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 net_device __rcu *cached_dev;
 299         struct packet_type      prot_hook ____cacheline_aligned_in_smp;
 300 };
 301
 302 #define PACKET_FANOUT_MAX       256
 303
 304 struct packet_fanout {
 305 #ifdef CONFIG_NET_NS
 306         struct net              *net;
 307 #endif
 308         unsigned int            num_members;
 309         u16                     id;
 310         u8                      type;
 311         u8                      defrag;
 312         atomic_t                rr_cur;
 313         struct list_head        list;
 314         struct sock             *arr[PACKET_FANOUT_MAX];
 315         spinlock_t              lock;
 316         atomic_t                sk_ref;
 317         struct packet_type      prot_hook ____cacheline_aligned_in_smp;
 318 };
 319
 320 struct packet_skb_cb {
 321         unsigned int origlen;
 322         union {
 323                 struct sockaddr_pkt pkt;
 324                 struct sockaddr_ll ll;
 325         } sa;
 326 };
 327
 328 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 329
 330 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 331 #define GET_PBLOCK_DESC(x, bid) \
 332         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 333 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 334         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 335 #define GET_NEXT_PRB_BLK_NUM(x) \
 336         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 337         ((x)->kactive_blk_num+1) : 0)
 338
 339 static struct packet_sock *pkt_sk(struct sock *sk)
 340 {
 341         return (struct packet_sock *)sk;
 342 }
 343
 344 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 345 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 346
 347 /* register_prot_hook must be invoked with the po->bind_lock held,
 348  * or from a context in which asynchronous accesses to the packet
 349  * socket is not possible (packet_create()).
 350  */
 351 static void register_prot_hook(struct sock *sk)
 352 {
 353         struct packet_sock *po = pkt_sk(sk);
 354
 355         if (!po->running) {
 356                 if (po->fanout) {
 357                         __fanout_link(sk, po);
 358                 } else {
 359                         dev_add_pack(&po->prot_hook);
 360                         rcu_assign_pointer(po->cached_dev, po->prot_hook.dev);
 361                 }
 362
 363                 sock_hold(sk);
 364                 po->running = 1;
 365         }
 366 }
 367
 368 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
 369  * held.   If the sync parameter is true, we will temporarily drop
 370  * the po->bind_lock and do a synchronize_net to make sure no
 371  * asynchronous packet processing paths still refer to the elements
 372  * of po->prot_hook.  If the sync parameter is false, it is the
 373  * callers responsibility to take care of this.
 374  */
 375 static void __unregister_prot_hook(struct sock *sk, bool sync)
 376 {
 377         struct packet_sock *po = pkt_sk(sk);
 378
 379         po->running = 0;
 380         if (po->fanout) {
 381                 __fanout_unlink(sk, po);
 382         } else {
 383                 __dev_remove_pack(&po->prot_hook);
 384                 RCU_INIT_POINTER(po->cached_dev, NULL);
 385         }
 386
 387         __sock_put(sk);
 388
 389         if (sync) {
 390                 spin_unlock(&po->bind_lock);
 391                 synchronize_net();
 392                 spin_lock(&po->bind_lock);
 393         }
 394 }
 395
 396 static void unregister_prot_hook(struct sock *sk, bool sync)
 397 {
 398         struct packet_sock *po = pkt_sk(sk);
 399
 400         if (po->running)
 401                 __unregister_prot_hook(sk, sync);
 402 }
 403
 404 static inline __pure struct page *pgv_to_page(void *addr)
 405 {
 406         if (is_vmalloc_addr(addr))
 407                 return vmalloc_to_page(addr);
 408         return virt_to_page(addr);
 409 }
 410
 411 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 412 {
 413         union {
 414                 struct tpacket_hdr *h1;
 415                 struct tpacket2_hdr *h2;
 416                 void *raw;
 417         } h;
 418
 419         h.raw = frame;
 420         switch (po->tp_version) {
 421         case TPACKET_V1:
 422                 h.h1->tp_status = status;
 423                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 424                 break;
 425         case TPACKET_V2:
 426                 h.h2->tp_status = status;
 427                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 428                 break;
 429         case TPACKET_V3:
 430         default:
 431                 WARN(1, "TPACKET version not supported.\n");
 432                 BUG();
 433         }
 434
 435         smp_wmb();
 436 }
 437
 438 static int __packet_get_status(struct packet_sock *po, void *frame)
 439 {
 440         union {
 441                 struct tpacket_hdr *h1;
 442                 struct tpacket2_hdr *h2;
 443                 void *raw;
 444         } h;
 445
 446         smp_rmb();
 447
 448         h.raw = frame;
 449         switch (po->tp_version) {
 450         case TPACKET_V1:
 451                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 452                 return h.h1->tp_status;
 453         case TPACKET_V2:
 454                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 455                 return h.h2->tp_status;
 456         case TPACKET_V3:
 457         default:
 458                 WARN(1, "TPACKET version not supported.\n");
 459                 BUG();
 460                 return 0;
 461         }
 462 }
 463
 464 static void *packet_lookup_frame(struct packet_sock *po,
 465                 struct packet_ring_buffer *rb,
 466                 unsigned int position,
 467                 int status)
 468 {
 469         unsigned int pg_vec_pos, frame_offset;
 470         union {
 471                 struct tpacket_hdr *h1;
 472                 struct tpacket2_hdr *h2;
 473                 void *raw;
 474         } h;
 475
 476         pg_vec_pos = position / rb->frames_per_block;
 477         frame_offset = position % rb->frames_per_block;
 478
 479         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 480                 (frame_offset * rb->frame_size);
 481
 482         if (status != __packet_get_status(po, h.raw))
 483                 return NULL;
 484
 485         return h.raw;
 486 }
 487
 488 static void *packet_current_frame(struct packet_sock *po,
 489                 struct packet_ring_buffer *rb,
 490                 int status)
 491 {
 492         return packet_lookup_frame(po, rb, rb->head, status);
 493 }
 494
 495 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 496 {
 497         del_timer_sync(&pkc->retire_blk_timer);
 498 }
 499
 500 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 501                 int tx_ring,
 502                 struct sk_buff_head *rb_queue)
 503 {
 504         struct tpacket_kbdq_core *pkc;
 505
 506         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 507
 508         spin_lock_bh(&rb_queue->lock);
 509         pkc->delete_blk_timer = 1;
 510         spin_unlock_bh(&rb_queue->lock);
 511
 512         prb_del_retire_blk_timer(pkc);
 513 }
 514
 515 static void prb_init_blk_timer(struct packet_sock *po,
 516                 struct tpacket_kbdq_core *pkc,
 517                 void (*func) (unsigned long))
 518 {
 519         init_timer(&pkc->retire_blk_timer);
 520         pkc->retire_blk_timer.data = (long)po;
 521         pkc->retire_blk_timer.function = func;
 522         pkc->retire_blk_timer.expires = jiffies;
 523 }
 524
 525 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
 526 {
 527         struct tpacket_kbdq_core *pkc;
 528
 529         if (tx_ring)
 530                 BUG();
 531
 532         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 533         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
 534 }
 535
 536 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 537                                 int blk_size_in_bytes)
 538 {
 539         struct net_device *dev;
 540         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 541         struct ethtool_cmd ecmd;
 542         int err;
 543
 544         rtnl_lock();
 545         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 546         if (unlikely(!dev)) {
 547                 rtnl_unlock();
 548                 return DEFAULT_PRB_RETIRE_TOV;
 549         }
 550         err = __ethtool_get_settings(dev, &ecmd);
 551         rtnl_unlock();
 552         if (!err) {
 553                 switch (ecmd.speed) {
 554                 case SPEED_10000:
 555                         msec = 1;
 556                         div = 10000/1000;
 557                         break;
 558                 case SPEED_1000:
 559                         msec = 1;
 560                         div = 1000/1000;
 561                         break;
 562                 /*
 563                  * If the link speed is so slow you don't really
 564                  * need to worry about perf anyways
 565                  */
 566                 case SPEED_100:
 567                 case SPEED_10:
 568                 default:
 569                         return DEFAULT_PRB_RETIRE_TOV;
 570                 }
 571         }
 572
 573         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 574
 575         if (div)
 576                 mbits /= div;
 577
 578         tmo = mbits * msec;
 579
 580         if (div)
 581                 return tmo+1;
 582         return tmo;
 583 }
 584
 585 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 586                         union tpacket_req_u *req_u)
 587 {
 588         p1->feature_req_word = req_u->req3.tp_feature_req_word;
 589 }
 590
 591 static void init_prb_bdqc(struct packet_sock *po,
 592                         struct packet_ring_buffer *rb,
 593                         struct pgv *pg_vec,
 594                         union tpacket_req_u *req_u, int tx_ring)
 595 {
 596         struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
 597         struct tpacket_block_desc *pbd;
 598
 599         memset(p1, 0x0, sizeof(*p1));
 600
 601         p1->knxt_seq_num = 1;
 602         p1->pkbdq = pg_vec;
 603         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 604         p1->pkblk_start = (char *)pg_vec[0].buffer;
 605         p1->kblk_size = req_u->req3.tp_block_size;
 606         p1->knum_blocks = req_u->req3.tp_block_nr;
 607         p1->hdrlen = po->tp_hdrlen;
 608         p1->version = po->tp_version;
 609         p1->last_kactive_blk_num = 0;
 610         po->stats_u.stats3.tp_freeze_q_cnt = 0;
 611         if (req_u->req3.tp_retire_blk_tov)
 612                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 613         else
 614                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 615                                                 req_u->req3.tp_block_size);
 616         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 617         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 618
 619         prb_init_ft_ops(p1, req_u);
 620         prb_setup_retire_blk_timer(po, tx_ring);
 621         prb_open_block(p1, pbd);
 622 }
 623
 624 /*  Do NOT update the last_blk_num first.
 625  *  Assumes sk_buff_head lock is held.
 626  */
 627 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 628 {
 629         mod_timer(&pkc->retire_blk_timer,
 630                         jiffies + pkc->tov_in_jiffies);
 631         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 632 }
 633
 634 /*
 635  * Timer logic:
 636  * 1) We refresh the timer only when we open a block.
 637  *    By doing this we don't waste cycles refreshing the timer
 638  *        on packet-by-packet basis.
 639  *
 640  * With a 1MB block-size, on a 1Gbps line, it will take
 641  * i) ~8 ms to fill a block + ii) memcpy etc.
 642  * In this cut we are not accounting for the memcpy time.
 643  *
 644  * So, if the user sets the 'tmo' to 10ms then the timer
 645  * will never fire while the block is still getting filled
 646  * (which is what we want). However, the user could choose
 647  * to close a block early and that's fine.
 648  *
 649  * But when the timer does fire, we check whether or not to refresh it.
 650  * Since the tmo granularity is in msecs, it is not too expensive
 651  * to refresh the timer, lets say every '8' msecs.
 652  * Either the user can set the 'tmo' or we can derive it based on
 653  * a) line-speed and b) block-size.
 654  * prb_calc_retire_blk_tmo() calculates the tmo.
 655  *
 656  */
 657 static void prb_retire_rx_blk_timer_expired(unsigned long data)
 658 {
 659         struct packet_sock *po = (struct packet_sock *)data;
 660         struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
 661         unsigned int frozen;
 662         struct tpacket_block_desc *pbd;
 663
 664         spin_lock(&po->sk.sk_receive_queue.lock);
 665
 666         frozen = prb_queue_frozen(pkc);
 667         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 668
 669         if (unlikely(pkc->delete_blk_timer))
 670                 goto out;
 671
 672         /* We only need to plug the race when the block is partially filled.
 673          * tpacket_rcv:
 674          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 675          *              copy_bits() is in progress ...
 676          *              timer fires on other cpu:
 677          *              we can't retire the current block because copy_bits
 678          *              is in progress.
 679          *
 680          */
 681         if (BLOCK_NUM_PKTS(pbd)) {
 682                 while (atomic_read(&pkc->blk_fill_in_prog)) {
 683                         /* Waiting for skb_copy_bits to finish... */
 684                         cpu_relax();
 685                 }
 686         }
 687
 688         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 689                 if (!frozen) {
 690                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 691                         if (!prb_dispatch_next_block(pkc, po))
 692                                 goto refresh_timer;
 693                         else
 694                                 goto out;
 695                 } else {
 696                         /* Case 1. Queue was frozen because user-space was
 697                          *         lagging behind.
 698                          */
 699                         if (prb_curr_blk_in_use(pkc, pbd)) {
 700                                 /*
 701                                  * Ok, user-space is still behind.
 702                                  * So just refresh the timer.
 703                                  */
 704                                 goto refresh_timer;
 705                         } else {
 706                                /* Case 2. queue was frozen,user-space caught up,
 707                                 * now the link went idle && the timer fired.
 708                                 * We don't have a block to close.So we open this
 709                                 * block and restart the timer.
 710                                 * opening a block thaws the queue,restarts timer
 711                                 * Thawing/timer-refresh is a side effect.
 712                                 */
 713                                 prb_open_block(pkc, pbd);
 714                                 goto out;
 715                         }
 716                 }
 717         }
 718
 719 refresh_timer:
 720         _prb_refresh_rx_retire_blk_timer(pkc);
 721
 722 out:
 723         spin_unlock(&po->sk.sk_receive_queue.lock);
 724 }
 725
 726 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 727                 struct tpacket_block_desc *pbd1, __u32 status)
 728 {
 729         /* Flush everything minus the block header */
 730
 731 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 732         u8 *start, *end;
 733
 734         start = (u8 *)pbd1;
 735
 736         /* Skip the block header(we know header WILL fit in 4K) */
 737         start += PAGE_SIZE;
 738
 739         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 740         for (; start < end; start += PAGE_SIZE)
 741                 flush_dcache_page(pgv_to_page(start));
 742
 743         smp_wmb();
 744 #endif
 745
 746         /* Now update the block status. */
 747
 748         BLOCK_STATUS(pbd1) = status;
 749
 750         /* Flush the block header */
 751
 752 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 753         start = (u8 *)pbd1;
 754         flush_dcache_page(pgv_to_page(start));
 755
 756         smp_wmb();
 757 #endif
 758 }
 759
 760 /*
 761  * Side effect:
 762  *
 763  * 1) flush the block
 764  * 2) Increment active_blk_num
 765  *
 766  * Note:We DONT refresh the timer on purpose.
