net/packet/af_packet.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              PACKET - implements raw packet sockets.
   7  *
   8  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  11  *
  12  * Fixes:
  13  *              Alan Cox        :       verify_area() now used correctly
  14  *              Alan Cox        :       new skbuff lists, look ma no backlogs!
  15  *              Alan Cox        :       tidied skbuff lists.
  16  *              Alan Cox        :       Now uses generic datagram routines I
  17  *                                      added. Also fixed the peek/read crash
  18  *                                      from all old Linux datagram code.
  19  *              Alan Cox        :       Uses the improved datagram code.
  20  *              Alan Cox        :       Added NULL's for socket options.
  21  *              Alan Cox        :       Re-commented the code.
  22  *              Alan Cox        :       Use new kernel side addressing
  23  *              Rob Janssen     :       Correct MTU usage.
  24  *              Dave Platt      :       Counter leaks caused by incorrect
  25  *                                      interrupt locking and some slightly
  26  *                                      dubious gcc output. Can you read
  27  *                                      compiler: it said _VOLATILE_
  28  *      Richard Kooijman        :       Timestamp fixes.
  29  *              Alan Cox        :       New buffers. Use sk->mac.raw.
  30  *              Alan Cox        :       sendmsg/recvmsg support.
  31  *              Alan Cox        :       Protocol setting support
  32  *      Alexey Kuznetsov        :       Untied from IPv4 stack.
  33  *      Cyrus Durgin            :       Fixed kerneld for kmod.
  34  *      Michal Ostrowski        :       Module initialization cleanup.
  35  *         Ulises Alonso        :       Frame number limit removal and
  36  *                                      packet_set_ring memory leak.
  37  *              Eric Biederman  :       Allow for > 8 byte hardware addresses.
  38  *                                      The convention is that longer addresses
  39  *                                      will simply extend the hardware address
  40  *                                      byte arrays at the end of sockaddr_ll
  41  *                                      and packet_mreq.
  42  *              Johann Baudy    :       Added TX RING.
  43  *              Chetan Loke     :       Implemented TPACKET_V3 block abstraction
  44  *                                      layer.
  45  *                                      Copyright (C) 2011, <lokec@ccs.neu.edu>
  46  *
  47  *
  48  *              This program is free software; you can redistribute it and/or
  49  *              modify it under the terms of the GNU General Public License
  50  *              as published by the Free Software Foundation; either version
  51  *              2 of the License, or (at your option) any later version.
  52  *
  53  */
  54
  55 #include <linux/types.h>
  56 #include <linux/mm.h>
  57 #include <linux/capability.h>
  58 #include <linux/fcntl.h>
  59 #include <linux/socket.h>
  60 #include <linux/in.h>
  61 #include <linux/inet.h>
  62 #include <linux/netdevice.h>
  63 #include <linux/if_packet.h>
  64 #include <linux/wireless.h>
  65 #include <linux/kernel.h>
  66 #include <linux/kmod.h>
  67 #include <linux/slab.h>
  68 #include <linux/vmalloc.h>
  69 #include <net/net_namespace.h>
  70 #include <net/ip.h>
  71 #include <net/protocol.h>
  72 #include <linux/skbuff.h>
  73 #include <net/sock.h>
  74 #include <linux/errno.h>
  75 #include <linux/timer.h>
  76 #include <asm/system.h>
  77 #include <asm/uaccess.h>
  78 #include <asm/ioctls.h>
  79 #include <asm/page.h>
  80 #include <asm/cacheflush.h>
  81 #include <asm/io.h>
  82 #include <linux/proc_fs.h>
  83 #include <linux/seq_file.h>
  84 #include <linux/poll.h>
  85 #include <linux/module.h>
  86 #include <linux/init.h>
  87 #include <linux/mutex.h>
  88 #include <linux/if_vlan.h>
  89 #include <linux/virtio_net.h>
  90 #include <linux/errqueue.h>
  91 #include <linux/net_tstamp.h>
  92
  93 #ifdef CONFIG_INET
  94 #include <net/inet_common.h>
  95 #endif
  96
  97 /*
  98    Assumptions:
  99    - if device has no dev->hard_header routine, it adds and removes ll header
 100      inside itself. In this case ll header is invisible outside of device,
 101      but higher levels still should reserve dev->hard_header_len.
 102      Some devices are enough clever to reallocate skb, when header
 103      will not fit to reserved space (tunnel), another ones are silly
 104      (PPP).
 105    - packet socket receives packets with pulled ll header,
 106      so that SOCK_RAW should push it back.
 107
 108 On receive:
 109 -----------
 110
 111 Incoming, dev->hard_header!=NULL
 112    mac_header -> ll header
 113    data       -> data
 114
 115 Outgoing, dev->hard_header!=NULL
 116    mac_header -> ll header
 117    data       -> ll header
 118
 119 Incoming, dev->hard_header==NULL
 120    mac_header -> UNKNOWN position. It is very likely, that it points to ll
 121                  header.  PPP makes it, that is wrong, because introduce
 122                  assymetry between rx and tx paths.
 123    data       -> data
 124
 125 Outgoing, dev->hard_header==NULL
 126    mac_header -> data. ll header is still not built!
 127    data       -> data
 128
 129 Resume
 130   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 131
 132
 133 On transmit:
 134 ------------
 135
 136 dev->hard_header != NULL
 137    mac_header -> ll header
 138    data       -> ll header
 139
 140 dev->hard_header == NULL (ll header is added by device, we cannot control it)
 141    mac_header -> data
 142    data       -> data
 143
 144    We should set nh.raw on output to correct posistion,
 145    packet classifier depends on it.
 146  */
 147
 148 /* Private packet socket structures. */
 149
 150 struct packet_mclist {
 151         struct packet_mclist    *next;
 152         int                     ifindex;
 153         int                     count;
 154         unsigned short          type;
 155         unsigned short          alen;
 156         unsigned char           addr[MAX_ADDR_LEN];
 157 };
 158 /* identical to struct packet_mreq except it has
 159  * a longer address field.
 160  */
 161 struct packet_mreq_max {
 162         int             mr_ifindex;
 163         unsigned short  mr_type;
 164         unsigned short  mr_alen;
 165         unsigned char   mr_address[MAX_ADDR_LEN];
 166 };
 167
 168 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 169                 int closing, int tx_ring);
 170
 171
 172 #define V3_ALIGNMENT    (8)
 173
 174 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 175
 176 #define BLK_PLUS_PRIV(sz_of_priv) \
 177         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 178
 179 /* kbdq - kernel block descriptor queue */
 180 struct tpacket_kbdq_core {
 181         struct pgv      *pkbdq;
 182         unsigned int    feature_req_word;
 183         unsigned int    hdrlen;
 184         unsigned char   reset_pending_on_curr_blk;
 185         unsigned char   delete_blk_timer;
 186         unsigned short  kactive_blk_num;
 187         unsigned short  blk_sizeof_priv;
 188
 189         /* last_kactive_blk_num:
 190          * trick to see if user-space has caught up
 191          * in order to avoid refreshing timer when every single pkt arrives.
 192          */
 193         unsigned short  last_kactive_blk_num;
 194
 195         char            *pkblk_start;
 196         char            *pkblk_end;
 197         int             kblk_size;
 198         unsigned int    knum_blocks;
 199         uint64_t        knxt_seq_num;
 200         char            *prev;
 201         char            *nxt_offset;
 202         struct sk_buff  *skb;
 203
 204         atomic_t        blk_fill_in_prog;
 205
 206         /* Default is set to 8ms */
 207 #define DEFAULT_PRB_RETIRE_TOV  (8)
 208
 209         unsigned short  retire_blk_tov;
 210         unsigned short  version;
 211         unsigned long   tov_in_jiffies;
 212
 213         /* timer to retire an outstanding block */
 214         struct timer_list retire_blk_timer;
 215 };
 216
 217 #define PGV_FROM_VMALLOC 1
 218 struct pgv {
 219         char *buffer;
 220 };
 221
 222 struct packet_ring_buffer {
 223         struct pgv              *pg_vec;
 224         unsigned int            head;
 225         unsigned int            frames_per_block;
 226         unsigned int            frame_size;
 227         unsigned int            frame_max;
 228
 229         unsigned int            pg_vec_order;
 230         unsigned int            pg_vec_pages;
 231         unsigned int            pg_vec_len;
 232
 233         struct tpacket_kbdq_core        prb_bdqc;
 234         atomic_t                pending;
 235 };
 236
 237 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
 238 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
 239 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
 240 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
 241 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
 242 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
 243 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
 244
 245 struct packet_sock;
 246 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
 247
 248 static void *packet_previous_frame(struct packet_sock *po,
 249                 struct packet_ring_buffer *rb,
 250                 int status);
 251 static void packet_increment_head(struct packet_ring_buffer *buff);
 252 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
 253                         struct tpacket_block_desc *);
 254 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 255                         struct packet_sock *);
 256 static void prb_retire_current_block(struct tpacket_kbdq_core *,
 257                 struct packet_sock *, unsigned int status);
 258 static int prb_queue_frozen(struct tpacket_kbdq_core *);
 259 static void prb_open_block(struct tpacket_kbdq_core *,
 260                 struct tpacket_block_desc *);
 261 static void prb_retire_rx_blk_timer_expired(unsigned long);
 262 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 263 static void prb_init_blk_timer(struct packet_sock *,
 264                 struct tpacket_kbdq_core *,
 265                 void (*func) (unsigned long));
 266 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 267 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 268                 struct tpacket3_hdr *);
 269 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 270                 struct tpacket3_hdr *);
 271 static void packet_flush_mclist(struct sock *sk);
 272
 273 struct packet_fanout;
 274 struct packet_sock {
 275         /* struct sock has to be the first member of packet_sock */
 276         struct sock             sk;
 277         struct packet_fanout    *fanout;
 278         struct tpacket_stats    stats;
 279         union  tpacket_stats_u  stats_u;
 280         struct packet_ring_buffer       rx_ring;
 281         struct packet_ring_buffer       tx_ring;
 282         int                     copy_thresh;
 283         spinlock_t              bind_lock;
 284         struct mutex            pg_vec_lock;
 285         unsigned int            running:1,      /* prot_hook is attached*/
 286                                 auxdata:1,
 287                                 origdev:1,
 288                                 has_vnet_hdr:1;
 289         int                     ifindex;        /* bound device         */
 290         __be16                  num;
 291         struct packet_mclist    *mclist;
 292         atomic_t                mapped;
 293         enum tpacket_versions   tp_version;
 294         unsigned int            tp_hdrlen;
 295         unsigned int            tp_reserve;
 296         unsigned int            tp_loss:1;
 297         unsigned int            tp_tstamp;
 298         struct packet_type      prot_hook ____cacheline_aligned_in_smp;
 299 };
 300
 301 #define PACKET_FANOUT_MAX       256
 302
 303 struct packet_fanout {
 304 #ifdef CONFIG_NET_NS
 305         struct net              *net;
 306 #endif
 307         unsigned int            num_members;
 308         u16                     id;
 309         u8                      type;
 310         u8                      defrag;
 311         atomic_t                rr_cur;
 312         struct list_head        list;
 313         struct sock             *arr[PACKET_FANOUT_MAX];
 314         spinlock_t              lock;
 315         atomic_t                sk_ref;
 316         struct packet_type      prot_hook ____cacheline_aligned_in_smp;
 317 };
 318
 319 struct packet_skb_cb {
 320         unsigned int origlen;
 321         union {
 322                 struct sockaddr_pkt pkt;
 323                 struct sockaddr_ll ll;
 324         } sa;
 325 };
 326
 327 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 328
 329 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 330 #define GET_PBLOCK_DESC(x, bid) \
 331         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 332 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 333         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 334 #define GET_NEXT_PRB_BLK_NUM(x) \
 335         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 336         ((x)->kactive_blk_num+1) : 0)
 337
 338 static struct packet_sock *pkt_sk(struct sock *sk)
 339 {
 340         return (struct packet_sock *)sk;
 341 }
 342
 343 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 344 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 345
 346 /* register_prot_hook must be invoked with the po->bind_lock held,
 347  * or from a context in which asynchronous accesses to the packet
 348  * socket is not possible (packet_create()).
 349  */
 350 static void register_prot_hook(struct sock *sk)
 351 {
 352         struct packet_sock *po = pkt_sk(sk);
 353         if (!po->running) {
 354                 if (po->fanout)
 355                         __fanout_link(sk, po);
 356                 else
 357                         dev_add_pack(&po->prot_hook);
 358                 sock_hold(sk);
 359                 po->running = 1;
 360         }
 361 }
 362
 363 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
 364  * held.   If the sync parameter is true, we will temporarily drop
 365  * the po->bind_lock and do a synchronize_net to make sure no
 366  * asynchronous packet processing paths still refer to the elements
 367  * of po->prot_hook.  If the sync parameter is false, it is the
 368  * callers responsibility to take care of this.
