Pileus Git - ~andy/linux/blob - net/packet/af_packet.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              PACKET - implements raw packet sockets.
   7  *
   8  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  11  *
  12  * Fixes:
  13  *              Alan Cox        :       verify_area() now used correctly
  14  *              Alan Cox        :       new skbuff lists, look ma no backlogs!
  15  *              Alan Cox        :       tidied skbuff lists.
  16  *              Alan Cox        :       Now uses generic datagram routines I
  17  *                                      added. Also fixed the peek/read crash
  18  *                                      from all old Linux datagram code.
  19  *              Alan Cox        :       Uses the improved datagram code.
  20  *              Alan Cox        :       Added NULL's for socket options.
  21  *              Alan Cox        :       Re-commented the code.
  22  *              Alan Cox        :       Use new kernel side addressing
  23  *              Rob Janssen     :       Correct MTU usage.
  24  *              Dave Platt      :       Counter leaks caused by incorrect
  25  *                                      interrupt locking and some slightly
  26  *                                      dubious gcc output. Can you read
  27  *                                      compiler: it said _VOLATILE_
  28  *      Richard Kooijman        :       Timestamp fixes.
  29  *              Alan Cox        :       New buffers. Use sk->mac.raw.
  30  *              Alan Cox        :       sendmsg/recvmsg support.
  31  *              Alan Cox        :       Protocol setting support
  32  *      Alexey Kuznetsov        :       Untied from IPv4 stack.
  33  *      Cyrus Durgin            :       Fixed kerneld for kmod.
  34  *      Michal Ostrowski        :       Module initialization cleanup.
  35  *         Ulises Alonso        :       Frame number limit removal and
  36  *                                      packet_set_ring memory leak.
  37  *              Eric Biederman  :       Allow for > 8 byte hardware addresses.
  38  *                                      The convention is that longer addresses
  39  *                                      will simply extend the hardware address
  40  *                                      byte arrays at the end of sockaddr_ll
  41  *                                      and packet_mreq.
  42  *              Johann Baudy    :       Added TX RING.
  43  *              Chetan Loke     :       Implemented TPACKET_V3 block abstraction
  44  *                                      layer.
  45  *                                      Copyright (C) 2011, <lokec@ccs.neu.edu>
  46  *
  47  *
  48  *              This program is free software; you can redistribute it and/or
  49  *              modify it under the terms of the GNU General Public License
  50  *              as published by the Free Software Foundation; either version
  51  *              2 of the License, or (at your option) any later version.
  52  *
  53  */
  54
  55 #include <linux/types.h>
  56 #include <linux/mm.h>
  57 #include <linux/capability.h>
  58 #include <linux/fcntl.h>
  59 #include <linux/socket.h>
  60 #include <linux/in.h>
  61 #include <linux/inet.h>
  62 #include <linux/netdevice.h>
  63 #include <linux/if_packet.h>
  64 #include <linux/wireless.h>
  65 #include <linux/kernel.h>
  66 #include <linux/kmod.h>
  67 #include <linux/slab.h>
  68 #include <linux/vmalloc.h>
  69 #include <net/net_namespace.h>
  70 #include <net/ip.h>
  71 #include <net/protocol.h>
  72 #include <linux/skbuff.h>
  73 #include <net/sock.h>
  74 #include <linux/errno.h>
  75 #include <linux/timer.h>
  76 #include <asm/uaccess.h>
  77 #include <asm/ioctls.h>
  78 #include <asm/page.h>
  79 #include <asm/cacheflush.h>
  80 #include <asm/io.h>
  81 #include <linux/proc_fs.h>
  82 #include <linux/seq_file.h>
  83 #include <linux/poll.h>
  84 #include <linux/module.h>
  85 #include <linux/init.h>
  86 #include <linux/mutex.h>
  87 #include <linux/if_vlan.h>
  88 #include <linux/virtio_net.h>
  89 #include <linux/errqueue.h>
  90 #include <linux/net_tstamp.h>
  91
  92 #ifdef CONFIG_INET
  93 #include <net/inet_common.h>
  94 #endif
  95
  96 #include "internal.h"
  97
  98 /*
  99    Assumptions:
 100    - if device has no dev->hard_header routine, it adds and removes ll header
 101      inside itself. In this case ll header is invisible outside of device,
 102      but higher levels still should reserve dev->hard_header_len.
 103      Some devices are enough clever to reallocate skb, when header
 104      will not fit to reserved space (tunnel), another ones are silly
 105      (PPP).
 106    - packet socket receives packets with pulled ll header,
 107      so that SOCK_RAW should push it back.
 108
 109 On receive:
 110 -----------
 111
 112 Incoming, dev->hard_header!=NULL
 113    mac_header -> ll header
 114    data       -> data
 115
 116 Outgoing, dev->hard_header!=NULL
 117    mac_header -> ll header
 118    data       -> ll header
 119
 120 Incoming, dev->hard_header==NULL
 121    mac_header -> UNKNOWN position. It is very likely, that it points to ll
 122                  header.  PPP makes it, that is wrong, because introduce
 123                  assymetry between rx and tx paths.
 124    data       -> data
 125
 126 Outgoing, dev->hard_header==NULL
 127    mac_header -> data. ll header is still not built!
 128    data       -> data
 129
 130 Resume
 131   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 132
 133
 134 On transmit:
 135 ------------
 136
 137 dev->hard_header != NULL
 138    mac_header -> ll header
 139    data       -> ll header
 140
 141 dev->hard_header == NULL (ll header is added by device, we cannot control it)
 142    mac_header -> data
 143    data       -> data
 144
 145    We should set nh.raw on output to correct posistion,
 146    packet classifier depends on it.
 147  */
 148
 149 /* Private packet socket structures. */
 150
 151 /* identical to struct packet_mreq except it has
 152  * a longer address field.
 153  */
 154 struct packet_mreq_max {
 155         int             mr_ifindex;
 156         unsigned short  mr_type;
 157         unsigned short  mr_alen;
 158         unsigned char   mr_address[MAX_ADDR_LEN];
 159 };
 160
 161 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 162                 int closing, int tx_ring);
 163
 164
 165 #define V3_ALIGNMENT    (8)
 166
 167 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 168
 169 #define BLK_PLUS_PRIV(sz_of_priv) \
 170         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 171
 172 #define PGV_FROM_VMALLOC 1
 173
 174 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
 175 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
 176 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
 177 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
 178 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
 179 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
 180 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
 181
 182 struct packet_sock;
 183 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
 184 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 185                        struct packet_type *pt, struct net_device *orig_dev);
 186
 187 static void *packet_previous_frame(struct packet_sock *po,
 188                 struct packet_ring_buffer *rb,
 189                 int status);
 190 static void packet_increment_head(struct packet_ring_buffer *buff);
 191 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
 192                         struct tpacket_block_desc *);
 193 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 194                         struct packet_sock *);
 195 static void prb_retire_current_block(struct tpacket_kbdq_core *,
 196                 struct packet_sock *, unsigned int status);
 197 static int prb_queue_frozen(struct tpacket_kbdq_core *);
 198 static void prb_open_block(struct tpacket_kbdq_core *,
 199                 struct tpacket_block_desc *);
 200 static void prb_retire_rx_blk_timer_expired(unsigned long);
 201 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 202 static void prb_init_blk_timer(struct packet_sock *,
 203                 struct tpacket_kbdq_core *,
 204                 void (*func) (unsigned long));
 205 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 206 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 207                 struct tpacket3_hdr *);
 208 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 209                 struct tpacket3_hdr *);
 210 static void packet_flush_mclist(struct sock *sk);
 211
 212 struct packet_skb_cb {
 213         unsigned int origlen;
 214         union {
 215                 struct sockaddr_pkt pkt;
 216                 struct sockaddr_ll ll;
 217         } sa;
 218 };
 219
 220 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 221
 222 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 223 #define GET_PBLOCK_DESC(x, bid) \
 224         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 225 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 226         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 227 #define GET_NEXT_PRB_BLK_NUM(x) \
 228         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 229         ((x)->kactive_blk_num+1) : 0)
 230
 231 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 232 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 233
 234 /* register_prot_hook must be invoked with the po->bind_lock held,
 235  * or from a context in which asynchronous accesses to the packet
 236  * socket is not possible (packet_create()).
 237  */
 238 static void register_prot_hook(struct sock *sk)
 239 {
 240         struct packet_sock *po = pkt_sk(sk);
 241         if (!po->running) {
 242                 if (po->fanout)
 243                         __fanout_link(sk, po);
 244                 else
 245                         dev_add_pack(&po->prot_hook);
 246                 sock_hold(sk);
 247                 po->running = 1;
 248         }
 249 }
 250
 251 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
 252  * held.   If the sync parameter is true, we will temporarily drop
 253  * the po->bind_lock and do a synchronize_net to make sure no
 254  * asynchronous packet processing paths still refer to the elements
 255  * of po->prot_hook.  If the sync parameter is false, it is the
 256  * callers responsibility to take care of this.