 767  *      Because almost always the next block will be opened.
 768  */
 769 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 770                 struct tpacket_block_desc *pbd1,
 771                 struct packet_sock *po, unsigned int stat)
 772 {
 773         __u32 status = TP_STATUS_USER | stat;
 774
 775         struct tpacket3_hdr *last_pkt;
 776         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 777
 778         if (po->stats.tp_drops)
 779                 status |= TP_STATUS_LOSING;
 780
 781         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 782         last_pkt->tp_next_offset = 0;
 783
 784         /* Get the ts of the last pkt */
 785         if (BLOCK_NUM_PKTS(pbd1)) {
 786                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 787                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 788         } else {
 789                 /* Ok, we tmo'd - so get the current time */
 790                 struct timespec ts;
 791                 getnstimeofday(&ts);
 792                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 793                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 794         }
 795
 796         smp_wmb();
 797
 798         /* Flush the block */
 799         prb_flush_block(pkc1, pbd1, status);
 800
 801         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 802 }
 803
 804 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 805 {
 806         pkc->reset_pending_on_curr_blk = 0;
 807 }
 808
 809 /*
 810  * Side effect of opening a block:
 811  *
 812  * 1) prb_queue is thawed.
 813  * 2) retire_blk_timer is refreshed.
 814  *
 815  */
 816 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 817         struct tpacket_block_desc *pbd1)
 818 {
 819         struct timespec ts;
 820         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 821
 822         smp_rmb();
 823
 824         /* We could have just memset this but we will lose the
 825          * flexibility of making the priv area sticky
 826          */
 827         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 828         BLOCK_NUM_PKTS(pbd1) = 0;
 829         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 830         getnstimeofday(&ts);
 831         h1->ts_first_pkt.ts_sec = ts.tv_sec;
 832         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 833         pkc1->pkblk_start = (char *)pbd1;
 834         pkc1->nxt_offset = (char *)(pkc1->pkblk_start +
 835                                     BLK_PLUS_PRIV(pkc1->blk_sizeof_priv));
 836         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 837         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 838         pbd1->version = pkc1->version;
 839         pkc1->prev = pkc1->nxt_offset;
 840         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 841         prb_thaw_queue(pkc1);
 842         _prb_refresh_rx_retire_blk_timer(pkc1);
 843
 844         smp_wmb();
 845 }
 846
 847 /*
 848  * Queue freeze logic:
 849  * 1) Assume tp_block_nr = 8 blocks.
 850  * 2) At time 't0', user opens Rx ring.
 851  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 852  * 4) user-space is either sleeping or processing block '0'.
 853  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 854  *    it will close block-7,loop around and try to fill block '0'.
 855  *    call-flow:
 856  *    __packet_lookup_frame_in_block
 857  *      prb_retire_current_block()
 858  *      prb_dispatch_next_block()
 859  *        |->(BLOCK_STATUS == USER) evaluates to true
 860  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 861  * 6) Now there are two cases:
 862  *    6.1) Link goes idle right after the queue is frozen.
 863  *         But remember, the last open_block() refreshed the timer.
 864  *         When this timer expires,it will refresh itself so that we can
 865  *         re-open block-0 in near future.
 866  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 867  *         case and __packet_lookup_frame_in_block will check if block-0
 868  *         is free and can now be re-used.
 869  */
 870 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 871                                   struct packet_sock *po)
 872 {
 873         pkc->reset_pending_on_curr_blk = 1;
 874         po->stats_u.stats3.tp_freeze_q_cnt++;
 875 }
 876
 877 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 878
 879 /*
 880  * If the next block is free then we will dispatch it
 881  * and return a good offset.
 882  * Else, we will freeze the queue.
 883  * So, caller must check the return value.
 884  */
 885 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 886                 struct packet_sock *po)
 887 {
 888         struct tpacket_block_desc *pbd;
 889
 890         smp_rmb();
 891
 892         /* 1. Get current block num */
 893         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 894
 895         /* 2. If this block is currently in_use then freeze the queue */
 896         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 897                 prb_freeze_queue(pkc, po);
 898                 return NULL;
 899         }
 900
 901         /*
 902          * 3.
 903          * open this block and return the offset where the first packet
 904          * needs to get stored.
 905          */
 906         prb_open_block(pkc, pbd);
 907         return (void *)pkc->nxt_offset;
 908 }
 909
 910 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 911                 struct packet_sock *po, unsigned int status)
 912 {
 913         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 914
 915         /* retire/close the current block */
 916         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 917                 /*
 918                  * Plug the case where copy_bits() is in progress on
 919                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 920                  * have space to copy the pkt in the current block and
 921                  * called prb_retire_current_block()
 922                  *
 923                  * We don't need to worry about the TMO case because
 924                  * the timer-handler already handled this case.
 925                  */
 926                 if (!(status & TP_STATUS_BLK_TMO)) {
 927                         while (atomic_read(&pkc->blk_fill_in_prog)) {
 928                                 /* Waiting for skb_copy_bits to finish... */
 929                                 cpu_relax();
 930                         }
 931                 }
 932                 prb_close_block(pkc, pbd, po, status);
 933                 return;
 934         }
 935 }
 936
 937 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
 938                                       struct tpacket_block_desc *pbd)
 939 {
 940         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 941 }
 942
 943 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 944 {
 945         return pkc->reset_pending_on_curr_blk;
 946 }
 947
 948 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 949 {
 950         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 951         atomic_dec(&pkc->blk_fill_in_prog);
 952 }
 953
 954 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 955                         struct tpacket3_hdr *ppd)
 956 {
 957         ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
 958 }
 959
 960 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 961                         struct tpacket3_hdr *ppd)
 962 {
 963         ppd->hv1.tp_rxhash = 0;
 964 }
 965
 966 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
 967                         struct tpacket3_hdr *ppd)
 968 {
 969         if (vlan_tx_tag_present(pkc->skb)) {
 970                 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
 971                 ppd->tp_status = TP_STATUS_VLAN_VALID;
 972         } else {
 973                 ppd->hv1.tp_vlan_tci = ppd->tp_status = 0;
 974         }
 975 }
 976
 977 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
 978                         struct tpacket3_hdr *ppd)
 979 {
 980         prb_fill_vlan_info(pkc, ppd);
 981
 982         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
 983                 prb_fill_rxhash(pkc, ppd);
 984         else
 985                 prb_clear_rxhash(pkc, ppd);
 986 }
 987
 988 static void prb_fill_curr_block(char *curr,
 989                                 struct tpacket_kbdq_core *pkc,
 990                                 struct tpacket_block_desc *pbd,
 991                                 unsigned int len)
 992 {
 993         struct tpacket3_hdr *ppd;
 994
 995         ppd  = (struct tpacket3_hdr *)curr;
 996         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
 997         pkc->prev = curr;
 998         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
 999         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1000         BLOCK_NUM_PKTS(pbd) += 1;
1001         atomic_inc(&pkc->blk_fill_in_prog);
1002         prb_run_all_ft_ops(pkc, ppd);
1003 }
1004
1005 /* Assumes caller has the sk->rx_queue.lock */
1006 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1007                                             struct sk_buff *skb,
1008                                                 int status,
1009                                             unsigned int len
1010                                             )
1011 {
1012         struct tpacket_kbdq_core *pkc;
1013         struct tpacket_block_desc *pbd;
1014         char *curr, *end;
1015
1016         pkc = GET_PBDQC_FROM_RB(((struct packet_ring_buffer *)&po->rx_ring));
1017         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1018
1019         /* Queue is frozen when user space is lagging behind */
1020         if (prb_queue_frozen(pkc)) {
1021                 /*
1022                  * Check if that last block which caused the queue to freeze,
1023                  * is still in_use by user-space.
1024                  */
1025                 if (prb_curr_blk_in_use(pkc, pbd)) {
1026                         /* Can't record this packet */
1027                         return NULL;
1028                 } else {
1029                         /*
1030                          * Ok, the block was released by user-space.
1031                          * Now let's open that block.
1032                          * opening a block also thaws the queue.
1033                          * Thawing is a side effect.
1034                          */
1035                         prb_open_block(pkc, pbd);
1036                 }
1037         }
1038
1039         smp_mb();
1040         curr = pkc->nxt_offset;
1041         pkc->skb = skb;
1042         end = (char *) ((char *)pbd + pkc->kblk_size);
1043
1044         /* first try the current block */
1045         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1046                 prb_fill_curr_block(curr, pkc, pbd, len);
1047                 return (void *)curr;
1048         }
1049
1050         /* Ok, close the current block */
1051         prb_retire_current_block(pkc, po, 0);
1052
1053         /* Now, try to dispatch the next block */
1054         curr = (char *)prb_dispatch_next_block(pkc, po);
1055         if (curr) {
1056                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1057                 prb_fill_curr_block(curr, pkc, pbd, len);
1058                 return (void *)curr;
1059         }
1060
1061         /*
1062          * No free blocks are available.user_space hasn't caught up yet.
1063          * Queue was just frozen and now this packet will get dropped.
1064          */
1065         return NULL;
1066 }
1067
1068 static void *packet_current_rx_frame(struct packet_sock *po,
1069                                             struct sk_buff *skb,
1070                                             int status, unsigned int len)
1071 {
1072         char *curr = NULL;
1073         switch (po->tp_version) {
1074         case TPACKET_V1:
1075         case TPACKET_V2:
1076                 curr = packet_lookup_frame(po, &po->rx_ring,
1077                                         po->rx_ring.head, status);
1078                 return curr;
1079         case TPACKET_V3:
1080                 return __packet_lookup_frame_in_block(po, skb, status, len);
1081         default:
1082                 WARN(1, "TPACKET version not supported\n");
1083                 BUG();
1084                 return 0;
1085         }
1086 }
1087
1088 static void *prb_lookup_block(struct packet_sock *po,
1089                                      struct packet_ring_buffer *rb,
1090                                      unsigned int previous,
1091                                      int status)
1092 {
1093         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1094         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, previous);
1095
1096         if (status != BLOCK_STATUS(pbd))
1097                 return NULL;
1098         return pbd;
1099 }
1100
1101 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1102 {
1103         unsigned int prev;
1104         if (rb->prb_bdqc.kactive_blk_num)
1105                 prev = rb->prb_bdqc.kactive_blk_num-1;
1106         else
1107                 prev = rb->prb_bdqc.knum_blocks-1;
1108         return prev;
1109 }
1110
1111 /* Assumes caller has held the rx_queue.lock */
1112 static void *__prb_previous_block(struct packet_sock *po,
1113                                          struct packet_ring_buffer *rb,
1114                                          int status)
1115 {
1116         unsigned int previous = prb_previous_blk_num(rb);
1117         return prb_lookup_block(po, rb, previous, status);
1118 }
1119
1120 static void *packet_previous_rx_frame(struct packet_sock *po,
1121                                              struct packet_ring_buffer *rb,
1122                                              int status)
1123 {
1124         if (po->tp_version <= TPACKET_V2)
1125                 return packet_previous_frame(po, rb, status);
1126
1127         return __prb_previous_block(po, rb, status);
1128 }
1129
1130 static void packet_increment_rx_head(struct packet_sock *po,
1131                                             struct packet_ring_buffer *rb)
1132 {
1133         switch (po->tp_version) {
1134         case TPACKET_V1:
1135         case TPACKET_V2:
1136                 return packet_increment_head(rb);
1137         case TPACKET_V3:
1138         default:
1139                 WARN(1, "TPACKET version not supported.\n");
1140                 BUG();
1141                 return;
1142         }
1143 }
1144
1145 static void *packet_previous_frame(struct packet_sock *po,
1146                 struct packet_ring_buffer *rb,
1147                 int status)
1148 {
1149         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1150         return packet_lookup_frame(po, rb, previous, status);
1151 }
1152
1153 static void packet_increment_head(struct packet_ring_buffer *buff)
1154 {
1155         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1156 }
1157
1158 static void packet_sock_destruct(struct sock *sk)