 369  */
 370 static void __unregister_prot_hook(struct sock *sk, bool sync)
 371 {
 372         struct packet_sock *po = pkt_sk(sk);
 373
 374         po->running = 0;
 375         if (po->fanout)
 376                 __fanout_unlink(sk, po);
 377         else
 378                 __dev_remove_pack(&po->prot_hook);
 379         __sock_put(sk);
 380
 381         if (sync) {
 382                 spin_unlock(&po->bind_lock);
 383                 synchronize_net();
 384                 spin_lock(&po->bind_lock);
 385         }
 386 }
 387
 388 static void unregister_prot_hook(struct sock *sk, bool sync)
 389 {
 390         struct packet_sock *po = pkt_sk(sk);
 391
 392         if (po->running)
 393                 __unregister_prot_hook(sk, sync);
 394 }
 395
 396 static inline __pure struct page *pgv_to_page(void *addr)
 397 {
 398         if (is_vmalloc_addr(addr))
 399                 return vmalloc_to_page(addr);
 400         return virt_to_page(addr);
 401 }
 402
 403 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 404 {
 405         union {
 406                 struct tpacket_hdr *h1;
 407                 struct tpacket2_hdr *h2;
 408                 void *raw;
 409         } h;
 410
 411         h.raw = frame;
 412         switch (po->tp_version) {
 413         case TPACKET_V1:
 414                 h.h1->tp_status = status;
 415                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 416                 break;
 417         case TPACKET_V2:
 418                 h.h2->tp_status = status;
 419                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 420                 break;
 421         case TPACKET_V3:
 422         default:
 423                 WARN(1, "TPACKET version not supported.\n");
 424                 BUG();
 425         }
 426
 427         smp_wmb();
 428 }
 429
 430 static int __packet_get_status(struct packet_sock *po, void *frame)
 431 {
 432         union {
 433                 struct tpacket_hdr *h1;
 434                 struct tpacket2_hdr *h2;
 435                 void *raw;
 436         } h;
 437
 438         smp_rmb();
 439
 440         h.raw = frame;
 441         switch (po->tp_version) {
 442         case TPACKET_V1:
 443                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 444                 return h.h1->tp_status;
 445         case TPACKET_V2:
 446                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 447                 return h.h2->tp_status;
 448         case TPACKET_V3:
 449         default:
 450                 WARN(1, "TPACKET version not supported.\n");
 451                 BUG();
 452                 return 0;
 453         }
 454 }
 455
 456 static void *packet_lookup_frame(struct packet_sock *po,
 457                 struct packet_ring_buffer *rb,
 458                 unsigned int position,
 459                 int status)
 460 {
 461         unsigned int pg_vec_pos, frame_offset;
 462         union {
 463                 struct tpacket_hdr *h1;
 464                 struct tpacket2_hdr *h2;
 465                 void *raw;
 466         } h;
 467
 468         pg_vec_pos = position / rb->frames_per_block;
 469         frame_offset = position % rb->frames_per_block;
 470
 471         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 472                 (frame_offset * rb->frame_size);
 473
 474         if (status != __packet_get_status(po, h.raw))
 475                 return NULL;
 476
 477         return h.raw;
 478 }
 479
 480 static void *packet_current_frame(struct packet_sock *po,
 481                 struct packet_ring_buffer *rb,
 482                 int status)
 483 {
 484         return packet_lookup_frame(po, rb, rb->head, status);
 485 }
 486
 487 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 488 {
 489         del_timer_sync(&pkc->retire_blk_timer);
 490 }
 491
 492 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 493                 int tx_ring,
 494                 struct sk_buff_head *rb_queue)
 495 {
 496         struct tpacket_kbdq_core *pkc;
 497
 498         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 499
 500         spin_lock(&rb_queue->lock);
 501         pkc->delete_blk_timer = 1;
 502         spin_unlock(&rb_queue->lock);
 503
 504         prb_del_retire_blk_timer(pkc);
 505 }
 506
 507 static void prb_init_blk_timer(struct packet_sock *po,
 508                 struct tpacket_kbdq_core *pkc,
 509                 void (*func) (unsigned long))
 510 {
 511         init_timer(&pkc->retire_blk_timer);
 512         pkc->retire_blk_timer.data = (long)po;
 513         pkc->retire_blk_timer.function = func;
 514         pkc->retire_blk_timer.expires = jiffies;
 515 }
 516
 517 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
 518 {
 519         struct tpacket_kbdq_core *pkc;
 520
 521         if (tx_ring)
 522                 BUG();
 523
 524         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 525         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
 526 }
 527
 528 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 529                                 int blk_size_in_bytes)
 530 {
 531         struct net_device *dev;
 532         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 533         struct ethtool_cmd ecmd;
 534         int err;
 535
 536         rtnl_lock();
 537         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 538         if (unlikely(!dev)) {
 539                 rtnl_unlock();
 540                 return DEFAULT_PRB_RETIRE_TOV;
 541         }
 542         err = __ethtool_get_settings(dev, &ecmd);
 543         rtnl_unlock();
 544         if (!err) {
 545                 switch (ecmd.speed) {
 546                 case SPEED_10000:
 547                         msec = 1;
 548                         div = 10000/1000;
 549                         break;
 550                 case SPEED_1000:
 551                         msec = 1;
 552                         div = 1000/1000;
 553                         break;
 554                 /*
 555                  * If the link speed is so slow you don't really
 556                  * need to worry about perf anyways
 557                  */
 558                 case SPEED_100:
 559                 case SPEED_10:
 560                 default:
 561                         return DEFAULT_PRB_RETIRE_TOV;
 562                 }
 563         }
 564
 565         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 566
 567         if (div)
 568                 mbits /= div;
 569
 570         tmo = mbits * msec;
 571
 572         if (div)
 573                 return tmo+1;
 574         return tmo;
 575 }
 576
 577 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 578                         union tpacket_req_u *req_u)
 579 {
 580         p1->feature_req_word = req_u->req3.tp_feature_req_word;
 581 }
 582
 583 static void init_prb_bdqc(struct packet_sock *po,
 584                         struct packet_ring_buffer *rb,
 585                         struct pgv *pg_vec,
 586                         union tpacket_req_u *req_u, int tx_ring)
 587 {
 588         struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
 589         struct tpacket_block_desc *pbd;
 590
 591         memset(p1, 0x0, sizeof(*p1));
 592
 593         p1->knxt_seq_num = 1;
 594         p1->pkbdq = pg_vec;
 595         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 596         p1->pkblk_start = (char *)pg_vec[0].buffer;
 597         p1->kblk_size = req_u->req3.tp_block_size;
 598         p1->knum_blocks = req_u->req3.tp_block_nr;
 599         p1->hdrlen = po->tp_hdrlen;
 600         p1->version = po->tp_version;
 601         p1->last_kactive_blk_num = 0;
 602         po->stats_u.stats3.tp_freeze_q_cnt = 0;
 603         if (req_u->req3.tp_retire_blk_tov)
 604                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 605         else
 606                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 607                                                 req_u->req3.tp_block_size);
 608         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 609         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 610
 611         prb_init_ft_ops(p1, req_u);
 612         prb_setup_retire_blk_timer(po, tx_ring);
 613         prb_open_block(p1, pbd);
 614 }
 615
 616 /*  Do NOT update the last_blk_num first.
 617  *  Assumes sk_buff_head lock is held.
 618  */
 619 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 620 {
 621         mod_timer(&pkc->retire_blk_timer,
 622                         jiffies + pkc->tov_in_jiffies);
 623         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 624 }
 625
 626 /*
 627  * Timer logic:
 628  * 1) We refresh the timer only when we open a block.
 629  *    By doing this we don't waste cycles refreshing the timer
 630  *        on packet-by-packet basis.
 631  *
 632  * With a 1MB block-size, on a 1Gbps line, it will take
 633  * i) ~8 ms to fill a block + ii) memcpy etc.
 634  * In this cut we are not accounting for the memcpy time.
 635  *
 636  * So, if the user sets the 'tmo' to 10ms then the timer
 637  * will never fire while the block is still getting filled
 638  * (which is what we want). However, the user could choose
 639  * to close a block early and that's fine.
 640  *
 641  * But when the timer does fire, we check whether or not to refresh it.
 642  * Since the tmo granularity is in msecs, it is not too expensive
 643  * to refresh the timer, lets say every '8' msecs.
 644  * Either the user can set the 'tmo' or we can derive it based on
 645  * a) line-speed and b) block-size.
 646  * prb_calc_retire_blk_tmo() calculates the tmo.
 647  *
 648  */
 649 static void prb_retire_rx_blk_timer_expired(unsigned long data)
 650 {
 651         struct packet_sock *po = (struct packet_sock *)data;
 652         struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
 653         unsigned int frozen;
 654         struct tpacket_block_desc *pbd;
 655
 656         spin_lock(&po->sk.sk_receive_queue.lock);
 657
 658         frozen = prb_queue_frozen(pkc);
 659         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 660
 661         if (unlikely(pkc->delete_blk_timer))
 662                 goto out;
 663
 664         /* We only need to plug the race when the block is partially filled.
 665          * tpacket_rcv:
 666          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 667          *              copy_bits() is in progress ...
 668          *              timer fires on other cpu:
 669          *              we can't retire the current block because copy_bits
 670          *              is in progress.
 671          *
 672          */
 673         if (BLOCK_NUM_PKTS(pbd)) {
 674                 while (atomic_read(&pkc->blk_fill_in_prog)) {
 675                         /* Waiting for skb_copy_bits to finish... */
 676                         cpu_relax();
 677                 }
 678         }
 679
 680         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 681                 if (!frozen) {
 682                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 683                         if (!prb_dispatch_next_block(pkc, po))
 684                                 goto refresh_timer;
 685                         else
 686                                 goto out;
 687                 } else {
 688                         /* Case 1. Queue was frozen because user-space was
 689                          *         lagging behind.
 690                          */
 691                         if (prb_curr_blk_in_use(pkc, pbd)) {
 692                                 /*
 693                                  * Ok, user-space is still behind.
 694                                  * So just refresh the timer.
 695                                  */
 696                                 goto refresh_timer;
 697                         } else {
 698                                /* Case 2. queue was frozen,user-space caught up,
 699                                 * now the link went idle && the timer fired.
 700                                 * We don't have a block to close.So we open this
 701                                 * block and restart the timer.
 702                                 * opening a block thaws the queue,restarts timer
 703                                 * Thawing/timer-refresh is a side effect.
 704                                 */
 705                                 prb_open_block(pkc, pbd);
 706                                 goto out;
 707                         }
 708                 }
 709         }
 710
 711 refresh_timer:
 712         _prb_refresh_rx_retire_blk_timer(pkc);
 713
 714 out:
 715         spin_unlock(&po->sk.sk_receive_queue.lock);
 716 }
 717
 718 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 719                 struct tpacket_block_desc *pbd1, __u32 status)
 720 {
 721         /* Flush everything minus the block header */
 722
 723 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 724         u8 *start, *end;
 725
 726         start = (u8 *)pbd1;
 727
 728         /* Skip the block header(we know header WILL fit in 4K) */
 729         start += PAGE_SIZE;
 730
 731         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 732         for (; start < end; start += PAGE_SIZE)
 733                 flush_dcache_page(pgv_to_page(start));
 734
 735         smp_wmb();
 736 #endif
 737
 738         /* Now update the block status. */
 739
 740         BLOCK_STATUS(pbd1) = status;
 741
 742         /* Flush the block header */
 743
 744 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 745         start = (u8 *)pbd1;
 746         flush_dcache_page(pgv_to_page(start));
 747
 748         smp_wmb();
 749 #endif
 750 }
 751
 752 /*
 753  * Side effect:
 754  *
 755  * 1) flush the block
 756  * 2) Increment active_blk_num
 757  *
 758  * Note:We DONT refresh the timer on purpose.
 759  *      Because almost always the next block will be opened.
 760  */
 761 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 762                 struct tpacket_block_desc *pbd1,
 763                 struct packet_sock *po, unsigned int stat)
 764 {
 765         __u32 status = TP_STATUS_USER | stat;
 766
 767         struct tpacket3_hdr *last_pkt;
 768         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 769
 770         if (po->stats.tp_drops)
 771                 status |= TP_STATUS_LOSING;
 772
 773         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 774         last_pkt->tp_next_offset = 0;
 775
 776         /* Get the ts of the last pkt */
 777         if (BLOCK_NUM_PKTS(pbd1)) {
 778                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 779                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 780         } else {
 781                 /* Ok, we tmo'd - so get the current time */
 782                 struct timespec ts;
 783                 getnstimeofday(&ts);
 784                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 785                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 786         }
 787
 788         smp_wmb();
 789
 790         /* Flush the block */
 791         prb_flush_block(pkc1, pbd1, status);
 792
 793         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 794 }
 795
 796 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 797 {
 798         pkc->reset_pending_on_curr_blk = 0;
 799 }
 800
 801 /*
 802  * Side effect of opening a block:
 803  *
 804  * 1) prb_queue is thawed.
 805  * 2) retire_blk_timer is refreshed.
 806  *
 807  */
 808 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 809         struct tpacket_block_desc *pbd1)
 810 {
 811         struct timespec ts;
 812         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 813
 814         smp_rmb();
 815
 816         /* We could have just memset this but we will lose the
 817          * flexibility of making the priv area sticky
 818          */
 819         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 820         BLOCK_NUM_PKTS(pbd1) = 0;
 821         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 822         getnstimeofday(&ts);
 823         h1->ts_first_pkt.ts_sec = ts.tv_sec;
 824         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 825         pkc1->pkblk_start = (char *)pbd1;
 826         pkc1->nxt_offset = (char *)(pkc1->pkblk_start +
 827                                     BLK_PLUS_PRIV(pkc1->blk_sizeof_priv));
 828         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 829         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 830         pbd1->version = pkc1->version;
 831         pkc1->prev = pkc1->nxt_offset;
 832         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 833         prb_thaw_queue(pkc1);
 834         _prb_refresh_rx_retire_blk_timer(pkc1);
 835
 836         smp_wmb();
 837 }
 838
 839 /*
 840  * Queue freeze logic:
 841  * 1) Assume tp_block_nr = 8 blocks.
 842  * 2) At time 't0', user opens Rx ring.
 843  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 844  * 4) user-space is either sleeping or processing block '0'.
 845  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 846  *    it will close block-7,loop around and try to fill block '0'.
 847  *    call-flow:
 848  *    __packet_lookup_frame_in_block
 849  *      prb_retire_current_block()
 850  *      prb_dispatch_next_block()
 851  *        |->(BLOCK_STATUS == USER) evaluates to true
 852  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 853  * 6) Now there are two cases:
 854  *    6.1) Link goes idle right after the queue is frozen.
 855  *         But remember, the last open_block() refreshed the timer.
 856  *         When this timer expires,it will refresh itself so that we can
 857  *         re-open block-0 in near future.
 858  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 859  *         case and __packet_lookup_frame_in_block will check if block-0
 860  *         is free and can now be re-used.
 861  */
 862 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 863                                   struct packet_sock *po)
 864 {
 865         pkc->reset_pending_on_curr_blk = 1;
 866         po->stats_u.stats3.tp_freeze_q_cnt++;
 867 }
 868
 869 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 870
 871 /*
 872  * If the next block is free then we will dispatch it
 873  * and return a good offset.
 874  * Else, we will freeze the queue.
 875  * So, caller must check the return value.
 876  */
 877 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 878                 struct packet_sock *po)
 879 {
 880         struct tpacket_block_desc *pbd;
 881
 882         smp_rmb();
 883
 884         /* 1. Get current block num */
 885         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 886
 887         /* 2. If this block is currently in_use then freeze the queue */
 888         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 889                 prb_freeze_queue(pkc, po);
 890                 return NULL;
 891         }
 892
 893         /*
 894          * 3.
 895          * open this block and return the offset where the first packet
 896          * needs to get stored.
 897          */
 898         prb_open_block(pkc, pbd);
 899         return (void *)pkc->nxt_offset;
 900 }
 901
 902 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 903                 struct packet_sock *po, unsigned int status)
 904 {
 905         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 906
 907         /* retire/close the current block */
 908         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 909                 /*
 910                  * Plug the case where copy_bits() is in progress on
 911                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 912                  * have space to copy the pkt in the current block and
 913                  * called prb_retire_current_block()
 914                  *
 915                  * We don't need to worry about the TMO case because
 916                  * the timer-handler already handled this case.