 257  */
 258 static void __unregister_prot_hook(struct sock *sk, bool sync)
 259 {
 260         struct packet_sock *po = pkt_sk(sk);
 261
 262         po->running = 0;
 263         if (po->fanout)
 264                 __fanout_unlink(sk, po);
 265         else
 266                 __dev_remove_pack(&po->prot_hook);
 267         __sock_put(sk);
 268
 269         if (sync) {
 270                 spin_unlock(&po->bind_lock);
 271                 synchronize_net();
 272                 spin_lock(&po->bind_lock);
 273         }
 274 }
 275
 276 static void unregister_prot_hook(struct sock *sk, bool sync)
 277 {
 278         struct packet_sock *po = pkt_sk(sk);
 279
 280         if (po->running)
 281                 __unregister_prot_hook(sk, sync);
 282 }
 283
 284 static inline __pure struct page *pgv_to_page(void *addr)
 285 {
 286         if (is_vmalloc_addr(addr))
 287                 return vmalloc_to_page(addr);
 288         return virt_to_page(addr);
 289 }
 290
 291 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 292 {
 293         union {
 294                 struct tpacket_hdr *h1;
 295                 struct tpacket2_hdr *h2;
 296                 void *raw;
 297         } h;
 298
 299         h.raw = frame;
 300         switch (po->tp_version) {
 301         case TPACKET_V1:
 302                 h.h1->tp_status = status;
 303                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 304                 break;
 305         case TPACKET_V2:
 306                 h.h2->tp_status = status;
 307                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 308                 break;
 309         case TPACKET_V3:
 310         default:
 311                 WARN(1, "TPACKET version not supported.\n");
 312                 BUG();
 313         }
 314
 315         smp_wmb();
 316 }
 317
 318 static int __packet_get_status(struct packet_sock *po, void *frame)
 319 {
 320         union {
 321                 struct tpacket_hdr *h1;
 322                 struct tpacket2_hdr *h2;
 323                 void *raw;
 324         } h;
 325
 326         smp_rmb();
 327
 328         h.raw = frame;
 329         switch (po->tp_version) {
 330         case TPACKET_V1:
 331                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 332                 return h.h1->tp_status;
 333         case TPACKET_V2:
 334                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 335                 return h.h2->tp_status;
 336         case TPACKET_V3:
 337         default:
 338                 WARN(1, "TPACKET version not supported.\n");
 339                 BUG();
 340                 return 0;
 341         }
 342 }
 343
 344 static void *packet_lookup_frame(struct packet_sock *po,
 345                 struct packet_ring_buffer *rb,
 346                 unsigned int position,
 347                 int status)
 348 {
 349         unsigned int pg_vec_pos, frame_offset;
 350         union {
 351                 struct tpacket_hdr *h1;
 352                 struct tpacket2_hdr *h2;
 353                 void *raw;
 354         } h;
 355
 356         pg_vec_pos = position / rb->frames_per_block;
 357         frame_offset = position % rb->frames_per_block;
 358
 359         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 360                 (frame_offset * rb->frame_size);
 361
 362         if (status != __packet_get_status(po, h.raw))
 363                 return NULL;
 364
 365         return h.raw;
 366 }
 367
 368 static void *packet_current_frame(struct packet_sock *po,
 369                 struct packet_ring_buffer *rb,
 370                 int status)
 371 {
 372         return packet_lookup_frame(po, rb, rb->head, status);
 373 }
 374
 375 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 376 {
 377         del_timer_sync(&pkc->retire_blk_timer);
 378 }
 379
 380 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 381                 int tx_ring,
 382                 struct sk_buff_head *rb_queue)
 383 {
 384         struct tpacket_kbdq_core *pkc;
 385
 386         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 387
 388         spin_lock(&rb_queue->lock);
 389         pkc->delete_blk_timer = 1;
 390         spin_unlock(&rb_queue->lock);
 391
 392         prb_del_retire_blk_timer(pkc);
 393 }
 394
 395 static void prb_init_blk_timer(struct packet_sock *po,
 396                 struct tpacket_kbdq_core *pkc,
 397                 void (*func) (unsigned long))
 398 {
 399         init_timer(&pkc->retire_blk_timer);
 400         pkc->retire_blk_timer.data = (long)po;
 401         pkc->retire_blk_timer.function = func;
 402         pkc->retire_blk_timer.expires = jiffies;
 403 }
 404
 405 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
 406 {
 407         struct tpacket_kbdq_core *pkc;
 408
 409         if (tx_ring)
 410                 BUG();
 411
 412         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 413         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
 414 }
 415
 416 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 417                                 int blk_size_in_bytes)
 418 {
 419         struct net_device *dev;
 420         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 421         struct ethtool_cmd ecmd;
 422         int err;
 423         u32 speed;
 424
 425         rtnl_lock();
 426         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 427         if (unlikely(!dev)) {
 428                 rtnl_unlock();
 429                 return DEFAULT_PRB_RETIRE_TOV;
 430         }
 431         err = __ethtool_get_settings(dev, &ecmd);
 432         speed = ethtool_cmd_speed(&ecmd);
 433         rtnl_unlock();
 434         if (!err) {
 435                 /*
 436                  * If the link speed is so slow you don't really
 437                  * need to worry about perf anyways
 438                  */
 439                 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
 440                         return DEFAULT_PRB_RETIRE_TOV;
 441                 } else {
 442                         msec = 1;
 443                         div = speed / 1000;
 444                 }
 445         }
 446
 447         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 448
 449         if (div)
 450                 mbits /= div;
 451
 452         tmo = mbits * msec;
 453
 454         if (div)
 455                 return tmo+1;
 456         return tmo;
 457 }
 458
 459 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 460                         union tpacket_req_u *req_u)
 461 {
 462         p1->feature_req_word = req_u->req3.tp_feature_req_word;
 463 }
 464
 465 static void init_prb_bdqc(struct packet_sock *po,
 466                         struct packet_ring_buffer *rb,
 467                         struct pgv *pg_vec,
 468                         union tpacket_req_u *req_u, int tx_ring)
 469 {
 470         struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
 471         struct tpacket_block_desc *pbd;
 472
 473         memset(p1, 0x0, sizeof(*p1));
 474
 475         p1->knxt_seq_num = 1;
 476         p1->pkbdq = pg_vec;
 477         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 478         p1->pkblk_start = pg_vec[0].buffer;
 479         p1->kblk_size = req_u->req3.tp_block_size;
 480         p1->knum_blocks = req_u->req3.tp_block_nr;
 481         p1->hdrlen = po->tp_hdrlen;
 482         p1->version = po->tp_version;
 483         p1->last_kactive_blk_num = 0;
 484         po->stats_u.stats3.tp_freeze_q_cnt = 0;
 485         if (req_u->req3.tp_retire_blk_tov)
 486                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 487         else
 488                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 489                                                 req_u->req3.tp_block_size);
 490         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 491         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 492
 493         prb_init_ft_ops(p1, req_u);
 494         prb_setup_retire_blk_timer(po, tx_ring);
 495         prb_open_block(p1, pbd);
 496 }
 497
 498 /*  Do NOT update the last_blk_num first.
 499  *  Assumes sk_buff_head lock is held.
 500  */
 501 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 502 {
 503         mod_timer(&pkc->retire_blk_timer,
 504                         jiffies + pkc->tov_in_jiffies);
 505         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 506 }
 507
 508 /*
 509  * Timer logic:
 510  * 1) We refresh the timer only when we open a block.
 511  *    By doing this we don't waste cycles refreshing the timer
 512  *        on packet-by-packet basis.
 513  *
 514  * With a 1MB block-size, on a 1Gbps line, it will take
 515  * i) ~8 ms to fill a block + ii) memcpy etc.
 516  * In this cut we are not accounting for the memcpy time.
 517  *
 518  * So, if the user sets the 'tmo' to 10ms then the timer
 519  * will never fire while the block is still getting filled
 520  * (which is what we want). However, the user could choose
 521  * to close a block early and that's fine.
 522  *
 523  * But when the timer does fire, we check whether or not to refresh it.
 524  * Since the tmo granularity is in msecs, it is not too expensive
 525  * to refresh the timer, lets say every '8' msecs.
 526  * Either the user can set the 'tmo' or we can derive it based on
 527  * a) line-speed and b) block-size.
 528  * prb_calc_retire_blk_tmo() calculates the tmo.
 529  *
 530  */
 531 static void prb_retire_rx_blk_timer_expired(unsigned long data)
 532 {
 533         struct packet_sock *po = (struct packet_sock *)data;
 534         struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
 535         unsigned int frozen;
 536         struct tpacket_block_desc *pbd;
 537
 538         spin_lock(&po->sk.sk_receive_queue.lock);
 539
 540         frozen = prb_queue_frozen(pkc);
 541         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 542
 543         if (unlikely(pkc->delete_blk_timer))
 544                 goto out;
 545
 546         /* We only need to plug the race when the block is partially filled.
 547          * tpacket_rcv:
 548          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 549          *              copy_bits() is in progress ...
 550          *              timer fires on other cpu:
 551          *              we can't retire the current block because copy_bits
 552          *              is in progress.
 553          *
 554          */
 555         if (BLOCK_NUM_PKTS(pbd)) {
 556                 while (atomic_read(&pkc->blk_fill_in_prog)) {
 557                         /* Waiting for skb_copy_bits to finish... */
 558                         cpu_relax();
 559                 }
 560         }
 561
 562         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 563                 if (!frozen) {
 564                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 565                         if (!prb_dispatch_next_block(pkc, po))
 566                                 goto refresh_timer;
 567                         else
 568                                 goto out;
 569                 } else {
 570                         /* Case 1. Queue was frozen because user-space was
 571                          *         lagging behind.
 572                          */
 573                         if (prb_curr_blk_in_use(pkc, pbd)) {
 574                                 /*
 575                                  * Ok, user-space is still behind.
 576                                  * So just refresh the timer.
 577                                  */
 578                                 goto refresh_timer;
 579                         } else {
 580                                /* Case 2. queue was frozen,user-space caught up,
 581                                 * now the link went idle && the timer fired.
 582                                 * We don't have a block to close.So we open this
 583                                 * block and restart the timer.
 584                                 * opening a block thaws the queue,restarts timer
 585                                 * Thawing/timer-refresh is a side effect.
 586                                 */
 587                                 prb_open_block(pkc, pbd);
 588                                 goto out;
 589                         }
 590                 }
 591         }
 592
 593 refresh_timer:
 594         _prb_refresh_rx_retire_blk_timer(pkc);
 595
 596 out:
 597         spin_unlock(&po->sk.sk_receive_queue.lock);
 598 }
 599
 600 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 601                 struct tpacket_block_desc *pbd1, __u32 status)
 602 {
 603         /* Flush everything minus the block header */
 604
 605 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 606         u8 *start, *end;
 607
 608         start = (u8 *)pbd1;
 609
 610         /* Skip the block header(we know header WILL fit in 4K) */
 611         start += PAGE_SIZE;
 612
 613         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 614         for (; start < end; start += PAGE_SIZE)
 615                 flush_dcache_page(pgv_to_page(start));
 616
 617         smp_wmb();
 618 #endif
 619
 620         /* Now update the block status. */
 621
 622         BLOCK_STATUS(pbd1) = status;
 623
 624         /* Flush the block header */
 625
 626 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 627         start = (u8 *)pbd1;
 628         flush_dcache_page(pgv_to_page(start));
 629
 630         smp_wmb();
 631 #endif
 632 }
 633
 634 /*
 635  * Side effect:
 636  *
 637  * 1) flush the block
 638  * 2) Increment active_blk_num
 639  *
 640  * Note:We DONT refresh the timer on purpose.
 641  *      Because almost always the next block will be opened.
 642  */
 643 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 644                 struct tpacket_block_desc *pbd1,
 645                 struct packet_sock *po, unsigned int stat)
 646 {
 647         __u32 status = TP_STATUS_USER | stat;
 648
 649         struct tpacket3_hdr *last_pkt;
 650         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 651
 652         if (po->stats.tp_drops)
 653                 status |= TP_STATUS_LOSING;
 654
 655         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 656         last_pkt->tp_next_offset = 0;
 657
 658         /* Get the ts of the last pkt */
 659         if (BLOCK_NUM_PKTS(pbd1)) {
 660                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 661                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 662         } else {
 663                 /* Ok, we tmo'd - so get the current time */
 664                 struct timespec ts;
 665                 getnstimeofday(&ts);
 666                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 667                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 668         }
 669
 670         smp_wmb();
 671
 672         /* Flush the block */
 673         prb_flush_block(pkc1, pbd1, status);
 674
 675         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 676 }
 677
 678 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 679 {
 680         pkc->reset_pending_on_curr_blk = 0;
 681 }
 682
 683 /*
 684  * Side effect of opening a block:
 685  *
 686  * 1) prb_queue is thawed.
 687  * 2) retire_blk_timer is refreshed.
 688  *
 689  */
 690 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 691         struct tpacket_block_desc *pbd1)
 692 {
 693         struct timespec ts;
 694         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 695
 696         smp_rmb();
 697
 698         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd1))) {
 699
 700                 /* We could have just memset this but we will lose the
 701                  * flexibility of making the priv area sticky
 702                  */
 703                 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 704                 BLOCK_NUM_PKTS(pbd1) = 0;
 705                 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 706                 getnstimeofday(&ts);
 707                 h1->ts_first_pkt.ts_sec = ts.tv_sec;
 708                 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 709                 pkc1->pkblk_start = (char *)pbd1;
 710                 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 711                 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 712                 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 713                 pbd1->version = pkc1->version;
 714                 pkc1->prev = pkc1->nxt_offset;
 715                 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 716                 prb_thaw_queue(pkc1);
 717                 _prb_refresh_rx_retire_blk_timer(pkc1);
 718
 719                 smp_wmb();
 720
 721                 return;
 722         }
 723
 724         WARN(1, "ERROR block:%p is NOT FREE status:%d kactive_blk_num:%d\n",
 725                 pbd1, BLOCK_STATUS(pbd1), pkc1->kactive_blk_num);
 726         dump_stack();
 727         BUG();
 728 }
 729
 730 /*
 731  * Queue freeze logic:
 732  * 1) Assume tp_block_nr = 8 blocks.
