_init(&po->pg_vec_lock);
2621         po->prot_hook.func = packet_rcv;
2622
2623         if (sock->type == SOCK_PACKET)
2624                 po->prot_hook.func = packet_rcv_spkt;
2625
2626         po->prot_hook.af_packet_priv = sk;
2627
2628         if (proto) {
2629                 po->prot_hook.type = proto;
2630                 register_prot_hook(sk);
2631         }
2632
2633         spin_lock_bh(&net->packet.sklist_lock);
2634         sk_add_node_rcu(sk, &net->packet.sklist);
2635         sock_prot_inuse_add(net, &packet_proto, 1);
2636         spin_unlock_bh(&net->packet.sklist_lock);
2637
2638         return 0;
2639 out:
2640         return err;
2641 }
2642
2643 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2644 {
2645         struct sock_exterr_skb *serr;
2646         struct sk_buff *skb, *skb2;
2647         int copied, err;
2648
2649         err = -EAGAIN;
2650         skb = skb_dequeue(&sk->sk_error_queue);
2651         if (skb == NULL)
2652                 goto out;
2653
2654         copied = skb->len;
2655         if (copied > len) {
2656                 msg->msg_flags |= MSG_TRUNC;
2657                 copied = len;
2658         }
2659         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2660         if (err)
2661                 goto out_free_skb;
2662
2663         sock_recv_timestamp(msg, sk, skb);
2664
2665         serr = SKB_EXT_ERR(skb);
2666         put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2667                  sizeof(serr->ee), &serr->ee);
2668
2669         msg->msg_flags |= MSG_ERRQUEUE;
2670         err = copied;
2671
2672         /* Reset and regenerate socket error */
2673         spin_lock_bh(&sk->sk_error_queue.lock);
2674         sk->sk_err = 0;
2675         if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2676                 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2677                 spin_unlock_bh(&sk->sk_error_queue.lock);
2678                 sk->sk_error_report(sk);
2679         } else
2680                 spin_unlock_bh(&sk->sk_error_queue.lock);
2681
2682 out_free_skb:
2683         kfree_skb(skb);
2684 out:
2685         return err;
2686 }
2687
2688 /*
2689  *      Pull a packet from our receive queue and hand it to the user.
2690  *      If necessary we block.
2691  */
2692
2693 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2694                           struct msghdr *msg, size_t len, int flags)
2695 {
2696         struct sock *sk = sock->sk;
2697         struct sk_buff *skb;
2698         int copied, err;
2699         struct sockaddr_ll *sll;
2700         int vnet_hdr_len = 0;
2701
2702         err = -EINVAL;
2703         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2704                 goto out;
2705
2706 #if 0
2707         /* What error should we return now? EUNATTACH? */
2708         if (pkt_sk(sk)->ifindex < 0)
2709                 return -ENODEV;
2710 #endif
2711
2712         if (flags & MSG_ERRQUEUE) {
2713                 err = packet_recv_error(sk, msg, len);
2714                 goto out;
2715         }
2716
2717         /*
2718          *      Call the generic datagram receiver. This handles all sorts
2719          *      of horrible races and re-entrancy so we can forget about it
2720          *      in the protocol layers.
2721          *
2722          *      Now it will return ENETDOWN, if device have just gone down,
2723          *      but then it will block.
2724          */
2725
2726         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2727
2728         /*
2729          *      An error occurred so return it. Because skb_recv_datagram()
2730          *      handles the blocking we don't see and worry about blocking
2731          *      retries.
2732          */
2733
2734         if (skb == NULL)
2735                 goto out;
2736
2737         if (pkt_sk(sk)->has_vnet_hdr) {
2738                 struct virtio_net_hdr vnet_hdr = { 0 };
2739
2740                 err = -EINVAL;
2741                 vnet_hdr_len = sizeof(vnet_hdr);
2742                 if (len < vnet_hdr_len)
2743                         goto out_free;
2744
2745                 len -= vnet_hdr_len;
2746
2747                 if (skb_is_gso(skb)) {
2748                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2749
2750                         /* This is a hint as to how much should be linear. */
2751                         vnet_hdr.hdr_len = skb_headlen(skb);
2752                         vnet_hdr.gso_size = sinfo->gso_size;
2753                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2754                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2755                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2756                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2757                         else if (sinfo->gso_type & SKB_GSO_UDP)
2758                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2759                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2760                                 goto out_free;
2761                         else
2762                                 BUG();
2763                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2764                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2765                 } else
2766                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2767
2768                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2769                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2770                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2771                         vnet_hdr.csum_offset = skb->csum_offset;
2772                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2773                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2774                 } /* else everything is zero */
2775
2776                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2777                                      vnet_hdr_len);
2778                 if (err < 0)
2779                         goto out_free;
2780         }
2781
2782         /*
2783          *      If the address length field is there to be filled in, we fill
2784          *      it in now.
2785          */
2786
2787         sll = &PACKET_SKB_CB(skb)->sa.ll;
2788         if (sock->type == SOCK_PACKET)
2789                 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2790         else
2791                 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2792
2793         /*
2794          *      You lose any data beyond the buffer you gave. If it worries a
2795          *      user program they can ask the device for its MTU anyway.