 917                  */
 918                 if (!(status & TP_STATUS_BLK_TMO)) {
 919                         while (atomic_read(&pkc->blk_fill_in_prog)) {
 920                                 /* Waiting for skb_copy_bits to finish... */
 921                                 cpu_relax();
 922                         }
 923                 }
 924                 prb_close_block(pkc, pbd, po, status);
 925                 return;
 926         }
 927 }
 928
 929 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
 930                                       struct tpacket_block_desc *pbd)
 931 {
 932         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 933 }
 934
 935 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 936 {
 937         return pkc->reset_pending_on_curr_blk;
 938 }
 939
 940 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 941 {
 942         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 943         atomic_dec(&pkc->blk_fill_in_prog);
 944 }
 945
 946 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 947                         struct tpacket3_hdr *ppd)
 948 {
 949         ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
 950 }
 951
 952 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 953                         struct tpacket3_hdr *ppd)
 954 {
 955         ppd->hv1.tp_rxhash = 0;
 956 }
 957
 958 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
 959                         struct tpacket3_hdr *ppd)
 960 {
 961         if (vlan_tx_tag_present(pkc->skb)) {
 962                 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
 963                 ppd->tp_status = TP_STATUS_VLAN_VALID;
 964         } else {
 965                 ppd->hv1.tp_vlan_tci = ppd->tp_status = 0;
 966         }
 967 }
 968
 969 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
 970                         struct tpacket3_hdr *ppd)
 971 {
 972         prb_fill_vlan_info(pkc, ppd);
 973
 974         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
 975                 prb_fill_rxhash(pkc, ppd);
 976         else
 977                 prb_clear_rxhash(pkc, ppd);
 978 }
 979
 980 static void prb_fill_curr_block(char *curr,
 981                                 struct tpacket_kbdq_core *pkc,
 982                                 struct tpacket_block_desc *pbd,
 983                                 unsigned int len)
 984 {
 985         struct tpacket3_hdr *ppd;
 986
 987         ppd  = (struct tpacket3_hdr *)curr;
 988         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
 989         pkc->prev = curr;
 990         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
 991         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
 992         BLOCK_NUM_PKTS(pbd) += 1;
 993         atomic_inc(&pkc->blk_fill_in_prog);
 994         prb_run_all_ft_ops(pkc, ppd);
 995 }
 996
 997 /* Assumes caller has the sk->rx_queue.lock */
 998 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
 999                                             struct sk_buff *skb,
1000                                                 int status,
1001                                             unsigned int len
1002                                             )
1003 {
1004         struct tpacket_kbdq_core *pkc;
1005         struct tpacket_block_desc *pbd;
1006         char *curr, *end;
1007
1008         pkc = GET_PBDQC_FROM_RB(((struct packet_ring_buffer *)&po->rx_ring));
1009         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1010
1011         /* Queue is frozen when user space is lagging behind */
1012         if (prb_queue_frozen(pkc)) {
1013                 /*
1014                  * Check if that last block which caused the queue to freeze,
1015                  * is still in_use by user-space.
1016                  */
1017                 if (prb_curr_blk_in_use(pkc, pbd)) {
1018                         /* Can't record this packet */
1019                         return NULL;
1020                 } else {
1021                         /*
1022                          * Ok, the block was released by user-space.
1023                          * Now let's open that block.
1024                          * opening a block also thaws the queue.
1025                          * Thawing is a side effect.
1026                          */
1027                         prb_open_block(pkc, pbd);
1028                 }
1029         }
1030
1031         smp_mb();
1032         curr = pkc->nxt_offset;
1033         pkc->skb = skb;
1034         end = (char *) ((char *)pbd + pkc->kblk_size);
1035
1036         /* first try the current block */
1037         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1038                 prb_fill_curr_block(curr, pkc, pbd, len);
1039                 return (void *)curr;
1040         }
1041
1042         /* Ok, close the current block */
1043         prb_retire_current_block(pkc, po, 0);
1044
1045         /* Now, try to dispatch the next block */
1046         curr = (char *)prb_dispatch_next_block(pkc, po);
1047         if (curr) {
1048                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1049                 prb_fill_curr_block(curr, pkc, pbd, len);
1050                 return (void *)curr;
1051         }
1052
1053         /*
1054          * No free blocks are available.user_space hasn't caught up yet.
1055          * Queue was just frozen and now this packet will get dropped.
1056          */
1057         return NULL;
1058 }
1059
1060 static void *packet_current_rx_frame(struct packet_sock *po,
1061                                             struct sk_buff *skb,
1062                                             int status, unsigned int len)
1063 {
1064         char *curr = NULL;
1065         switch (po->tp_version) {
1066         case TPACKET_V1:
1067         case TPACKET_V2:
1068                 curr = packet_lookup_frame(po, &po->rx_ring,
1069                                         po->rx_ring.head, status);
1070                 return curr;
1071         case TPACKET_V3:
1072                 return __packet_lookup_frame_in_block(po, skb, status, len);
1073         default:
1074                 WARN(1, "TPACKET version not supported\n");
1075                 BUG();
1076                 return 0;
1077         }
1078 }
1079
1080 static void *prb_lookup_block(struct packet_sock *po,
1081                                      struct packet_ring_buffer *rb,
1082                                      unsigned int previous,
1083                                      int status)
1084 {
1085         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1086         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, previous);
1087
1088         if (status != BLOCK_STATUS(pbd))
1089                 return NULL;
1090         return pbd;
1091 }
1092
1093 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1094 {
1095         unsigned int prev;
1096         if (rb->prb_bdqc.kactive_blk_num)
1097                 prev = rb->prb_bdqc.kactive_blk_num-1;
1098         else
1099                 prev = rb->prb_bdqc.knum_blocks-1;
1100         return prev;
1101 }
1102
1103 /* Assumes caller has held the rx_queue.lock */
1104 static void *__prb_previous_block(struct packet_sock *po,
1105                                          struct packet_ring_buffer *rb,
1106                                          int status)
1107 {
1108         unsigned int previous = prb_previous_blk_num(rb);
1109         return prb_lookup_block(po, rb, previous, status);
1110 }
1111
1112 static void *packet_previous_rx_frame(struct packet_sock *po,
1113                                              struct packet_ring_buffer *rb,
1114                                              int status)
1115 {
1116         if (po->tp_version <= TPACKET_V2)
1117                 return packet_previous_frame(po, rb, status);
1118
1119         return __prb_previous_block(po, rb, status);
1120 }
1121
1122 static void packet_increment_rx_head(struct packet_sock *po,
1123                                             struct packet_ring_buffer *rb)
1124 {
1125         switch (po->tp_version) {
1126         case TPACKET_V1:
1127         case TPACKET_V2:
1128                 return packet_increment_head(rb);
1129         case TPACKET_V3:
1130         default:
1131                 WARN(1, "TPACKET version not supported.\n");
1132                 BUG();
1133                 return;
1134         }
1135 }
1136
1137 static void *packet_previous_frame(struct packet_sock *po,
1138                 struct packet_ring_buffer *rb,
1139                 int status)
1140 {
1141         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1142         return packet_lookup_frame(po, rb, previous, status);
1143 }
1144
1145 static void packet_increment_head(struct packet_ring_buffer *buff)
1146 {
1147         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1148 }
1149
1150 static void packet_sock_destruct(struct sock *sk)
1151 {
1152         skb_queue_purge(&sk->sk_error_queue);
1153
1154         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1155         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1156
1157         if (!sock_flag(sk, SOCK_DEAD)) {
1158                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1159                 return;
1160         }
1161
1162         sk_refcnt_debug_dec(sk);
1163 }
1164
1165 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1166 {
1167         int x = atomic_read(&f->rr_cur) + 1;
1168
1169         if (x >= num)
1170                 x = 0;
1171
1172         return x;
1173 }
1174
1175 static struct sock *fanout_demux_hash(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1176 {
1177         u32 idx, hash = skb->rxhash;
1178
1179         idx = ((u64)hash * num) >> 32;
1180
1181         return f->arr[idx];
1182 }
1183
1184 static struct sock *fanout_demux_lb(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1185 {
1186         int cur, old;
1187
1188         cur = atomic_read(&f->rr_cur);
1189         while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1190                                      fanout_rr_next(f, num))) != cur)
1191                 cur = old;
1192         return f->arr[cur];
1193 }
1194
1195 static struct sock *fanout_demux_cpu(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1196 {
1197         unsigned int cpu = smp_processor_id();
1198
1199         return f->arr[cpu % num];
1200 }
1201
1202 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1203                              struct packet_type *pt, struct net_device *orig_dev)
1204 {
1205         struct packet_fanout *f = pt->af_packet_priv;
1206         unsigned int num = f->num_members;
1207         struct packet_sock *po;
1208         struct sock *sk;
1209
1210         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1211             !num) {
1212                 kfree_skb(skb);
1213                 return 0;
1214         }
1215
1216         switch (f->type) {
1217         case PACKET_FANOUT_HASH:
1218         default:
1219                 if (f->defrag) {
1220                         skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1221                         if (!skb)
1222                                 return 0;
1223                 }
1224                 skb_get_rxhash(skb);
1225                 sk = fanout_demux_hash(f, skb, num);
1226                 break;
1227         case PACKET_FANOUT_LB:
1228                 sk = fanout_demux_lb(f, skb, num);
1229                 break;
1230         case PACKET_FANOUT_CPU:
1231                 sk = fanout_demux_cpu(f, skb, num);
1232                 break;
1233         }
1234
1235         po = pkt_sk(sk);
1236
1237         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1238 }
1239
1240 static DEFINE_MUTEX(fanout_mutex);
1241 static LIST_HEAD(fanout_list);
1242
1243 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1244 {
1245         struct packet_fanout *f = po->fanout;
1246
1247         spin_lock(&f->lock);
1248         f->arr[f->num_members] = sk;
1249         smp_wmb();
1250         f->num_members++;
1251         spin_unlock(&f->lock);
1252 }
1253
1254 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1255 {
1256         struct packet_fanout *f = po->fanout;
1257         int i;
1258
1259         spin_lock(&f->lock);
1260         for (i = 0; i < f->num_members; i++) {
1261                 if (f->arr[i] == sk)
1262                         break;
1263         }
1264         BUG_ON(i >= f->num_members);
1265         f->arr[i] = f->arr[f->num_members - 1];
1266         f->num_members--;
1267         spin_unlock(&f->lock);
1268 }
1269
1270 bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1271 {
1272         if (ptype->af_packet_priv == (void*)((struct packet_sock *)sk)->fanout)
1273                 return true;
1274
1275         return false;
1276 }
1277
1278 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1279 {
1280         struct packet_sock *po = pkt_sk(sk);
1281         struct packet_fanout *f, *match;
1282         u8 type = type_flags & 0xff;
1283         u8 defrag = (type_flags & PACKET_FANOUT_FLAG_DEFRAG) ? 1 : 0;
1284         int err;
1285
1286         switch (type) {
1287         case PACKET_FANOUT_HASH:
1288         case PACKET_FANOUT_LB:
1289         case PACKET_FANOUT_CPU:
1290                 break;
1291         default:
1292                 return -EINVAL;
1293         }
1294
1295         if (!po->running)
1296                 return -EINVAL;
1297
1298         if (po->fanout)
1299                 return -EALREADY;
1300
1301         mutex_lock(&fanout_mutex);
1302         match = NULL;
1303         list_for_each_entry(f, &fanout_list, list) {
1304                 if (f->id == id &&
1305                     read_pnet(&f->net) == sock_net(sk)) {
1306                         match = f;
1307                         break;
1308                 }
1309         }
1310         err = -EINVAL;
1311         if (match && match->defrag != defrag)
1312                 goto out;
1313         if (!match) {
1314                 err = -ENOMEM;
1315                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1316                 if (!match)
1317                         goto out;
1318                 write_pnet(&match->net, sock_net(sk));
1319                 match->id = id;
1320                 match->type = type;
1321                 match->defrag = defrag;
1322                 atomic_set(&match->rr_cur, 0);
1323                 INIT_LIST_HEAD(&match->list);
1324                 spin_lock_init(&match->lock);
1325                 atomic_set(&match->sk_ref, 0);
1326                 match->prot_hook.type = po->prot_hook.type;
1327                 match->prot_hook.dev = po->prot_hook.dev;
1328                 match->prot_hook.func = packet_rcv_fanout;
1329                 match->prot_hook.af_packet_priv = match;
1330                 match->prot_hook.id_match = match_fanout_group;
1331                 dev_add_pack(&match->prot_hook);
1332                 list_add(&match->list, &fanout_list);
1333         }
1334         err = -EINVAL;
1335         if (match->type == type &&
1336             match->prot_hook.type == po->prot_hook.type &&
1337             match->prot_hook.dev == po->prot_hook.dev) {
1338                 err = -ENOSPC;
1339                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1340                         __dev_remove_pack(&po->prot_hook);
1341                         po->fanout = match;
1342                         atomic_inc(&match->sk_ref);
1343                         __fanout_link(sk, po);
1344                         err = 0;
1345                 }
1346         }
1347 out:
1348         mutex_unlock(&fanout_mutex);
1349         return err;
1350 }
1351
1352 static void fanout_release(struct sock *sk)
1353 {
1354         struct packet_sock *po = pkt_sk(sk);
1355         struct packet_fanout *f;
1356
1357         f = po->fanout;
1358         if (!f)
1359                 return;
1360
1361         po->fanout = NULL;
1362
1363         mutex_lock(&fanout_mutex);
1364         if (atomic_dec_and_test(&f->sk_ref)) {
1365                 list_del(&f->list);
1366                 dev_remove_pack(&f->prot_hook);
1367                 kfree(f);
1368         }
1369         mutex_unlock(&fanout_mutex);
1370 }
1371
1372 static const struct proto_ops packet_ops;
1373
1374 static const struct proto_ops packet_ops_spkt;
1375
1376 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1377                            struct packet_type *pt, struct net_device *orig_dev)
1378 {
1379         struct sock *sk;
1380         struct sockaddr_pkt *spkt;
1381
1382         /*
1383          *      When we registered the protocol we saved the socket in the data
1384          *      field for just this event.
1385          */
1386
1387         sk = pt->af_packet_priv;
1388
1389         /*
1390          *      Yank back the headers [hope the device set this
1391          *      right or kerboom...]
1392          *
1393          *      Incoming packets have ll header pulled,
1394          *      push it back.
1395          *
1396          *      For outgoing ones skb->data == skb_mac_header(skb)
1397          *      so that this procedure is noop.
1398          */
1399
1400         if (skb->pkt_type == PACKET_LOOPBACK)
1401                 goto out;
1402
1403         if (!net_eq(dev_net(dev), sock_net(sk)))
1404                 goto out;
1405
1406         skb = skb_share_check(skb, GFP_ATOMIC);
1407         if (skb == NULL)
1408                 goto oom;
1409
1410         /* drop any routing info */
1411         skb_dst_drop(skb);
1412
1413         /* drop conntrack reference */
1414         nf_reset(skb);
1415
1416         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1417
1418         skb_push(skb, skb->data - skb_mac_header(skb));
1419
1420         /*
1421          *      The SOCK_PACKET socket receives _all_ frames.
1422          */
1423
1424         spkt->spkt_family = dev->type;
1425         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1426         spkt->spkt_protocol = skb->protocol;
1427
1428         /*
1429          *      Charge the memory to the socket. This is done specifically
1430          *      to prevent sockets using all the memory up.
1431          */
1432
1433         if (sock_queue_rcv_skb(sk, skb) == 0)
1434                 return 0;
1435
1436 out:
1437         kfree_skb(skb);
1438 oom:
1439         return 0;
1440 }
1441
1442
1443 /*
1444  *      Output a raw packet to a device layer. This bypasses all the other
1445  *      protocol layers and you must therefore supply it with a complete frame
1446  */
1447
1448 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1449                                struct msghdr *msg, size_t len)
1450 {
1451         struct sock *sk = sock->sk;
1452         struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1453         struct sk_buff *skb = NULL;
1454         struct net_device *dev;
1455         __be16 proto = 0;
1456         int err;
1457
1458         /*
1459          *      Get and verify the address.