 733  * 2) At time 't0', user opens Rx ring.
 734  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 735  * 4) user-space is either sleeping or processing block '0'.
 736  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 737  *    it will close block-7,loop around and try to fill block '0'.
 738  *    call-flow:
 739  *    __packet_lookup_frame_in_block
 740  *      prb_retire_current_block()
 741  *      prb_dispatch_next_block()
 742  *        |->(BLOCK_STATUS == USER) evaluates to true
 743  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 744  * 6) Now there are two cases:
 745  *    6.1) Link goes idle right after the queue is frozen.
 746  *         But remember, the last open_block() refreshed the timer.
 747  *         When this timer expires,it will refresh itself so that we can
 748  *         re-open block-0 in near future.
 749  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 750  *         case and __packet_lookup_frame_in_block will check if block-0
 751  *         is free and can now be re-used.
 752  */
 753 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 754                                   struct packet_sock *po)
 755 {
 756         pkc->reset_pending_on_curr_blk = 1;
 757         po->stats_u.stats3.tp_freeze_q_cnt++;
 758 }
 759
 760 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 761
 762 /*
 763  * If the next block is free then we will dispatch it
 764  * and return a good offset.
 765  * Else, we will freeze the queue.
 766  * So, caller must check the return value.
 767  */
 768 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 769                 struct packet_sock *po)
 770 {
 771         struct tpacket_block_desc *pbd;
 772
 773         smp_rmb();
 774
 775         /* 1. Get current block num */
 776         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 777
 778         /* 2. If this block is currently in_use then freeze the queue */
 779         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 780                 prb_freeze_queue(pkc, po);
 781                 return NULL;
 782         }
 783
 784         /*
 785          * 3.
 786          * open this block and return the offset where the first packet
 787          * needs to get stored.
 788          */
 789         prb_open_block(pkc, pbd);
 790         return (void *)pkc->nxt_offset;
 791 }
 792
 793 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 794                 struct packet_sock *po, unsigned int status)
 795 {
 796         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 797
 798         /* retire/close the current block */
 799         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 800                 /*
 801                  * Plug the case where copy_bits() is in progress on
 802                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 803                  * have space to copy the pkt in the current block and
 804                  * called prb_retire_current_block()
 805                  *
 806                  * We don't need to worry about the TMO case because
 807                  * the timer-handler already handled this case.
 808                  */
 809                 if (!(status & TP_STATUS_BLK_TMO)) {
 810                         while (atomic_read(&pkc->blk_fill_in_prog)) {
 811                                 /* Waiting for skb_copy_bits to finish... */
 812                                 cpu_relax();
 813                         }
 814                 }
 815                 prb_close_block(pkc, pbd, po, status);
 816                 return;
 817         }
 818
 819         WARN(1, "ERROR-pbd[%d]:%p\n", pkc->kactive_blk_num, pbd);
 820         dump_stack();
 821         BUG();
 822 }
 823
 824 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
 825                                       struct tpacket_block_desc *pbd)
 826 {
 827         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 828 }
 829
 830 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 831 {
 832         return pkc->reset_pending_on_curr_blk;
 833 }
 834
 835 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 836 {
 837         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 838         atomic_dec(&pkc->blk_fill_in_prog);
 839 }
 840
 841 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 842                         struct tpacket3_hdr *ppd)
 843 {
 844         ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
 845 }
 846
 847 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 848                         struct tpacket3_hdr *ppd)
 849 {
 850         ppd->hv1.tp_rxhash = 0;
 851 }
 852
 853 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
 854                         struct tpacket3_hdr *ppd)
 855 {
 856         if (vlan_tx_tag_present(pkc->skb)) {
 857                 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
 858                 ppd->tp_status = TP_STATUS_VLAN_VALID;
 859         } else {
 860                 ppd->hv1.tp_vlan_tci = 0;
 861                 ppd->tp_status = TP_STATUS_AVAILABLE;
 862         }
 863 }
 864
 865 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
 866                         struct tpacket3_hdr *ppd)
 867 {
 868         prb_fill_vlan_info(pkc, ppd);
 869
 870         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
 871                 prb_fill_rxhash(pkc, ppd);
 872         else
 873                 prb_clear_rxhash(pkc, ppd);
 874 }
 875
 876 static void prb_fill_curr_block(char *curr,
 877                                 struct tpacket_kbdq_core *pkc,
 878                                 struct tpacket_block_desc *pbd,
 879                                 unsigned int len)
 880 {
 881         struct tpacket3_hdr *ppd;
 882
 883         ppd  = (struct tpacket3_hdr *)curr;
 884         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
 885         pkc->prev = curr;
 886         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
 887         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
 888         BLOCK_NUM_PKTS(pbd) += 1;
 889         atomic_inc(&pkc->blk_fill_in_prog);
 890         prb_run_all_ft_ops(pkc, ppd);
 891 }
 892
 893 /* Assumes caller has the sk->rx_queue.lock */
 894 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
 895                                             struct sk_buff *skb,
 896                                                 int status,
 897                                             unsigned int len
 898                                             )
 899 {
 900         struct tpacket_kbdq_core *pkc;
 901         struct tpacket_block_desc *pbd;
 902         char *curr, *end;
 903
 904         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 905         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 906
 907         /* Queue is frozen when user space is lagging behind */
 908         if (prb_queue_frozen(pkc)) {
 909                 /*
 910                  * Check if that last block which caused the queue to freeze,
 911                  * is still in_use by user-space.
 912                  */
 913                 if (prb_curr_blk_in_use(pkc, pbd)) {
 914                         /* Can't record this packet */
 915                         return NULL;
 916                 } else {
 917                         /*
 918                          * Ok, the block was released by user-space.
 919                          * Now let's open that block.
 920                          * opening a block also thaws the queue.
 921                          * Thawing is a side effect.
 922                          */
 923                         prb_open_block(pkc, pbd);
 924                 }
 925         }
 926
 927         smp_mb();
 928         curr = pkc->nxt_offset;
 929         pkc->skb = skb;
 930         end = (char *)pbd + pkc->kblk_size;
 931
 932         /* first try the current block */
 933         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
 934                 prb_fill_curr_block(curr, pkc, pbd, len);
 935                 return (void *)curr;
 936         }
 937
 938         /* Ok, close the current block */
 939         prb_retire_current_block(pkc, po, 0);
 940
 941         /* Now, try to dispatch the next block */
 942         curr = (char *)prb_dispatch_next_block(pkc, po);
 943         if (curr) {
 944                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 945                 prb_fill_curr_block(curr, pkc, pbd, len);
 946                 return (void *)curr;
 947         }
 948
 949         /*
 950          * No free blocks are available.user_space hasn't caught up yet.
 951          * Queue was just frozen and now this packet will get dropped.
 952          */
 953         return NULL;
 954 }
 955
 956 static void *packet_current_rx_frame(struct packet_sock *po,
 957                                             struct sk_buff *skb,
 958                                             int status, unsigned int len)
 959 {
 960         char *curr = NULL;
 961         switch (po->tp_version) {
 962         case TPACKET_V1:
 963         case TPACKET_V2:
 964                 curr = packet_lookup_frame(po, &po->rx_ring,
 965                                         po->rx_ring.head, status);
 966                 return curr;
 967         case TPACKET_V3:
 968                 return __packet_lookup_frame_in_block(po, skb, status, len);
 969         default:
 970                 WARN(1, "TPACKET version not supported\n");
 971                 BUG();
 972                 return NULL;
 973         }
 974 }
 975
 976 static void *prb_lookup_block(struct packet_sock *po,
 977                                      struct packet_ring_buffer *rb,
 978                                      unsigned int idx,
 979                                      int status)
 980 {
 981         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 982         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
 983
 984         if (status != BLOCK_STATUS(pbd))
 985                 return NULL;
 986         return pbd;
 987 }
 988
 989 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
 990 {
 991         unsigned int prev;
 992         if (rb->prb_bdqc.kactive_blk_num)
 993                 prev = rb->prb_bdqc.kactive_blk_num-1;
 994         else
 995                 prev = rb->prb_bdqc.knum_blocks-1;
 996         return prev;
 997 }
 998
 999 /* Assumes caller has held the rx_queue.lock */
1000 static void *__prb_previous_block(struct packet_sock *po,
1001                                          struct packet_ring_buffer *rb,
1002                                          int status)
1003 {
1004         unsigned int previous = prb_previous_blk_num(rb);
1005         return prb_lookup_block(po, rb, previous, status);
1006 }
1007
1008 static void *packet_previous_rx_frame(struct packet_sock *po,
1009                                              struct packet_ring_buffer *rb,
1010                                              int status)
1011 {
1012         if (po->tp_version <= TPACKET_V2)
1013                 return packet_previous_frame(po, rb, status);
1014
1015         return __prb_previous_block(po, rb, status);
1016 }
1017
1018 static void packet_increment_rx_head(struct packet_sock *po,
1019                                             struct packet_ring_buffer *rb)
1020 {
1021         switch (po->tp_version) {
1022         case TPACKET_V1:
1023         case TPACKET_V2:
1024                 return packet_increment_head(rb);
1025         case TPACKET_V3:
1026         default:
1027                 WARN(1, "TPACKET version not supported.\n");
1028                 BUG();
1029                 return;
1030         }
1031 }
1032
1033 static void *packet_previous_frame(struct packet_sock *po,
1034                 struct packet_ring_buffer *rb,
1035                 int status)
1036 {
1037         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1038         return packet_lookup_frame(po, rb, previous, status);
1039 }
1040
1041 static void packet_increment_head(struct packet_ring_buffer *buff)
1042 {
1043         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1044 }
1045
1046 static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1047 {
1048         struct sock *sk = &po->sk;
1049         bool has_room;
1050
1051         if (po->prot_hook.func != tpacket_rcv)
1052                 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1053                         <= sk->sk_rcvbuf;
1054
1055         spin_lock(&sk->sk_receive_queue.lock);
1056         if (po->tp_version == TPACKET_V3)
1057                 has_room = prb_lookup_block(po, &po->rx_ring,
1058                                             po->rx_ring.prb_bdqc.kactive_blk_num,
1059                                             TP_STATUS_KERNEL);
1060         else
1061                 has_room = packet_lookup_frame(po, &po->rx_ring,
1062                                                po->rx_ring.head,
1063                                                TP_STATUS_KERNEL);
1064         spin_unlock(&sk->sk_receive_queue.lock);
1065
1066         return has_room;
1067 }
1068
1069 static void packet_sock_destruct(struct sock *sk)
1070 {
1071         skb_queue_purge(&sk->sk_error_queue);
1072
1073         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1074         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1075
1076         if (!sock_flag(sk, SOCK_DEAD)) {
1077                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1078                 return;
1079         }
1080
1081         sk_refcnt_debug_dec(sk);
1082 }
1083
1084 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1085 {
1086         int x = atomic_read(&f->rr_cur) + 1;
1087
1088         if (x >= num)
1089                 x = 0;
1090
1091         return x;
1092 }
1093
1094 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1095                                       struct sk_buff *skb,
1096                                       unsigned int num)
1097 {
1098         return (((u64)skb->rxhash) * num) >> 32;