2796          */
2797
2798         copied = skb->len;
2799         if (copied > len) {
2800                 copied = len;
2801                 msg->msg_flags |= MSG_TRUNC;
2802         }
2803
2804         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2805         if (err)
2806                 goto out_free;
2807
2808         sock_recv_ts_and_drops(msg, sk, skb);
2809
2810         if (msg->msg_name)
2811                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2812                        msg->msg_namelen);
2813
2814         if (pkt_sk(sk)->auxdata) {
2815                 struct tpacket_auxdata aux;
2816
2817                 aux.tp_status = TP_STATUS_USER;
2818                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2819                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2820                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2821                 aux.tp_snaplen = skb->len;
2822                 aux.tp_mac = 0;
2823                 aux.tp_net = skb_network_offset(skb);
2824                 if (vlan_tx_tag_present(skb)) {
2825                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2826                         aux.tp_status |= TP_STATUS_VLAN_VALID;
2827                 } else {
2828                         aux.tp_vlan_tci = 0;
2829                 }
2830                 aux.tp_padding = 0;
2831                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2832         }
2833
2834         /*
2835          *      Free or return the buffer as appropriate. Again this
2836          *      hides all the races and re-entrancy issues from us.
2837          */
2838         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2839
2840 out_free:
2841         skb_free_datagram(sk, skb);
2842 out:
2843         return err;
2844 }
2845
2846 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2847                                int *uaddr_len, int peer)
2848 {
2849         struct net_device *dev;
2850         struct sock *sk = sock->sk;
2851
2852         if (peer)
2853                 return -EOPNOTSUPP;
2854
2855         uaddr->sa_family = AF_PACKET;
2856         rcu_read_lock();
2857         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2858         if (dev)
2859                 strncpy(uaddr->sa_data, dev->name, 14);
2860         else
2861                 memset(uaddr->sa_data, 0, 14);
2862         rcu_read_unlock();
2863         *uaddr_len = sizeof(*uaddr);
2864
2865         return 0;
2866 }
2867
2868 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2869                           int *uaddr_len, int peer)
2870 {
2871         struct net_device *dev;
2872         struct sock *sk = sock->sk;
2873         struct packet_sock *po = pkt_sk(sk);
2874         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2875
2876         if (peer)
2877                 return -EOPNOTSUPP;
2878
2879         sll->sll_family = AF_PACKET;
2880         sll->sll_ifindex = po->ifindex;
2881         sll->sll_protocol = po->num;
2882         sll->sll_pkttype = 0;
2883         rcu_read_lock();
2884         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2885         if (dev) {
2886                 sll->sll_hatype = dev->type;
2887                 sll->sll_halen = dev->addr_len;
2888                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2889         } else {
2890                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
2891                 sll->sll_halen = 0;
2892         }
2893         rcu_read_unlock();
2894         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2895
2896         return 0;
2897 }
2898
2899 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2900                          int what)
2901 {
2902         switch (i->type) {
2903         case PACKET_MR_MULTICAST:
2904                 if (i->alen != dev->addr_len)
2905                         return -EINVAL;
2906                 if (what > 0)
2907                         return dev_mc_add(dev, i->addr);
2908                 else
2909                         return dev_mc_del(dev, i->addr);
2910                 break;
2911         case PACKET_MR_PROMISC:
2912                 return dev_set_promiscuity(dev, what);
2913                 break;
2914         case PACKET_MR_ALLMULTI:
2915                 return dev_set_allmulti(dev, what);
2916                 break;
2917         case PACKET_MR_UNICAST:
2918                 if (i->alen != dev->addr_len)
2919                         return -EINVAL;
2920                 if (what > 0)
2921                         return dev_uc_add(dev, i->addr);
2922                 else
2923                         return dev_uc_del(dev, i->addr);
2924                 break;
2925         default:
2926                 break;
2927         }
2928         return 0;
2929 }
2930
2931 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2932 {
2933         for ( ; i; i = i->next) {
2934                 if (i->ifindex == dev->ifindex)
2935                         packet_dev_mc(dev, i, what);
2936         }
2937 }
2938
2939 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2940 {
2941         struct packet_sock *po = pkt_sk(sk);
2942         struct packet_mclist *ml, *i;
2943         struct net_device *dev;
2944         int err;
2945
2946         rtnl_lock();
2947
2948         err = -ENODEV;
2949         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2950         if (!dev)
2951                 goto done;
2952
2953         err = -EINVAL;
2954         if (mreq->mr_alen > dev->addr_len)
2955                 goto done;
2956
2957         err = -ENOBUFS;
2958         i = kmalloc(sizeof(*i), GFP_KERNEL);
2959         if (i == NULL)
2960                 goto done;
2961
2962         err = 0;
2963         for (ml = po->mclist; ml; ml = ml->next) {
2964                 if (ml->ifindex == mreq->mr_ifindex &&
2965                     ml->type == mreq->mr_type &&
2966                     ml->alen == mreq->mr_alen &&
2967                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2968                         ml->count++;
2969                         /* Free the new element ... */