1460          */
1461
1462         if (saddr) {
1463                 if (msg->msg_namelen < sizeof(struct sockaddr))
1464                         return -EINVAL;
1465                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1466                         proto = saddr->spkt_protocol;
1467         } else
1468                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1469
1470         /*
1471          *      Find the device first to size check it
1472          */
1473
1474         saddr->spkt_device[13] = 0;
1475 retry:
1476         rcu_read_lock();
1477         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1478         err = -ENODEV;
1479         if (dev == NULL)
1480                 goto out_unlock;
1481
1482         err = -ENETDOWN;
1483         if (!(dev->flags & IFF_UP))
1484                 goto out_unlock;
1485
1486         /*
1487          * You may not queue a frame bigger than the mtu. This is the lowest level
1488          * raw protocol and you must do your own fragmentation at this level.
1489          */
1490
1491         err = -EMSGSIZE;
1492         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN)
1493                 goto out_unlock;
1494
1495         if (!skb) {
1496                 size_t reserved = LL_RESERVED_SPACE(dev);
1497                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1498
1499                 rcu_read_unlock();
1500                 skb = sock_wmalloc(sk, len + reserved, 0, GFP_KERNEL);
1501                 if (skb == NULL)
1502                         return -ENOBUFS;
1503                 /* FIXME: Save some space for broken drivers that write a hard
1504                  * header at transmission time by themselves. PPP is the notable
1505                  * one here. This should really be fixed at the driver level.
1506                  */
1507                 skb_reserve(skb, reserved);
1508                 skb_reset_network_header(skb);
1509
1510                 /* Try to align data part correctly */
1511                 if (hhlen) {
1512                         skb->data -= hhlen;
1513                         skb->tail -= hhlen;
1514                         if (len < hhlen)
1515                                 skb_reset_network_header(skb);
1516                 }
1517                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1518                 if (err)
1519                         goto out_free;
1520                 goto retry;
1521         }
1522
1523         if (len > (dev->mtu + dev->hard_header_len)) {
1524                 /* Earlier code assumed this would be a VLAN pkt,
1525                  * double-check this now that we have the actual
1526                  * packet in hand.
1527                  */
1528                 struct ethhdr *ehdr;
1529                 skb_reset_mac_header(skb);
1530                 ehdr = eth_hdr(skb);
1531                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1532                         err = -EMSGSIZE;
1533                         goto out_unlock;
1534                 }
1535         }
1536
1537         skb->protocol = proto;
1538         skb->dev = dev;
1539         skb->priority = sk->sk_priority;
1540         skb->mark = sk->sk_mark;
1541         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1542         if (err < 0)
1543                 goto out_unlock;
1544
1545         dev_queue_xmit(skb);
1546         rcu_read_unlock();
1547         return len;
1548
1549 out_unlock:
1550         rcu_read_unlock();
1551 out_free:
1552         kfree_skb(skb);
1553         return err;
1554 }
1555
1556 static unsigned int run_filter(const struct sk_buff *skb,
1557                                       const struct sock *sk,
1558                                       unsigned int res)
1559 {
1560         struct sk_filter *filter;
1561
1562         rcu_read_lock();
1563         filter = rcu_dereference(sk->sk_filter);
1564         if (filter != NULL)
1565                 res = SK_RUN_FILTER(filter, skb);
1566         rcu_read_unlock();
1567
1568         return res;
1569 }
1570
1571 /*
1572  * This function makes lazy skb cloning in hope that most of packets
1573  * are discarded by BPF.
1574  *
1575  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1576  * and skb->cb are mangled. It works because (and until) packets
1577  * falling here are owned by current CPU. Output packets are cloned
1578  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1579  * sequencially, so that if we return skb to original state on exit,
1580  * we will not harm anyone.
1581  */
1582
1583 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1584                       struct packet_type *pt, struct net_device *orig_dev)
1585 {
1586         struct sock *sk;
1587         struct sockaddr_ll *sll;
1588         struct packet_sock *po;
1589         u8 *skb_head = skb->data;
1590         int skb_len = skb->len;
1591         unsigned int snaplen, res;
1592
1593         if (skb->pkt_type == PACKET_LOOPBACK)
1594                 goto drop;
1595
1596         sk = pt->af_packet_priv;
1597         po = pkt_sk(sk);
1598
1599         if (!net_eq(dev_net(dev), sock_net(sk)))
1600                 goto drop;
1601
1602         skb->dev = dev;
1603
1604         if (dev->header_ops) {
1605                 /* The device has an explicit notion of ll header,
1606                  * exported to higher levels.
1607                  *
1608                  * Otherwise, the device hides details of its frame
1609                  * structure, so that corresponding packet head is
1610                  * never delivered to user.
1611                  */
1612                 if (sk->sk_type != SOCK_DGRAM)
1613                         skb_push(skb, skb->data - skb_mac_header(skb));
1614                 else if (skb->pkt_type == PACKET_OUTGOING) {
1615                         /* Special case: outgoing packets have ll header at head */
1616                         skb_pull(skb, skb_network_offset(skb));
1617                 }
1618         }
1619
1620         snaplen = skb->len;
1621
1622         res = run_filter(skb, sk, snaplen);
1623         if (!res)
1624                 goto drop_n_restore;
1625         if (snaplen > res)
1626                 snaplen = res;
1627
1628         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1629                 goto drop_n_acct;
1630
1631         if (skb_shared(skb)) {
1632                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1633                 if (nskb == NULL)
1634                         goto drop_n_acct;
1635
1636                 if (skb_head != skb->data) {
1637                         skb->data = skb_head;
1638                         skb->len = skb_len;
1639                 }
1640                 kfree_skb(skb);
1641                 skb = nskb;
1642         }
1643
1644         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1645                      sizeof(skb->cb));
1646
1647         sll = &PACKET_SKB_CB(skb)->sa.ll;
1648         sll->sll_family = AF_PACKET;
1649         sll->sll_hatype = dev->type;
1650         sll->sll_protocol = skb->protocol;
1651         sll->sll_pkttype = skb->pkt_type;
1652         if (unlikely(po->origdev))
1653                 sll->sll_ifindex = orig_dev->ifindex;
1654         else
1655                 sll->sll_ifindex = dev->ifindex;
1656
1657         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1658
1659         PACKET_SKB_CB(skb)->origlen = skb->len;
1660
1661         if (pskb_trim(skb, snaplen))
1662                 goto drop_n_acct;
1663
1664         skb_set_owner_r(skb, sk);
1665         skb->dev = NULL;
1666         skb_dst_drop(skb);
1667
1668         /* drop conntrack reference */
1669         nf_reset(skb);
1670
1671         spin_lock(&sk->sk_receive_queue.lock);
1672         po->stats.tp_packets++;
1673         skb->dropcount = atomic_read(&sk->sk_drops);
1674         __skb_queue_tail(&sk->sk_receive_queue, skb);
1675         spin_unlock(&sk->sk_receive_queue.lock);
1676         sk->sk_data_ready(sk, skb->len);
1677         return 0;
1678
1679 drop_n_acct:
1680         spin_lock(&sk->sk_receive_queue.lock);
1681         po->stats.tp_drops++;
1682         atomic_inc(&sk->sk_drops);
1683         spin_unlock(&sk->sk_receive_queue.lock);
1684
1685 drop_n_restore:
1686         if (skb_head != skb->data && skb_shared(skb)) {
1687                 skb->data = skb_head;
1688                 skb->len = skb_len;
1689         }
1690 drop:
1691         consume_skb(skb);
1692         return 0;
1693 }
1694
1695 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1696                        struct packet_type *pt, struct net_device *orig_dev)
1697 {
1698         struct sock *sk;
1699         struct packet_sock *po;
1700         struct sockaddr_ll *sll;
1701         union {
1702                 struct tpacket_hdr *h1;
1703                 struct tpacket2_hdr *h2;
1704                 struct tpacket3_hdr *h3;
1705                 void *raw;
1706         } h;
1707         u8 *skb_head = skb->data;
1708         int skb_len = skb->len;
1709         unsigned int snaplen, res;
1710         unsigned long status = TP_STATUS_USER;
1711         unsigned short macoff, netoff, hdrlen;
1712         struct sk_buff *copy_skb = NULL;
1713         struct timeval tv;
1714         struct timespec ts;
1715         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1716
1717         if (skb->pkt_type == PACKET_LOOPBACK)
1718                 goto drop;
1719
1720         sk = pt->af_packet_priv;
1721         po = pkt_sk(sk);
1722
1723         if (!net_eq(dev_net(dev), sock_net(sk)))
1724                 goto drop;
1725
1726         if (dev->header_ops) {
1727                 if (sk->sk_type != SOCK_DGRAM)
1728                         skb_push(skb, skb->data - skb_mac_header(skb));
1729                 else if (skb->pkt_type == PACKET_OUTGOING) {
1730                         /* Special case: outgoing packets have ll header at head */
1731                         skb_pull(skb, skb_network_offset(skb));
1732                 }
1733         }
1734
1735         if (skb->ip_summed == CHECKSUM_PARTIAL)
1736                 status |= TP_STATUS_CSUMNOTREADY;
1737
1738         snaplen = skb->len;
1739
1740         res = run_filter(skb, sk, snaplen);
1741         if (!res)
1742                 goto drop_n_restore;
1743         if (snaplen > res)
1744                 snaplen = res;
1745
1746         if (sk->sk_type == SOCK_DGRAM) {
1747                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1748                                   po->tp_reserve;
1749         } else {
1750                 unsigned maclen = skb_network_offset(skb);
1751                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1752                                        (maclen < 16 ? 16 : maclen)) +
1753                         po->tp_reserve;
1754                 macoff = netoff - maclen;
1755         }
1756         if (po->tp_version <= TPACKET_V2) {
1757                 if (macoff + snaplen > po->rx_ring.frame_size) {
1758                         if (po->copy_thresh &&
1759                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1760                                 if (skb_shared(skb)) {
1761                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1762                                 } else {
1763                                         copy_skb = skb_get(skb);
1764                                         skb_head = skb->data;
1765                                 }
1766                                 if (copy_skb)
1767                                         skb_set_owner_r(copy_skb, sk);
1768                         }
1769                         snaplen = po->rx_ring.frame_size - macoff;
1770                         if ((int)snaplen < 0)
1771                                 snaplen = 0;
1772                 }
1773         }
1774         spin_lock(&sk->sk_receive_queue.lock);
1775         h.raw = packet_current_rx_frame(po, skb,
1776                                         TP_STATUS_KERNEL, (macoff+snaplen));
1777         if (!h.raw)
1778                 goto ring_is_full;
1779         if (po->tp_version <= TPACKET_V2) {
1780                 packet_increment_rx_head(po, &po->rx_ring);
1781         /*
1782          * LOSING will be reported till you read the stats,
1783          * because it's COR - Clear On Read.
1784          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1785          * at packet level.
1786          */
1787                 if (po->stats.tp_drops)
1788                         status |= TP_STATUS_LOSING;
1789         }
1790         po->stats.tp_packets++;
1791         if (copy_skb) {
1792                 status |= TP_STATUS_COPY;
1793                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1794         }
1795         spin_unlock(&sk->sk_receive_queue.lock);
1796
1797         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1798
1799         switch (po->tp_version) {
1800         case TPACKET_V1:
1801                 h.h1->tp_len = skb->len;
1802                 h.h1->tp_snaplen = snaplen;
1803                 h.h1->tp_mac = macoff;
1804                 h.h1->tp_net = netoff;
1805                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1806                                 && shhwtstamps->syststamp.tv64)
1807                         tv = ktime_to_timeval(shhwtstamps->syststamp);
1808                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1809                                 && shhwtstamps->hwtstamp.tv64)
1810                         tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1811                 else if (skb->tstamp.tv64)
1812                         tv = ktime_to_timeval(skb->tstamp);
1813                 else
1814                         do_gettimeofday(&tv);
1815                 h.h1->tp_sec = tv.tv_sec;
1816                 h.h1->tp_usec = tv.tv_usec;
1817                 hdrlen = sizeof(*h.h1);
1818                 break;
1819         case TPACKET_V2:
1820                 h.h2->tp_len = skb->len;
1821                 h.h2->tp_snaplen = snaplen;
1822                 h.h2->tp_mac = macoff;
1823                 h.h2->tp_net = netoff;
1824                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1825                                 && shhwtstamps->syststamp.tv64)
1826                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1827                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1828                                 && shhwtstamps->hwtstamp.tv64)
1829                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1830                 else if (skb->tstamp.tv64)
1831                         ts = ktime_to_timespec(skb->tstamp);
1832                 else
1833                         getnstimeofday(&ts);
1834                 h.h2->tp_sec = ts.tv_sec;
1835                 h.h2->tp_nsec = ts.tv_nsec;
1836                 if (vlan_tx_tag_present(skb)) {
1837                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1838                         status |= TP_STATUS_VLAN_VALID;
1839                 } else {
1840                         h.h2->tp_vlan_tci = 0;
1841                 }
1842                 h.h2->tp_padding = 0;
1843                 hdrlen = sizeof(*h.h2);
1844                 break;
1845         case TPACKET_V3:
1846                 /* tp_nxt_offset,vlan are already populated above.