1159 {
1160         skb_queue_purge(&sk->sk_error_queue);
1161
1162         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1163         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1164
1165         if (!sock_flag(sk, SOCK_DEAD)) {
1166                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1167                 return;
1168         }
1169
1170         sk_refcnt_debug_dec(sk);
1171 }
1172
1173 static struct sock *fanout_demux_hash(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1174 {
1175         u32 idx, hash = skb->rxhash;
1176
1177         idx = ((u64)hash * num) >> 32;
1178
1179         return f->arr[idx];
1180 }
1181
1182 static struct sock *fanout_demux_lb(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1183 {
1184         unsigned int val = atomic_inc_return(&f->rr_cur);
1185
1186         return f->arr[val % num];
1187 }
1188
1189 static struct sock *fanout_demux_cpu(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1190 {
1191         unsigned int cpu = smp_processor_id();
1192
1193         return f->arr[cpu % num];
1194 }
1195
1196 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1197                              struct packet_type *pt, struct net_device *orig_dev)
1198 {
1199         struct packet_fanout *f = pt->af_packet_priv;
1200         unsigned int num = ACCESS_ONCE(f->num_members);
1201         struct packet_sock *po;
1202         struct sock *sk;
1203
1204         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1205             !num) {
1206                 kfree_skb(skb);
1207                 return 0;
1208         }
1209
1210         switch (f->type) {
1211         case PACKET_FANOUT_HASH:
1212         default:
1213                 if (f->defrag) {
1214                         skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1215                         if (!skb)
1216                                 return 0;
1217                 }
1218                 skb_get_rxhash(skb);
1219                 sk = fanout_demux_hash(f, skb, num);
1220                 break;
1221         case PACKET_FANOUT_LB:
1222                 sk = fanout_demux_lb(f, skb, num);
1223                 break;
1224         case PACKET_FANOUT_CPU:
1225                 sk = fanout_demux_cpu(f, skb, num);
1226                 break;
1227         }
1228
1229         po = pkt_sk(sk);
1230
1231         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1232 }
1233
1234 static DEFINE_MUTEX(fanout_mutex);
1235 static LIST_HEAD(fanout_list);
1236
1237 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1238 {
1239         struct packet_fanout *f = po->fanout;
1240
1241         spin_lock(&f->lock);
1242         f->arr[f->num_members] = sk;
1243         smp_wmb();
1244         f->num_members++;
1245         if (f->num_members == 1)
1246                 dev_add_pack(&f->prot_hook);
1247         spin_unlock(&f->lock);
1248 }
1249
1250 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1251 {
1252         struct packet_fanout *f = po->fanout;
1253         int i;
1254
1255         spin_lock(&f->lock);
1256         for (i = 0; i < f->num_members; i++) {
1257                 if (f->arr[i] == sk)
1258                         break;
1259         }
1260         BUG_ON(i >= f->num_members);
1261         f->arr[i] = f->arr[f->num_members - 1];
1262         f->num_members--;
1263         if (f->num_members == 0)
1264                 __dev_remove_pack(&f->prot_hook);
1265         spin_unlock(&f->lock);
1266 }
1267
1268 bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1269 {
1270         if (sk->sk_family != PF_PACKET)
1271                 return false;
1272
1273         return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1274 }
1275
1276 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1277 {
1278         struct packet_sock *po = pkt_sk(sk);
1279         struct packet_fanout *f, *match;
1280         u8 type = type_flags & 0xff;
1281         u8 defrag = (type_flags & PACKET_FANOUT_FLAG_DEFRAG) ? 1 : 0;
1282         int err;
1283
1284         switch (type) {
1285         case PACKET_FANOUT_HASH:
1286         case PACKET_FANOUT_LB:
1287         case PACKET_FANOUT_CPU:
1288                 break;
1289         default:
1290                 return -EINVAL;
1291         }
1292
1293         mutex_lock(&fanout_mutex);
1294
1295         err = -EINVAL;
1296         if (!po->running)
1297                 goto out;
1298
1299         err = -EALREADY;
1300         if (po->fanout)
1301                 goto out;
1302
1303         match = NULL;
1304         list_for_each_entry(f, &fanout_list, list) {
1305                 if (f->id == id &&
1306                     read_pnet(&f->net) == sock_net(sk)) {
1307                         match = f;
1308                         break;
1309                 }
1310         }
1311         err = -EINVAL;
1312         if (match && match->defrag != defrag)
1313                 goto out;
1314         if (!match) {
1315                 err = -ENOMEM;
1316                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1317                 if (!match)
1318                         goto out;
1319                 write_pnet(&match->net, sock_net(sk));
1320                 match->id = id;
1321                 match->type = type;
1322                 match->defrag = defrag;
1323                 atomic_set(&match->rr_cur, 0);
1324                 INIT_LIST_HEAD(&match->list);
1325                 spin_lock_init(&match->lock);
1326                 atomic_set(&match->sk_ref, 0);
1327                 match->prot_hook.type = po->prot_hook.type;
1328                 match->prot_hook.dev = po->prot_hook.dev;
1329                 match->prot_hook.func = packet_rcv_fanout;
1330                 match->prot_hook.af_packet_priv = match;
1331                 match->prot_hook.id_match = match_fanout_group;
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 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1353  * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1354  * It is the responsibility of the caller to call fanout_release_data() and
1355  * free the returned packet_fanout (after synchronize_net())
1356  */
1357 static struct packet_fanout *fanout_release(struct sock *sk)
1358 {
1359         struct packet_sock *po = pkt_sk(sk);
1360         struct packet_fanout *f;
1361
1362         mutex_lock(&fanout_mutex);
1363         f = po->fanout;
1364         if (f) {
1365                 po->fanout = NULL;
1366
1367                 if (atomic_dec_and_test(&f->sk_ref))
1368                         list_del(&f->list);
1369                 else
1370                         f = NULL;
1371         }
1372         mutex_unlock(&fanout_mutex);
1373
1374         return f;
1375 }
1376
1377 static const struct proto_ops packet_ops;
1378
1379 static const struct proto_ops packet_ops_spkt;
1380
1381 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1382                            struct packet_type *pt, struct net_device *orig_dev)
1383 {
1384         struct sock *sk;
1385         struct sockaddr_pkt *spkt;
1386
1387         /*
1388          *      When we registered the protocol we saved the socket in the data
1389          *      field for just this event.
1390          */
1391
1392         sk = pt->af_packet_priv;
1393
1394         /*
1395          *      Yank back the headers [hope the device set this
1396          *      right or kerboom...]
1397          *
1398          *      Incoming packets have ll header pulled,
1399          *      push it back.
1400          *
1401          *      For outgoing ones skb->data == skb_mac_header(skb)
1402          *      so that this procedure is noop.
1403          */
1404
1405         if (skb->pkt_type == PACKET_LOOPBACK)
1406                 goto out;
1407
1408         if (!net_eq(dev_net(dev), sock_net(sk)))
1409                 goto out;
1410
1411         skb = skb_share_check(skb, GFP_ATOMIC);
1412         if (skb == NULL)
1413                 goto oom;
1414
1415         /* drop any routing info */
1416         skb_dst_drop(skb);
1417
1418         /* drop conntrack reference */
1419         nf_reset(skb);
1420
1421         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1422
1423         skb_push(skb, skb->data - skb_mac_header(skb));
1424
1425         /*
1426          *      The SOCK_PACKET socket receives _all_ frames.
1427          */
1428
1429         spkt->spkt_family = dev->type;
1430         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1431         spkt->spkt_protocol = skb->protocol;
1432
1433         /*
1434          *      Charge the memory to the socket. This is done specifically
1435          *      to prevent sockets using all the memory up.
1436          */
1437
1438         if (sock_queue_rcv_skb(sk, skb) == 0)
1439                 return 0;
1440
1441 out:
1442         kfree_skb(skb);
1443 oom:
1444         return 0;
1445 }
1446
1447
1448 /*
1449  *      Output a raw packet to a device layer. This bypasses all the other
1450  *      protocol layers and you must therefore supply it with a complete frame
1451  */
1452
1453 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1454                                struct msghdr *msg, size_t len)
1455 {
1456         struct sock *sk = sock->sk;
1457         struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1458         struct sk_buff *skb = NULL;
1459         struct net_device *dev;
1460         __be16 proto = 0;
1461         int err;
1462
1463         /*
1464          *      Get and verify the address.
1465          */
1466
1467         if (saddr) {
1468                 if (msg->msg_namelen < sizeof(struct sockaddr))
1469                         return -EINVAL;
1470                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1471                         proto = saddr->spkt_protocol;
1472         } else
1473                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1474
1475         /*
1476          *      Find the device first to size check it
1477          */
1478
1479         saddr->spkt_device[13] = 0;
1480 retry:
1481         rcu_read_lock();
1482         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1483         err = -ENODEV;
1484         if (dev == NULL)
1485                 goto out_unlock;
1486
1487         err = -ENETDOWN;
1488         if (!(dev->flags & IFF_UP))
1489                 goto out_unlock;
1490
1491         /*
1492          * You may not queue a frame bigger than the mtu. This is the lowest level
1493          * raw protocol and you must do your own fragmentation at this level.
1494          */
1495
1496         err = -EMSGSIZE;
1497         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN)
1498                 goto out_unlock;
1499
1500         if (!skb) {
1501                 size_t reserved = LL_RESERVED_SPACE(dev);
1502                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1503
1504                 rcu_read_unlock();
1505                 skb = sock_wmalloc(sk, len + reserved, 0, GFP_KERNEL);
1506                 if (skb == NULL)
1507                         return -ENOBUFS;
1508                 /* FIXME: Save some space for broken drivers that write a hard
1509                  * header at transmission time by themselves. PPP is the notable
1510                  * one here. This should really be fixed at the driver level.
1511                  */
1512                 skb_reserve(skb, reserved);
1513                 skb_reset_network_header(skb);
1514
1515                 /* Try to align data part correctly */
1516                 if (hhlen) {
1517                         skb->data -= hhlen;
1518                         skb->tail -= hhlen;
1519                         if (len < hhlen)
1520                                 skb_reset_network_header(skb);
1521                 }
1522                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1523                 if (err)
1524                         goto out_free;
1525                 goto retry;
1526         }
1527
1528         if (len > (dev->mtu + dev->hard_header_len)) {
1529                 /* Earlier code assumed this would be a VLAN pkt,
1530                  * double-check this now that we have the actual
1531                  * packet in hand.
1532                  */
1533                 struct ethhdr *ehdr;
1534                 skb_reset_mac_header(skb);
1535                 ehdr = eth_hdr(skb);
1536                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1537                         err = -EMSGSIZE;
1538                         goto out_unlock;
1539                 }
1540         }
1541
1542         skb->protocol = proto;
1543         skb->dev = dev;
1544         skb->priority = sk->sk_priority;
1545         skb->mark = sk->sk_mark;
1546         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1547         if (err < 0)
1548                 goto out_unlock;
1549
1550         dev_queue_xmit(skb);
1551         rcu_read_unlock();
1552         return len;
1553
1554 out_unlock:
1555         rcu_read_unlock();
1556 out_free:
1557         kfree_skb(skb);
1558         return err;
1559 }
1560
1561 static unsigned int run_filter(const struct sk_buff *skb,
1562                                       const struct sock *sk,
1563                                       unsigned int res)
1564 {
1565         struct sk_filter *filter;
1566
1567         rcu_read_lock();
1568         filter = rcu_dereference(sk->sk_filter);
1569         if (filter != NULL)
1570                 res = SK_RUN_FILTER(filter, skb);
1571         rcu_read_unlock();
1572
1573         return res;
1574 }
1575
1576 /*
1577  * This function makes lazy skb cloning in hope that most of packets
1578  * are discarded by BPF.
1579  *
1580  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1581  * and skb->cb are mangled. It works because (and until) packets
1582  * falling here are owned by current CPU. Output packets are cloned
1583  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1584  * sequencially, so that if we return skb to original state on exit,
1585  * we will not harm anyone.
1586  */
1587
1588 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1589                       struct packet_type *pt, struct net_device *orig_dev)
1590 {
1591         struct sock *sk;
1592         struct sockaddr_ll *sll;
1593         struct packet_sock *po;
1594         u8 *skb_head = skb->data;
1595         int skb_len = skb->len;
1596         unsigned int snaplen, res;
1597
1598         if (skb->pkt_type == PACKET_LOOPBACK)
1599                 goto drop;
1600
1601         sk = pt->af_packet_priv;
1602         po = pkt_sk(sk);
1603
1604         if (!net_eq(dev_net(dev), sock_net(sk)))
1605                 goto drop;
1606
1607         skb->dev = dev;
1608
1609         if (dev->header_ops) {
1610                 /* The device has an explicit notion of ll header,
1611                  * exported to higher levels.
1612                  *
1613                  * Otherwise, the device hides details of its frame
1614                  * structure, so that corresponding packet head is
1615                  * never delivered to user.