1099 }
1100
1101 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1102                                     struct sk_buff *skb,
1103                                     unsigned int num)
1104 {
1105         int cur, old;
1106
1107         cur = atomic_read(&f->rr_cur);
1108         while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1109                                      fanout_rr_next(f, num))) != cur)
1110                 cur = old;
1111         return cur;
1112 }
1113
1114 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1115                                      struct sk_buff *skb,
1116                                      unsigned int num)
1117 {
1118         return smp_processor_id() % num;
1119 }
1120
1121 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1122                                           struct sk_buff *skb,
1123                                           unsigned int idx, unsigned int skip,
1124                                           unsigned int num)
1125 {
1126         unsigned int i, j;
1127
1128         i = j = min_t(int, f->next[idx], num - 1);
1129         do {
1130                 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1131                         if (i != j)
1132                                 f->next[idx] = i;
1133                         return i;
1134                 }
1135                 if (++i == num)
1136                         i = 0;
1137         } while (i != j);
1138
1139         return idx;
1140 }
1141
1142 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1143 {
1144         return f->flags & (flag >> 8);
1145 }
1146
1147 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1148                              struct packet_type *pt, struct net_device *orig_dev)
1149 {
1150         struct packet_fanout *f = pt->af_packet_priv;
1151         unsigned int num = f->num_members;
1152         struct packet_sock *po;
1153         unsigned int idx;
1154
1155         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1156             !num) {
1157                 kfree_skb(skb);
1158                 return 0;
1159         }
1160
1161         switch (f->type) {
1162         case PACKET_FANOUT_HASH:
1163         default:
1164                 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1165                         skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1166                         if (!skb)
1167                                 return 0;
1168                 }
1169                 skb_get_rxhash(skb);
1170                 idx = fanout_demux_hash(f, skb, num);
1171                 break;
1172         case PACKET_FANOUT_LB:
1173                 idx = fanout_demux_lb(f, skb, num);
1174                 break;
1175         case PACKET_FANOUT_CPU:
1176                 idx = fanout_demux_cpu(f, skb, num);
1177                 break;
1178         case PACKET_FANOUT_ROLLOVER:
1179                 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1180                 break;
1181         }
1182
1183         po = pkt_sk(f->arr[idx]);
1184         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1185             unlikely(!packet_rcv_has_room(po, skb))) {
1186                 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1187                 po = pkt_sk(f->arr[idx]);
1188         }
1189
1190         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1191 }
1192
1193 DEFINE_MUTEX(fanout_mutex);
1194 EXPORT_SYMBOL_GPL(fanout_mutex);
1195 static LIST_HEAD(fanout_list);
1196
1197 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1198 {
1199         struct packet_fanout *f = po->fanout;
1200
1201         spin_lock(&f->lock);
1202         f->arr[f->num_members] = sk;
1203         smp_wmb();
1204         f->num_members++;
1205         spin_unlock(&f->lock);
1206 }
1207
1208 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1209 {
1210         struct packet_fanout *f = po->fanout;
1211         int i;
1212
1213         spin_lock(&f->lock);
1214         for (i = 0; i < f->num_members; i++) {
1215                 if (f->arr[i] == sk)
1216                         break;
1217         }
1218         BUG_ON(i >= f->num_members);
1219         f->arr[i] = f->arr[f->num_members - 1];
1220         f->num_members--;
1221         spin_unlock(&f->lock);
1222 }
1223
1224 static bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1225 {
1226         if (ptype->af_packet_priv == (void*)((struct packet_sock *)sk)->fanout)
1227                 return true;
1228
1229         return false;
1230 }
1231
1232 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1233 {
1234         struct packet_sock *po = pkt_sk(sk);
1235         struct packet_fanout *f, *match;
1236         u8 type = type_flags & 0xff;
1237         u8 flags = type_flags >> 8;
1238         int err;
1239
1240         switch (type) {
1241         case PACKET_FANOUT_ROLLOVER:
1242                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1243                         return -EINVAL;
1244         case PACKET_FANOUT_HASH:
1245         case PACKET_FANOUT_LB:
1246         case PACKET_FANOUT_CPU:
1247                 break;
1248         default:
1249                 return -EINVAL;
1250         }
1251
1252         if (!po->running)
1253                 return -EINVAL;
1254
1255         if (po->fanout)
1256                 return -EALREADY;
1257
1258         mutex_lock(&fanout_mutex);
1259         match = NULL;
1260         list_for_each_entry(f, &fanout_list, list) {
1261                 if (f->id == id &&
1262                     read_pnet(&f->net) == sock_net(sk)) {
1263                         match = f;
1264                         break;
1265                 }
1266         }
1267         err = -EINVAL;
1268         if (match && match->flags != flags)
1269                 goto out;
1270         if (!match) {
1271                 err = -ENOMEM;
1272                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1273                 if (!match)
1274                         goto out;
1275                 write_pnet(&match->net, sock_net(sk));
1276                 match->id = id;
1277                 match->type = type;
1278                 match->flags = flags;
1279                 atomic_set(&match->rr_cur, 0);
1280                 INIT_LIST_HEAD(&match->list);
1281                 spin_lock_init(&match->lock);
1282                 atomic_set(&match->sk_ref, 0);
1283                 match->prot_hook.type = po->prot_hook.type;
1284                 match->prot_hook.dev = po->prot_hook.dev;
1285                 match->prot_hook.func = packet_rcv_fanout;
1286                 match->prot_hook.af_packet_priv = match;
1287                 match->prot_hook.id_match = match_fanout_group;
1288                 dev_add_pack(&match->prot_hook);
1289                 list_add(&match->list, &fanout_list);
1290         }
1291         err = -EINVAL;
1292         if (match->type == type &&
1293             match->prot_hook.type == po->prot_hook.type &&
1294             match->prot_hook.dev == po->prot_hook.dev) {
1295                 err = -ENOSPC;
1296                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1297                         __dev_remove_pack(&po->prot_hook);
1298                         po->fanout = match;
1299                         atomic_inc(&match->sk_ref);
1300                         __fanout_link(sk, po);
1301                         err = 0;
1302                 }
1303         }
1304 out:
1305         mutex_unlock(&fanout_mutex);
1306         return err;
1307 }
1308
1309 static void fanout_release(struct sock *sk)
1310 {
1311         struct packet_sock *po = pkt_sk(sk);
1312         struct packet_fanout *f;
1313
1314         f = po->fanout;
1315         if (!f)
1316                 return;
1317
1318         mutex_lock(&fanout_mutex);
1319         po->fanout = NULL;
1320
1321         if (atomic_dec_and_test(&f->sk_ref)) {
1322                 list_del(&f->list);
1323                 dev_remove_pack(&f->prot_hook);
1324                 kfree(f);
1325         }
1326         mutex_unlock(&fanout_mutex);
1327 }
1328
1329 static const struct proto_ops packet_ops;
1330
1331 static const struct proto_ops packet_ops_spkt;
1332
1333 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1334                            struct packet_type *pt, struct net_device *orig_dev)
1335 {
1336         struct sock *sk;
1337         struct sockaddr_pkt *spkt;
1338
1339         /*
1340          *      When we registered the protocol we saved the socket in the data
1341          *      field for just this event.
1342          */
1343
1344         sk = pt->af_packet_priv;
1345
1346         /*
1347          *      Yank back the headers [hope the device set this
1348          *      right or kerboom...]
1349          *
1350          *      Incoming packets have ll header pulled,
1351          *      push it back.
1352          *
1353          *      For outgoing ones skb->data == skb_mac_header(skb)
1354          *      so that this procedure is noop.
1355          */
1356
1357         if (skb->pkt_type == PACKET_LOOPBACK)
1358                 goto out;
1359
1360         if (!net_eq(dev_net(dev), sock_net(sk)))
1361                 goto out;
1362
1363         skb = skb_share_check(skb, GFP_ATOMIC);
1364         if (skb == NULL)
1365                 goto oom;
1366
1367         /* drop any routing info */
1368         skb_dst_drop(skb);
1369
1370         /* drop conntrack reference */
1371         nf_reset(skb);
1372
1373         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1374
1375         skb_push(skb, skb->data - skb_mac_header(skb));
1376
1377         /*
1378          *      The SOCK_PACKET socket receives _all_ frames.
1379          */
1380
1381         spkt->spkt_family = dev->type;
1382         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1383         spkt->spkt_protocol = skb->protocol;
1384
1385         /*
1386          *      Charge the memory to the socket. This is done specifically
1387          *      to prevent sockets using all the memory up.
1388          */
1389
1390         if (sock_queue_rcv_skb(sk, skb) == 0)
1391                 return 0;
1392
1393 out:
1394         kfree_skb(skb);
1395 oom:
1396         return 0;
1397 }
1398
1399
1400 /*
1401  *      Output a raw packet to a device layer. This bypasses all the other
1402  *      protocol layers and you must therefore supply it with a complete frame
1403  */
1404
1405 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1406                                struct msghdr *msg, size_t len)
1407 {
1408         struct sock *sk = sock->sk;
1409         struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1410         struct sk_buff *skb = NULL;
1411         struct net_device *dev;
1412         __be16 proto = 0;
1413         int err;
1414         int extra_len = 0;
1415
1416         /*
1417          *      Get and verify the address.
1418          */
1419
1420         if (saddr) {
1421                 if (msg->msg_namelen < sizeof(struct sockaddr))
1422                         return -EINVAL;
1423                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1424                         proto = saddr->spkt_protocol;
1425         } else
1426                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1427
1428         /*
1429          *      Find the device first to size check it
1430          */
1431
1432         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1433 retry:
1434         rcu_read_lock();
1435         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1436         err = -ENODEV;
1437         if (dev == NULL)
1438                 goto out_unlock;
1439
1440         err = -ENETDOWN;
1441         if (!(dev->flags & IFF_UP))
1442                 goto out_unlock;
1443
1444         /*
1445          * You may not queue a frame bigger than the mtu. This is the lowest level
1446          * raw protocol and you must do your own fragmentation at this level.
1447          */
1448
1449         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1450                 if (!netif_supports_nofcs(dev)) {
1451                         err = -EPROTONOSUPPORT;
1452                         goto out_unlock;
1453                 }
1454                 extra_len = 4; /* We're doing our own CRC */
1455         }
1456
1457         err = -EMSGSIZE;
1458         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1459                 goto out_unlock;
1460
1461         if (!skb) {
1462                 size_t reserved = LL_RESERVED_SPACE(dev);
1463                 int tlen = dev->needed_tailroom;
1464                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1465
1466                 rcu_read_unlock();
1467                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1468                 if (skb == NULL)
1469                         return -ENOBUFS;
1470                 /* FIXME: Save some space for broken drivers that write a hard
1471                  * header at transmission time by themselves. PPP is the notable
1472                  * one here. This should really be fixed at the driver level.
1473                  */
1474                 skb_reserve(skb, reserved);
1475                 skb_reset_network_header(skb);
1476
1477                 /* Try to align data part correctly */
1478                 if (hhlen) {
1479                         skb->data -= hhlen;
1480                         skb->tail -= hhlen;
1481                         if (len < hhlen)
1482                                 skb_reset_network_header(skb);
1483                 }
1484                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1485                 if (err)
1486                         goto out_free;
1487                 goto retry;
1488         }
1489
1490         if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1491                 /* Earlier code assumed this would be a VLAN pkt,
1492                  * double-check this now that we have the actual
1493                  * packet in hand.