1847                  * So DONT clear those fields here
1848                  */
1849                 h.h3->tp_status |= status;
1850                 h.h3->tp_len = skb->len;
1851                 h.h3->tp_snaplen = snaplen;
1852                 h.h3->tp_mac = macoff;
1853                 h.h3->tp_net = netoff;
1854                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1855                                 && shhwtstamps->syststamp.tv64)
1856                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1857                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1858                                 && shhwtstamps->hwtstamp.tv64)
1859                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1860                 else if (skb->tstamp.tv64)
1861                         ts = ktime_to_timespec(skb->tstamp);
1862                 else
1863                         getnstimeofday(&ts);
1864                 h.h3->tp_sec  = ts.tv_sec;
1865                 h.h3->tp_nsec = ts.tv_nsec;
1866                 hdrlen = sizeof(*h.h3);
1867                 break;
1868         default:
1869                 BUG();
1870         }
1871
1872         sll = h.raw + TPACKET_ALIGN(hdrlen);
1873         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1874         sll->sll_family = AF_PACKET;
1875         sll->sll_hatype = dev->type;
1876         sll->sll_protocol = skb->protocol;
1877         sll->sll_pkttype = skb->pkt_type;
1878         if (unlikely(po->origdev))
1879                 sll->sll_ifindex = orig_dev->ifindex;
1880         else
1881                 sll->sll_ifindex = dev->ifindex;
1882
1883         smp_mb();
1884 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1885         {
1886                 u8 *start, *end;
1887
1888                 if (po->tp_version <= TPACKET_V2) {
1889                         end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1890                                 + macoff + snaplen);
1891                         for (start = h.raw; start < end; start += PAGE_SIZE)
1892                                 flush_dcache_page(pgv_to_page(start));
1893                 }
1894                 smp_wmb();
1895         }
1896 #endif
1897         if (po->tp_version <= TPACKET_V2)
1898                 __packet_set_status(po, h.raw, status);
1899         else
1900                 prb_clear_blk_fill_status(&po->rx_ring);
1901
1902         sk->sk_data_ready(sk, 0);
1903
1904 drop_n_restore:
1905         if (skb_head != skb->data && skb_shared(skb)) {
1906                 skb->data = skb_head;
1907                 skb->len = skb_len;
1908         }
1909 drop:
1910         kfree_skb(skb);
1911         return 0;
1912
1913 ring_is_full:
1914         po->stats.tp_drops++;
1915         spin_unlock(&sk->sk_receive_queue.lock);
1916
1917         sk->sk_data_ready(sk, 0);
1918         kfree_skb(copy_skb);
1919         goto drop_n_restore;
1920 }
1921
1922 static void tpacket_destruct_skb(struct sk_buff *skb)
1923 {
1924         struct packet_sock *po = pkt_sk(skb->sk);
1925         void *ph;
1926
1927         if (likely(po->tx_ring.pg_vec)) {
1928                 ph = skb_shinfo(skb)->destructor_arg;
1929                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1930                 atomic_dec(&po->tx_ring.pending);
1931                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1932         }
1933
1934         sock_wfree(skb);
1935 }
1936
1937 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1938                 void *frame, struct net_device *dev, int size_max,
1939                 __be16 proto, unsigned char *addr)
1940 {
1941         union {
1942                 struct tpacket_hdr *h1;
1943                 struct tpacket2_hdr *h2;
1944                 void *raw;
1945         } ph;
1946         int to_write, offset, len, tp_len, nr_frags, len_max;
1947         struct socket *sock = po->sk.sk_socket;
1948         struct page *page;
1949         void *data;
1950         int err;
1951
1952         ph.raw = frame;
1953
1954         skb->protocol = proto;
1955         skb->dev = dev;
1956         skb->priority = po->sk.sk_priority;
1957         skb->mark = po->sk.sk_mark;
1958         skb_shinfo(skb)->destructor_arg = ph.raw;
1959
1960         switch (po->tp_version) {
1961         case TPACKET_V2:
1962                 tp_len = ph.h2->tp_len;
1963                 break;
1964         default:
1965                 tp_len = ph.h1->tp_len;
1966                 break;
1967         }
1968         if (unlikely(tp_len > size_max)) {
1969                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1970                 return -EMSGSIZE;
1971         }
1972
1973         skb_reserve(skb, LL_RESERVED_SPACE(dev));
1974         skb_reset_network_header(skb);
1975
1976         data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1977         to_write = tp_len;
1978
1979         if (sock->type == SOCK_DGRAM) {
1980                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
1981                                 NULL, tp_len);
1982                 if (unlikely(err < 0))
1983                         return -EINVAL;
1984         } else if (dev->hard_header_len) {
1985                 /* net device doesn't like empty head */
1986                 if (unlikely(tp_len <= dev->hard_header_len)) {
1987                         pr_err("packet size is too short (%d < %d)\n",
1988                                tp_len, dev->hard_header_len);
1989                         return -EINVAL;
1990                 }
1991
1992                 skb_push(skb, dev->hard_header_len);
1993                 err = skb_store_bits(skb, 0, data,
1994                                 dev->hard_header_len);
1995                 if (unlikely(err))
1996                         return err;
1997
1998                 data += dev->hard_header_len;
1999                 to_write -= dev->hard_header_len;
2000         }
2001
2002         err = -EFAULT;
2003         offset = offset_in_page(data);
2004         len_max = PAGE_SIZE - offset;
2005         len = ((to_write > len_max) ? len_max : to_write);
2006
2007         skb->data_len = to_write;
2008         skb->len += to_write;
2009         skb->truesize += to_write;
2010         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2011
2012         while (likely(to_write)) {
2013                 nr_frags = skb_shinfo(skb)->nr_frags;
2014
2015                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2016                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2017                                MAX_SKB_FRAGS);
2018                         return -EFAULT;
2019                 }
2020
2021                 page = pgv_to_page(data);
2022                 data += len;
2023                 flush_dcache_page(page);
2024                 get_page(page);
2025                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2026                 to_write -= len;
2027                 offset = 0;
2028                 len_max = PAGE_SIZE;
2029                 len = ((to_write > len_max) ? len_max : to_write);
2030         }
2031
2032         return tp_len;
2033 }
2034
2035 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2036 {
2037         struct sk_buff *skb;
2038         struct net_device *dev;
2039         __be16 proto;
2040         bool need_rls_dev = false;
2041         int err, reserve = 0;
2042         void *ph;
2043         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2044         int tp_len, size_max;
2045         unsigned char *addr;
2046         int len_sum = 0;
2047         int status = 0;
2048
2049         mutex_lock(&po->pg_vec_lock);
2050
2051         err = -EBUSY;
2052         if (saddr == NULL) {
2053                 dev = po->prot_hook.dev;
2054                 proto   = po->num;
2055                 addr    = NULL;
2056         } else {
2057                 err = -EINVAL;
2058                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2059                         goto out;
2060                 if (msg->msg_namelen < (saddr->sll_halen
2061                                         + offsetof(struct sockaddr_ll,
2062                                                 sll_addr)))
2063                         goto out;
2064                 proto   = saddr->sll_protocol;
2065                 addr    = saddr->sll_addr;
2066                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2067                 need_rls_dev = true;
2068         }
2069
2070         err = -ENXIO;
2071         if (unlikely(dev == NULL))
2072                 goto out;
2073
2074         reserve = dev->hard_header_len;
2075
2076         err = -ENETDOWN;
2077         if (unlikely(!(dev->flags & IFF_UP)))
2078                 goto out_put;
2079
2080         size_max = po->tx_ring.frame_size
2081                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2082
2083         if (size_max > dev->mtu + reserve)
2084                 size_max = dev->mtu + reserve;
2085
2086         do {
2087                 ph = packet_current_frame(po, &po->tx_ring,
2088                                 TP_STATUS_SEND_REQUEST);
2089
2090                 if (unlikely(ph == NULL)) {
2091                         schedule();
2092                         continue;
2093                 }
2094
2095                 status = TP_STATUS_SEND_REQUEST;
2096                 skb = sock_alloc_send_skb(&po->sk,
2097                                 LL_ALLOCATED_SPACE(dev)
2098                                 + sizeof(struct sockaddr_ll),
2099                                 0, &err);
2100
2101                 if (unlikely(skb == NULL))
2102                         goto out_status;
2103
2104                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2105                                 addr);
2106
2107                 if (unlikely(tp_len < 0)) {
2108                         if (po->tp_loss) {
2109                                 __packet_set_status(po, ph,
2110                                                 TP_STATUS_AVAILABLE);
2111                                 packet_increment_head(&po->tx_ring);
2112                                 kfree_skb(skb);
2113                                 continue;
2114                         } else {
2115                                 status = TP_STATUS_WRONG_FORMAT;
2116                                 err = tp_len;
2117                                 goto out_status;
2118                         }
2119                 }
2120
2121                 skb->destructor = tpacket_destruct_skb;
2122                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2123                 atomic_inc(&po->tx_ring.pending);
2124
2125                 status = TP_STATUS_SEND_REQUEST;
2126                 err = dev_queue_xmit(skb);
2127                 if (unlikely(err > 0)) {
2128                         err = net_xmit_errno(err);
2129                         if (err && __packet_get_status(po, ph) ==
2130                                    TP_STATUS_AVAILABLE) {
2131                                 /* skb was destructed already */
2132                                 skb = NULL;
2133                                 goto out_status;
2134                         }
2135                         /*
2136                          * skb was dropped but not destructed yet;
2137                          * let's treat it like congestion or err < 0
2138                          */
2139                         err = 0;
2140                 }
2141                 packet_increment_head(&po->tx_ring);
2142                 len_sum += tp_len;
2143         } while (likely((ph != NULL) ||
2144                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2145                          (atomic_read(&po->tx_ring.pending))))
2146                 );
2147
2148         err = len_sum;
2149         goto out_put;
2150
2151 out_status:
2152         __packet_set_status(po, ph, status);
2153         kfree_skb(skb);
2154 out_put:
2155         if (need_rls_dev)
2156                 dev_put(dev);
2157 out:
2158         mutex_unlock(&po->pg_vec_lock);
2159         return err;
2160 }
2161
2162 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2163                                         size_t reserve, size_t len,
2164                                         size_t linear, int noblock,
2165                                         int *err)
2166 {
2167         struct sk_buff *skb;
2168
2169         /* Under a page?  Don't bother with paged skb. */
2170         if (prepad + len < PAGE_SIZE || !linear)
2171                 linear = len;
2172
2173         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2174                                    err);
2175         if (!skb)
2176                 return NULL;
2177
2178         skb_reserve(skb, reserve);
2179         skb_put(skb, linear);
2180         skb->data_len = len - linear;
2181         skb->len += len - linear;
2182
2183         return skb;
2184 }
2185
2186 static int packet_snd(struct socket *sock,
2187                           struct msghdr *msg, size_t len)
2188 {
2189         struct sock *sk = sock->sk;
2190         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2191         struct sk_buff *skb;
2192         struct net_device *dev;
2193         __be16 proto;
2194         bool need_rls_dev = false;
2195         unsigned char *addr;
2196         int err, reserve = 0;
2197         struct virtio_net_hdr vnet_hdr = { 0 };
2198         int offset = 0;
2199         int vnet_hdr_len;
2200         struct packet_sock *po = pkt_sk(sk);
2201         unsigned short gso_type = 0;
2202
2203         /*
2204          *      Get and verify the address.
2205          */
2206
2207         if (saddr == NULL) {
2208                 dev = po->prot_hook.dev;
2209                 proto   = po->num;
2210                 addr    = NULL;
2211         } else {
2212                 err = -EINVAL;
2213                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2214                         goto out;
2215                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2216                         goto out;
2217                 proto   = saddr->sll_protocol;
2218                 addr    = saddr->sll_addr;
2219                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2220                 need_rls_dev = true;
2221         }
2222
2223         err = -ENXIO;
2224         if (dev == NULL)
2225                 goto out_unlock;
2226         if (sock->type == SOCK_RAW)
2227                 reserve = dev->hard_header_len;
2228
2229         err = -ENETDOWN;
2230         if (!(dev->flags & IFF_UP))
2231                 goto out_unlock;
2232
2233         if (po->has_vnet_hdr) {
2234                 vnet_hdr_len = sizeof(vnet_hdr);
2235
2236                 err = -EINVAL;
2237                 if (len < vnet_hdr_len)
2238                         goto out_unlock;
2239
2240                 len -= vnet_hdr_len;
2241
2242                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2243                                        vnet_hdr_len);
2244                 if (err < 0)
2245                         goto out_unlock;
2246
2247                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2248                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2249                       vnet_hdr.hdr_len))
2250                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
2251                                                  vnet_hdr.csum_offset + 2;
2252
2253                 err = -EINVAL;
2254                 if (vnet_hdr.hdr_len > len)
2255                         goto out_unlock;
2256
2257                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2258                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2259                         case VIRTIO_NET_HDR_GSO_TCPV4:
2260                                 gso_type = SKB_GSO_TCPV4;
2261                                 break;
2262                         case VIRTIO_NET_HDR_GSO_TCPV6:
2263                                 gso_type = SKB_GSO_TCPV6;
2264                                 break;
2265                         case VIRTIO_NET_HDR_GSO_UDP:
2266                                 gso_type = SKB_GSO_UDP;
2267                                 break;
2268                         default:
2269                                 goto out_unlock;
2270                         }
2271
2272                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2273                                 gso_type |= SKB_GSO_TCP_ECN;
2274
2275                         if (vnet_hdr.gso_size == 0)
2276                                 goto out_unlock;
2277
2278                 }
2279         }
2280
2281         err = -EMSGSIZE;
2282         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN))
2283                 goto out_unlock;
2284
2285         err = -ENOBUFS;
2286         skb = packet_alloc_skb(sk, LL_ALLOCATED_SPACE(dev),
2287                                LL_RESERVED_SPACE(dev), len, vnet_hdr.hdr_len,
2288                                msg->msg_flags & MSG_DONTWAIT, &err);
2289         if (skb == NULL)
2290                 goto out_unlock;
2291
2292         skb_set_network_header(skb, reserve);
2293
2294         err = -EINVAL;
2295         if (sock->type == SOCK_DGRAM &&
2296             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2297                 goto out_free;
2298
2299         /* Returns -EFAULT on error */
2300         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2301         if (err)
2302                 goto out_free;
2303         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2304         if (err < 0)
2305                 goto out_free;
2306
2307         if (!gso_type && (len > dev->mtu + reserve)) {
2308                 /* Earlier code assumed this would be a VLAN pkt,
2309                  * double-check this now that we have the actual
2310                  * packet in hand.
2311                  */
2312                 struct ethhdr *ehdr;
2313                 skb_reset_mac_header(skb);
2314                 ehdr = eth_hdr(skb);
2315                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2316                         err = -EMSGSIZE;
2317                         goto out_free;
2318                 }
2319         }
2320
2321         skb->protocol = proto;
2322         skb->dev = dev;
2323         skb->priority = sk->sk_priority;
2324         skb->mark = sk->sk_mark;
2325
2326         if (po->has_vnet_hdr) {
2327                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2328                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2329                                                   vnet_hdr.csum_offset)) {
2330                                 err = -EINVAL;
2331                                 goto out_free;
2332                         }
2333                 }
2334
2335                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2336                 skb_shinfo(skb)->gso_type = gso_type;
2337
2338                 /* Header must be checked, and gso_segs computed. */
2339                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2340                 skb_shinfo(skb)->gso_segs = 0;
2341
2342                 len += vnet_hdr_len;
2343         }
2344
2345         /*
2346          *      Now send it
2347          */
2348
2349         err = dev_queue_xmit(skb);
2350         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2351                 goto out_unlock;
2352
2353         if (need_rls_dev)
2354                 dev_put(dev);
2355
2356         return len;
2357
2358 out_free:
2359         kfree_skb(skb);
2360 out_unlock:
2361         if (dev && need_rls_dev)
2362                 dev_put(dev);
2363 out:
2364         return err;
2365 }
2366
2367 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2368                 struct msghdr *msg, size_t len)
2369 {
2370         struct sock *sk = sock->sk;
2371         struct packet_sock *po = pkt_sk(sk);
2372         if (po->tx_ring.pg_vec)
2373                 return tpacket_snd(po, msg);
2374         else
2375                 return packet_snd(sock, msg, len);
2376 }
2377
2378 /*
2379  *      Close a PACKET socket. This is fairly simple. We immediately go
2380  *      to 'closed' state and remove our protocol entry in the device list.
2381  */
2382
2383 static int packet_release(struct socket *sock)
2384 {
2385         struct sock *sk = sock->sk;
2386         struct packet_sock *po;
2387         struct net *net;
2388         union tpacket_req_u req_u;
2389
2390         if (!sk)
2391                 return 0;
2392
2393         net = sock_net(sk);
2394         po = pkt_sk(sk);
2395
2396         spin_lock_bh(&net->packet.sklist_lock);
2397         sk_del_node_init_rcu(sk);
2398         sock_prot_inuse_add(net, sk->sk_prot, -1);
2399         spin_unlock_bh(&net->packet.sklist_lock);
2400
2401         spin_lock(&po->bind_lock);
2402         unregister_prot_hook(sk, false);
2403         if (po->prot_hook.dev) {
2404                 dev_put(po->prot_hook.dev);
2405                 po->prot_hook.dev = NULL;
2406         }
2407         spin_unlock(&po->bind_lock);
2408
2409         packet_flush_mclist(sk);
2410
2411         if (po->rx_ring.pg_vec) {
2412                 memset(&req_u, 0, sizeof(req_u));
2413                 packet_set_ring(sk, &req_u, 1, 0);
2414         }
2415
2416         if (po->tx_ring.pg_vec) {
2417                 memset(&req_u, 0, sizeof(req_u));
2418                 packet_set_ring(sk, &req_u, 1, 1);
2419         }
2420
2421         fanout_release(sk);
2422
2423         synchronize_net();
2424         /*
2425          *      Now the socket is dead. No more input will appear.