1616                  */
1617                 if (sk->sk_type != SOCK_DGRAM)
1618                         skb_push(skb, skb->data - skb_mac_header(skb));
1619                 else if (skb->pkt_type == PACKET_OUTGOING) {
1620                         /* Special case: outgoing packets have ll header at head */
1621                         skb_pull(skb, skb_network_offset(skb));
1622                 }
1623         }
1624
1625         snaplen = skb->len;
1626
1627         res = run_filter(skb, sk, snaplen);
1628         if (!res)
1629                 goto drop_n_restore;
1630         if (snaplen > res)
1631                 snaplen = res;
1632
1633         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1634                 goto drop_n_acct;
1635
1636         if (skb_shared(skb)) {
1637                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1638                 if (nskb == NULL)
1639                         goto drop_n_acct;
1640
1641                 if (skb_head != skb->data) {
1642                         skb->data = skb_head;
1643                         skb->len = skb_len;
1644                 }
1645                 kfree_skb(skb);
1646                 skb = nskb;
1647         }
1648
1649         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1650                      sizeof(skb->cb));
1651
1652         sll = &PACKET_SKB_CB(skb)->sa.ll;
1653         sll->sll_family = AF_PACKET;
1654         sll->sll_hatype = dev->type;
1655         sll->sll_protocol = skb->protocol;
1656         sll->sll_pkttype = skb->pkt_type;
1657         if (unlikely(po->origdev))
1658                 sll->sll_ifindex = orig_dev->ifindex;
1659         else
1660                 sll->sll_ifindex = dev->ifindex;
1661
1662         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1663
1664         PACKET_SKB_CB(skb)->origlen = skb->len;
1665
1666         if (pskb_trim(skb, snaplen))
1667                 goto drop_n_acct;
1668
1669         skb_set_owner_r(skb, sk);
1670         skb->dev = NULL;
1671         skb_dst_drop(skb);
1672
1673         /* drop conntrack reference */
1674         nf_reset(skb);
1675
1676         spin_lock(&sk->sk_receive_queue.lock);
1677         po->stats.tp_packets++;
1678         skb->dropcount = atomic_read(&sk->sk_drops);
1679         __skb_queue_tail(&sk->sk_receive_queue, skb);
1680         spin_unlock(&sk->sk_receive_queue.lock);
1681         sk->sk_data_ready(sk, skb->len);
1682         return 0;
1683
1684 drop_n_acct:
1685         spin_lock(&sk->sk_receive_queue.lock);
1686         po->stats.tp_drops++;
1687         atomic_inc(&sk->sk_drops);
1688         spin_unlock(&sk->sk_receive_queue.lock);
1689
1690 drop_n_restore:
1691         if (skb_head != skb->data && skb_shared(skb)) {
1692                 skb->data = skb_head;
1693                 skb->len = skb_len;
1694         }
1695 drop:
1696         consume_skb(skb);
1697         return 0;
1698 }
1699
1700 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1701                        struct packet_type *pt, struct net_device *orig_dev)
1702 {
1703         struct sock *sk;
1704         struct packet_sock *po;
1705         struct sockaddr_ll *sll;
1706         union {
1707                 struct tpacket_hdr *h1;
1708                 struct tpacket2_hdr *h2;
1709                 struct tpacket3_hdr *h3;
1710                 void *raw;
1711         } h;
1712         u8 *skb_head = skb->data;
1713         int skb_len = skb->len;
1714         unsigned int snaplen, res;
1715         unsigned long status = TP_STATUS_USER;
1716         unsigned short macoff, netoff, hdrlen;
1717         struct sk_buff *copy_skb = NULL;
1718         struct timeval tv;
1719         struct timespec ts;
1720         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1721
1722         if (skb->pkt_type == PACKET_LOOPBACK)
1723                 goto drop;
1724
1725         sk = pt->af_packet_priv;
1726         po = pkt_sk(sk);
1727
1728         if (!net_eq(dev_net(dev), sock_net(sk)))
1729                 goto drop;
1730
1731         if (dev->header_ops) {
1732                 if (sk->sk_type != SOCK_DGRAM)
1733                         skb_push(skb, skb->data - skb_mac_header(skb));
1734                 else if (skb->pkt_type == PACKET_OUTGOING) {
1735                         /* Special case: outgoing packets have ll header at head */
1736                         skb_pull(skb, skb_network_offset(skb));
1737                 }
1738         }
1739
1740         if (skb->ip_summed == CHECKSUM_PARTIAL)
1741                 status |= TP_STATUS_CSUMNOTREADY;
1742
1743         snaplen = skb->len;
1744
1745         res = run_filter(skb, sk, snaplen);
1746         if (!res)
1747                 goto drop_n_restore;
1748         if (snaplen > res)
1749                 snaplen = res;
1750
1751         if (sk->sk_type == SOCK_DGRAM) {
1752                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1753                                   po->tp_reserve;
1754         } else {
1755                 unsigned maclen = skb_network_offset(skb);
1756                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1757                                        (maclen < 16 ? 16 : maclen)) +
1758                         po->tp_reserve;
1759                 macoff = netoff - maclen;
1760         }
1761         if (po->tp_version <= TPACKET_V2) {
1762                 if (macoff + snaplen > po->rx_ring.frame_size) {
1763                         if (po->copy_thresh &&
1764                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1765                                 if (skb_shared(skb)) {
1766                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1767                                 } else {
1768                                         copy_skb = skb_get(skb);
1769                                         skb_head = skb->data;
1770                                 }
1771                                 if (copy_skb)
1772                                         skb_set_owner_r(copy_skb, sk);
1773                         }
1774                         snaplen = po->rx_ring.frame_size - macoff;
1775                         if ((int)snaplen < 0)
1776                                 snaplen = 0;
1777                 }
1778         }
1779         spin_lock(&sk->sk_receive_queue.lock);
1780         h.raw = packet_current_rx_frame(po, skb,
1781                                         TP_STATUS_KERNEL, (macoff+snaplen));
1782         if (!h.raw)
1783                 goto ring_is_full;
1784         if (po->tp_version <= TPACKET_V2) {
1785                 packet_increment_rx_head(po, &po->rx_ring);
1786         /*
1787          * LOSING will be reported till you read the stats,
1788          * because it's COR - Clear On Read.
1789          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1790          * at packet level.
1791          */
1792                 if (po->stats.tp_drops)
1793                         status |= TP_STATUS_LOSING;
1794         }
1795         po->stats.tp_packets++;
1796         if (copy_skb) {
1797                 status |= TP_STATUS_COPY;
1798                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1799         }
1800         spin_unlock(&sk->sk_receive_queue.lock);
1801
1802         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1803
1804         switch (po->tp_version) {
1805         case TPACKET_V1:
1806                 h.h1->tp_len = skb->len;
1807                 h.h1->tp_snaplen = snaplen;
1808                 h.h1->tp_mac = macoff;
1809                 h.h1->tp_net = netoff;
1810                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1811                                 && shhwtstamps->syststamp.tv64)
1812                         tv = ktime_to_timeval(shhwtstamps->syststamp);
1813                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1814                                 && shhwtstamps->hwtstamp.tv64)
1815                         tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1816                 else if (skb->tstamp.tv64)
1817                         tv = ktime_to_timeval(skb->tstamp);
1818                 else
1819                         do_gettimeofday(&tv);
1820                 h.h1->tp_sec = tv.tv_sec;
1821                 h.h1->tp_usec = tv.tv_usec;
1822                 hdrlen = sizeof(*h.h1);
1823                 break;
1824         case TPACKET_V2:
1825                 h.h2->tp_len = skb->len;
1826                 h.h2->tp_snaplen = snaplen;
1827                 h.h2->tp_mac = macoff;
1828                 h.h2->tp_net = netoff;
1829                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1830                                 && shhwtstamps->syststamp.tv64)
1831                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1832                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1833                                 && shhwtstamps->hwtstamp.tv64)
1834                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1835                 else if (skb->tstamp.tv64)
1836                         ts = ktime_to_timespec(skb->tstamp);
1837                 else
1838                         getnstimeofday(&ts);
1839                 h.h2->tp_sec = ts.tv_sec;
1840                 h.h2->tp_nsec = ts.tv_nsec;
1841                 if (vlan_tx_tag_present(skb)) {
1842                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1843                         status |= TP_STATUS_VLAN_VALID;
1844                 } else {
1845                         h.h2->tp_vlan_tci = 0;
1846                 }
1847                 h.h2->tp_padding = 0;
1848                 hdrlen = sizeof(*h.h2);
1849                 break;
1850         case TPACKET_V3:
1851                 /* tp_nxt_offset,vlan are already populated above.
1852                  * So DONT clear those fields here
1853                  */
1854                 h.h3->tp_status |= status;
1855                 h.h3->tp_len = skb->len;
1856                 h.h3->tp_snaplen = snaplen;
1857                 h.h3->tp_mac = macoff;
1858                 h.h3->tp_net = netoff;
1859                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1860                                 && shhwtstamps->syststamp.tv64)
1861                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1862                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1863                                 && shhwtstamps->hwtstamp.tv64)
1864                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1865                 else if (skb->tstamp.tv64)
1866                         ts = ktime_to_timespec(skb->tstamp);
1867                 else
1868                         getnstimeofday(&ts);
1869                 h.h3->tp_sec  = ts.tv_sec;
1870                 h.h3->tp_nsec = ts.tv_nsec;
1871                 hdrlen = sizeof(*h.h3);
1872                 break;
1873         default:
1874                 BUG();
1875         }
1876
1877         sll = h.raw + TPACKET_ALIGN(hdrlen);
1878         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1879         sll->sll_family = AF_PACKET;
1880         sll->sll_hatype = dev->type;
1881         sll->sll_protocol = skb->protocol;
1882         sll->sll_pkttype = skb->pkt_type;
1883         if (unlikely(po->origdev))
1884                 sll->sll_ifindex = orig_dev->ifindex;
1885         else
1886                 sll->sll_ifindex = dev->ifindex;
1887
1888         smp_mb();
1889 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1890         {
1891                 u8 *start, *end;
1892
1893                 if (po->tp_version <= TPACKET_V2) {
1894                         end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1895                                 + macoff + snaplen);
1896                         for (start = h.raw; start < end; start += PAGE_SIZE)
1897                                 flush_dcache_page(pgv_to_page(start));
1898                 }
1899                 smp_wmb();
1900         }
1901 #endif
1902         if (po->tp_version <= TPACKET_V2)
1903                 __packet_set_status(po, h.raw, status);
1904         else
1905                 prb_clear_blk_fill_status(&po->rx_ring);
1906
1907         sk->sk_data_ready(sk, 0);
1908
1909 drop_n_restore:
1910         if (skb_head != skb->data && skb_shared(skb)) {
1911                 skb->data = skb_head;
1912                 skb->len = skb_len;
1913         }
1914 drop:
1915         kfree_skb(skb);
1916         return 0;
1917
1918 ring_is_full:
1919         po->stats.tp_drops++;
1920         spin_unlock(&sk->sk_receive_queue.lock);
1921
1922         sk->sk_data_ready(sk, 0);
1923         kfree_skb(copy_skb);
1924         goto drop_n_restore;
1925 }
1926
1927 static void tpacket_destruct_skb(struct sk_buff *skb)
1928 {
1929         struct packet_sock *po = pkt_sk(skb->sk);
1930         void *ph;
1931
1932         if (likely(po->tx_ring.pg_vec)) {
1933                 ph = skb_shinfo(skb)->destructor_arg;
1934                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1935                 atomic_dec(&po->tx_ring.pending);
1936                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1937         }
1938
1939         sock_wfree(skb);
1940 }
1941
1942 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1943                 void *frame, struct net_device *dev, int size_max,
1944                 __be16 proto, unsigned char *addr)
1945 {
1946         union {
1947                 struct tpacket_hdr *h1;
1948                 struct tpacket2_hdr *h2;
1949                 void *raw;
1950         } ph;
1951         int to_write, offset, len, tp_len, nr_frags, len_max;
1952         struct socket *sock = po->sk.sk_socket;
1953         struct page *page;
1954         void *data;
1955         int err;
1956
1957         ph.raw = frame;
1958
1959         skb->protocol = proto;
1960         skb->dev = dev;
1961         skb->priority = po->sk.sk_priority;
1962         skb->mark = po->sk.sk_mark;
1963         skb_shinfo(skb)->destructor_arg = ph.raw;
1964
1965         switch (po->tp_version) {
1966         case TPACKET_V2:
1967                 tp_len = ph.h2->tp_len;
1968                 break;
1969         default:
1970                 tp_len = ph.h1->tp_len;
1971                 break;
1972         }
1973         if (unlikely(tp_len > size_max)) {
1974                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1975                 return -EMSGSIZE;
1976         }
1977
1978         skb_reserve(skb, LL_RESERVED_SPACE(dev));
1979         skb_reset_network_header(skb);
1980
1981         data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1982         to_write = tp_len;
1983
1984         if (sock->type == SOCK_DGRAM) {
1985                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
1986                                 NULL, tp_len);
1987                 if (unlikely(err < 0))
1988                         return -EINVAL;
1989         } else if (dev->hard_header_len) {
1990                 /* net device doesn't like empty head */
1991                 if (unlikely(tp_len <= dev->hard_header_len)) {
1992                         pr_err("packet size is too short (%d < %d)\n",
1993                                tp_len, dev->hard_header_len);
1994                         return -EINVAL;
1995                 }
1996
1997                 skb_push(skb, dev->hard_header_len);
1998                 err = skb_store_bits(skb, 0, data,
1999                                 dev->hard_header_len);
2000                 if (unlikely(err))
2001                         return err;
2002
2003                 data += dev->hard_header_len;
2004                 to_write -= dev->hard_header_len;
2005         }
2006
2007         err = -EFAULT;
2008         offset = offset_in_page(data);
2009         len_max = PAGE_SIZE - offset;
2010         len = ((to_write > len_max) ? len_max : to_write);
2011
2012         skb->data_len = to_write;
2013         skb->len += to_write;
2014         skb->truesize += to_write;
2015         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2016
2017         while (likely(to_write)) {
2018                 nr_frags = skb_shinfo(skb)->nr_frags;
2019
2020                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2021                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2022                                MAX_SKB_FRAGS);
2023                         return -EFAULT;
2024                 }
2025
2026                 page = pgv_to_page(data);
2027                 data += len;
2028                 flush_dcache_page(page);
2029                 get_page(page);
2030                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2031                 to_write -= len;
2032                 offset = 0;
2033                 len_max = PAGE_SIZE;
2034                 len = ((to_write > len_max) ? len_max : to_write);
2035         }
2036
2037         return tp_len;
2038 }
2039
2040 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
2041 {
2042         struct net_device *dev;
2043
2044         rcu_read_lock();
2045         dev = rcu_dereference(po->cached_dev);
2046         if (dev)
2047                 dev_hold(dev);
2048         rcu_read_unlock();
2049
2050         return dev;
2051 }
2052
2053 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2054 {
2055         struct sk_buff *skb;
2056         struct net_device *dev;
2057         __be16 proto;
2058         int err, reserve = 0;
2059         void *ph;
2060         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2061         int tp_len, size_max;
2062         unsigned char *addr;
2063         int len_sum = 0;
2064         int status = 0;
2065
2066         mutex_lock(&po->pg_vec_lock);
2067
2068         err = -EBUSY;
2069         if (saddr == NULL) {
2070                 dev     = packet_cached_dev_get(po);
2071                 proto   = po->num;
2072                 addr    = NULL;
2073         } else {
2074                 err = -EINVAL;
2075                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2076                         goto out;
2077                 if (msg->msg_namelen < (saddr->sll_halen
2078                                         + offsetof(struct sockaddr_ll,
2079                                                 sll_addr)))
2080                         goto out;
2081                 proto   = saddr->sll_protocol;
2082                 addr    = saddr->sll_addr;
2083                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2084         }
2085
2086         err = -ENXIO;
2087         if (unlikely(dev == NULL))
2088                 goto out;
2089         err = -ENETDOWN;
2090         if (unlikely(!(dev->flags & IFF_UP)))
2091                 goto out_put;
2092
2093         reserve = dev->hard_header_len;
2094
2095         size_max = po->tx_ring.frame_size
2096                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2097
2098         if (size_max > dev->mtu + reserve)
2099                 size_max = dev->mtu + reserve;
2100
2101         do {
2102                 ph = packet_current_frame(po, &po->tx_ring,
2103                                 TP_STATUS_SEND_REQUEST);
2104
2105                 if (unlikely(ph == NULL)) {
2106                         schedule();
2107                         continue;
2108                 }
2109
2110                 status = TP_STATUS_SEND_REQUEST;
2111                 skb = sock_alloc_send_skb(&po->sk,
2112                                 LL_ALLOCATED_SPACE(dev)
2113                                 + sizeof(struct sockaddr_ll),
2114                                 0, &err);
2115
2116                 if (unlikely(skb == NULL))
2117                         goto out_status;
2118
2119                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2120                                 addr);
2121
2122                 if (unlikely(tp_len < 0)) {
2123                         if (po->tp_loss) {
2124                                 __packet_set_status(po, ph,
2125                                                 TP_STATUS_AVAILABLE);
2126                                 packet_increment_head(&po->tx_ring);
2127                                 kfree_skb(skb);
2128                                 continue;
2129                         } else {
2130                                 status = TP_STATUS_WRONG_FORMAT;
2131                                 err = tp_len;
2132                                 goto out_status;
2133                         }
2134                 }
2135
2136                 skb->destructor = tpacket_destruct_skb;
2137                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2138                 atomic_inc(&po->tx_ring.pending);
2139
2140                 status = TP_STATUS_SEND_REQUEST;
2141                 err = dev_queue_xmit(skb);
2142                 if (unlikely(err > 0)) {
2143                         err = net_xmit_errno(err);
2144                         if (err && __packet_get_status(po, ph) ==
2145                                    TP_STATUS_AVAILABLE) {
2146                                 /* skb was destructed already */
2147                                 skb = NULL;
2148                                 goto out_status;
2149                         }
2150                         /*
2151                          * skb was dropped but not destructed yet;
2152                          * let's treat it like congestion or err < 0
2153                          */
2154                         err = 0;
2155                 }
2156                 packet_increment_head(&po->tx_ring);
2157                 len_sum += tp_len;
2158         } while (likely((ph != NULL) ||
2159                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2160                          (atomic_read(&po->tx_ring.pending))))
2161                 );
2162
2163         err = len_sum;
2164         goto out_put;
2165
2166 out_status:
2167         __packet_set_status(po, ph, status);
2168         kfree_skb(skb);
2169 out_put:
2170         dev_put(dev);
2171 out:
2172         mutex_unlock(&po->pg_vec_lock);
2173         return err;
2174 }
2175
2176 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2177                                         size_t reserve, size_t len,
2178                                         size_t linear, int noblock,
2179                                         int *err)
2180 {
2181         struct sk_buff *skb;
2182
2183         /* Under a page?  Don't bother with paged skb. */
2184         if (prepad + len < PAGE_SIZE || !linear)
2185                 linear = len;
2186
2187         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2188                                    err);
2189         if (!skb)
2190                 return NULL;
2191
2192         skb_reserve(skb, reserve);
2193         skb_put(skb, linear);
2194         skb->data_len = len - linear;
2195         skb->len += len - linear;
2196
2197         return skb;
2198 }
2199
2200 static int packet_snd(struct socket *sock,
2201                           struct msghdr *msg, size_t len)
2202 {
2203         struct sock *sk = sock->sk;
2204         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2205         struct sk_buff *skb;
2206         struct net_device *dev;
2207         __be16 proto;
2208         unsigned char *addr;
2209         int err, reserve = 0;
2210         struct virtio_net_hdr vnet_hdr = { 0 };
2211         int offset = 0;
2212         int vnet_hdr_len;
2213         struct packet_sock *po = pkt_sk(sk);
2214         unsigned short gso_type = 0;
2215
2216         /*
2217          *      Get and verify the address.