1494                  */
1495                 struct ethhdr *ehdr;
1496                 skb_reset_mac_header(skb);
1497                 ehdr = eth_hdr(skb);
1498                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1499                         err = -EMSGSIZE;
1500                         goto out_unlock;
1501                 }
1502         }
1503
1504         skb->protocol = proto;
1505         skb->dev = dev;
1506         skb->priority = sk->sk_priority;
1507         skb->mark = sk->sk_mark;
1508         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1509         if (err < 0)
1510                 goto out_unlock;
1511
1512         if (unlikely(extra_len == 4))
1513                 skb->no_fcs = 1;
1514
1515         dev_queue_xmit(skb);
1516         rcu_read_unlock();
1517         return len;
1518
1519 out_unlock:
1520         rcu_read_unlock();
1521 out_free:
1522         kfree_skb(skb);
1523         return err;
1524 }
1525
1526 static unsigned int run_filter(const struct sk_buff *skb,
1527                                       const struct sock *sk,
1528                                       unsigned int res)
1529 {
1530         struct sk_filter *filter;
1531
1532         rcu_read_lock();
1533         filter = rcu_dereference(sk->sk_filter);
1534         if (filter != NULL)
1535                 res = SK_RUN_FILTER(filter, skb);
1536         rcu_read_unlock();
1537
1538         return res;
1539 }
1540
1541 /*
1542  * This function makes lazy skb cloning in hope that most of packets
1543  * are discarded by BPF.
1544  *
1545  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1546  * and skb->cb are mangled. It works because (and until) packets
1547  * falling here are owned by current CPU. Output packets are cloned
1548  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1549  * sequencially, so that if we return skb to original state on exit,
1550  * we will not harm anyone.
1551  */
1552
1553 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1554                       struct packet_type *pt, struct net_device *orig_dev)
1555 {
1556         struct sock *sk;
1557         struct sockaddr_ll *sll;
1558         struct packet_sock *po;
1559         u8 *skb_head = skb->data;
1560         int skb_len = skb->len;
1561         unsigned int snaplen, res;
1562
1563         if (skb->pkt_type == PACKET_LOOPBACK)
1564                 goto drop;
1565
1566         sk = pt->af_packet_priv;
1567         po = pkt_sk(sk);
1568
1569         if (!net_eq(dev_net(dev), sock_net(sk)))
1570                 goto drop;
1571
1572         skb->dev = dev;
1573
1574         if (dev->header_ops) {
1575                 /* The device has an explicit notion of ll header,
1576                  * exported to higher levels.
1577                  *
1578                  * Otherwise, the device hides details of its frame
1579                  * structure, so that corresponding packet head is
1580                  * never delivered to user.
1581                  */
1582                 if (sk->sk_type != SOCK_DGRAM)
1583                         skb_push(skb, skb->data - skb_mac_header(skb));
1584                 else if (skb->pkt_type == PACKET_OUTGOING) {
1585                         /* Special case: outgoing packets have ll header at head */
1586                         skb_pull(skb, skb_network_offset(skb));
1587                 }
1588         }
1589
1590         snaplen = skb->len;
1591
1592         res = run_filter(skb, sk, snaplen);
1593         if (!res)
1594                 goto drop_n_restore;
1595         if (snaplen > res)
1596                 snaplen = res;
1597
1598         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1599                 goto drop_n_acct;
1600
1601         if (skb_shared(skb)) {
1602                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1603                 if (nskb == NULL)
1604                         goto drop_n_acct;
1605
1606                 if (skb_head != skb->data) {
1607                         skb->data = skb_head;
1608                         skb->len = skb_len;
1609                 }
1610                 consume_skb(skb);
1611                 skb = nskb;
1612         }
1613
1614         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1615                      sizeof(skb->cb));
1616
1617         sll = &PACKET_SKB_CB(skb)->sa.ll;
1618         sll->sll_family = AF_PACKET;
1619         sll->sll_hatype = dev->type;
1620         sll->sll_protocol = skb->protocol;
1621         sll->sll_pkttype = skb->pkt_type;
1622         if (unlikely(po->origdev))
1623                 sll->sll_ifindex = orig_dev->ifindex;
1624         else
1625                 sll->sll_ifindex = dev->ifindex;
1626
1627         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1628
1629         PACKET_SKB_CB(skb)->origlen = skb->len;
1630
1631         if (pskb_trim(skb, snaplen))
1632                 goto drop_n_acct;
1633
1634         skb_set_owner_r(skb, sk);
1635         skb->dev = NULL;
1636         skb_dst_drop(skb);
1637
1638         /* drop conntrack reference */
1639         nf_reset(skb);
1640
1641         spin_lock(&sk->sk_receive_queue.lock);
1642         po->stats.tp_packets++;
1643         skb->dropcount = atomic_read(&sk->sk_drops);
1644         __skb_queue_tail(&sk->sk_receive_queue, skb);
1645         spin_unlock(&sk->sk_receive_queue.lock);
1646         sk->sk_data_ready(sk, skb->len);
1647         return 0;
1648
1649 drop_n_acct:
1650         spin_lock(&sk->sk_receive_queue.lock);
1651         po->stats.tp_drops++;
1652         atomic_inc(&sk->sk_drops);
1653         spin_unlock(&sk->sk_receive_queue.lock);
1654
1655 drop_n_restore:
1656         if (skb_head != skb->data && skb_shared(skb)) {
1657                 skb->data = skb_head;
1658                 skb->len = skb_len;
1659         }
1660 drop:
1661         consume_skb(skb);
1662         return 0;
1663 }
1664
1665 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1666                        struct packet_type *pt, struct net_device *orig_dev)
1667 {
1668         struct sock *sk;
1669         struct packet_sock *po;
1670         struct sockaddr_ll *sll;
1671         union {
1672                 struct tpacket_hdr *h1;
1673                 struct tpacket2_hdr *h2;
1674                 struct tpacket3_hdr *h3;
1675                 void *raw;
1676         } h;
1677         u8 *skb_head = skb->data;
1678         int skb_len = skb->len;
1679         unsigned int snaplen, res;
1680         unsigned long status = TP_STATUS_USER;
1681         unsigned short macoff, netoff, hdrlen;
1682         struct sk_buff *copy_skb = NULL;
1683         struct timeval tv;
1684         struct timespec ts;
1685         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1686
1687         if (skb->pkt_type == PACKET_LOOPBACK)
1688                 goto drop;
1689
1690         sk = pt->af_packet_priv;
1691         po = pkt_sk(sk);
1692
1693         if (!net_eq(dev_net(dev), sock_net(sk)))
1694                 goto drop;
1695
1696         if (dev->header_ops) {
1697                 if (sk->sk_type != SOCK_DGRAM)
1698                         skb_push(skb, skb->data - skb_mac_header(skb));
1699                 else if (skb->pkt_type == PACKET_OUTGOING) {
1700                         /* Special case: outgoing packets have ll header at head */
1701                         skb_pull(skb, skb_network_offset(skb));
1702                 }
1703         }
1704
1705         if (skb->ip_summed == CHECKSUM_PARTIAL)
1706                 status |= TP_STATUS_CSUMNOTREADY;
1707
1708         snaplen = skb->len;
1709
1710         res = run_filter(skb, sk, snaplen);
1711         if (!res)
1712                 goto drop_n_restore;
1713         if (snaplen > res)
1714                 snaplen = res;
1715
1716         if (sk->sk_type == SOCK_DGRAM) {
1717                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1718                                   po->tp_reserve;
1719         } else {
1720                 unsigned int maclen = skb_network_offset(skb);
1721                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1722                                        (maclen < 16 ? 16 : maclen)) +
1723                         po->tp_reserve;
1724                 macoff = netoff - maclen;
1725         }
1726         if (po->tp_version <= TPACKET_V2) {
1727                 if (macoff + snaplen > po->rx_ring.frame_size) {
1728                         if (po->copy_thresh &&
1729                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1730                                 if (skb_shared(skb)) {
1731                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1732                                 } else {
1733                                         copy_skb = skb_get(skb);
1734                                         skb_head = skb->data;
1735                                 }
1736                                 if (copy_skb)
1737                                         skb_set_owner_r(copy_skb, sk);
1738                         }
1739                         snaplen = po->rx_ring.frame_size - macoff;
1740                         if ((int)snaplen < 0)
1741                                 snaplen = 0;
1742                 }
1743         }
1744         spin_lock(&sk->sk_receive_queue.lock);
1745         h.raw = packet_current_rx_frame(po, skb,
1746                                         TP_STATUS_KERNEL, (macoff+snaplen));
1747         if (!h.raw)
1748                 goto ring_is_full;
1749         if (po->tp_version <= TPACKET_V2) {
1750                 packet_increment_rx_head(po, &po->rx_ring);
1751         /*
1752          * LOSING will be reported till you read the stats,
1753          * because it's COR - Clear On Read.
1754          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1755          * at packet level.
1756          */
1757                 if (po->stats.tp_drops)
1758                         status |= TP_STATUS_LOSING;
1759         }
1760         po->stats.tp_packets++;
1761         if (copy_skb) {
1762                 status |= TP_STATUS_COPY;
1763                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1764         }
1765         spin_unlock(&sk->sk_receive_queue.lock);
1766
1767         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1768
1769         switch (po->tp_version) {
1770         case TPACKET_V1:
1771                 h.h1->tp_len = skb->len;
1772                 h.h1->tp_snaplen = snaplen;
1773                 h.h1->tp_mac = macoff;
1774                 h.h1->tp_net = netoff;
1775                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1776                                 && shhwtstamps->syststamp.tv64)
1777                         tv = ktime_to_timeval(shhwtstamps->syststamp);
1778                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1779                                 && shhwtstamps->hwtstamp.tv64)
1780                         tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1781                 else if (skb->tstamp.tv64)
1782                         tv = ktime_to_timeval(skb->tstamp);
1783                 else
1784                         do_gettimeofday(&tv);
1785                 h.h1->tp_sec = tv.tv_sec;
1786                 h.h1->tp_usec = tv.tv_usec;
1787                 hdrlen = sizeof(*h.h1);
1788                 break;
1789         case TPACKET_V2:
1790                 h.h2->tp_len = skb->len;
1791                 h.h2->tp_snaplen = snaplen;
1792                 h.h2->tp_mac = macoff;
1793                 h.h2->tp_net = netoff;
1794                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1795                                 && shhwtstamps->syststamp.tv64)
1796                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1797                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1798                                 && shhwtstamps->hwtstamp.tv64)
1799                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1800                 else if (skb->tstamp.tv64)
1801                         ts = ktime_to_timespec(skb->tstamp);
1802                 else
1803                         getnstimeofday(&ts);
1804                 h.h2->tp_sec = ts.tv_sec;
1805                 h.h2->tp_nsec = ts.tv_nsec;
1806                 if (vlan_tx_tag_present(skb)) {
1807                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1808                         status |= TP_STATUS_VLAN_VALID;
1809                 } else {
1810                         h.h2->tp_vlan_tci = 0;
1811                 }
1812                 h.h2->tp_padding = 0;
1813                 hdrlen = sizeof(*h.h2);
1814                 break;
1815         case TPACKET_V3:
1816                 /* tp_nxt_offset,vlan are already populated above.