2426          */
2427         sock_orphan(sk);
2428         sock->sk = NULL;
2429
2430         /* Purge queues */
2431
2432         skb_queue_purge(&sk->sk_receive_queue);
2433         sk_refcnt_debug_release(sk);
2434
2435         sock_put(sk);
2436         return 0;
2437 }
2438
2439 /*
2440  *      Attach a packet hook.
2441  */
2442
2443 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2444 {
2445         struct packet_sock *po = pkt_sk(sk);
2446
2447         if (po->fanout) {
2448                 if (dev)
2449                         dev_put(dev);
2450
2451                 return -EINVAL;
2452         }
2453
2454         lock_sock(sk);
2455
2456         spin_lock(&po->bind_lock);
2457         unregister_prot_hook(sk, true);
2458         po->num = protocol;
2459         po->prot_hook.type = protocol;
2460         if (po->prot_hook.dev)
2461                 dev_put(po->prot_hook.dev);
2462         po->prot_hook.dev = dev;
2463
2464         po->ifindex = dev ? dev->ifindex : 0;
2465
2466         if (protocol == 0)
2467                 goto out_unlock;
2468
2469         if (!dev || (dev->flags & IFF_UP)) {
2470                 register_prot_hook(sk);
2471         } else {
2472                 sk->sk_err = ENETDOWN;
2473                 if (!sock_flag(sk, SOCK_DEAD))
2474                         sk->sk_error_report(sk);
2475         }
2476
2477 out_unlock:
2478         spin_unlock(&po->bind_lock);
2479         release_sock(sk);
2480         return 0;
2481 }
2482
2483 /*
2484  *      Bind a packet socket to a device
2485  */
2486
2487 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2488                             int addr_len)
2489 {
2490         struct sock *sk = sock->sk;
2491         char name[15];
2492         struct net_device *dev;
2493         int err = -ENODEV;
2494
2495         /*
2496          *      Check legality
2497          */
2498
2499         if (addr_len != sizeof(struct sockaddr))
2500                 return -EINVAL;
2501         strlcpy(name, uaddr->sa_data, sizeof(name));
2502
2503         dev = dev_get_by_name(sock_net(sk), name);
2504         if (dev)
2505                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2506         return err;
2507 }
2508
2509 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2510 {
2511         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2512         struct sock *sk = sock->sk;
2513         struct net_device *dev = NULL;
2514         int err;
2515
2516
2517         /*
2518          *      Check legality
2519          */
2520
2521         if (addr_len < sizeof(struct sockaddr_ll))
2522                 return -EINVAL;
2523         if (sll->sll_family != AF_PACKET)
2524                 return -EINVAL;
2525
2526         if (sll->sll_ifindex) {
2527                 err = -ENODEV;
2528                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2529                 if (dev == NULL)
2530                         goto out;
2531         }
2532         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2533
2534 out:
2535         return err;
2536 }
2537
2538 static struct proto packet_proto = {
2539         .name     = "PACKET",
2540         .owner    = THIS_MODULE,
2541         .obj_size = sizeof(struct packet_sock),
2542 };
2543
2544 /*
2545  *      Create a packet of type SOCK_PACKET.
2546  */
2547
2548 static int packet_create(struct net *net, struct socket *sock, int protocol,
2549                          int kern)
2550 {
2551         struct sock *sk;
2552         struct packet_sock *po;
2553         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2554         int err;
2555
2556         if (!capable(CAP_NET_RAW))
2557                 return -EPERM;
2558         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2559             sock->type != SOCK_PACKET)
2560                 return -ESOCKTNOSUPPORT;
2561
2562         sock->state = SS_UNCONNECTED;
2563
2564         err = -ENOBUFS;
2565         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2566         if (sk == NULL)
2567                 goto out;
2568
2569         sock->ops = &packet_ops;
2570         if (sock->type == SOCK_PACKET)
2571                 sock->ops = &packet_ops_spkt;
2572
2573         sock_init_data(sock, sk);
2574
2575         po = pkt_sk(sk);
2576         sk->sk_family = PF_PACKET;
2577         po->num = proto;
2578
2579         sk->sk_destruct = packet_sock_destruct;
2580         sk_refcnt_debug_inc(sk);
2581
2582         /*
2583          *      Attach a protocol block
2584          */
2585
2586         spin_lock_init(&po->bind_lock);
2587         mutex_init(&po->pg_vec_lock);
2588         po->prot_hook.func = packet_rcv;
2589
2590         if (sock->type == SOCK_PACKET)
2591                 po->prot_hook.func = packet_rcv_spkt;
2592
2593         po->prot_hook.af_packet_priv = sk;
2594
2595         if (proto) {
2596                 po->prot_hook.type = proto;
2597                 register_prot_hook(sk);
2598         }
2599
2600         spin_lock_bh(&net->packet.sklist_lock);
2601         sk_add_node_rcu(sk, &net->packet.sklist);
2602         sock_prot_inuse_add(net, &packet_proto, 1);
2603         spin_unlock_bh(&net->packet.sklist_lock);
2604
2605         return 0;
2606 out:
2607         return err;
2608 }
2609
2610 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2611 {
2612         struct sock_exterr_skb *serr;
2613         struct sk_buff *skb, *skb2;
2614         int copied, err;
2615
2616         err = -EAGAIN;
2617         skb = skb_dequeue(&sk->sk_error_queue);
2618         if (skb == NULL)
2619                 goto out;
2620
2621         copied = skb->len;
2622         if (copied > len) {
2623                 msg->msg_flags |= MSG_TRUNC;
2624                 copied = len;
2625         }
2626         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2627         if (err)
2628                 goto out_free_skb;
2629
2630         sock_recv_timestamp(msg, sk, skb);
2631
2632         serr = SKB_EXT_ERR(skb);
2633         put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2634                  sizeof(serr->ee), &serr->ee);
2635
2636         msg->msg_flags |= MSG_ERRQUEUE;
2637         err = copied;
2638
2639         /* Reset and regenerate socket error */
2640         spin_lock_bh(&sk->sk_error_queue.lock);
2641         sk->sk_err = 0;
2642         if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2643                 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2644                 spin_unlock_bh(&sk->sk_error_queue.lock);
2645                 sk->sk_error_report(sk);
2646         } else
2647                 spin_unlock_bh(&sk->sk_error_queue.lock);
2648
2649 out_free_skb:
2650         kfree_skb(skb);
2651 out:
2652         return err;
2653 }
2654
2655 /*
2656  *      Pull a packet from our receive queue and hand it to the user.
2657  *      If necessary we block.
2658  */
2659
2660 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2661                           struct msghdr *msg, size_t len, int flags)
2662 {
2663         struct sock *sk = sock->sk;
2664         struct sk_buff *skb;
2665         int copied, err;
2666         int vnet_hdr_len = 0;
2667
2668         err = -EINVAL;
2669         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2670                 goto out;
2671
2672 #if 0
2673         /* What error should we return now? EUNATTACH? */
2674         if (pkt_sk(sk)->ifindex < 0)
2675                 return -ENODEV;
2676 #endif
2677
2678         if (flags & MSG_ERRQUEUE) {
2679                 err = packet_recv_error(sk, msg, len);
2680                 goto out;
2681         }
2682
2683         /*
2684          *      Call the generic datagram receiver. This handles all sorts
2685          *      of horrible races and re-entrancy so we can forget about it
2686          *      in the protocol layers.
2687          *
2688          *      Now it will return ENETDOWN, if device have just gone down,
2689          *      but then it will block.
2690          */
2691
2692         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2693
2694         /*
2695          *      An error occurred so return it. Because skb_recv_datagram()
2696          *      handles the blocking we don't see and worry about blocking
2697          *      retries.
2698          */
2699
2700         if (skb == NULL)
2701                 goto out;
2702
2703         if (pkt_sk(sk)->has_vnet_hdr) {
2704                 struct virtio_net_hdr vnet_hdr = { 0 };
2705
2706                 err = -EINVAL;
2707                 vnet_hdr_len = sizeof(vnet_hdr);
2708                 if (len < vnet_hdr_len)
2709                         goto out_free;
2710
2711                 len -= vnet_hdr_len;
2712
2713                 if (skb_is_gso(skb)) {
2714                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2715
2716                         /* This is a hint as to how much should be linear. */
2717                         vnet_hdr.hdr_len = skb_headlen(skb);
2718                         vnet_hdr.gso_size = sinfo->gso_size;
2719                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2720                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2721                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2722                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2723                         else if (sinfo->gso_type & SKB_GSO_UDP)
2724                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2725                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2726                                 goto out_free;
2727                         else
2728                                 BUG();
2729                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2730                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2731                 } else
2732                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2733
2734                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2735                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2736                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2737                         vnet_hdr.csum_offset = skb->csum_offset;
2738                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2739                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2740                 } /* else everything is zero */
2741
2742                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2743                                      vnet_hdr_len);
2744                 if (err < 0)
2745                         goto out_free;
2746         }
2747
2748         /* You lose any data beyond the buffer you gave. If it worries
2749          * a user program they can ask the device for its MTU
2750          * anyway.
2751          */
2752         copied = skb->len;
2753         if (copied > len) {
2754                 copied = len;
2755                 msg->msg_flags |= MSG_TRUNC;
2756         }
2757
2758         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2759         if (err)
2760                 goto out_free;
2761
2762         sock_recv_ts_and_drops(msg, sk, skb);
2763
2764         if (msg->msg_name) {
2765                 /* If the address length field is there to be filled
2766                  * in, we fill it in now.
2767                  */
2768                 if (sock->type == SOCK_PACKET) {
2769                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
2770                 } else {
2771                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2772                         msg->msg_namelen = sll->sll_halen +
2773                                 offsetof(struct sockaddr_ll, sll_addr);
2774                 }
2775                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2776                        msg->msg_namelen);
2777         }
2778
2779         if (pkt_sk(sk)->auxdata) {
2780                 struct tpacket_auxdata aux;
2781
2782                 aux.tp_status = TP_STATUS_USER;
2783                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2784                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2785                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2786                 aux.tp_snaplen = skb->len;
2787                 aux.tp_mac = 0;
2788                 aux.tp_net = skb_network_offset(skb);
2789                 if (vlan_tx_tag_present(skb)) {
2790                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2791                         aux.tp_status |= TP_STATUS_VLAN_VALID;
2792                 } else {
2793                         aux.tp_vlan_tci = 0;
2794                 }
2795                 aux.tp_padding = 0;
2796                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2797         }
2798
2799         /*
2800          *      Free or return the buffer as appropriate. Again this
2801          *      hides all the races and re-entrancy issues from us.