2218          */
2219
2220         if (saddr == NULL) {
2221                 dev     = packet_cached_dev_get(po);
2222                 proto   = po->num;
2223                 addr    = NULL;
2224         } else {
2225                 err = -EINVAL;
2226                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2227                         goto out;
2228                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2229                         goto out;
2230                 proto   = saddr->sll_protocol;
2231                 addr    = saddr->sll_addr;
2232                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2233         }
2234
2235         err = -ENXIO;
2236         if (unlikely(dev == NULL))
2237                 goto out_unlock;
2238         err = -ENETDOWN;
2239         if (unlikely(!(dev->flags & IFF_UP)))
2240                 goto out_unlock;
2241
2242         if (sock->type == SOCK_RAW)
2243                 reserve = dev->hard_header_len;
2244         if (po->has_vnet_hdr) {
2245                 vnet_hdr_len = sizeof(vnet_hdr);
2246
2247                 err = -EINVAL;
2248                 if (len < vnet_hdr_len)
2249                         goto out_unlock;
2250
2251                 len -= vnet_hdr_len;
2252
2253                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2254                                        vnet_hdr_len);
2255                 if (err < 0)
2256                         goto out_unlock;
2257
2258                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2259                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2260                       vnet_hdr.hdr_len))
2261                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
2262                                                  vnet_hdr.csum_offset + 2;
2263
2264                 err = -EINVAL;
2265                 if (vnet_hdr.hdr_len > len)
2266                         goto out_unlock;
2267
2268                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2269                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2270                         case VIRTIO_NET_HDR_GSO_TCPV4:
2271                                 gso_type = SKB_GSO_TCPV4;
2272                                 break;
2273                         case VIRTIO_NET_HDR_GSO_TCPV6:
2274                                 gso_type = SKB_GSO_TCPV6;
2275                                 break;
2276                         case VIRTIO_NET_HDR_GSO_UDP:
2277                                 gso_type = SKB_GSO_UDP;
2278                                 break;
2279                         default:
2280                                 goto out_unlock;
2281                         }
2282
2283                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2284                                 gso_type |= SKB_GSO_TCP_ECN;
2285
2286                         if (vnet_hdr.gso_size == 0)
2287                                 goto out_unlock;
2288
2289                 }
2290         }
2291
2292         err = -EMSGSIZE;
2293         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN))
2294                 goto out_unlock;
2295
2296         err = -ENOBUFS;
2297         skb = packet_alloc_skb(sk, LL_ALLOCATED_SPACE(dev),
2298                                LL_RESERVED_SPACE(dev), len, vnet_hdr.hdr_len,
2299                                msg->msg_flags & MSG_DONTWAIT, &err);
2300         if (skb == NULL)
2301                 goto out_unlock;
2302
2303         skb_set_network_header(skb, reserve);
2304
2305         err = -EINVAL;
2306         if (sock->type == SOCK_DGRAM &&
2307             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2308                 goto out_free;
2309
2310         /* Returns -EFAULT on error */
2311         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2312         if (err)
2313                 goto out_free;
2314         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2315         if (err < 0)
2316                 goto out_free;
2317
2318         if (!gso_type && (len > dev->mtu + reserve)) {
2319                 /* Earlier code assumed this would be a VLAN pkt,
2320                  * double-check this now that we have the actual
2321                  * packet in hand.
2322                  */
2323                 struct ethhdr *ehdr;
2324                 skb_reset_mac_header(skb);
2325                 ehdr = eth_hdr(skb);
2326                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2327                         err = -EMSGSIZE;
2328                         goto out_free;
2329                 }
2330         }
2331
2332         skb->protocol = proto;
2333         skb->dev = dev;
2334         skb->priority = sk->sk_priority;
2335         skb->mark = sk->sk_mark;
2336
2337         if (po->has_vnet_hdr) {
2338                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2339                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2340                                                   vnet_hdr.csum_offset)) {
2341                                 err = -EINVAL;
2342                                 goto out_free;
2343                         }
2344                 }
2345
2346                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2347                 skb_shinfo(skb)->gso_type = gso_type;
2348
2349                 /* Header must be checked, and gso_segs computed. */
2350                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2351                 skb_shinfo(skb)->gso_segs = 0;
2352
2353                 len += vnet_hdr_len;
2354         }
2355
2356         /*
2357          *      Now send it
2358          */
2359
2360         err = dev_queue_xmit(skb);
2361         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2362                 goto out_unlock;
2363
2364         dev_put(dev);
2365
2366         return len;
2367
2368 out_free:
2369         kfree_skb(skb);
2370 out_unlock:
2371         if (dev)
2372                 dev_put(dev);
2373 out:
2374         return err;
2375 }
2376
2377 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2378                 struct msghdr *msg, size_t len)
2379 {
2380         struct sock *sk = sock->sk;
2381         struct packet_sock *po = pkt_sk(sk);
2382         if (po->tx_ring.pg_vec)
2383                 return tpacket_snd(po, msg);
2384         else
2385                 return packet_snd(sock, msg, len);
2386 }
2387
2388 /*
2389  *      Close a PACKET socket. This is fairly simple. We immediately go
2390  *      to 'closed' state and remove our protocol entry in the device list.
2391  */
2392
2393 static int packet_release(struct socket *sock)
2394 {
2395         struct sock *sk = sock->sk;
2396         struct packet_sock *po;
2397         struct packet_fanout *f;
2398         struct net *net;
2399         union tpacket_req_u req_u;
2400
2401         if (!sk)
2402                 return 0;
2403
2404         net = sock_net(sk);
2405         po = pkt_sk(sk);
2406
2407         spin_lock_bh(&net->packet.sklist_lock);
2408         sk_del_node_init_rcu(sk);
2409         sock_prot_inuse_add(net, sk->sk_prot, -1);
2410         spin_unlock_bh(&net->packet.sklist_lock);
2411
2412         spin_lock(&po->bind_lock);
2413         unregister_prot_hook(sk, false);
2414         if (po->prot_hook.dev) {
2415                 dev_put(po->prot_hook.dev);
2416                 po->prot_hook.dev = NULL;
2417         }
2418         spin_unlock(&po->bind_lock);
2419
2420         packet_flush_mclist(sk);
2421
2422         if (po->rx_ring.pg_vec) {
2423                 memset(&req_u, 0, sizeof(req_u));
2424                 packet_set_ring(sk, &req_u, 1, 0);
2425         }
2426
2427         if (po->tx_ring.pg_vec) {
2428                 memset(&req_u, 0, sizeof(req_u));
2429                 packet_set_ring(sk, &req_u, 1, 1);
2430         }
2431
2432         f = fanout_release(sk);
2433
2434         synchronize_net();
2435
2436         kfree(f);
2437
2438         /*
2439          *      Now the socket is dead. No more input will appear.
2440          */
2441         sock_orphan(sk);
2442         sock->sk = NULL;
2443
2444         /* Purge queues */
2445
2446         skb_queue_purge(&sk->sk_receive_queue);
2447         sk_refcnt_debug_release(sk);
2448
2449         sock_put(sk);
2450         return 0;
2451 }
2452
2453 /*
2454  *      Attach a packet hook.
2455  */
2456
2457 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2458 {
2459         struct packet_sock *po = pkt_sk(sk);
2460
2461         if (po->fanout) {
2462                 if (dev)
2463                         dev_put(dev);
2464
2465                 return -EINVAL;
2466         }
2467
2468         lock_sock(sk);
2469
2470         spin_lock(&po->bind_lock);
2471         unregister_prot_hook(sk, true);
2472         po->num = protocol;
2473         po->prot_hook.type = protocol;
2474         if (po->prot_hook.dev)
2475                 dev_put(po->prot_hook.dev);
2476         po->prot_hook.dev = dev;
2477
2478         po->ifindex = dev ? dev->ifindex : 0;
2479
2480         if (protocol == 0)
2481                 goto out_unlock;
2482
2483         if (!dev || (dev->flags & IFF_UP)) {
2484                 register_prot_hook(sk);
2485         } else {
2486                 sk->sk_err = ENETDOWN;
2487                 if (!sock_flag(sk, SOCK_DEAD))
2488                         sk->sk_error_report(sk);
2489         }
2490
2491 out_unlock:
2492         spin_unlock(&po->bind_lock);
2493         release_sock(sk);
2494         return 0;
2495 }
2496
2497 /*
2498  *      Bind a packet socket to a device
2499  */
2500
2501 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2502                             int addr_len)
2503 {
2504         struct sock *sk = sock->sk;
2505         char name[15];
2506         struct net_device *dev;
2507         int err = -ENODEV;
2508
2509         /*
2510          *      Check legality
2511          */
2512
2513         if (addr_len != sizeof(struct sockaddr))
2514                 return -EINVAL;
2515         strlcpy(name, uaddr->sa_data, sizeof(name));
2516
2517         dev = dev_get_by_name(sock_net(sk), name);
2518         if (dev)
2519                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2520         return err;
2521 }
2522
2523 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2524 {
2525         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2526         struct sock *sk = sock->sk;
2527         struct net_device *dev = NULL;
2528         int err;
2529
2530
2531         /*
2532          *      Check legality
2533          */
2534
2535         if (addr_len < sizeof(struct sockaddr_ll))
2536                 return -EINVAL;
2537         if (sll->sll_family != AF_PACKET)
2538                 return -EINVAL;
2539
2540         if (sll->sll_ifindex) {
2541                 err = -ENODEV;
2542                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2543                 if (dev == NULL)
2544                         goto out;
2545         }
2546         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2547
2548 out:
2549         return err;
2550 }
2551
2552 static struct proto packet_proto = {
2553         .name     = "PACKET",
2554         .owner    = THIS_MODULE,
2555         .obj_size = sizeof(struct packet_sock),
2556 };
2557
2558 /*
2559  *      Create a packet of type SOCK_PACKET.