2970                         kfree(i);
2971                         goto done;
2972                 }
2973         }
2974
2975         i->type = mreq->mr_type;
2976         i->ifindex = mreq->mr_ifindex;
2977         i->alen = mreq->mr_alen;
2978         memcpy(i->addr, mreq->mr_address, i->alen);
2979         i->count = 1;
2980         i->next = po->mclist;
2981         po->mclist = i;
2982         err = packet_dev_mc(dev, i, 1);
2983         if (err) {
2984                 po->mclist = i->next;
2985                 kfree(i);
2986         }
2987
2988 done:
2989         rtnl_unlock();
2990         return err;
2991 }
2992
2993 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2994 {
2995         struct packet_mclist *ml, **mlp;
2996
2997         rtnl_lock();
2998
2999         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3000                 if (ml->ifindex == mreq->mr_ifindex &&
3001                     ml->type == mreq->mr_type &&
3002                     ml->alen == mreq->mr_alen &&
3003                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3004                         if (--ml->count == 0) {
3005                                 struct net_device *dev;
3006                                 *mlp = ml->next;
3007                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3008                                 if (dev)
3009                                         packet_dev_mc(dev, ml, -1);
3010                                 kfree(ml);
3011                         }
3012                         rtnl_unlock();
3013                         return 0;
3014                 }
3015         }
3016         rtnl_unlock();
3017         return -EADDRNOTAVAIL;
3018 }
3019
3020 static void packet_flush_mclist(struct sock *sk)
3021 {
3022         struct packet_sock *po = pkt_sk(sk);
3023         struct packet_mclist *ml;
3024
3025         if (!po->mclist)
3026                 return;
3027
3028         rtnl_lock();
3029         while ((ml = po->mclist) != NULL) {
3030                 struct net_device *dev;
3031
3032                 po->mclist = ml->next;
3033                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3034                 if (dev != NULL)
3035                         packet_dev_mc(dev, ml, -1);
3036                 kfree(ml);
3037         }
3038         rtnl_unlock();
3039 }
3040
3041 static int
3042 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3043 {
3044         struct sock *sk = sock->sk;
3045         struct packet_sock *po = pkt_sk(sk);
3046         int ret;
3047
3048         if (level != SOL_PACKET)
3049                 return -ENOPROTOOPT;
3050
3051         switch (optname) {
3052         case PACKET_ADD_MEMBERSHIP:
3053         case PACKET_DROP_MEMBERSHIP:
3054         {
3055                 struct packet_mreq_max mreq;
3056                 int len = optlen;
3057                 memset(&mreq, 0, sizeof(mreq));
3058                 if (len < sizeof(struct packet_mreq))
3059                         return -EINVAL;
3060                 if (len > sizeof(mreq))
3061                         len = sizeof(mreq);
3062                 if (copy_from_user(&mreq, optval, len))
3063                         return -EFAULT;
3064                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3065                         return -EINVAL;
3066                 if (optname == PACKET_ADD_MEMBERSHIP)
3067                         ret = packet_mc_add(sk, &mreq);
3068                 else
3069                         ret = packet_mc_drop(sk, &mreq);
3070                 return ret;
3071         }
3072
3073         case PACKET_RX_RING:
3074         case PACKET_TX_RING:
3075         {
3076                 union tpacket_req_u req_u;
3077                 int len;
3078
3079                 switch (po->tp_version) {
3080                 case TPACKET_V1:
3081                 case TPACKET_V2:
3082                         len = sizeof(req_u.req);
3083                         break;
3084                 case TPACKET_V3:
3085                 default:
3086                         len = sizeof(req_u.req3);
3087                         break;
3088                 }
3089                 if (optlen < len)
3090                         return -EINVAL;
3091                 if (pkt_sk(sk)->has_vnet_hdr)
3092                         return -EINVAL;
3093                 if (copy_from_user(&req_u.req, optval, len))
3094                         return -EFAULT;
3095                 return packet_set_ring(sk, &req_u, 0,
3096                         optname == PACKET_TX_RING);
3097         }
3098         case PACKET_COPY_THRESH:
3099         {
3100                 int val;
3101
3102                 if (optlen != sizeof(val))
3103                         return -EINVAL;
3104                 if (copy_from_user(&val, optval, sizeof(val)))
3105                         return -EFAULT;
3106
3107                 pkt_sk(sk)->copy_thresh = val;
3108                 return 0;
3109         }
3110         case PACKET_VERSION:
3111         {
3112                 int val;
3113
3114                 if (optlen != sizeof(val))
3115                         return -EINVAL;
3116                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3117                         return -EBUSY;
3118                 if (copy_from_user(&val, optval, sizeof(val)))
3119                         return -EFAULT;
3120                 switch (val) {
3121                 case TPACKET_V1:
3122                 case TPACKET_V2:
3123                 case TPACKET_V3:
3124                         po->tp_version = val;
3125                         return 0;
3126                 default:
3127                         return -EINVAL;
3128                 }
3129         }
3130         case PACKET_RESERVE:
3131         {
3132                 unsigned int val;
3133
3134                 if (optlen != sizeof(val))
3135                         return -EINVAL;
3136                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3137                         return -EBUSY;
3138                 if (copy_from_user(&val, optval, sizeof(val)))
3139                         return -EFAULT;