2802          */
2803         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2804
2805 out_free:
2806         skb_free_datagram(sk, skb);
2807 out:
2808         return err;
2809 }
2810
2811 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2812                                int *uaddr_len, int peer)
2813 {
2814         struct net_device *dev;
2815         struct sock *sk = sock->sk;
2816
2817         if (peer)
2818                 return -EOPNOTSUPP;
2819
2820         uaddr->sa_family = AF_PACKET;
2821         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
2822         rcu_read_lock();
2823         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2824         if (dev)
2825                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
2826         rcu_read_unlock();
2827         *uaddr_len = sizeof(*uaddr);
2828
2829         return 0;
2830 }
2831
2832 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2833                           int *uaddr_len, int peer)
2834 {
2835         struct net_device *dev;
2836         struct sock *sk = sock->sk;
2837         struct packet_sock *po = pkt_sk(sk);
2838         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2839
2840         if (peer)
2841                 return -EOPNOTSUPP;
2842
2843         sll->sll_family = AF_PACKET;
2844         sll->sll_ifindex = po->ifindex;
2845         sll->sll_protocol = po->num;
2846         sll->sll_pkttype = 0;
2847         rcu_read_lock();
2848         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2849         if (dev) {
2850                 sll->sll_hatype = dev->type;
2851                 sll->sll_halen = dev->addr_len;
2852                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2853         } else {
2854                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
2855                 sll->sll_halen = 0;
2856         }
2857         rcu_read_unlock();
2858         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2859
2860         return 0;
2861 }
2862
2863 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2864                          int what)
2865 {
2866         switch (i->type) {
2867         case PACKET_MR_MULTICAST:
2868                 if (i->alen != dev->addr_len)
2869                         return -EINVAL;
2870                 if (what > 0)
2871                         return dev_mc_add(dev, i->addr);
2872                 else
2873                         return dev_mc_del(dev, i->addr);
2874                 break;
2875         case PACKET_MR_PROMISC:
2876                 return dev_set_promiscuity(dev, what);
2877                 break;
2878         case PACKET_MR_ALLMULTI:
2879                 return dev_set_allmulti(dev, what);
2880                 break;
2881         case PACKET_MR_UNICAST:
2882                 if (i->alen != dev->addr_len)
2883                         return -EINVAL;
2884                 if (what > 0)
2885                         return dev_uc_add(dev, i->addr);
2886                 else
2887                         return dev_uc_del(dev, i->addr);
2888                 break;
2889         default:
2890                 break;
2891         }
2892         return 0;
2893 }
2894
2895 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2896 {
2897         for ( ; i; i = i->next) {
2898                 if (i->ifindex == dev->ifindex)
2899                         packet_dev_mc(dev, i, what);
2900         }
2901 }
2902
2903 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2904 {
2905         struct packet_sock *po = pkt_sk(sk);
2906         struct packet_mclist *ml, *i;
2907         struct net_device *dev;
2908         int err;
2909
2910         rtnl_lock();
2911
2912         err = -ENODEV;
2913         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2914         if (!dev)
2915                 goto done;
2916
2917         err = -EINVAL;
2918         if (mreq->mr_alen > dev->addr_len)
2919                 goto done;
2920
2921         err = -ENOBUFS;
2922         i = kmalloc(sizeof(*i), GFP_KERNEL);
2923         if (i == NULL)
2924                 goto done;
2925
2926         err = 0;
2927         for (ml = po->mclist; ml; ml = ml->next) {
2928                 if (ml->ifindex == mreq->mr_ifindex &&
2929                     ml->type == mreq->mr_type &&
2930                     ml->alen == mreq->mr_alen &&
2931                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2932                         ml->count++;
2933                         /* Free the new element ... */
2934                         kfree(i);
2935                         goto done;
2936                 }
2937         }
2938
2939         i->type = mreq->mr_type;
2940         i->ifindex = mreq->mr_ifindex;
2941         i->alen = mreq->mr_alen;
2942         memcpy(i->addr, mreq->mr_address, i->alen);
2943         i->count = 1;
2944         i->next = po->mclist;
2945         po->mclist = i;
2946         err = packet_dev_mc(dev, i, 1);
2947         if (err) {
2948                 po->mclist = i->next;
2949                 kfree(i);
2950         }
2951
2952 done:
2953         rtnl_unlock();
2954         return err;
2955 }
2956
2957 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2958 {
2959         struct packet_mclist *ml, **mlp;
2960
2961         rtnl_lock();
2962
2963         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &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                         if (--ml->count == 0) {
2969                                 struct net_device *dev;
2970                                 *mlp = ml->next;
2971                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2972                                 if (dev)
2973                                         packet_dev_mc(dev, ml, -1);
2974                                 kfree(ml);
2975                         }
2976                         rtnl_unlock();
2977                         return 0;
2978                 }
2979         }
2980         rtnl_unlock();
2981         return -EADDRNOTAVAIL;
2982 }
2983
2984 static void packet_flush_mclist(struct sock *sk)
2985 {
2986         struct packet_sock *po = pkt_sk(sk);
2987         struct packet_mclist *ml;
2988
2989         if (!po->mclist)
2990                 return;
2991
2992         rtnl_lock();
2993         while ((ml = po->mclist) != NULL) {
2994                 struct net_device *dev;
2995
2996                 po->mclist = ml->next;
2997                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2998                 if (dev != NULL)
2999                         packet_dev_mc(dev, ml, -1);
3000                 kfree(ml);
3001         }
3002         rtnl_unlock();
3003 }
3004
3005 static int
3006 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3007 {
3008         struct sock *sk = sock->sk;
3009         struct packet_sock *po = pkt_sk(sk);
3010         int ret;
3011
3012         if (level != SOL_PACKET)
3013                 return -ENOPROTOOPT;
3014
3015         switch (optname) {
3016         case PACKET_ADD_MEMBERSHIP:
3017         case PACKET_DROP_MEMBERSHIP:
3018         {
3019                 struct packet_mreq_max mreq;
3020                 int len = optlen;
3021                 memset(&mreq, 0, sizeof(mreq));
3022                 if (len < sizeof(struct packet_mreq))
3023                         return -EINVAL;
3024                 if (len > sizeof(mreq))
3025                         len = sizeof(mreq);
3026                 if (copy_from_user(&mreq, optval, len))
3027                         return -EFAULT;
3028                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3029                         return -EINVAL;
3030                 if (optname == PACKET_ADD_MEMBERSHIP)
3031                         ret = packet_mc_add(sk, &mreq);
3032                 else
3033                         ret = packet_mc_drop(sk, &mreq);
3034                 return ret;
3035         }
3036
3037         case PACKET_RX_RING:
3038         case PACKET_TX_RING:
3039         {
3040                 union tpacket_req_u req_u;
3041                 int len;
3042
3043                 switch (po->tp_version) {
3044                 case TPACKET_V1:
3045                 case TPACKET_V2:
3046                         len = sizeof(req_u.req);
3047                         break;
3048                 case TPACKET_V3:
3049                 default:
3050                         len = sizeof(req_u.req3);
3051                         break;
3052                 }
3053                 if (optlen < len)
3054                         return -EINVAL;
3055                 if (pkt_sk(sk)->has_vnet_hdr)
3056                         return -EINVAL;
3057                 if (copy_from_user(&req_u.req, optval, len))
3058                         return -EFAULT;
3059                 return packet_set_ring(sk, &req_u, 0,
3060                         optname == PACKET_TX_RING);
3061         }
3062         case PACKET_COPY_THRESH:
3063         {
3064                 int val;
3065
3066                 if (optlen != sizeof(val))
3067                         return -EINVAL;
3068                 if (copy_from_user(&val, optval, sizeof(val)))
3069                         return -EFAULT;
3070
3071                 pkt_sk(sk)->copy_thresh = val;
3072                 return 0;
3073         }
3074         case PACKET_VERSION:
3075         {
3076                 int val;
3077
3078                 if (optlen != sizeof(val))
3079                         return -EINVAL;
3080                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3081                         return -EBUSY;
3082                 if (copy_from_user(&val, optval, sizeof(val)))
3083                         return -EFAULT;
3084                 switch (val) {
3085                 case TPACKET_V1:
3086                 case TPACKET_V2:
3087                 case TPACKET_V3:
3088                         po->tp_version = val;
3089                         return 0;
3090                 default:
3091                         return -EINVAL;
3092                 }
3093         }
3094         case PACKET_RESERVE:
3095         {
3096                 unsigned int val;
3097
3098                 if (optlen != sizeof(val))
3099                         return -EINVAL;
3100                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3101                         return -EBUSY;
3102                 if (copy_from_user(&val, optval, sizeof(val)))
3103                         return -EFAULT;
3104                 po->tp_reserve = val;
3105                 return 0;
3106         }
3107         case PACKET_LOSS:
3108         {
3109                 unsigned int val;
3110
3111                 if (optlen != sizeof(val))
3112                         return -EINVAL;
3113                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3114                         return -EBUSY;
3115                 if (copy_from_user(&val, optval, sizeof(val)))
3116                         return -EFAULT;
3117                 po->tp_loss = !!val;
3118                 return 0;
3119         }
3120         case PACKET_AUXDATA:
3121         {
3122                 int val;
3123
3124                 if (optlen < sizeof(val))
3125                         return -EINVAL;
3126                 if (copy_from_user(&val, optval, sizeof(val)))
3127                         return -EFAULT;
3128
3129                 po->auxdata = !!val;
3130                 return 0;
3131         }
3132         case PACKET_ORIGDEV:
3133         {
3134                 int val;
3135
3136                 if (optlen < sizeof(val))
3137                         return -EINVAL;
3138                 if (copy_from_user(&val, optval, sizeof(val)))
3139                         return -EFAULT;
3140
3141                 po->origdev = !!val;
3142                 return 0;
3143         }
3144         case PACKET_VNET_HDR:
3145         {
3146                 int val;
3147
3148                 if (sock->type != SOCK_RAW)
3149                         return -EINVAL;
3150                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3151                         return -EBUSY;
3152                 if (optlen < sizeof(val))
3153                         return -EINVAL;
3154                 if (copy_from_user(&val, optval, sizeof(val)))
3155                         return -EFAULT;
3156
3157                 po->has_vnet_hdr = !!val;
3158                 return 0;
3159         }
3160         case PACKET_TIMESTAMP:
3161         {
3162                 int val;
3163
3164                 if (optlen != sizeof(val))
3165                         return -EINVAL;
3166                 if (copy_from_user(&val, optval, sizeof(val)))
3167                         return -EFAULT;
3168
3169                 po->tp_tstamp = val;
3170                 return 0;
3171         }
3172         case PACKET_FANOUT:
3173         {
3174                 int val;
3175
3176                 if (optlen != sizeof(val))
3177                         return -EINVAL;
3178                 if (copy_from_user(&val, optval, sizeof(val)))
3179                         return -EFAULT;
3180
3181                 return fanout_add(sk, val & 0xffff, val >> 16);
3182         }
3183         default:
3184                 return -ENOPROTOOPT;
3185         }
3186 }
3187
3188 static int packet_getsockopt(struct socket *sock, int level, int optname,
3189                              char __user *optval, int __user *optlen)
3190 {
3191         int len;
3192         int val;
3193         struct sock *sk = sock->sk;
3194         struct packet_sock *po = pkt_sk(sk);
3195         void *data;
3196         struct tpacket_stats st;
3197         union tpacket_stats_u st_u;
3198
3199         if (level != SOL_PACKET)
3200                 return -ENOPROTOOPT;
3201
3202         if (get_user(len, optlen))
3203                 return -EFAULT;
3204
3205         if (len < 0)
3206                 return -EINVAL;
3207
3208         switch (optname) {
3209         case PACKET_STATISTICS:
3210                 if (po->tp_version == TPACKET_V3) {
3211                         len = sizeof(struct tpacket_stats_v3);
3212                 } else {
3213                         if (len > sizeof(struct tpacket_stats))
3214                                 len = sizeof(struct tpacket_stats);
3215                 }
3216                 spin_lock_bh(&sk->sk_receive_queue.lock);
3217                 if (po->tp_version == TPACKET_V3) {
3218                         memcpy(&st_u.stats3, &po->stats,
3219                         sizeof(struct tpacket_stats));
3220                         st_u.stats3.tp_freeze_q_cnt =
3221                         po->stats_u.stats3.tp_freeze_q_cnt;
3222                         st_u.stats3.tp_packets += po->stats.tp_drops;
3223                         data = &st_u.stats3;
3224                 } else {
3225                         st = po->stats;
3226                         st.tp_packets += st.tp_drops;
3227                         data = &st;
3228                 }
3229                 memset(&po->stats, 0, sizeof(st));
3230                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3231                 break;
3232         case PACKET_AUXDATA:
3233                 if (len > sizeof(int))
3234                         len = sizeof(int);
3235                 val = po->auxdata;
3236
3237                 data = &val;
3238                 break;
3239         case PACKET_ORIGDEV:
3240                 if (len > sizeof(int))
3241                         len = sizeof(int);
3242                 val = po->origdev;
3243
3244                 data = &val;
3245                 break;
3246         case PACKET_VNET_HDR:
3247                 if (len > sizeof(int))
3248                         len = sizeof(int);
3249                 val = po->has_vnet_hdr;
3250
3251                 data = &val;
3252                 break;
3253         case PACKET_VERSION:
3254                 if (len > sizeof(int))
3255                         len = sizeof(int);
3256                 val = po->tp_version;
3257                 data = &val;
3258                 break;
3259         case PACKET_HDRLEN:
3260                 if (len > sizeof(int))
3261                         len = sizeof(int);
3262                 if (copy_from_user(&val, optval, len))
3263                         return -EFAULT;
3264                 switch (val) {
3265                 case TPACKET_V1:
3266                         val = sizeof(struct tpacket_hdr);
3267                         break;
3268                 case TPACKET_V2:
3269                         val = sizeof(struct tpacket2_hdr);
3270                         break;
3271                 case TPACKET_V3:
3272                         val = sizeof(struct tpacket3_hdr);
3273                         break;
3274                 default:
3275                         return -EINVAL;
3276                 }
3277                 data = &val;
3278                 break;
3279         case PACKET_RESERVE:
3280                 if (len > sizeof(unsigned int))
3281                         len = sizeof(unsigned int);
3282                 val = po->tp_reserve;
3283                 data = &val;
3284                 break;
3285         case PACKET_LOSS:
3286                 if (len > sizeof(unsigned int))
3287                         len = sizeof(unsigned int);
3288                 val = po->tp_loss;
3289                 data = &val;
3290                 break;
3291         case PACKET_TIMESTAMP:
3292                 if (len > sizeof(int))
3293                         len = sizeof(int);
3294                 val = po->tp_tstamp;
3295                 data = &val;
3296                 break;
3297         case PACKET_FANOUT:
3298                 if (len > sizeof(int))
3299                         len = sizeof(int);
3300                 val = (po->fanout ?
3301                        ((u32)po->fanout->id |
3302                         ((u32)po->fanout->type << 16)) :
3303                        0);
3304                 data = &val;
3305                 break;
3306         default:
3307                 return -ENOPROTOOPT;
3308         }
3309
3310         if (put_user(len, optlen))
3311                 return -EFAULT;
3312         if (copy_to_user(optval, data, len))
3313                 return -EFAULT;
3314         return 0;
3315 }
3316
3317
3318 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3319 {
3320         struct sock *sk;
3321         struct hlist_node *node;
3322         struct net_device *dev = data;
3323         struct net *net = dev_net(dev);
3324
3325         rcu_read_lock();
3326         sk_for_each_rcu(sk, node, &net->packet.sklist) {
3327                 struct packet_sock *po = pkt_sk(sk);
3328
3329                 switch (msg) {
3330                 case NETDEV_UNREGISTER:
3331                         if (po->mclist)
3332                                 packet_dev_mclist(dev, po->mclist, -1);
3333                         /* fallthrough */
3334
3335                 case NETDEV_DOWN:
3336                         if (dev->ifindex == po->ifindex) {
3337                                 spin_lock(&po->bind_lock);
3338                                 if (po->running) {
3339                                         __unregister_prot_hook(sk, false);
3340                                         sk->sk_err = ENETDOWN;
3341                                         if (!sock_flag(sk, SOCK_DEAD))
3342                                                 sk->sk_error_report(sk);
3343                                 }
3344                                 if (msg == NETDEV_UNREGISTER) {
3345                                         po->ifindex = -1;
3346                                         if (po->prot_hook.dev)
3347                                                 dev_put(po->prot_hook.dev);
3348                                         po->prot_hook.dev = NULL;
3349                                 }
3350                                 spin_unlock(&po->bind_lock);
3351                         }
3352                         break;
3353                 case NETDEV_UP:
3354                         if (dev->ifindex == po->ifindex) {
3355                                 spin_lock(&po->bind_lock);
3356                                 if (po->num)
3357                                         register_prot_hook(sk);
3358                                 spin_unlock(&po->bind_lock);
3359                         }
3360                         break;
3361                 }
3362         }
3363         rcu_read_unlock();
3364         return NOTIFY_DONE;
3365 }
3366
3367
3368 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3369                         unsigned long arg)
3370 {
3371         struct sock *sk = sock->sk;
3372
3373         switch (cmd) {
3374         case SIOCOUTQ:
3375         {
3376                 int amount = sk_wmem_alloc_get(sk);
3377
3378                 return put_user(amount, (int __user *)arg);
3379         }
3380         case SIOCINQ:
3381         {
3382                 struct sk_buff *skb;
3383                 int amount = 0;
3384
3385                 spin_lock_bh(&sk->sk_receive_queue.lock);
3386                 skb = skb_peek(&sk->sk_receive_queue);
3387                 if (skb)
3388                         amount = skb->len;
3389                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3390                 return put_user(amount, (int __user *)arg);
3391         }
3392         case SIOCGSTAMP:
3393                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3394         case SIOCGSTAMPNS:
3395                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3396
3397 #ifdef CONFIG_INET
3398         case SIOCADDRT:
3399         case SIOCDELRT:
3400         case SIOCDARP:
3401         case SIOCGARP:
3402         case SIOCSARP:
3403         case SIOCGIFADDR:
3404         case SIOCSIFADDR:
3405         case SIOCGIFBRDADDR:
3406         case SIOCSIFBRDADDR:
3407         case SIOCGIFNETMASK:
3408         case SIOCSIFNETMASK:
3409         case SIOCGIFDSTADDR:
3410         case SIOCSIFDSTADDR:
3411         case SIOCSIFFLAGS:
3412                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3413 #endif
3414
3415         default:
3416                 return -ENOIOCTLCMD;
3417         }
3418         return 0;
3419 }
3420
3421 static unsigned int packet_poll(struct file *file, struct socket *sock,
3422                                 poll_table *wait)
3423 {
3424         struct sock *sk = sock->sk;
3425         struct packet_sock *po = pkt_sk(sk);
3426         unsigned int mask = datagram_poll(file, sock, wait);
3427
3428         spin_lock_bh(&sk->sk_receive_queue.lock);
3429         if (po->rx_ring.pg_vec) {
3430                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3431                         TP_STATUS_KERNEL))
3432                         mask |= POLLIN | POLLRDNORM;
3433         }
3434         spin_unlock_bh(&sk->sk_receive_queue.lock);
3435         spin_lock_bh(&sk->sk_write_queue.lock);
3436         if (po->tx_ring.pg_vec) {
3437                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3438                         mask |= POLLOUT | POLLWRNORM;
3439         }
3440         spin_unlock_bh(&sk->sk_write_queue.lock);
3441         return mask;
3442 }
3443
3444
3445 /* Dirty? Well, I still did not learn better way to account
3446  * for user mmaps.