2560  */
2561
2562 static int packet_create(struct net *net, struct socket *sock, int protocol,
2563                          int kern)
2564 {
2565         struct sock *sk;
2566         struct packet_sock *po;
2567         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2568         int err;
2569
2570         if (!capable(CAP_NET_RAW))
2571                 return -EPERM;
2572         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2573             sock->type != SOCK_PACKET)
2574                 return -ESOCKTNOSUPPORT;
2575
2576         sock->state = SS_UNCONNECTED;
2577
2578         err = -ENOBUFS;
2579         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2580         if (sk == NULL)
2581                 goto out;
2582
2583         sock->ops = &packet_ops;
2584         if (sock->type == SOCK_PACKET)
2585                 sock->ops = &packet_ops_spkt;
2586
2587         sock_init_data(sock, sk);
2588
2589         po = pkt_sk(sk);
2590         sk->sk_family = PF_PACKET;
2591         po->num = proto;
2592         RCU_INIT_POINTER(po->cached_dev, NULL);
2593
2594         sk->sk_destruct = packet_sock_destruct;
2595         sk_refcnt_debug_inc(sk);
2596
2597         /*
2598          *      Attach a protocol block
2599          */
2600
2601         spin_lock_init(&po->bind_lock);
2602         mutex_init(&po->pg_vec_lock);
2603         po->prot_hook.func = packet_rcv;
2604
2605         if (sock->type == SOCK_PACKET)
2606                 po->prot_hook.func = packet_rcv_spkt;
2607
2608         po->prot_hook.af_packet_priv = sk;
2609
2610         if (proto) {
2611                 po->prot_hook.type = proto;
2612                 register_prot_hook(sk);
2613         }
2614
2615         spin_lock_bh(&net->packet.sklist_lock);
2616         sk_add_node_rcu(sk, &net->packet.sklist);
2617         sock_prot_inuse_add(net, &packet_proto, 1);
2618         spin_unlock_bh(&net->packet.sklist_lock);
2619
2620         return 0;
2621 out:
2622         return err;
2623 }
2624
2625 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2626 {
2627         struct sock_exterr_skb *serr;
2628         struct sk_buff *skb, *skb2;
2629         int copied, err;
2630
2631         err = -EAGAIN;
2632         skb = skb_dequeue(&sk->sk_error_queue);
2633         if (skb == NULL)
2634                 goto out;
2635
2636         copied = skb->len;
2637         if (copied > len) {
2638                 msg->msg_flags |= MSG_TRUNC;
2639                 copied = len;
2640         }
2641         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2642         if (err)
2643                 goto out_free_skb;
2644
2645         sock_recv_timestamp(msg, sk, skb);
2646
2647         serr = SKB_EXT_ERR(skb);
2648         put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2649                  sizeof(serr->ee), &serr->ee);
2650
2651         msg->msg_flags |= MSG_ERRQUEUE;
2652         err = copied;
2653
2654         /* Reset and regenerate socket error */
2655         spin_lock_bh(&sk->sk_error_queue.lock);
2656         sk->sk_err = 0;
2657         if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2658                 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2659                 spin_unlock_bh(&sk->sk_error_queue.lock);
2660                 sk->sk_error_report(sk);
2661         } else
2662                 spin_unlock_bh(&sk->sk_error_queue.lock);
2663
2664 out_free_skb:
2665         kfree_skb(skb);
2666 out:
2667         return err;
2668 }
2669
2670 /*
2671  *      Pull a packet from our receive queue and hand it to the user.
2672  *      If necessary we block.
2673  */
2674
2675 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2676                           struct msghdr *msg, size_t len, int flags)
2677 {
2678         struct sock *sk = sock->sk;
2679         struct sk_buff *skb;
2680         int copied, err;
2681         int vnet_hdr_len = 0;
2682
2683         err = -EINVAL;
2684         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2685                 goto out;
2686
2687 #if 0
2688         /* What error should we return now? EUNATTACH? */
2689         if (pkt_sk(sk)->ifindex < 0)
2690                 return -ENODEV;
2691 #endif
2692
2693         if (flags & MSG_ERRQUEUE) {
2694                 err = packet_recv_error(sk, msg, len);
2695                 goto out;
2696         }
2697
2698         /*
2699          *      Call the generic datagram receiver. This handles all sorts
2700          *      of horrible races and re-entrancy so we can forget about it
2701          *      in the protocol layers.
2702          *
2703          *      Now it will return ENETDOWN, if device have just gone down,
2704          *      but then it will block.
2705          */
2706
2707         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2708
2709         /*
2710          *      An error occurred so return it. Because skb_recv_datagram()
2711          *      handles the blocking we don't see and worry about blocking
2712          *      retries.
2713          */
2714
2715         if (skb == NULL)
2716                 goto out;
2717
2718         if (pkt_sk(sk)->has_vnet_hdr) {
2719                 struct virtio_net_hdr vnet_hdr = { 0 };
2720
2721                 err = -EINVAL;
2722                 vnet_hdr_len = sizeof(vnet_hdr);
2723                 if (len < vnet_hdr_len)
2724                         goto out_free;
2725
2726                 len -= vnet_hdr_len;
2727
2728                 if (skb_is_gso(skb)) {
2729                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2730
2731                         /* This is a hint as to how much should be linear. */
2732                         vnet_hdr.hdr_len = skb_headlen(skb);
2733                         vnet_hdr.gso_size = sinfo->gso_size;
2734                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2735                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2736                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2737                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2738                         else if (sinfo->gso_type & SKB_GSO_UDP)
2739                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2740                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2741                                 goto out_free;
2742                         else
2743                                 BUG();
2744                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2745                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2746                 } else
2747                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2748
2749                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2750                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2751                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2752                         vnet_hdr.csum_offset = skb->csum_offset;
2753                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2754                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2755                 } /* else everything is zero */
2756
2757                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2758                                      vnet_hdr_len);
2759                 if (err < 0)
2760                         goto out_free;
2761         }
2762
2763         /* You lose any data beyond the buffer you gave. If it worries
2764          * a user program they can ask the device for its MTU
2765          * anyway.
2766          */
2767         copied = skb->len;
2768         if (copied > len) {
2769                 copied = len;
2770                 msg->msg_flags |= MSG_TRUNC;
2771         }
2772
2773         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2774         if (err)
2775                 goto out_free;
2776
2777         sock_recv_ts_and_drops(msg, sk, skb);
2778
2779         if (msg->msg_name) {
2780                 /* If the address length field is there to be filled
2781                  * in, we fill it in now.
2782                  */
2783                 if (sock->type == SOCK_PACKET) {
2784                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
2785                 } else {
2786                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2787                         msg->msg_namelen = sll->sll_halen +
2788                                 offsetof(struct sockaddr_ll, sll_addr);
2789                 }
2790                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2791                        msg->msg_namelen);
2792         }
2793
2794         if (pkt_sk(sk)->auxdata) {
2795                 struct tpacket_auxdata aux;
2796
2797                 aux.tp_status = TP_STATUS_USER;
2798                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2799                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2800                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2801                 aux.tp_snaplen = skb->len;
2802                 aux.tp_mac = 0;
2803                 aux.tp_net = skb_network_offset(skb);
2804                 if (vlan_tx_tag_present(skb)) {
2805                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2806                         aux.tp_status |= TP_STATUS_VLAN_VALID;
2807                 } else {
2808                         aux.tp_vlan_tci = 0;
2809                 }
2810                 aux.tp_padding = 0;
2811                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2812         }
2813
2814         /*
2815          *      Free or return the buffer as appropriate. Again this
2816          *      hides all the races and re-entrancy issues from us.
2817          */
2818         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2819
2820 out_free:
2821         skb_free_datagram(sk, skb);
2822 out:
2823         return err;
2824 }
2825
2826 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2827                                int *uaddr_len, int peer)
2828 {
2829         struct net_device *dev;
2830         struct sock *sk = sock->sk;
2831
2832         if (peer)
2833                 return -EOPNOTSUPP;
2834
2835         uaddr->sa_family = AF_PACKET;
2836         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
2837         rcu_read_lock();
2838         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2839         if (dev)
2840                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
2841         rcu_read_unlock();
2842         *uaddr_len = sizeof(*uaddr);
2843
2844         return 0;
2845 }
2846
2847 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2848                           int *uaddr_len, int peer)
2849 {
2850         struct net_device *dev;
2851         struct sock *sk = sock->sk;
2852         struct packet_sock *po = pkt_sk(sk);
2853         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2854
2855         if (peer)
2856                 return -EOPNOTSUPP;
2857
2858         sll->sll_family = AF_PACKET;
2859         sll->sll_ifindex = po->ifindex;
2860         sll->sll_protocol = po->num;
2861         sll->sll_pkttype = 0;
2862         rcu_read_lock();
2863         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2864         if (dev) {
2865                 sll->sll_hatype = dev->type;
2866                 sll->sll_halen = dev->addr_len;
2867                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2868         } else {
2869                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
2870                 sll->sll_halen = 0;
2871         }
2872         rcu_read_unlock();
2873         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2874
2875         return 0;
2876 }
2877
2878 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2879                          int what)
2880 {
2881         switch (i->type) {
2882         case PACKET_MR_MULTICAST:
2883                 if (i->alen != dev->addr_len)
2884                         return -EINVAL;
2885                 if (what > 0)
2886                         return dev_mc_add(dev, i->addr);
2887                 else
2888                         return dev_mc_del(dev, i->addr);
2889                 break;
2890         case PACKET_MR_PROMISC:
2891                 return dev_set_promiscuity(dev, what);
2892                 break;
2893         case PACKET_MR_ALLMULTI:
2894                 return dev_set_allmulti(dev, what);
2895                 break;
2896         case PACKET_MR_UNICAST:
2897                 if (i->alen != dev->addr_len)
2898                         return -EINVAL;
2899                 if (what > 0)
2900                         return dev_uc_add(dev, i->addr);
2901                 else
2902                         return dev_uc_del(dev, i->addr);
2903                 break;
2904         default:
2905                 break;
2906         }
2907         return 0;
2908 }
2909
2910 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2911 {
2912         for ( ; i; i = i->next) {
2913                 if (i->ifindex == dev->ifindex)
2914                         packet_dev_mc(dev, i, what);
2915         }
2916 }
2917
2918 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2919 {
2920         struct packet_sock *po = pkt_sk(sk);
2921         struct packet_mclist *ml, *i;
2922         struct net_device *dev;
2923         int err;
2924
2925         rtnl_lock();
2926
2927         err = -ENODEV;
2928         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2929         if (!dev)
2930                 goto done;
2931
2932         err = -EINVAL;
2933         if (mreq->mr_alen > dev->addr_len)
2934                 goto done;
2935
2936         err = -ENOBUFS;
2937         i = kmalloc(sizeof(*i), GFP_KERNEL);
2938         if (i == NULL)
2939                 goto done;
2940
2941         err = 0;
2942         for (ml = po->mclist; ml; ml = ml->next) {
2943                 if (ml->ifindex == mreq->mr_ifindex &&
2944                     ml->type == mreq->mr_type &&
2945                     ml->alen == mreq->mr_alen &&
2946                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2947                         ml->count++;
2948                         /* Free the new element ... */
2949                         kfree(i);
2950                         goto done;
2951                 }
2952         }
2953
2954         i->type = mreq->mr_type;
2955         i->ifindex = mreq->mr_ifindex;
2956         i->alen = mreq->mr_alen;
2957         memcpy(i->addr, mreq->mr_address, i->alen);
2958         i->count = 1;
2959         i->next = po->mclist;
2960         po->mclist = i;
2961         err = packet_dev_mc(dev, i, 1);
2962         if (err) {
2963                 po->mclist = i->next;
2964                 kfree(i);
2965         }
2966
2967 done:
2968         rtnl_unlock();
2969         return err;
2970 }
2971
2972 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2973 {
2974         struct packet_mclist *ml, **mlp;
2975
2976         rtnl_lock();
2977
2978         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
2979                 if (ml->ifindex == mreq->mr_ifindex &&
2980                     ml->type == mreq->mr_type &&
2981                     ml->alen == mreq->mr_alen &&
2982                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2983                         if (--ml->count == 0) {
2984                                 struct net_device *dev;
2985                                 *mlp = ml->next;
2986                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2987                                 if (dev)
2988                                         packet_dev_mc(dev, ml, -1);
2989                                 kfree(ml);
2990                         }
2991                         rtnl_unlock();
2992                         return 0;
2993                 }
2994         }
2995         rtnl_unlock();
2996         return -EADDRNOTAVAIL;
2997 }
2998
2999 static void packet_flush_mclist(struct sock *sk)
3000 {
3001         struct packet_sock *po = pkt_sk(sk);
3002         struct packet_mclist *ml;
3003
3004         if (!po->mclist)
3005                 return;
3006
3007         rtnl_lock();
3008         while ((ml = po->mclist) != NULL) {
3009                 struct net_device *dev;
3010
3011                 po->mclist = ml->next;
3012                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3013                 if (dev != NULL)
3014                         packet_dev_mc(dev, ml, -1);
3015                 kfree(ml);
3016         }
3017         rtnl_unlock();
3018 }
3019
3020 static int
3021 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3022 {
3023         struct sock *sk = sock->sk;