1817                  * So DONT clear those fields here
1818                  */
1819                 h.h3->tp_status |= status;
1820                 h.h3->tp_len = skb->len;
1821                 h.h3->tp_snaplen = snaplen;
1822                 h.h3->tp_mac = macoff;
1823                 h.h3->tp_net = netoff;
1824                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1825                                 && shhwtstamps->syststamp.tv64)
1826                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1827                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1828                                 && shhwtstamps->hwtstamp.tv64)
1829                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1830                 else if (skb->tstamp.tv64)
1831                         ts = ktime_to_timespec(skb->tstamp);
1832                 else
1833                         getnstimeofday(&ts);
1834                 h.h3->tp_sec  = ts.tv_sec;
1835                 h.h3->tp_nsec = ts.tv_nsec;
1836                 hdrlen = sizeof(*h.h3);
1837                 break;
1838         default:
1839                 BUG();
1840         }
1841
1842         sll = h.raw + TPACKET_ALIGN(hdrlen);
1843         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1844         sll->sll_family = AF_PACKET;
1845         sll->sll_hatype = dev->type;
1846         sll->sll_protocol = skb->protocol;
1847         sll->sll_pkttype = skb->pkt_type;
1848         if (unlikely(po->origdev))
1849                 sll->sll_ifindex = orig_dev->ifindex;
1850         else
1851                 sll->sll_ifindex = dev->ifindex;
1852
1853         smp_mb();
1854 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1855         {
1856                 u8 *start, *end;
1857
1858                 if (po->tp_version <= TPACKET_V2) {
1859                         end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1860                                 + macoff + snaplen);
1861                         for (start = h.raw; start < end; start += PAGE_SIZE)
1862                                 flush_dcache_page(pgv_to_page(start));
1863                 }
1864                 smp_wmb();
1865         }
1866 #endif
1867         if (po->tp_version <= TPACKET_V2)
1868                 __packet_set_status(po, h.raw, status);
1869         else
1870                 prb_clear_blk_fill_status(&po->rx_ring);
1871
1872         sk->sk_data_ready(sk, 0);
1873
1874 drop_n_restore:
1875         if (skb_head != skb->data && skb_shared(skb)) {
1876                 skb->data = skb_head;
1877                 skb->len = skb_len;
1878         }
1879 drop:
1880         kfree_skb(skb);
1881         return 0;
1882
1883 ring_is_full:
1884         po->stats.tp_drops++;
1885         spin_unlock(&sk->sk_receive_queue.lock);
1886
1887         sk->sk_data_ready(sk, 0);
1888         kfree_skb(copy_skb);
1889         goto drop_n_restore;
1890 }
1891
1892 static void tpacket_destruct_skb(struct sk_buff *skb)
1893 {
1894         struct packet_sock *po = pkt_sk(skb->sk);
1895         void *ph;
1896
1897         if (likely(po->tx_ring.pg_vec)) {
1898                 ph = skb_shinfo(skb)->destructor_arg;
1899                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1900                 atomic_dec(&po->tx_ring.pending);
1901                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1902         }
1903
1904         sock_wfree(skb);
1905 }
1906
1907 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1908                 void *frame, struct net_device *dev, int size_max,
1909                 __be16 proto, unsigned char *addr, int hlen)
1910 {
1911         union {
1912                 struct tpacket_hdr *h1;
1913                 struct tpacket2_hdr *h2;
1914                 void *raw;
1915         } ph;
1916         int to_write, offset, len, tp_len, nr_frags, len_max;
1917         struct socket *sock = po->sk.sk_socket;
1918         struct page *page;
1919         void *data;
1920         int err;
1921
1922         ph.raw = frame;
1923
1924         skb->protocol = proto;
1925         skb->dev = dev;
1926         skb->priority = po->sk.sk_priority;
1927         skb->mark = po->sk.sk_mark;
1928         skb_shinfo(skb)->destructor_arg = ph.raw;
1929
1930         switch (po->tp_version) {
1931         case TPACKET_V2:
1932                 tp_len = ph.h2->tp_len;
1933                 break;
1934         default:
1935                 tp_len = ph.h1->tp_len;
1936                 break;
1937         }
1938         if (unlikely(tp_len > size_max)) {
1939                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1940                 return -EMSGSIZE;
1941         }
1942
1943         skb_reserve(skb, hlen);
1944         skb_reset_network_header(skb);
1945
1946         if (po->tp_tx_has_off) {
1947                 int off_min, off_max, off;
1948                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
1949                 off_max = po->tx_ring.frame_size - tp_len;
1950                 if (sock->type == SOCK_DGRAM) {
1951                         switch (po->tp_version) {
1952                         case TPACKET_V2:
1953                                 off = ph.h2->tp_net;
1954                                 break;
1955                         default:
1956                                 off = ph.h1->tp_net;
1957                                 break;
1958                         }
1959                 } else {
1960                         switch (po->tp_version) {
1961                         case TPACKET_V2:
1962                                 off = ph.h2->tp_mac;
1963                                 break;
1964                         default:
1965                                 off = ph.h1->tp_mac;
1966                                 break;
1967                         }
1968                 }
1969                 if (unlikely((off < off_min) || (off_max < off)))
1970                         return -EINVAL;
1971                 data = ph.raw + off;
1972         } else {
1973                 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1974         }
1975         to_write = tp_len;
1976
1977         if (sock->type == SOCK_DGRAM) {
1978                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
1979                                 NULL, tp_len);
1980                 if (unlikely(err < 0))
1981                         return -EINVAL;
1982         } else if (dev->hard_header_len) {
1983                 /* net device doesn't like empty head */
1984                 if (unlikely(tp_len <= dev->hard_header_len)) {
1985                         pr_err("packet size is too short (%d < %d)\n",
1986                                tp_len, dev->hard_header_len);
1987                         return -EINVAL;
1988                 }
1989
1990                 skb_push(skb, dev->hard_header_len);
1991                 err = skb_store_bits(skb, 0, data,
1992                                 dev->hard_header_len);
1993                 if (unlikely(err))
1994                         return err;
1995
1996                 data += dev->hard_header_len;
1997                 to_write -= dev->hard_header_len;
1998         }
1999
2000         offset = offset_in_page(data);
2001         len_max = PAGE_SIZE - offset;
2002         len = ((to_write > len_max) ? len_max : to_write);
2003
2004         skb->data_len = to_write;
2005         skb->len += to_write;
2006         skb->truesize += to_write;
2007         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2008
2009         while (likely(to_write)) {
2010                 nr_frags = skb_shinfo(skb)->nr_frags;
2011
2012                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2013                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2014                                MAX_SKB_FRAGS);
2015                         return -EFAULT;
2016                 }
2017
2018                 page = pgv_to_page(data);
2019                 data += len;
2020                 flush_dcache_page(page);
2021                 get_page(page);
2022                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2023                 to_write -= len;
2024                 offset = 0;
2025                 len_max = PAGE_SIZE;
2026                 len = ((to_write > len_max) ? len_max : to_write);
2027         }
2028
2029         return tp_len;
2030 }
2031
2032 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2033 {
2034         struct sk_buff *skb;
2035         struct net_device *dev;
2036         __be16 proto;
2037         bool need_rls_dev = false;
2038         int err, reserve = 0;
2039         void *ph;
2040         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2041         int tp_len, size_max;
2042         unsigned char *addr;
2043         int len_sum = 0;
2044         int status = TP_STATUS_AVAILABLE;
2045         int hlen, tlen;
2046
2047         mutex_lock(&po->pg_vec_lock);
2048
2049         if (saddr == NULL) {
2050                 dev = po->prot_hook.dev;
2051                 proto   = po->num;
2052                 addr    = NULL;
2053         } else {
2054                 err = -EINVAL;
2055                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2056                         goto out;
2057                 if (msg->msg_namelen < (saddr->sll_halen
2058                                         + offsetof(struct sockaddr_ll,
2059                                                 sll_addr)))
2060                         goto out;
2061                 proto   = saddr->sll_protocol;
2062                 addr    = saddr->sll_addr;
2063                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2064                 need_rls_dev = true;
2065         }
2066
2067         err = -ENXIO;
2068         if (unlikely(dev == NULL))
2069                 goto out;
2070
2071         reserve = dev->hard_header_len;
2072
2073         err = -ENETDOWN;
2074         if (unlikely(!(dev->flags & IFF_UP)))
2075                 goto out_put;
2076
2077         size_max = po->tx_ring.frame_size
2078                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2079
2080         if (size_max > dev->mtu + reserve)
2081                 size_max = dev->mtu + reserve;
2082
2083         do {
2084                 ph = packet_current_frame(po, &po->tx_ring,
2085                                 TP_STATUS_SEND_REQUEST);
2086
2087                 if (unlikely(ph == NULL)) {
2088                         schedule();
2089                         continue;
2090                 }
2091
2092                 status = TP_STATUS_SEND_REQUEST;
2093                 hlen = LL_RESERVED_SPACE(dev);
2094                 tlen = dev->needed_tailroom;
2095                 skb = sock_alloc_send_skb(&po->sk,
2096                                 hlen + tlen + sizeof(struct sockaddr_ll),
2097                                 0, &err);
2098
2099                 if (unlikely(skb == NULL))
2100                         goto out_status;
2101
2102                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2103                                 addr, hlen);
2104
2105                 if (unlikely(tp_len < 0)) {
2106                         if (po->tp_loss) {
2107                                 __packet_set_status(po, ph,
2108                                                 TP_STATUS_AVAILABLE);
2109                                 packet_increment_head(&po->tx_ring);
2110                                 kfree_skb(skb);
2111                                 continue;
2112                         } else {
2113                                 status = TP_STATUS_WRONG_FORMAT;
2114                                 err = tp_len;
2115                                 goto out_status;
2116                         }
2117                 }
2118
2119                 skb->destructor = tpacket_destruct_skb;
2120                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2121                 atomic_inc(&po->tx_ring.pending);
2122
2123                 status = TP_STATUS_SEND_REQUEST;
2124                 err = dev_queue_xmit(skb);
2125                 if (unlikely(err > 0)) {
2126                         err = net_xmit_errno(err);
2127                         if (err && __packet_get_status(po, ph) ==
2128                                    TP_STATUS_AVAILABLE) {
2129                                 /* skb was destructed already */
2130                                 skb = NULL;
2131                                 goto out_status;
2132                         }
2133                         /*
2134                          * skb was dropped but not destructed yet;
2135                          * let's treat it like congestion or err < 0
2136                          */
2137                         err = 0;
2138                 }
2139                 packet_increment_head(&po->tx_ring);
2140                 len_sum += tp_len;
2141         } while (likely((ph != NULL) ||
2142                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2143                          (atomic_read(&po->tx_ring.pending))))
2144                 );
2145
2146         err = len_sum;
2147         goto out_put;
2148
2149 out_status:
2150         __packet_set_status(po, ph, status);
2151         kfree_skb(skb);
2152 out_put:
2153         if (need_rls_dev)
2154                 dev_put(dev);
2155 out:
2156         mutex_unlock(&po->pg_vec_lock);
2157         return err;
2158 }
2159
2160 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2161                                         size_t reserve, size_t len,
2162                                         size_t linear, int noblock,
2163                                         int *err)
2164 {
2165         struct sk_buff *skb;
2166
2167         /* Under a page?  Don't bother with paged skb. */
2168         if (prepad + len < PAGE_SIZE || !linear)
2169                 linear = len;
2170
2171         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2172                                    err);
2173         if (!skb)
2174                 return NULL;
2175
2176         skb_reserve(skb, reserve);
2177         skb_put(skb, linear);
2178         skb->data_len = len - linear;
2179         skb->len += len - linear;
2180
2181         return skb;
2182 }
2183
2184 static int packet_snd(struct socket *sock,
2185                           struct msghdr *msg, size_t len)
2186 {
2187         struct sock *sk = sock->sk;
2188         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2189         struct sk_buff *skb;
2190         struct net_device *dev;
2191         __be16 proto;
2192         bool need_rls_dev = false;
2193         unsigned char *addr;
2194         int err, reserve = 0;
2195         struct virtio_net_hdr vnet_hdr = { 0 };
2196         int offset = 0;
2197         int vnet_hdr_len;
2198         struct packet_sock *po = pkt_sk(sk);
2199         unsigned short gso_type = 0;
2200         int hlen, tlen;
2201         int extra_len = 0;
2202
2203         /*
2204          *      Get and verify the address.