3140                 po->tp_reserve = val;
3141                 return 0;
3142         }
3143         case PACKET_LOSS:
3144         {
3145                 unsigned int val;
3146
3147                 if (optlen != sizeof(val))
3148                         return -EINVAL;
3149                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3150                         return -EBUSY;
3151                 if (copy_from_user(&val, optval, sizeof(val)))
3152                         return -EFAULT;
3153                 po->tp_loss = !!val;
3154                 return 0;
3155         }
3156         case PACKET_AUXDATA:
3157         {
3158                 int val;
3159
3160                 if (optlen < sizeof(val))
3161                         return -EINVAL;
3162                 if (copy_from_user(&val, optval, sizeof(val)))
3163                         return -EFAULT;
3164
3165                 po->auxdata = !!val;
3166                 return 0;
3167         }
3168         case PACKET_ORIGDEV:
3169         {
3170                 int val;
3171
3172                 if (optlen < sizeof(val))
3173                         return -EINVAL;
3174                 if (copy_from_user(&val, optval, sizeof(val)))
3175                         return -EFAULT;
3176
3177                 po->origdev = !!val;
3178                 return 0;
3179         }
3180         case PACKET_VNET_HDR:
3181         {
3182                 int val;
3183
3184                 if (sock->type != SOCK_RAW)
3185                         return -EINVAL;
3186                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3187                         return -EBUSY;
3188                 if (optlen < sizeof(val))
3189                         return -EINVAL;
3190                 if (copy_from_user(&val, optval, sizeof(val)))
3191                         return -EFAULT;
3192
3193                 po->has_vnet_hdr = !!val;
3194                 return 0;
3195         }
3196         case PACKET_TIMESTAMP:
3197         {
3198                 int val;
3199
3200                 if (optlen != sizeof(val))
3201                         return -EINVAL;
3202                 if (copy_from_user(&val, optval, sizeof(val)))
3203                         return -EFAULT;
3204
3205                 po->tp_tstamp = val;
3206                 return 0;
3207         }
3208         case PACKET_FANOUT:
3209         {
3210                 int val;
3211
3212                 if (optlen != sizeof(val))
3213                         return -EINVAL;
3214                 if (copy_from_user(&val, optval, sizeof(val)))
3215                         return -EFAULT;
3216
3217                 return fanout_add(sk, val & 0xffff, val >> 16);
3218         }
3219         default:
3220                 return -ENOPROTOOPT;
3221         }
3222 }
3223
3224 static int packet_getsockopt(struct socket *sock, int level, int optname,
3225                              char __user *optval, int __user *optlen)
3226 {
3227         int len;
3228         int val, lv = sizeof(val);
3229         struct sock *sk = sock->sk;
3230         struct packet_sock *po = pkt_sk(sk);
3231         void *data = &val;
3232         struct tpacket_stats st;
3233         union tpacket_stats_u st_u;
3234
3235         if (level != SOL_PACKET)
3236                 return -ENOPROTOOPT;
3237
3238         if (get_user(len, optlen))
3239                 return -EFAULT;
3240
3241         if (len < 0)
3242                 return -EINVAL;
3243
3244         switch (optname) {
3245         case PACKET_STATISTICS:
3246                 spin_lock_bh(&sk->sk_receive_queue.lock);
3247                 if (po->tp_version == TPACKET_V3) {
3248                         lv = sizeof(struct tpacket_stats_v3);
3249                         memcpy(&st_u.stats3, &po->stats,
3250                                sizeof(struct tpacket_stats));
3251                         st_u.stats3.tp_freeze_q_cnt =
3252                                         po->stats_u.stats3.tp_freeze_q_cnt;
3253                         st_u.stats3.tp_packets += po->stats.tp_drops;
3254                         data = &st_u.stats3;
3255                 } else {
3256                         lv = sizeof(struct tpacket_stats);
3257                         st = po->stats;
3258                         st.tp_packets += st.tp_drops;
3259                         data = &st;
3260                 }
3261                 memset(&po->stats, 0, sizeof(st));
3262                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3263                 break;
3264         case PACKET_AUXDATA:
3265                 val = po->auxdata;
3266                 break;
3267         case PACKET_ORIGDEV:
3268                 val = po->origdev;
3269                 break;
3270         case PACKET_VNET_HDR:
3271                 val = po->has_vnet_hdr;
3272                 break;
3273         case PACKET_VERSION:
3274                 val = po->tp_version;
3275                 break;
3276         case PACKET_HDRLEN:
3277                 if (len > sizeof(int))
3278                         len = sizeof(int);
3279                 if (copy_from_user(&val, optval, len))
3280                         return -EFAULT;
3281                 switch (val) {
3282                 case TPACKET_V1:
3283                         val = sizeof(struct tpacket_hdr);
3284                         break;
3285                 case TPACKET_V2:
3286                         val = sizeof(struct tpacket2_hdr);
3287                         break;
3288                 case TPACKET_V3:
3289                         val = sizeof(struct tpacket3_hdr);
3290                         break;
3291                 default:
3292                         return -EINVAL;
3293                 }
3294                 break;
3295         case PACKET_RESERVE:
3296                 val = po->tp_reserve;
3297                 break;
3298         case PACKET_LOSS:
3299                 val = po->tp_loss;
3300                 break;
3301         case PACKET_TIMESTAMP:
3302                 val = po->tp_tstamp;
3303                 break;
3304         case PACKET_FANOUT:
3305                 val = (po->fanout ?