3024         struct packet_sock *po = pkt_sk(sk);
3025         int ret;
3026
3027         if (level != SOL_PACKET)
3028                 return -ENOPROTOOPT;
3029
3030         switch (optname) {
3031         case PACKET_ADD_MEMBERSHIP:
3032         case PACKET_DROP_MEMBERSHIP:
3033         {
3034                 struct packet_mreq_max mreq;
3035                 int len = optlen;
3036                 memset(&mreq, 0, sizeof(mreq));
3037                 if (len < sizeof(struct packet_mreq))
3038                         return -EINVAL;
3039                 if (len > sizeof(mreq))
3040                         len = sizeof(mreq);
3041                 if (copy_from_user(&mreq, optval, len))
3042                         return -EFAULT;
3043                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3044                         return -EINVAL;
3045                 if (optname == PACKET_ADD_MEMBERSHIP)
3046                         ret = packet_mc_add(sk, &mreq);
3047                 else
3048                         ret = packet_mc_drop(sk, &mreq);
3049                 return ret;
3050         }
3051
3052         case PACKET_RX_RING:
3053         case PACKET_TX_RING:
3054         {
3055                 union tpacket_req_u req_u;
3056                 int len;
3057
3058                 switch (po->tp_version) {
3059                 case TPACKET_V1:
3060                 case TPACKET_V2:
3061                         len = sizeof(req_u.req);
3062                         break;
3063                 case TPACKET_V3:
3064                 default:
3065                         len = sizeof(req_u.req3);
3066                         break;
3067                 }
3068                 if (optlen < len)
3069                         return -EINVAL;
3070                 if (pkt_sk(sk)->has_vnet_hdr)
3071                         return -EINVAL;
3072                 if (copy_from_user(&req_u.req, optval, len))
3073                         return -EFAULT;
3074                 return packet_set_ring(sk, &req_u, 0,
3075                         optname == PACKET_TX_RING);
3076         }
3077         case PACKET_COPY_THRESH:
3078         {
3079                 int val;
3080
3081                 if (optlen != sizeof(val))
3082                         return -EINVAL;
3083                 if (copy_from_user(&val, optval, sizeof(val)))
3084                         return -EFAULT;
3085
3086                 pkt_sk(sk)->copy_thresh = val;
3087                 return 0;
3088         }
3089         case PACKET_VERSION:
3090         {
3091                 int val;
3092
3093                 if (optlen != sizeof(val))
3094                         return -EINVAL;
3095                 if (copy_from_user(&val, optval, sizeof(val)))
3096                         return -EFAULT;
3097                 switch (val) {
3098                 case TPACKET_V1:
3099                 case TPACKET_V2:
3100                 case TPACKET_V3:
3101                         break;
3102                 default:
3103                         return -EINVAL;
3104                 }
3105                 lock_sock(sk);
3106                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3107                         ret = -EBUSY;
3108                 } else {
3109                         po->tp_version = val;
3110                         ret = 0;
3111                 }
3112                 release_sock(sk);
3113                 return ret;
3114         }
3115         case PACKET_RESERVE:
3116         {
3117                 unsigned int val;
3118
3119                 if (optlen != sizeof(val))
3120                         return -EINVAL;
3121                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3122                         return -EBUSY;
3123                 if (copy_from_user(&val, optval, sizeof(val)))
3124                         return -EFAULT;
3125                 po->tp_reserve = val;
3126                 return 0;
3127         }
3128         case PACKET_LOSS:
3129         {
3130                 unsigned int val;
3131
3132                 if (optlen != sizeof(val))
3133                         return -EINVAL;
3134                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3135                         return -EBUSY;
3136                 if (copy_from_user(&val, optval, sizeof(val)))
3137                         return -EFAULT;
3138                 po->tp_loss = !!val;
3139                 return 0;
3140         }
3141         case PACKET_AUXDATA:
3142         {
3143                 int val;
3144
3145                 if (optlen < sizeof(val))
3146                         return -EINVAL;
3147                 if (copy_from_user(&val, optval, sizeof(val)))
3148                         return -EFAULT;
3149
3150                 po->auxdata = !!val;
3151                 return 0;
3152         }
3153         case PACKET_ORIGDEV:
3154         {
3155                 int val;
3156
3157                 if (optlen < sizeof(val))
3158                         return -EINVAL;
3159                 if (copy_from_user(&val, optval, sizeof(val)))
3160                         return -EFAULT;
3161
3162                 po->origdev = !!val;
3163                 return 0;
3164         }
3165         case PACKET_VNET_HDR:
3166         {
3167                 int val;
3168
3169                 if (sock->type != SOCK_RAW)
3170                         return -EINVAL;
3171                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3172                         return -EBUSY;
3173                 if (optlen < sizeof(val))
3174                         return -EINVAL;
3175                 if (copy_from_user(&val, optval, sizeof(val)))
3176                         return -EFAULT;
3177
3178                 po->has_vnet_hdr = !!val;
3179                 return 0;
3180         }
3181         case PACKET_TIMESTAMP:
3182         {
3183                 int val;
3184
3185                 if (optlen != sizeof(val))
3186                         return -EINVAL;
3187                 if (copy_from_user(&val, optval, sizeof(val)))
3188                         return -EFAULT;
3189
3190                 po->tp_tstamp = val;
3191                 return 0;
3192         }
3193         case PACKET_FANOUT:
3194         {
3195                 int val;
3196
3197                 if (optlen != sizeof(val))
3198                         return -EINVAL;
3199                 if (copy_from_user(&val, optval, sizeof(val)))
3200                         return -EFAULT;
3201
3202                 return fanout_add(sk, val & 0xffff, val >> 16);
3203         }
3204         default:
3205                 return -ENOPROTOOPT;
3206         }
3207 }
3208
3209 static int packet_getsockopt(struct socket *sock, int level, int optname,
3210                              char __user *optval, int __user *optlen)
3211 {
3212         int len;
3213         int val;
3214         struct sock *sk = sock->sk;
3215         struct packet_sock *po = pkt_sk(sk);
3216         void *data;
3217         struct tpacket_stats st;
3218         union tpacket_stats_u st_u;
3219
3220         if (level != SOL_PACKET)
3221                 return -ENOPROTOOPT;
3222
3223         if (get_user(len, optlen))
3224                 return -EFAULT;
3225
3226         if (len < 0)
3227                 return -EINVAL;
3228
3229         switch (optname) {
3230         case PACKET_STATISTICS:
3231                 if (po->tp_version == TPACKET_V3) {
3232                         len = sizeof(struct tpacket_stats_v3);
3233                 } else {
3234                         if (len > sizeof(struct tpacket_stats))
3235                                 len = sizeof(struct tpacket_stats);
3236                 }
3237                 spin_lock_bh(&sk->sk_receive_queue.lock);
3238                 if (po->tp_version == TPACKET_V3) {
3239                         memcpy(&st_u.stats3, &po->stats,
3240                         sizeof(struct tpacket_stats));
3241                         st_u.stats3.tp_freeze_q_cnt =
3242                         po->stats_u.stats3.tp_freeze_q_cnt;
3243                         st_u.stats3.tp_packets += po->stats.tp_drops;
3244                         data = &st_u.stats3;
3245                 } else {
3246                         st = po->stats;
3247                         st.tp_packets += st.tp_drops;
3248                         data = &st;
3249                 }
3250                 memset(&po->stats, 0, sizeof(st));
3251                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3252                 break;
3253         case PACKET_AUXDATA:
3254                 if (len > sizeof(int))
3255                         len = sizeof(int);
3256                 val = po->auxdata;
3257
3258                 data = &val;
3259                 break;
3260         case PACKET_ORIGDEV:
3261                 if (len > sizeof(int))
3262                         len = sizeof(int);
3263                 val = po->origdev;
3264
3265                 data = &val;
3266                 break;
3267         case PACKET_VNET_HDR:
3268                 if (len > sizeof(int))
3269                         len = sizeof(int);
3270                 val = po->has_vnet_hdr;
3271
3272                 data = &val;
3273                 break;
3274         case PACKET_VERSION:
3275                 if (len > sizeof(int))
3276                         len = sizeof(int);
3277                 val = po->tp_version;
3278                 data = &val;
3279                 break;
3280         case PACKET_HDRLEN:
3281                 if (len > sizeof(int))
3282                         len = sizeof(int);
3283                 if (copy_from_user(&val, optval, len))
3284                         return -EFAULT;
3285                 switch (val) {
3286                 case TPACKET_V1:
3287                         val = sizeof(struct tpacket_hdr);
3288                         break;
3289                 case TPACKET_V2:
3290                         val = sizeof(struct tpacket2_hdr);
3291                         break;
3292                 case TPACKET_V3:
3293                         val = sizeof(struct tpacket3_hdr);
3294                         break;
3295                 default:
3296                         return -EINVAL;
3297                 }
3298                 data = &val;
3299                 break;
3300         case PACKET_RESERVE:
3301                 if (len > sizeof(unsigned int))
3302                         len = sizeof(unsigned int);
3303                 val = po->tp_reserve;
3304                 data = &val;
3305                 break;
3306         case PACKET_LOSS:
3307                 if (len > sizeof(unsigned int))
3308                         len = sizeof(unsigned int);
3309                 val = po->tp_loss;
3310                 data = &val;
3311                 break;
3312         case PACKET_TIMESTAMP:
3313                 if (len > sizeof(int))
3314                         len = sizeof(int);
3315                 val = po->tp_tstamp;
3316                 data = &val;
3317                 break;
3318         case PACKET_FANOUT:
3319                 if (len > sizeof(int))
3320                         len = sizeof(int);
3321                 val = (po->fanout ?
3322                        ((u32)po->fanout->id |
3323                         ((u32)po->fanout->type << 16)) :
3324                        0);
3325                 data = &val;
3326                 break;
3327         default:
3328                 return -ENOPROTOOPT;
3329         }
3330
3331         if (put_user(len, optlen))
3332                 return -EFAULT;
3333         if (copy_to_user(optval, data, len))
3334                 return -EFAULT;
3335         return 0;
3336 }
3337
3338
3339 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3340 {
3341         struct sock *sk;
3342         struct hlist_node *node;
3343         struct net_device *dev = data;
3344         struct net *net = dev_net(dev);
3345
3346         rcu_read_lock();
3347         sk_for_each_rcu(sk, node, &net->packet.sklist) {
3348                 struct packet_sock *po = pkt_sk(sk);
3349
3350                 switch (msg) {
3351                 case NETDEV_UNREGISTER:
3352                         if (po->mclist)
3353                                 packet_dev_mclist(dev, po->mclist, -1);
3354                         /* fallthrough */
3355
3356                 case NETDEV_DOWN:
3357                         if (dev->ifindex == po->ifindex) {
3358                                 spin_lock(&po->bind_lock);
3359                                 if (po->running) {
3360                                         __unregister_prot_hook(sk, false);
3361                                         sk->sk_err = ENETDOWN;
3362                                         if (!sock_flag(sk, SOCK_DEAD))
3363                                                 sk->sk_error_report(sk);
3364                                 }
3365                                 if (msg == NETDEV_UNREGISTER) {
3366                                         po->ifindex = -1;
3367                                         if (po->prot_hook.dev)
3368                                                 dev_put(po->prot_hook.dev);
3369                                         po->prot_hook.dev = NULL;
3370                                 }
3371                                 spin_unlock(&po->bind_lock);
3372                         }
3373                         break;
3374                 case NETDEV_UP:
3375                         if (dev->ifindex == po->ifindex) {
3376                                 spin_lock(&po->bind_lock);
3377                                 if (po->num)
3378                                         register_prot_hook(sk);
3379                                 spin_unlock(&po->bind_lock);
3380                         }
3381                         break;
3382                 }
3383         }
3384         rcu_read_unlock();
3385         return NOTIFY_DONE;
3386 }
3387
3388
3389 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3390                         unsigned long arg)
3391 {
3392         struct sock *sk = sock->sk;
3393
3394         switch (cmd) {
3395         case SIOCOUTQ:
3396         {
3397                 int amount = sk_wmem_alloc_get(sk);
3398
3399                 return put_user(amount, (int __user *)arg);
3400         }
3401         case SIOCINQ:
3402         {
3403                 struct sk_buff *skb;
3404                 int amount = 0;
3405
3406                 spin_lock_bh(&sk->sk_receive_queue.lock);
3407                 skb = skb_peek(&sk->sk_receive_queue);
3408                 if (skb)
3409                         amount = skb->len;
3410                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3411                 return put_user(amount, (int __user *)arg);
3412         }
3413         case SIOCGSTAMP:
3414                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3415         case SIOCGSTAMPNS:
3416                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3417
3418 #ifdef CONFIG_INET
3419         case SIOCADDRT:
3420         case SIOCDELRT:
3421         case SIOCDARP:
3422         case SIOCGARP:
3423         case SIOCSARP:
3424         case SIOCGIFADDR:
3425         case SIOCSIFADDR:
3426         case SIOCGIFBRDADDR:
3427         case SIOCSIFBRDADDR:
3428         case SIOCGIFNETMASK:
3429         case SIOCSIFNETMASK:
3430         case SIOCGIFDSTADDR:
3431         case SIOCSIFDSTADDR:
3432         case SIOCSIFFLAGS:
3433                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3434 #endif
3435
3436         default:
3437                 return -ENOIOCTLCMD;
3438         }
3439         return 0;
3440 }
3441
3442 static unsigned int packet_poll(struct file *file, struct socket *sock,
3443                                 poll_table *wait)
3444 {
3445         struct sock *sk = sock->sk;
3446         struct packet_sock *po = pkt_sk(sk);
3447         unsigned int mask = datagram_poll(file, sock, wait);
3448
3449         spin_lock_bh(&sk->sk_receive_queue.lock);
3450         if (po->rx_ring.pg_vec) {
3451                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3452                         TP_STATUS_KERNEL))
3453                         mask |= POLLIN | POLLRDNORM;
3454         }
3455         spin_unlock_bh(&sk->sk_receive_queue.lock);
3456         spin_lock_bh(&sk->sk_write_queue.lock);
3457         if (po->tx_ring.pg_vec) {
3458                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3459                         mask |= POLLOUT | POLLWRNORM;
3460         }
3461         spin_unlock_bh(&sk->sk_write_queue.lock);
3462         return mask;
3463 }
3464
3465
3466 /* Dirty? Well, I still did not learn better way to account
3467  * for user mmaps.