2205          */
2206
2207         if (saddr == NULL) {
2208                 dev = po->prot_hook.dev;
2209                 proto   = po->num;
2210                 addr    = NULL;
2211         } else {
2212                 err = -EINVAL;
2213                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2214                         goto out;
2215                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2216                         goto out;
2217                 proto   = saddr->sll_protocol;
2218                 addr    = saddr->sll_addr;
2219                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2220                 need_rls_dev = true;
2221         }
2222
2223         err = -ENXIO;
2224         if (dev == NULL)
2225                 goto out_unlock;
2226         if (sock->type == SOCK_RAW)
2227                 reserve = dev->hard_header_len;
2228
2229         err = -ENETDOWN;
2230         if (!(dev->flags & IFF_UP))
2231                 goto out_unlock;
2232
2233         if (po->has_vnet_hdr) {
2234                 vnet_hdr_len = sizeof(vnet_hdr);
2235
2236                 err = -EINVAL;
2237                 if (len < vnet_hdr_len)
2238                         goto out_unlock;
2239
2240                 len -= vnet_hdr_len;
2241
2242                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2243                                        vnet_hdr_len);
2244                 if (err < 0)
2245                         goto out_unlock;
2246
2247                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2248                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2249                       vnet_hdr.hdr_len))
2250                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
2251                                                  vnet_hdr.csum_offset + 2;
2252
2253                 err = -EINVAL;
2254                 if (vnet_hdr.hdr_len > len)
2255                         goto out_unlock;
2256
2257                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2258                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2259                         case VIRTIO_NET_HDR_GSO_TCPV4:
2260                                 gso_type = SKB_GSO_TCPV4;
2261                                 break;
2262                         case VIRTIO_NET_HDR_GSO_TCPV6:
2263                                 gso_type = SKB_GSO_TCPV6;
2264                                 break;
2265                         case VIRTIO_NET_HDR_GSO_UDP:
2266                                 gso_type = SKB_GSO_UDP;
2267                                 break;
2268                         default:
2269                                 goto out_unlock;
2270                         }
2271
2272                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2273                                 gso_type |= SKB_GSO_TCP_ECN;
2274
2275                         if (vnet_hdr.gso_size == 0)
2276                                 goto out_unlock;
2277
2278                 }
2279         }
2280
2281         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2282                 if (!netif_supports_nofcs(dev)) {
2283                         err = -EPROTONOSUPPORT;
2284                         goto out_unlock;
2285                 }
2286                 extra_len = 4; /* We're doing our own CRC */
2287         }
2288
2289         err = -EMSGSIZE;
2290         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2291                 goto out_unlock;
2292
2293         err = -ENOBUFS;
2294         hlen = LL_RESERVED_SPACE(dev);
2295         tlen = dev->needed_tailroom;
2296         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2297                                msg->msg_flags & MSG_DONTWAIT, &err);
2298         if (skb == NULL)
2299                 goto out_unlock;
2300
2301         skb_set_network_header(skb, reserve);
2302
2303         err = -EINVAL;
2304         if (sock->type == SOCK_DGRAM &&
2305             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2306                 goto out_free;
2307
2308         /* Returns -EFAULT on error */
2309         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2310         if (err)
2311                 goto out_free;
2312         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2313         if (err < 0)
2314                 goto out_free;
2315
2316         if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2317                 /* Earlier code assumed this would be a VLAN pkt,
2318                  * double-check this now that we have the actual
2319                  * packet in hand.
2320                  */
2321                 struct ethhdr *ehdr;
2322                 skb_reset_mac_header(skb);
2323                 ehdr = eth_hdr(skb);
2324                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2325                         err = -EMSGSIZE;
2326                         goto out_free;
2327                 }
2328         }
2329
2330         skb->protocol = proto;
2331         skb->dev = dev;
2332         skb->priority = sk->sk_priority;
2333         skb->mark = sk->sk_mark;
2334
2335         if (po->has_vnet_hdr) {
2336                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2337                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2338                                                   vnet_hdr.csum_offset)) {
2339                                 err = -EINVAL;
2340                                 goto out_free;
2341                         }
2342                 }
2343
2344                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2345                 skb_shinfo(skb)->gso_type = gso_type;
2346
2347                 /* Header must be checked, and gso_segs computed. */
2348                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2349                 skb_shinfo(skb)->gso_segs = 0;
2350
2351                 len += vnet_hdr_len;
2352         }
2353
2354         if (unlikely(extra_len == 4))
2355                 skb->no_fcs = 1;
2356
2357         /*
2358          *      Now send it
2359          */
2360
2361         err = dev_queue_xmit(skb);
2362         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2363                 goto out_unlock;
2364
2365         if (need_rls_dev)
2366                 dev_put(dev);
2367
2368         return len;
2369
2370 out_free:
2371         kfree_skb(skb);
2372 out_unlock:
2373         if (dev && need_rls_dev)
2374                 dev_put(dev);
2375 out:
2376         return err;
2377 }
2378
2379 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2380                 struct msghdr *msg, size_t len)
2381 {
2382         struct sock *sk = sock->sk;
2383         struct packet_sock *po = pkt_sk(sk);
2384         if (po->tx_ring.pg_vec)
2385                 return tpacket_snd(po, msg);
2386         else
2387                 return packet_snd(sock, msg, len);
2388 }
2389
2390 /*
2391  *      Close a PACKET socket. This is fairly simple. We immediately go
2392  *      to 'closed' state and remove our protocol entry in the device list.
2393  */
2394
2395 static int packet_release(struct socket *sock)
2396 {
2397         struct sock *sk = sock->sk;
2398         struct packet_sock *po;
2399         struct net *net;
2400         union tpacket_req_u req_u;
2401
2402         if (!sk)
2403                 return 0;
2404
2405         net = sock_net(sk);
2406         po = pkt_sk(sk);
2407
2408         mutex_lock(&net->packet.sklist_lock);
2409         sk_del_node_init_rcu(sk);
2410         mutex_unlock(&net->packet.sklist_lock);
2411
2412         preempt_disable();
2413         sock_prot_inuse_add(net, sk->sk_prot, -1);
2414         preempt_enable();
2415
2416         spin_lock(&po->bind_lock);
2417         unregister_prot_hook(sk, false);
2418         if (po->prot_hook.dev) {
2419                 dev_put(po->prot_hook.dev);
2420                 po->prot_hook.dev = NULL;
2421         }
2422         spin_unlock(&po->bind_lock);
2423
2424         packet_flush_mclist(sk);
2425
2426         if (po->rx_ring.pg_vec) {
2427                 memset(&req_u, 0, sizeof(req_u));
2428                 packet_set_ring(sk, &req_u, 1, 0);
2429         }
2430
2431         if (po->tx_ring.pg_vec) {
2432                 memset(&req_u, 0, sizeof(req_u));
2433                 packet_set_ring(sk, &req_u, 1, 1);
2434         }
2435
2436         fanout_release(sk);
2437
2438         synchronize_net();
2439         /*
2440          *      Now the socket is dead. No more input will appear.
2441          */
2442         sock_orphan(sk);
2443         sock->sk = NULL;
2444
2445         /* Purge queues */
2446
2447         skb_queue_purge(&sk->sk_receive_queue);
2448         sk_refcnt_debug_release(sk);
2449
2450         sock_put(sk);
2451         return 0;
2452 }
2453
2454 /*
2455  *      Attach a packet hook.
2456  */
2457
2458 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2459 {
2460         struct packet_sock *po = pkt_sk(sk);
2461
2462         if (po->fanout) {
2463                 if (dev)
2464                         dev_put(dev);
2465
2466                 return -EINVAL;
2467         }
2468
2469         lock_sock(sk);
2470
2471         spin_lock(&po->bind_lock);
2472         unregister_prot_hook(sk, true);
2473         po->num = protocol;
2474         po->prot_hook.type = protocol;
2475         if (po->prot_hook.dev)
2476                 dev_put(po->prot_hook.dev);
2477         po->prot_hook.dev = dev;
2478
2479         po->ifindex = dev ? dev->ifindex : 0;
2480
2481         if (protocol == 0)
2482                 goto out_unlock;
2483
2484         if (!dev || (dev->flags & IFF_UP)) {
2485                 register_prot_hook(sk);
2486         } else {
2487                 sk->sk_err = ENETDOWN;
2488                 if (!sock_flag(sk, SOCK_DEAD))
2489                         sk->sk_error_report(sk);
2490         }
2491
2492 out_unlock:
2493         spin_unlock(&po->bind_lock);
2494         release_sock(sk);
2495         return 0;
2496 }
2497
2498 /*
2499  *      Bind a packet socket to a device
2500  */
2501
2502 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2503                             int addr_len)
2504 {
2505         struct sock *sk = sock->sk;
2506         char name[15];
2507         struct net_device *dev;
2508         int err = -ENODEV;
2509
2510         /*
2511          *      Check legality
2512          */
2513
2514         if (addr_len != sizeof(struct sockaddr))
2515                 return -EINVAL;
2516         strlcpy(name, uaddr->sa_data, sizeof(name));
2517
2518         dev = dev_get_by_name(sock_net(sk), name);
2519         if (dev)
2520                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2521         return err;
2522 }
2523
2524 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2525 {
2526         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2527         struct sock *sk = sock->sk;
2528         struct net_device *dev = NULL;
2529         int err;
2530
2531
2532         /*
2533          *      Check legality
2534          */
2535
2536         if (addr_len < sizeof(struct sockaddr_ll))
2537                 return -EINVAL;
2538         if (sll->sll_family != AF_PACKET)
2539                 return -EINVAL;
2540
2541         if (sll->sll_ifindex) {
2542                 err = -ENODEV;
2543                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2544                 if (dev == NULL)
2545                         goto out;
2546         }
2547         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2548
2549 out:
2550         return err;
2551 }
2552
2553 static struct proto packet_proto = {
2554         .name     = "PACKET",
2555         .owner    = THIS_MODULE,
2556         .obj_size = sizeof(struct packet_sock),
2557 };
2558
2559 /*
2560  *      Create a packet of type SOCK_PACKET.
2561  */
2562
2563 static int packet_create(struct net *net, struct socket *sock, int protocol,
2564                          int kern)
2565 {
2566         struct sock *sk;
2567         struct packet_sock *po;
2568         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2569         int err;
2570
2571         if (!ns_capable(net->user_ns, CAP_NET_RAW))
2572                 return -EPERM;
2573         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2574             sock->type != SOCK_PACKET)
2575                 return -ESOCKTNOSUPPORT;
2576
2577         sock->state = SS_UNCONNECTED;
2578
2579         err = -ENOBUFS;
2580         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2581         if (sk == NULL)
2582                 goto out;
2583
2584         sock->ops = &packet_ops;
2585         if (sock->type == SOCK_PACKET)
2586                 sock->ops = &packet_ops_spkt;
2587
2588         sock_init_data(sock, sk);
2589
2590         po = pkt_sk(sk);
2591         sk->sk_family = PF_PACKET;
2592         po->num = proto;
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         mutex_lock(&net->packet.sklist_lock);
2616         sk_add_node_rcu(sk, &net->packet.sklist);
2617         mutex_unlock(&net->packet.sklist_lock);
2618
2619         preempt_disable();
2620         sock_prot_inuse_add(net, &packet_proto, 1);
2621         preempt_enable();
2622
2623         return 0;
2624 out:
2625         return err;
2626 }
2627
2628 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2629 {
2630         struct sock_exterr_skb *serr;
2631         struct sk_buff *skb, *skb2;
2632         int copied, err;
2633
2634         err = -EAGAIN;
2635         skb = skb_dequeue(&sk->sk_error_queue);
2636         if (skb == NULL)
2637                 goto out;
2638
2639         copied = skb->len;
2640         if (copied > len) {
2641                 msg->msg_flags |= MSG_TRUNC;
2642                 copied = len;
2643         }
2644         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2645         if (err)
2646                 goto out_free_skb;
2647
2648         sock_recv_timestamp(msg, sk, skb);
2649
2650         serr = SKB_EXT_ERR(skb);
2651         put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2652                  sizeof(serr->ee), &serr->ee);
2653
2654         msg->msg_flags |= MSG_ERRQUEUE;
2655         err = copied;
2656
2657         /* Reset and regenerate socket error */
2658         spin_lock_bh(&sk->sk_error_queue.lock);
2659         sk->sk_err = 0;
2660         if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2661                 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2662                 spin_unlock_bh(&sk->sk_error_queue.lock);
2663                 sk->sk_error_report(sk);
2664         } else
2665                 spin_unlock_bh(&sk->sk_error_queue.lock);
2666
2667 out_free_skb:
2668         kfree_skb(skb);
2669 out:
2670         return err;
2671 }
2672
2673 /*
2674  *      Pull a packet from our receive queue and hand it to the user.