3306                        ((u32)po->fanout->id |
3307                         ((u32)po->fanout->type << 16)) :
3308                        0);
3309                 break;
3310         default:
3311                 return -ENOPROTOOPT;
3312         }
3313
3314         if (len > lv)
3315                 len = lv;
3316         if (put_user(len, optlen))
3317                 return -EFAULT;
3318         if (copy_to_user(optval, data, len))
3319                 return -EFAULT;
3320         return 0;
3321 }
3322
3323
3324 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3325 {
3326         struct sock *sk;
3327         struct hlist_node *node;
3328         struct net_device *dev = data;
3329         struct net *net = dev_net(dev);
3330
3331         rcu_read_lock();
3332         sk_for_each_rcu(sk, node, &net->packet.sklist) {
3333                 struct packet_sock *po = pkt_sk(sk);
3334
3335                 switch (msg) {
3336                 case NETDEV_UNREGISTER:
3337                         if (po->mclist)
3338                                 packet_dev_mclist(dev, po->mclist, -1);
3339                         /* fallthrough */
3340
3341                 case NETDEV_DOWN:
3342                         if (dev->ifindex == po->ifindex) {
3343                                 spin_lock(&po->bind_lock);
3344                                 if (po->running) {
3345                                         __unregister_prot_hook(sk, false);
3346                                         sk->sk_err = ENETDOWN;
3347                                         if (!sock_flag(sk, SOCK_DEAD))
3348                                                 sk->sk_error_report(sk);
3349                                 }
3350                                 if (msg == NETDEV_UNREGISTER) {
3351                                         po->ifindex = -1;
3352                                         if (po->prot_hook.dev)
3353                                                 dev_put(po->prot_hook.dev);
3354                                         po->prot_hook.dev = NULL;
3355                                 }
3356                                 spin_unlock(&po->bind_lock);
3357                         }
3358                         break;
3359                 case NETDEV_UP:
3360                         if (dev->ifindex == po->ifindex) {
3361                                 spin_lock(&po->bind_lock);
3362                                 if (po->num)
3363                                         register_prot_hook(sk);
3364                                 spin_unlock(&po->bind_lock);
3365                         }
3366                         break;
3367                 }
3368         }
3369         rcu_read_unlock();
3370         return NOTIFY_DONE;
3371 }
3372
3373
3374 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3375                         unsigned long arg)
3376 {
3377         struct sock *sk = sock->sk;
3378
3379         switch (cmd) {
3380         case SIOCOUTQ:
3381         {
3382                 int amount = sk_wmem_alloc_get(sk);
3383
3384                 return put_user(amount, (int __user *)arg);
3385         }
3386         case SIOCINQ:
3387         {
3388                 struct sk_buff *skb;
3389                 int amount = 0;
3390
3391                 spin_lock_bh(&sk->sk_receive_queue.lock);
3392                 skb = skb_peek(&sk->sk_receive_queue);
3393                 if (skb)
3394                         amount = skb->len;
3395                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3396                 return put_user(amount, (int __user *)arg);
3397         }
3398         case SIOCGSTAMP:
3399                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3400         case SIOCGSTAMPNS:
3401                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3402
3403 #ifdef CONFIG_INET
3404         case SIOCADDRT:
3405         case SIOCDELRT:
3406         case SIOCDARP:
3407         case SIOCGARP:
3408         case SIOCSARP:
3409         case SIOCGIFADDR:
3410         case SIOCSIFADDR:
3411         case SIOCGIFBRDADDR:
3412         case SIOCSIFBRDADDR:
3413         case SIOCGIFNETMASK:
3414         case SIOCSIFNETMASK:
3415         case SIOCGIFDSTADDR:
3416         case SIOCSIFDSTADDR:
3417         case SIOCSIFFLAGS:
3418                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3419 #endif
3420
3421         default:
3422                 return -ENOIOCTLCMD;
3423         }
3424         return 0;
3425 }
3426
3427 static unsigned int packet_poll(struct file *file, struct socket *sock,
3428                                 poll_table *wait)
3429 {
3430         struct sock *sk = sock->sk;
3431         struct packet_sock *po = pkt_sk(sk);
3432         unsigned int mask = datagram_poll(file, sock, wait);
3433
3434         spin_lock_bh(&sk->sk_receive_queue.lock);
3435         if (po->rx_ring.pg_vec) {
3436                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3437                         TP_STATUS_KERNEL))
3438                         mask |= POLLIN | POLLRDNORM;
3439         }
3440         spin_unlock_bh(&sk->sk_receive_queue.lock);
3441         spin_lock_bh(&sk->sk_write_queue.lock);
3442         if (po->tx_ring.pg_vec) {
3443                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3444                         mask |= POLLOUT | POLLWRNORM;
3445         }
3446         spin_unlock_bh(&sk->sk_write_queue.lock);
3447         return mask;
3448 }
3449
3450
3451 /* Dirty? Well, I still did not learn better way to account
3452  * for user mmaps.