3468  */
3469
3470 static void packet_mm_open(struct vm_area_struct *vma)
3471 {
3472         struct file *file = vma->vm_file;
3473         struct socket *sock = file->private_data;
3474         struct sock *sk = sock->sk;
3475
3476         if (sk)
3477                 atomic_inc(&pkt_sk(sk)->mapped);
3478 }
3479
3480 static void packet_mm_close(struct vm_area_struct *vma)
3481 {
3482         struct file *file = vma->vm_file;
3483         struct socket *sock = file->private_data;
3484         struct sock *sk = sock->sk;
3485
3486         if (sk)
3487                 atomic_dec(&pkt_sk(sk)->mapped);
3488 }
3489
3490 static const struct vm_operations_struct packet_mmap_ops = {
3491         .open   =       packet_mm_open,
3492         .close  =       packet_mm_close,
3493 };
3494
3495 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3496                         unsigned int len)
3497 {
3498         int i;
3499
3500         for (i = 0; i < len; i++) {
3501                 if (likely(pg_vec[i].buffer)) {
3502                         if (is_vmalloc_addr(pg_vec[i].buffer))
3503                                 vfree(pg_vec[i].buffer);
3504                         else
3505                                 free_pages((unsigned long)pg_vec[i].buffer,
3506                                            order);
3507                         pg_vec[i].buffer = NULL;
3508                 }
3509         }
3510         kfree(pg_vec);
3511 }
3512
3513 static char *alloc_one_pg_vec_page(unsigned long order)
3514 {
3515         char *buffer = NULL;
3516         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3517                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3518
3519         buffer = (char *) __get_free_pages(gfp_flags, order);
3520
3521         if (buffer)
3522                 return buffer;
3523
3524         /*
3525          * __get_free_pages failed, fall back to vmalloc
3526          */
3527         buffer = vzalloc((1 << order) * PAGE_SIZE);
3528
3529         if (buffer)
3530                 return buffer;
3531
3532         /*
3533          * vmalloc failed, lets dig into swap here
3534          */
3535         gfp_flags &= ~__GFP_NORETRY;
3536         buffer = (char *)__get_free_pages(gfp_flags, order);
3537         if (buffer)
3538                 return buffer;
3539
3540         /*
3541          * complete and utter failure
3542          */
3543         return NULL;
3544 }
3545
3546 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3547 {
3548         unsigned int block_nr = req->tp_block_nr;
3549         struct pgv *pg_vec;
3550         int i;
3551
3552         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3553         if (unlikely(!pg_vec))
3554                 goto out;
3555
3556         for (i = 0; i < block_nr; i++) {
3557                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3558                 if (unlikely(!pg_vec[i].buffer))
3559                         goto out_free_pgvec;
3560         }
3561
3562 out:
3563         return pg_vec;
3564
3565 out_free_pgvec:
3566         free_pg_vec(pg_vec, order, block_nr);
3567         pg_vec = NULL;
3568         goto out;
3569 }
3570
3571 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3572                 int closing, int tx_ring)
3573 {
3574         struct pgv *pg_vec = NULL;
3575         struct packet_sock *po = pkt_sk(sk);
3576         int was_running, order = 0;
3577         struct packet_ring_buffer *rb;
3578         struct sk_buff_head *rb_queue;
3579         __be16 num;
3580         int err = -EINVAL;
3581         /* Added to avoid minimal code churn */
3582         struct tpacket_req *req = &req_u->req;
3583
3584         lock_sock(sk);
3585         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3586         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3587                 WARN(1, "Tx-ring is not supported.\n");
3588                 goto out;
3589         }
3590
3591         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3592         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3593
3594         err = -EBUSY;
3595         if (!closing) {
3596                 if (atomic_read(&po->mapped))
3597                         goto out;
3598                 if (atomic_read(&rb->pending))
3599                         goto out;
3600         }
3601
3602         if (req->tp_block_nr) {
3603                 /* Sanity tests and some calculations */
3604                 err = -EBUSY;
3605                 if (unlikely(rb->pg_vec))
3606                         goto out;
3607
3608                 switch (po->tp_version) {
3609                 case TPACKET_V1:
3610                         po->tp_hdrlen = TPACKET_HDRLEN;
3611                         break;
3612                 case TPACKET_V2:
3613                         po->tp_hdrlen = TPACKET2_HDRLEN;
3614                         break;
3615                 case TPACKET_V3:
3616                         po->tp_hdrlen = TPACKET3_HDRLEN;
3617                         break;
3618                 }
3619
3620                 err = -EINVAL;
3621                 if (unlikely((int)req->tp_block_size <= 0))
3622                         goto out;
3623                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3624                         goto out;
3625                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3626                                         po->tp_reserve))
3627                         goto out;
3628                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3629                         goto out;
3630
3631                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3632                 if (unlikely(rb->frames_per_block <= 0))
3633                         goto out;
3634                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3635                                         req->tp_frame_nr))
3636                         goto out;
3637
3638                 err = -ENOMEM;
3639                 order = get_order(req->tp_block_size);
3640                 pg_vec = alloc_pg_vec(req, order);
3641                 if (unlikely(!pg_vec))
3642                         goto out;
3643                 switch (po->tp_version) {
3644                 case TPACKET_V3:
3645                 /* Transmit path is not supported. We checked
3646                  * it above but just being paranoid
3647                  */
3648                         if (!tx_ring)
3649                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3650                                 break;
3651                 default:
3652                         break;
3653                 }
3654         }
3655         /* Done */
3656         else {
3657                 err = -EINVAL;
3658                 if (unlikely(req->tp_frame_nr))
3659                         goto out;
3660         }
3661
3662
3663         /* Detach socket from network */
3664         spin_lock(&po->bind_lock);
3665         was_running = po->running;
3666         num = po->num;
3667         if (was_running) {
3668                 po->num = 0;
3669                 __unregister_prot_hook(sk, false);
3670         }
3671         spin_unlock(&po->bind_lock);
3672
3673         synchronize_net();
3674
3675         err = -EBUSY;
3676         mutex_lock(&po->pg_vec_lock);
3677         if (closing || atomic_read(&po->mapped) == 0) {
3678                 err = 0;
3679                 spin_lock_bh(&rb_queue->lock);
3680                 swap(rb->pg_vec, pg_vec);
3681                 rb->frame_max = (req->tp_frame_nr - 1);
3682                 rb->head = 0;
3683                 rb->frame_size = req->tp_frame_size;
3684                 spin_unlock_bh(&rb_queue->lock);
3685
3686                 swap(rb->pg_vec_order, order);
3687                 swap(rb->pg_vec_len, req->tp_block_nr);
3688
3689                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3690                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3691                                                 tpacket_rcv : packet_rcv;
3692                 skb_queue_purge(rb_queue);
3693                 if (atomic_read(&po->mapped))
3694                         pr_err("packet_mmap: vma is busy: %d\n",
3695                                atomic_read(&po->mapped));
3696         }
3697         mutex_unlock(&po->pg_vec_lock);
3698
3699         spin_lock(&po->bind_lock);
3700         if (was_running) {
3701                 po->num = num;
3702                 register_prot_hook(sk);
3703         }
3704         spin_unlock(&po->bind_lock);
3705         if (closing && (po->tp_version > TPACKET_V2)) {
3706                 /* Because we don't support block-based V3 on tx-ring */
3707                 if (!tx_ring)
3708                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3709         }
3710
3711         if (pg_vec)
3712                 free_pg_vec(pg_vec, order, req->tp_block_nr);
3713 out:
3714         release_sock(sk);
3715         return err;
3716 }
3717
3718 static int packet_mmap(struct file *file, struct socket *sock,
3719                 struct vm_area_struct *vma)
3720 {
3721         struct sock *sk = sock->sk;
3722         struct packet_sock *po = pkt_sk(sk);
3723         unsigned long size, expected_size;
3724         struct packet_ring_buffer *rb;
3725         unsigned long start;
3726         int err = -EINVAL;
3727         int i;
3728
3729         if (vma->vm_pgoff)
3730                 return -EINVAL;
3731
3732         mutex_lock(&po->pg_vec_lock);
3733
3734         expected_size = 0;
3735         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3736                 if (rb->pg_vec) {
3737                         expected_size += rb->pg_vec_len
3738                                                 * rb->pg_vec_pages
3739                                                 * PAGE_SIZE;
3740                 }
3741         }
3742
3743         if (expected_size == 0)
3744                 goto out;
3745
3746         size = vma->vm_end - vma->vm_start;
3747         if (size != expected_size)
3748                 goto out;
3749
3750         start = vma->vm_start;
3751         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3752                 if (rb->pg_vec == NULL)
3753                         continue;
3754
3755                 for (i = 0; i < rb->pg_vec_len; i++) {
3756                         struct page *page;
3757                         void *kaddr = rb->pg_vec[i].buffer;
3758                         int pg_num;
3759
3760                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3761                                 page = pgv_to_page(kaddr);
3762                                 err = vm_insert_page(vma, start, page);
3763                                 if (unlikely(err))
3764                                         goto out;
3765                                 start += PAGE_SIZE;
3766                                 kaddr += PAGE_SIZE;
3767                         }
3768                 }
3769         }
3770
3771         atomic_inc(&po->mapped);
3772         vma->vm_ops = &packet_mmap_ops;
3773         err = 0;
3774
3775 out:
3776         mutex_unlock(&po->pg_vec_lock);
3777         return err;
3778 }
3779
3780 static const struct proto_ops packet_ops_spkt = {
3781         .family =       PF_PACKET,
3782         .owner =        THIS_MODULE,
3783         .release =      packet_release,
3784         .bind =         packet_bind_spkt,
3785         .connect =      sock_no_connect,
3786         .socketpair =   sock_no_socketpair,
3787         .accept =       sock_no_accept,
3788         .getname =      packet_getname_spkt,
3789         .poll =         datagram_poll,
3790         .ioctl =        packet_ioctl,
3791         .listen =       sock_no_listen,
3792         .shutdown =     sock_no_shutdown,
3793         .setsockopt =   sock_no_setsockopt,
3794         .getsockopt =   sock_no_getsockopt,
3795         .sendmsg =      packet_sendmsg_spkt,
3796         .recvmsg =      packet_recvmsg,
3797         .mmap =         sock_no_mmap,
3798         .sendpage =     sock_no_sendpage,
3799 };
3800
3801 static const struct proto_ops packet_ops = {
3802         .family =       PF_PACKET,
3803         .owner =        THIS_MODULE,
3804         .release =      packet_release,
3805         .bind =         packet_bind,
3806         .connect =      sock_no_connect,
3807         .socketpair =   sock_no_socketpair,
3808         .accept =       sock_no_accept,
3809         .getname =      packet_getname,
3810         .poll =         packet_poll,
3811         .ioctl =        packet_ioctl,
3812         .listen =       sock_no_listen,
3813         .shutdown =     sock_no_shutdown,
3814         .setsockopt =   packet_setsockopt,
3815         .getsockopt =   packet_getsockopt,
3816         .sendmsg =      packet_sendmsg,
3817         .recvmsg =      packet_recvmsg,
3818         .mmap =         packet_mmap,
3819         .sendpage =     sock_no_sendpage,
3820 };
3821
3822 static const struct net_proto_family packet_family_ops = {
3823         .family =       PF_PACKET,
3824         .create =       packet_create,
3825         .owner  =       THIS_MODULE,
3826 };
3827
3828 static struct notifier_block packet_netdev_notifier = {
3829         .notifier_call =        packet_notifier,
3830 };
3831
3832 #ifdef CONFIG_PROC_FS
3833
3834 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3835         __acquires(RCU)
3836 {
3837         struct net *net = seq_file_net(seq);
3838
3839         rcu_read_lock();
3840         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3841 }
3842
3843 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3844 {
3845         struct net *net = seq_file_net(seq);
3846         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3847 }
3848
3849 static void packet_seq_stop(struct seq_file *seq, void *v)
3850         __releases(RCU)
3851 {
3852         rcu_read_unlock();
3853 }
3854
3855 static int packet_seq_show(struct seq_file *seq, void *v)
3856 {
3857         if (v == SEQ_START_TOKEN)
3858                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
3859         else {
3860                 struct sock *s = sk_entry(v);
3861                 const struct packet_sock *po = pkt_sk(s);
3862
3863                 seq_printf(seq,
3864                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
3865                            s,
3866                            atomic_read(&s->sk_refcnt),
3867                            s->sk_type,
3868                            ntohs(po->num),
3869                            po->ifindex,
3870                            po->running,
3871                            atomic_read(&s->sk_rmem_alloc),
3872                            sock_i_uid(s),
3873                            sock_i_ino(s));
3874         }
3875
3876         return 0;
3877 }
3878
3879 static const struct seq_operations packet_seq_ops = {
3880         .start  = packet_seq_start,
3881         .next   = packet_seq_next,
3882         .stop   = packet_seq_stop,
3883         .show   = packet_seq_show,
3884 };
3885
3886 static int packet_seq_open(struct inode *inode, struct file *file)
3887 {
3888         return seq_open_net(inode, file, &packet_seq_ops,
3889                             sizeof(struct seq_net_private));
3890 }
3891
3892 static const struct file_operations packet_seq_fops = {
3893         .owner          = THIS_MODULE,
3894         .open           = packet_seq_open,
3895         .read           = seq_read,
3896         .llseek         = seq_lseek,
3897         .release        = seq_release_net,
3898 };
3899
3900 #endif
3901
3902 static int __net_init packet_net_init(struct net *net)
3903 {
3904         spin_lock_init(&net->packet.sklist_lock);
3905         INIT_HLIST_HEAD(&net->packet.sklist);
3906
3907         if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
3908                 return -ENOMEM;
3909
3910         return 0;
3911 }
3912
3913 static void __net_exit packet_net_exit(struct net *net)
3914 {
3915         proc_net_remove(net, "packet");
3916 }
3917
3918 static struct pernet_operations packet_net_ops = {
3919         .init = packet_net_init,
3920         .exit = packet_net_exit,
3921 };
3922
3923
3924 static void __exit packet_exit(void)
3925 {
3926         unregister_netdevice_notifier(&packet_netdev_notifier);
3927         unregister_pernet_subsys(&packet_net_ops);
3928         sock_unregister(PF_PACKET);
3929         proto_unregister(&packet_proto);
3930 }
3931
3932 static int __init packet_init(void)
3933 {
3934         int rc = proto_register(&packet_proto, 0);
3935
3936         if (rc != 0)
3937                 goto out;
3938
3939         sock_register(&packet_family_ops);
3940         register_pernet_subsys(&packet_net_ops);
3941         register_netdevice_notifier(&packet_netdev_notifier);
3942 out:
3943         return rc;
3944 }
3945
3946 module_init(packet_init);
3947 module_exit(packet_exit);
3948 MODULE_LICENSE("GPL");
3949 MODULE_ALIAS_NETPROTO(PF_PACKET);