3447  */
3448
3449 static void packet_mm_open(struct vm_area_struct *vma)
3450 {
3451         struct file *file = vma->vm_file;
3452         struct socket *sock = file->private_data;
3453         struct sock *sk = sock->sk;
3454
3455         if (sk)
3456                 atomic_inc(&pkt_sk(sk)->mapped);
3457 }
3458
3459 static void packet_mm_close(struct vm_area_struct *vma)
3460 {
3461         struct file *file = vma->vm_file;
3462         struct socket *sock = file->private_data;
3463         struct sock *sk = sock->sk;
3464
3465         if (sk)
3466                 atomic_dec(&pkt_sk(sk)->mapped);
3467 }
3468
3469 static const struct vm_operations_struct packet_mmap_ops = {
3470         .open   =       packet_mm_open,
3471         .close  =       packet_mm_close,
3472 };
3473
3474 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3475                         unsigned int len)
3476 {
3477         int i;
3478
3479         for (i = 0; i < len; i++) {
3480                 if (likely(pg_vec[i].buffer)) {
3481                         if (is_vmalloc_addr(pg_vec[i].buffer))
3482                                 vfree(pg_vec[i].buffer);
3483                         else
3484                                 free_pages((unsigned long)pg_vec[i].buffer,
3485                                            order);
3486                         pg_vec[i].buffer = NULL;
3487                 }
3488         }
3489         kfree(pg_vec);
3490 }
3491
3492 static char *alloc_one_pg_vec_page(unsigned long order)
3493 {
3494         char *buffer = NULL;
3495         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3496                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3497
3498         buffer = (char *) __get_free_pages(gfp_flags, order);
3499
3500         if (buffer)
3501                 return buffer;
3502
3503         /*
3504          * __get_free_pages failed, fall back to vmalloc
3505          */
3506         buffer = vzalloc((1 << order) * PAGE_SIZE);
3507
3508         if (buffer)
3509                 return buffer;
3510
3511         /*
3512          * vmalloc failed, lets dig into swap here
3513          */
3514         gfp_flags &= ~__GFP_NORETRY;
3515         buffer = (char *)__get_free_pages(gfp_flags, order);
3516         if (buffer)
3517                 return buffer;
3518
3519         /*
3520          * complete and utter failure
3521          */
3522         return NULL;
3523 }
3524
3525 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3526 {
3527         unsigned int block_nr = req->tp_block_nr;
3528         struct pgv *pg_vec;
3529         int i;
3530
3531         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3532         if (unlikely(!pg_vec))
3533                 goto out;
3534
3535         for (i = 0; i < block_nr; i++) {
3536                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3537                 if (unlikely(!pg_vec[i].buffer))
3538                         goto out_free_pgvec;
3539         }
3540
3541 out:
3542         return pg_vec;
3543
3544 out_free_pgvec:
3545         free_pg_vec(pg_vec, order, block_nr);
3546         pg_vec = NULL;
3547         goto out;
3548 }
3549
3550 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3551                 int closing, int tx_ring)
3552 {
3553         struct pgv *pg_vec = NULL;
3554         struct packet_sock *po = pkt_sk(sk);
3555         int was_running, order = 0;
3556         struct packet_ring_buffer *rb;
3557         struct sk_buff_head *rb_queue;
3558         __be16 num;
3559         int err = -EINVAL;
3560         /* Added to avoid minimal code churn */
3561         struct tpacket_req *req = &req_u->req;
3562
3563         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3564         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3565                 WARN(1, "Tx-ring is not supported.\n");
3566                 goto out;
3567         }
3568
3569         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3570         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3571
3572         err = -EBUSY;
3573         if (!closing) {
3574                 if (atomic_read(&po->mapped))
3575                         goto out;
3576                 if (atomic_read(&rb->pending))
3577                         goto out;
3578         }
3579
3580         if (req->tp_block_nr) {
3581                 /* Sanity tests and some calculations */
3582                 err = -EBUSY;
3583                 if (unlikely(rb->pg_vec))
3584                         goto out;
3585
3586                 switch (po->tp_version) {
3587                 case TPACKET_V1:
3588                         po->tp_hdrlen = TPACKET_HDRLEN;
3589                         break;
3590                 case TPACKET_V2:
3591                         po->tp_hdrlen = TPACKET2_HDRLEN;
3592                         break;
3593                 case TPACKET_V3:
3594                         po->tp_hdrlen = TPACKET3_HDRLEN;
3595                         break;
3596                 }
3597
3598                 err = -EINVAL;
3599                 if (unlikely((int)req->tp_block_size <= 0))
3600                         goto out;
3601                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3602                         goto out;
3603                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3604                                         po->tp_reserve))
3605                         goto out;
3606                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3607                         goto out;
3608
3609                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3610                 if (unlikely(rb->frames_per_block <= 0))
3611                         goto out;
3612                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3613                                         req->tp_frame_nr))
3614                         goto out;
3615
3616                 err = -ENOMEM;
3617                 order = get_order(req->tp_block_size);
3618                 pg_vec = alloc_pg_vec(req, order);
3619                 if (unlikely(!pg_vec))
3620                         goto out;
3621                 switch (po->tp_version) {
3622                 case TPACKET_V3:
3623                 /* Transmit path is not supported. We checked
3624                  * it above but just being paranoid
3625                  */
3626                         if (!tx_ring)
3627                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3628                                 break;
3629                 default:
3630                         break;
3631                 }
3632         }
3633         /* Done */
3634         else {
3635                 err = -EINVAL;
3636                 if (unlikely(req->tp_frame_nr))
3637                         goto out;
3638         }
3639
3640         lock_sock(sk);
3641
3642         /* Detach socket from network */
3643         spin_lock(&po->bind_lock);
3644         was_running = po->running;
3645         num = po->num;
3646         if (was_running) {
3647                 po->num = 0;
3648                 __unregister_prot_hook(sk, false);
3649         }
3650         spin_unlock(&po->bind_lock);
3651
3652         synchronize_net();
3653
3654         err = -EBUSY;
3655         mutex_lock(&po->pg_vec_lock);
3656         if (closing || atomic_read(&po->mapped) == 0) {
3657                 err = 0;
3658                 spin_lock_bh(&rb_queue->lock);
3659                 swap(rb->pg_vec, pg_vec);
3660                 rb->frame_max = (req->tp_frame_nr - 1);
3661                 rb->head = 0;
3662                 rb->frame_size = req->tp_frame_size;
3663                 spin_unlock_bh(&rb_queue->lock);
3664
3665                 swap(rb->pg_vec_order, order);
3666                 swap(rb->pg_vec_len, req->tp_block_nr);
3667
3668                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3669                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3670                                                 tpacket_rcv : packet_rcv;
3671                 skb_queue_purge(rb_queue);
3672                 if (atomic_read(&po->mapped))
3673                         pr_err("packet_mmap: vma is busy: %d\n",
3674                                atomic_read(&po->mapped));
3675         }
3676         mutex_unlock(&po->pg_vec_lock);
3677
3678         spin_lock(&po->bind_lock);
3679         if (was_running) {
3680                 po->num = num;
3681                 register_prot_hook(sk);
3682         }
3683         spin_unlock(&po->bind_lock);
3684         if (closing && (po->tp_version > TPACKET_V2)) {
3685                 /* Because we don't support block-based V3 on tx-ring */
3686                 if (!tx_ring)
3687                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3688         }
3689         release_sock(sk);
3690
3691         if (pg_vec)
3692                 free_pg_vec(pg_vec, order, req->tp_block_nr);
3693 out:
3694         return err;
3695 }
3696
3697 static int packet_mmap(struct file *file, struct socket *sock,
3698                 struct vm_area_struct *vma)
3699 {
3700         struct sock *sk = sock->sk;
3701         struct packet_sock *po = pkt_sk(sk);
3702         unsigned long size, expected_size;
3703         struct packet_ring_buffer *rb;
3704         unsigned long start;
3705         int err = -EINVAL;
3706         int i;
3707
3708         if (vma->vm_pgoff)
3709                 return -EINVAL;
3710
3711         mutex_lock(&po->pg_vec_lock);
3712
3713         expected_size = 0;
3714         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3715                 if (rb->pg_vec) {
3716                         expected_size += rb->pg_vec_len
3717                                                 * rb->pg_vec_pages
3718                                                 * PAGE_SIZE;
3719                 }
3720         }
3721
3722         if (expected_size == 0)
3723                 goto out;
3724
3725         size = vma->vm_end - vma->vm_start;
3726         if (size != expected_size)
3727                 goto out;
3728
3729         start = vma->vm_start;
3730         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3731                 if (rb->pg_vec == NULL)
3732                         continue;
3733
3734                 for (i = 0; i < rb->pg_vec_len; i++) {
3735                         struct page *page;
3736                         void *kaddr = rb->pg_vec[i].buffer;
3737                         int pg_num;
3738
3739                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3740                                 page = pgv_to_page(kaddr);
3741                                 err = vm_insert_page(vma, start, page);
3742                                 if (unlikely(err))
3743                                         goto out;
3744                                 start += PAGE_SIZE;
3745                                 kaddr += PAGE_SIZE;
3746                         }
3747                 }
3748         }
3749
3750         atomic_inc(&po->mapped);
3751         vma->vm_ops = &packet_mmap_ops;
3752         err = 0;
3753
3754 out:
3755         mutex_unlock(&po->pg_vec_lock);
3756         return err;
3757 }
3758
3759 static const struct proto_ops packet_ops_spkt = {
3760         .family =       PF_PACKET,
3761         .owner =        THIS_MODULE,
3762         .release =      packet_release,
3763         .bind =         packet_bind_spkt,
3764         .connect =      sock_no_connect,
3765         .socketpair =   sock_no_socketpair,
3766         .accept =       sock_no_accept,
3767         .getname =      packet_getname_spkt,
3768         .poll =         datagram_poll,
3769         .ioctl =        packet_ioctl,
3770         .listen =       sock_no_listen,
3771         .shutdown =     sock_no_shutdown,
3772         .setsockopt =   sock_no_setsockopt,
3773         .getsockopt =   sock_no_getsockopt,
3774         .sendmsg =      packet_sendmsg_spkt,
3775         .recvmsg =      packet_recvmsg,
3776         .mmap =         sock_no_mmap,
3777         .sendpage =     sock_no_sendpage,
3778 };
3779
3780 static const struct proto_ops packet_ops = {
3781         .family =       PF_PACKET,
3782         .owner =        THIS_MODULE,
3783         .release =      packet_release,
3784         .bind =         packet_bind,
3785         .connect =      sock_no_connect,
3786         .socketpair =   sock_no_socketpair,
3787         .accept =       sock_no_accept,
3788         .getname =      packet_getname,
3789         .poll =         packet_poll,
3790         .ioctl =        packet_ioctl,
3791         .listen =       sock_no_listen,
3792         .shutdown =     sock_no_shutdown,
3793         .setsockopt =   packet_setsockopt,
3794         .getsockopt =   packet_getsockopt,
3795         .sendmsg =      packet_sendmsg,
3796         .recvmsg =      packet_recvmsg,
3797         .mmap =         packet_mmap,
3798         .sendpage =     sock_no_sendpage,
3799 };
3800
3801 static const struct net_proto_family packet_family_ops = {
3802         .family =       PF_PACKET,
3803         .create =       packet_create,
3804         .owner  =       THIS_MODULE,
3805 };
3806
3807 static struct notifier_block packet_netdev_notifier = {
3808         .notifier_call =        packet_notifier,
3809 };
3810
3811 #ifdef CONFIG_PROC_FS
3812
3813 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3814         __acquires(RCU)
3815 {
3816         struct net *net = seq_file_net(seq);
3817
3818         rcu_read_lock();
3819         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3820 }
3821
3822 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3823 {
3824         struct net *net = seq_file_net(seq);
3825         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3826 }
3827
3828 static void packet_seq_stop(struct seq_file *seq, void *v)
3829         __releases(RCU)
3830 {
3831         rcu_read_unlock();
3832 }
3833
3834 static int packet_seq_show(struct seq_file *seq, void *v)
3835 {
3836         if (v == SEQ_START_TOKEN)
3837                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
3838         else {
3839                 struct sock *s = sk_entry(v);
3840                 const struct packet_sock *po = pkt_sk(s);
3841
3842                 seq_printf(seq,
3843                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
3844                            s,
3845                            atomic_read(&s->sk_refcnt),
3846                            s->sk_type,
3847                            ntohs(po->num),
3848                            po->ifindex,
3849                            po->running,
3850                            atomic_read(&s->sk_rmem_alloc),
3851                            sock_i_uid(s),
3852                            sock_i_ino(s));
3853         }
3854
3855         return 0;
3856 }
3857
3858 static const struct seq_operations packet_seq_ops = {
3859         .start  = packet_seq_start,
3860         .next   = packet_seq_next,
3861         .stop   = packet_seq_stop,
3862         .show   = packet_seq_show,
3863 };
3864
3865 static int packet_seq_open(struct inode *inode, struct file *file)
3866 {
3867         return seq_open_net(inode, file, &packet_seq_ops,
3868                             sizeof(struct seq_net_private));
3869 }
3870
3871 static const struct file_operations packet_seq_fops = {
3872         .owner          = THIS_MODULE,
3873         .open           = packet_seq_open,
3874         .read           = seq_read,
3875         .llseek         = seq_lseek,
3876         .release        = seq_release_net,
3877 };
3878
3879 #endif
3880
3881 static int __net_init packet_net_init(struct net *net)
3882 {
3883         spin_lock_init(&net->packet.sklist_lock);
3884         INIT_HLIST_HEAD(&net->packet.sklist);
3885
3886         if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
3887                 return -ENOMEM;
3888
3889         return 0;
3890 }
3891
3892 static void __net_exit packet_net_exit(struct net *net)
3893 {
3894         proc_net_remove(net, "packet");
3895 }
3896
3897 static struct pernet_operations packet_net_ops = {
3898         .init = packet_net_init,
3899         .exit = packet_net_exit,
3900 };
3901
3902
3903 static void __exit packet_exit(void)
3904 {
3905         unregister_netdevice_notifier(&packet_netdev_notifier);
3906         unregister_pernet_subsys(&packet_net_ops);
3907         sock_unregister(PF_PACKET);
3908         proto_unregister(&packet_proto);
3909 }
3910
3911 static int __init packet_init(void)
3912 {
3913         int rc = proto_register(&packet_proto, 0);
3914
3915         if (rc != 0)
3916                 goto out;
3917
3918         sock_register(&packet_family_ops);
3919         register_pernet_subsys(&packet_net_ops);
3920         register_netdevice_notifier(&packet_netdev_notifier);
3921 out:
3922         return rc;
3923 }
3924
3925 module_init(packet_init);
3926 module_exit(packet_exit);
3927 MODULE_LICENSE("GPL");
3928 MODULE_ALIAS_NETPROTO(PF_PACKET);