3453  */
3454
3455 static void packet_mm_open(struct vm_area_struct *vma)
3456 {
3457         struct file *file = vma->vm_file;
3458         struct socket *sock = file->private_data;
3459         struct sock *sk = sock->sk;
3460
3461         if (sk)
3462                 atomic_inc(&pkt_sk(sk)->mapped);
3463 }
3464
3465 static void packet_mm_close(struct vm_area_struct *vma)
3466 {
3467         struct file *file = vma->vm_file;
3468         struct socket *sock = file->private_data;
3469         struct sock *sk = sock->sk;
3470
3471         if (sk)
3472                 atomic_dec(&pkt_sk(sk)->mapped);
3473 }
3474
3475 static const struct vm_operations_struct packet_mmap_ops = {
3476         .open   =       packet_mm_open,
3477         .close  =       packet_mm_close,
3478 };
3479
3480 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3481                         unsigned int len)
3482 {
3483         int i;
3484
3485         for (i = 0; i < len; i++) {
3486                 if (likely(pg_vec[i].buffer)) {
3487                         if (is_vmalloc_addr(pg_vec[i].buffer))
3488                                 vfree(pg_vec[i].buffer);
3489                         else
3490                                 free_pages((unsigned long)pg_vec[i].buffer,
3491                                            order);
3492                         pg_vec[i].buffer = NULL;
3493                 }
3494         }
3495         kfree(pg_vec);
3496 }
3497
3498 static char *alloc_one_pg_vec_page(unsigned long order)
3499 {
3500         char *buffer = NULL;
3501         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3502                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3503
3504         buffer = (char *) __get_free_pages(gfp_flags, order);
3505
3506         if (buffer)
3507                 return buffer;
3508
3509         /*
3510          * __get_free_pages failed, fall back to vmalloc
3511          */
3512         buffer = vzalloc((1 << order) * PAGE_SIZE);
3513
3514         if (buffer)
3515                 return buffer;
3516
3517         /*
3518          * vmalloc failed, lets dig into swap here
3519          */
3520         gfp_flags &= ~__GFP_NORETRY;
3521         buffer = (char *)__get_free_pages(gfp_flags, order);
3522         if (buffer)
3523                 return buffer;
3524
3525         /*
3526          * complete and utter failure
3527          */
3528         return NULL;
3529 }
3530
3531 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3532 {
3533         unsigned int block_nr = req->tp_block_nr;
3534         struct pgv *pg_vec;
3535         int i;
3536
3537         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3538         if (unlikely(!pg_vec))
3539                 goto out;
3540
3541         for (i = 0; i < block_nr; i++) {
3542                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3543                 if (unlikely(!pg_vec[i].buffer))
3544                         goto out_free_pgvec;
3545         }
3546
3547 out:
3548         return pg_vec;
3549
3550 out_free_pgvec:
3551         free_pg_vec(pg_vec, order, block_nr);
3552         pg_vec = NULL;
3553         goto out;
3554 }
3555
3556 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3557                 int closing, int tx_ring)
3558 {
3559         struct pgv *pg_vec = NULL;
3560         struct packet_sock *po = pkt_sk(sk);
3561         int was_running, order = 0;
3562         struct packet_ring_buffer *rb;
3563         struct sk_buff_head *rb_queue;
3564         __be16 num;
3565         int err = -EINVAL;
3566         /* Added to avoid minimal code churn */
3567         struct tpacket_req *req = &req_u->req;
3568
3569         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3570         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3571                 WARN(1, "Tx-ring is not supported.\n");
3572                 goto out;
3573         }
3574
3575         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3576         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3577
3578         err = -EBUSY;
3579         if (!closing) {
3580                 if (atomic_read(&po->mapped))
3581                         goto out;
3582                 if (atomic_read(&rb->pending))
3583                         goto out;
3584         }
3585
3586         if (req->tp_block_nr) {
3587                 /* Sanity tests and some calculations */
3588                 err = -EBUSY;
3589                 if (unlikely(rb->pg_vec))
3590                         goto out;
3591
3592                 switch (po->tp_version) {
3593                 case TPACKET_V1:
3594                         po->tp_hdrlen = TPACKET_HDRLEN;
3595                         break;
3596                 case TPACKET_V2:
3597                         po->tp_hdrlen = TPACKET2_HDRLEN;
3598                         break;
3599                 case TPACKET_V3:
3600                         po->tp_hdrlen = TPACKET3_HDRLEN;
3601                         break;
3602                 }
3603
3604                 err = -EINVAL;
3605                 if (unlikely((int)req->tp_block_size <= 0))
3606                         goto out;
3607                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3608                         goto out;
3609                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3610                                         po->tp_reserve))
3611                         goto out;
3612                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3613                         goto out;
3614
3615                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3616                 if (unlikely(rb->frames_per_block <= 0))
3617                         goto out;
3618                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3619                                         req->tp_frame_nr))
3620                         goto out;
3621
3622                 err = -ENOMEM;
3623                 order = get_order(req->tp_block_size);
3624                 pg_vec = alloc_pg_vec(req, order);
3625                 if (unlikely(!pg_vec))
3626                         goto out;
3627                 switch (po->tp_version) {
3628                 case TPACKET_V3:
3629                 /* Transmit path is not supported. We checked
3630                  * it above but just being paranoid
3631                  */
3632                         if (!tx_ring)
3633                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3634                                 break;
3635                 default:
3636                         break;
3637                 }
3638         }
3639         /* Done */
3640         else {
3641                 err = -EINVAL;
3642                 if (unlikely(req->tp_frame_nr))
3643                         goto out;
3644         }
3645
3646         lock_sock(sk);
3647
3648         /* Detach socket from network */
3649         spin_lock(&po->bind_lock);
3650         was_running = po->running;
3651         num = po->num;
3652         if (was_running) {
3653                 po->num = 0;
3654                 __unregister_prot_hook(sk, false);
3655         }
3656         spin_unlock(&po->bind_lock);
3657
3658         synchronize_net();
3659
3660         err = -EBUSY;
3661         mutex_lock(&po->pg_vec_lock);
3662         if (closing || atomic_read(&po->mapped) == 0) {
3663                 err = 0;
3664                 spin_lock_bh(&rb_queue->lock);
3665                 swap(rb->pg_vec, pg_vec);
3666                 rb->frame_max = (req->tp_frame_nr - 1);
3667                 rb->head = 0;
3668                 rb->frame_size = req->tp_frame_size;
3669                 spin_unlock_bh(&rb_queue->lock);
3670
3671                 swap(rb->pg_vec_order, order);
3672                 swap(rb->pg_vec_len, req->tp_block_nr);
3673
3674                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3675                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3676                                                 tpacket_rcv : packet_rcv;
3677                 skb_queue_purge(rb_queue);
3678                 if (atomic_read(&po->mapped))
3679                         pr_err("packet_mmap: vma is busy: %d\n",
3680                                atomic_read(&po->mapped));
3681         }
3682         mutex_unlock(&po->pg_vec_lock);
3683
3684         spin_lock(&po->bind_lock);
3685         if (was_running) {
3686                 po->num = num;
3687                 register_prot_hook(sk);
3688         }
3689         spin_unlock(&po->bind_lock);
3690         if (closing && (po->tp_version > TPACKET_V2)) {
3691                 /* Because we don't support block-based V3 on tx-ring */
3692                 if (!tx_ring)
3693                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3694         }
3695         release_sock(sk);
3696
3697         if (pg_vec)
3698                 free_pg_vec(pg_vec, order, req->tp_block_nr);
3699 out:
3700         return err;
3701 }
3702
3703 static int packet_mmap(struct file *file, struct socket *sock,
3704                 struct vm_area_struct *vma)
3705 {
3706         struct sock *sk = sock->sk;
3707         struct packet_sock *po = pkt_sk(sk);
3708         unsigned long size, expected_size;
3709         struct packet_ring_buffer *rb;
3710         unsigned long start;
3711         int err = -EINVAL;
3712         int i;
3713
3714         if (vma->vm_pgoff)
3715                 return -EINVAL;
3716
3717         mutex_lock(&po->pg_vec_lock);
3718
3719         expected_size = 0;
3720         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3721                 if (rb->pg_vec) {
3722                         expected_size += rb->pg_vec_len
3723                                                 * rb->pg_vec_pages
3724                                                 * PAGE_SIZE;
3725                 }
3726         }
3727
3728         if (expected_size == 0)
3729                 goto out;
3730
3731         size = vma->vm_end - vma->vm_start;
3732         if (size != expected_size)
3733                 goto out;
3734
3735         start = vma->vm_start;
3736         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3737                 if (rb->pg_vec == NULL)
3738                         continue;
3739
3740                 for (i = 0; i < rb->pg_vec_len; i++) {
3741                         struct page *page;
3742                         void *kaddr = rb->pg_vec[i].buffer;
3743                         int pg_num;
3744
3745                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3746                                 page = pgv_to_page(kaddr);
3747                                 err = vm_insert_page(vma, start, page);
3748                                 if (unlikely(err))
3749                                         goto out;
3750                                 start += PAGE_SIZE;
3751                                 kaddr += PAGE_SIZE;
3752                         }
3753                 }
3754         }
3755
3756         atomic_inc(&po->mapped);
3757         vma->vm_ops = &packet_mmap_ops;
3758         err = 0;
3759
3760 out:
3761         mutex_unlock(&po->pg_vec_lock);
3762         return err;
3763 }
3764
3765 static const struct proto_ops packet_ops_spkt = {
3766         .family =       PF_PACKET,
3767         .owner =        THIS_MODULE,
3768         .release =      packet_release,
3769         .bind =         packet_bind_spkt,
3770         .connect =      sock_no_connect,
3771         .socketpair =   sock_no_socketpair,
3772         .accept =       sock_no_accept,
3773         .getname =      packet_getname_spkt,
3774         .poll =         datagram_poll,
3775         .ioctl =        packet_ioctl,
3776         .listen =       sock_no_listen,
3777         .shutdown =     sock_no_shutdown,
3778         .setsockopt =   sock_no_setsockopt,
3779         .getsockopt =   sock_no_getsockopt,
3780         .sendmsg =      packet_sendmsg_spkt,
3781         .recvmsg =      packet_recvmsg,
3782         .mmap =         sock_no_mmap,
3783         .sendpage =     sock_no_sendpage,
3784 };
3785
3786 static const struct proto_ops packet_ops = {
3787         .family =       PF_PACKET,
3788         .owner =        THIS_MODULE,
3789         .release =      packet_release,
3790         .bind =         packet_bind,
3791         .connect =      sock_no_connect,
3792         .socketpair =   sock_no_socketpair,
3793         .accept =       sock_no_accept,
3794         .getname =      packet_getname,
3795         .poll =         packet_poll,
3796         .ioctl =        packet_ioctl,
3797         .listen =       sock_no_listen,
3798         .shutdown =     sock_no_shutdown,
3799         .setsockopt =   packet_setsockopt,
3800         .getsockopt =   packet_getsockopt,
3801         .sendmsg =      packet_sendmsg,
3802         .recvmsg =      packet_recvmsg,
3803         .mmap =         packet_mmap,
3804         .sendpage =     sock_no_sendpage,
3805 };
3806
3807 static const struct net_proto_family packet_family_ops = {
3808         .family =       PF_PACKET,
3809         .create =       packet_create,
3810         .owner  =       THIS_MODULE,
3811 };
3812
3813 static struct notifier_block packet_netdev_notifier = {
3814         .notifier_call =        packet_notifier,
3815 };
3816
3817 #ifdef CONFIG_PROC_FS
3818
3819 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3820         __acquires(RCU)
3821 {
3822         struct net *net = seq_file_net(seq);
3823
3824         rcu_read_lock();
3825         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3826 }
3827
3828 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3829 {
3830         struct net *net = seq_file_net(seq);
3831         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3832 }
3833
3834 static void packet_seq_stop(struct seq_file *seq, void *v)
3835         __releases(RCU)
3836 {
3837         rcu_read_unlock();
3838 }
3839
3840 static int packet_seq_show(struct seq_file *seq, void *v)
3841 {
3842         if (v == SEQ_START_TOKEN)
3843                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
3844         else {
3845                 struct sock *s = sk_entry(v);
3846                 const struct packet_sock *po = pkt_sk(s);
3847
3848                 seq_printf(seq,
3849                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
3850                            s,
3851                            atomic_read(&s->sk_refcnt),
3852                            s->sk_type,
3853                            ntohs(po->num),
3854                            po->ifindex,
3855                            po->running,
3856                            atomic_read(&s->sk_rmem_alloc),
3857                            sock_i_uid(s),
3858                            sock_i_ino(s));
3859         }
3860
3861         return 0;
3862 }
3863
3864 static const struct seq_operations packet_seq_ops = {
3865         .start  = packet_seq_start,
3866         .next   = packet_seq_next,
3867         .stop   = packet_seq_stop,
3868         .show   = packet_seq_show,
3869 };
3870
3871 static int packet_seq_open(struct inode *inode, struct file *file)
3872 {
3873         return seq_open_net(inode, file, &packet_seq_ops,
3874                             sizeof(struct seq_net_private));
3875 }
3876
3877 static const struct file_operations packet_seq_fops = {
3878         .owner          = THIS_MODULE,
3879         .open           = packet_seq_open,
3880         .read           = seq_read,
3881         .llseek         = seq_lseek,
3882         .release        = seq_release_net,
3883 };
3884
3885 #endif
3886
3887 static int __net_init packet_net_init(struct net *net)
3888 {
3889         spin_lock_init(&net->packet.sklist_lock);
3890         INIT_HLIST_HEAD(&net->packet.sklist);
3891
3892         if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
3893                 return -ENOMEM;
3894
3895         return 0;
3896 }
3897
3898 static void __net_exit packet_net_exit(struct net *net)
3899 {
3900         proc_net_remove(net, "packet");
3901 }
3902
3903 static struct pernet_operations packet_net_ops = {
3904         .init = packet_net_init,
3905         .exit = packet_net_exit,
3906 };
3907
3908
3909 static void __exit packet_exit(void)
3910 {
3911         unregister_netdevice_notifier(&packet_netdev_notifier);
3912         unregister_pernet_subsys(&packet_net_ops);
3913         sock_unregister(PF_PACKET);
3914         proto_unregister(&packet_proto);
3915 }
3916
3917 static int __init packet_init(void)
3918 {
3919         int rc = proto_register(&packet_proto, 0);
3920
3921         if (rc != 0)
3922                 goto out;
3923
3924         sock_register(&packet_family_ops);
3925         register_pernet_subsys(&packet_net_ops);
3926         register_netdevice_notifier(&packet_netdev_notifier);
3927 out:
3928         return rc;
3929 }
3930
3931 module_init(packet_init);
3932 module_exit(packet_exit);
3933 MODULE_LICENSE("GPL");
3934 MODULE_ALIAS_NETPROTO(PF_PACKET);