2675  *      If necessary we block.
2676  */
2677
2678 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2679                           struct msghdr *msg, size_t len, int flags)
2680 {
2681         struct sock *sk = sock->sk;
2682         struct sk_buff *skb;
2683         int copied, err;
2684         struct sockaddr_ll *sll;
2685         int vnet_hdr_len = 0;
2686
2687         err = -EINVAL;
2688         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2689                 goto out;
2690
2691 #if 0
2692         /* What error should we return now? EUNATTACH? */
2693         if (pkt_sk(sk)->ifindex < 0)
2694                 return -ENODEV;
2695 #endif
2696
2697         if (flags & MSG_ERRQUEUE) {
2698                 err = packet_recv_error(sk, msg, len);
2699                 goto out;
2700         }
2701
2702         /*
2703          *      Call the generic datagram receiver. This handles all sorts
2704          *      of horrible races and re-entrancy so we can forget about it
2705          *      in the protocol layers.
2706          *
2707          *      Now it will return ENETDOWN, if device have just gone down,
2708          *      but then it will block.
2709          */
2710
2711         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2712
2713         /*
2714          *      An error occurred so return it. Because skb_recv_datagram()
2715          *      handles the blocking we don't see and worry about blocking
2716          *      retries.
2717          */
2718
2719         if (skb == NULL)
2720                 goto out;
2721
2722         if (pkt_sk(sk)->has_vnet_hdr) {
2723                 struct virtio_net_hdr vnet_hdr = { 0 };
2724
2725                 err = -EINVAL;
2726                 vnet_hdr_len = sizeof(vnet_hdr);
2727                 if (len < vnet_hdr_len)
2728                         goto out_free;
2729
2730                 len -= vnet_hdr_len;
2731
2732                 if (skb_is_gso(skb)) {
2733                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2734
2735                         /* This is a hint as to how much should be linear. */
2736                         vnet_hdr.hdr_len = skb_headlen(skb);
2737                         vnet_hdr.gso_size = sinfo->gso_size;
2738                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2739                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2740                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2741                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2742                         else if (sinfo->gso_type & SKB_GSO_UDP)
2743                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2744                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2745                                 goto out_free;
2746                         else
2747                                 BUG();
2748                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2749                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2750                 } else
2751                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2752
2753                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2754                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2755                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2756                         vnet_hdr.csum_offset = skb->csum_offset;
2757                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2758                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2759                 } /* else everything is zero */
2760
2761                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2762                                      vnet_hdr_len);
2763                 if (err < 0)
2764                         goto out_free;
2765         }
2766
2767         /*
2768          *      If the address length field is there to be filled in, we fill
2769          *      it in now.
2770          */
2771
2772         sll = &PACKET_SKB_CB(skb)->sa.ll;
2773         if (sock->type == SOCK_PACKET)
2774                 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2775         else
2776                 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2777
2778         /*
2779          *      You lose any data beyond the buffer you gave. If it worries a
2780          *      user program they can ask the device for its MTU anyway.
2781          */
2782
2783         copied = skb->len;
2784         if (copied > len) {
2785                 copied = len;
2786                 msg->msg_flags |= MSG_TRUNC;
2787         }
2788
2789         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2790         if (err)
2791                 goto out_free;
2792
2793         sock_recv_ts_and_drops(msg, sk, skb);
2794
2795         if (msg->msg_name)
2796                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2797                        msg->msg_namelen);
2798
2799         if (pkt_sk(sk)->auxdata) {
2800                 struct tpacket_auxdata aux;
2801
2802                 aux.tp_status = TP_STATUS_USER;
2803                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2804                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2805                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2806                 aux.tp_snaplen = skb->len;
2807                 aux.tp_mac = 0;
2808                 aux.tp_net = skb_network_offset(skb);
2809                 if (vlan_tx_tag_present(skb)) {
2810                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2811                         aux.tp_status |= TP_STATUS_VLAN_VALID;
2812                 } else {
2813                         aux.tp_vlan_tci = 0;
2814                 }
2815                 aux.tp_padding = 0;
2816                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2817         }
2818
2819         /*
2820          *      Free or return the buffer as appropriate. Again this
2821          *      hides all the races and re-entrancy issues from us.
2822          */
2823         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2824
2825 out_free:
2826         skb_free_datagram(sk, skb);
2827 out:
2828         return err;
2829 }
2830
2831 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2832                                int *uaddr_len, int peer)
2833 {
2834         struct net_device *dev;
2835         struct sock *sk = sock->sk;
2836
2837         if (peer)
2838                 return -EOPNOTSUPP;
2839
2840         uaddr->sa_family = AF_PACKET;
2841         rcu_read_lock();
2842         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2843         if (dev)
2844                 strncpy(uaddr->sa_data, dev->name, 14);
2845         else
2846                 memset(uaddr->sa_data, 0, 14);
2847         rcu_read_unlock();
2848         *uaddr_len = sizeof(*uaddr);
2849
2850         return 0;
2851 }
2852
2853 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2854                           int *uaddr_len, int peer)
2855 {
2856         struct net_device *dev;
2857         struct sock *sk = sock->sk;
2858         struct packet_sock *po = pkt_sk(sk);
2859         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2860
2861         if (peer)
2862                 return -EOPNOTSUPP;
2863
2864         sll->sll_family = AF_PACKET;
2865         sll->sll_ifindex = po->ifindex;
2866         sll->sll_protocol = po->num;
2867         sll->sll_pkttype = 0;
2868         rcu_read_lock();
2869         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2870         if (dev) {
2871                 sll->sll_hatype = dev->type;
2872                 sll->sll_halen = dev->addr_len;
2873                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2874         } else {
2875                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
2876                 sll->sll_halen = 0;
2877         }
2878         rcu_read_unlock();
2879         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2880
2881         return 0;
2882 }
2883
2884 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2885                          int what)
2886 {
2887         switch (i->type) {
2888         case PACKET_MR_MULTICAST:
2889                 if (i->alen != dev->addr_len)
2890                         return -EINVAL;
2891                 if (what > 0)
2892                         return dev_mc_add(dev, i->addr);
2893                 else
2894                         return dev_mc_del(dev, i->addr);
2895                 break;
2896         case PACKET_MR_PROMISC:
2897                 return dev_set_promiscuity(dev, what);
2898                 break;
2899         case PACKET_MR_ALLMULTI:
2900                 return dev_set_allmulti(dev, what);
2901                 break;
2902         case PACKET_MR_UNICAST:
2903                 if (i->alen != dev->addr_len)
2904                         return -EINVAL;
2905                 if (what > 0)
2906                         return dev_uc_add(dev, i->addr);
2907                 else
2908                         return dev_uc_del(dev, i->addr);
2909                 break;
2910         default:
2911                 break;
2912         }
2913         return 0;
2914 }
2915
2916 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2917 {
2918         for ( ; i; i = i->next) {
2919                 if (i->ifindex == dev->ifindex)
2920                         packet_dev_mc(dev, i, what);
2921         }
2922 }
2923
2924 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2925 {
2926         struct packet_sock *po = pkt_sk(sk);
2927         struct packet_mclist *ml, *i;
2928         struct net_device *dev;
2929         int err;
2930
2931         rtnl_lock();
2932
2933         err = -ENODEV;
2934         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2935         if (!dev)
2936                 goto done;
2937
2938         err = -EINVAL;
2939         if (mreq->mr_alen > dev->addr_len)
2940                 goto done;
2941
2942         err = -ENOBUFS;
2943         i = kmalloc(sizeof(*i), GFP_KERNEL);
2944         if (i == NULL)
2945                 goto done;
2946
2947         err = 0;
2948         for (ml = po->mclist; ml; ml = ml->next) {
2949                 if (ml->ifindex == mreq->mr_ifindex &&
2950                     ml->type == mreq->mr_type &&
2951                     ml->alen == mreq->mr_alen &&
2952                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2953                         ml->count++;
2954                         /* Free the new element ... */
2955                         kfree(i);
2956                         goto done;
2957                 }
2958         }
2959
2960         i->type = mreq->mr_type;
2961         i->ifindex = mreq->mr_ifindex;
2962         i->alen = mreq->mr_alen;
2963         memcpy(i->addr, mreq->mr_address, i->alen);
2964         i->count = 1;
2965         i->next = po->mclist;
2966         po->mclist = i;
2967         err = packet_dev_mc(dev, i, 1);
2968         if (err) {
2969                 po->mclist = i->next;
2970                 kfree(i);
2971         }
2972
2973 done:
2974         rtnl_unlock();
2975         return err;
2976 }
2977
2978 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2979 {
2980         struct packet_mclist *ml, **mlp;
2981
2982         rtnl_lock();
2983
2984         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
2985                 if (ml->ifindex == mreq->mr_ifindex &&
2986                     ml->type == mreq->mr_type &&
2987                     ml->alen == mreq->mr_alen &&
2988                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2989                         if (--ml->count == 0) {
2990                                 struct net_device *dev;
2991                                 *mlp = ml->next;
2992                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2993                                 if (dev)
2994                                         packet_dev_mc(dev, ml, -1);
2995                                 kfree(ml);
2996                         }
2997                         rtnl_unlock();
2998                         return 0;
2999                 }
3000         }
3001         rtnl_unlock();
3002         return -EADDRNOTAVAIL;
3003 }
3004
3005 static void packet_flush_mclist(struct sock *sk)
3006 {
3007         struct packet_sock *po = pkt_sk(sk);
3008         struct packet_mclist *ml;
3009
3010         if (!po->mclist)
3011                 return;
3012
3013         rtnl_lock();
3014         while ((ml = po->mclist) != NULL) {
3015                 struct net_device *dev;
3016
3017                 po->mclist = ml->next;
3018                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3019                 if (dev != NULL)
3020                         packet_dev_mc(dev, ml, -1);
3021                 kfree(ml);
3022         }
3023         rtnl_unlock();
3024 }
3025
3026 static int
3027 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3028 {
3029         struct sock *sk = sock->sk;
3030         struct packet_sock *po = pkt_sk(sk);
3031         int ret;
3032
3033         if (level != SOL_PACKET)
3034                 return -ENOPROTOOPT;
3035
3036         switch (optname) {
3037         case PACKET_ADD_MEMBERSHIP:
3038         case PACKET_DROP_MEMBERSHIP:
3039         {
3040                 struct packet_mreq_max mreq;
3041                 int len = optlen;
3042                 memset(&mreq, 0, sizeof(mreq));
3043                 if (len < sizeof(struct packet_mreq))
3044                         return -EINVAL;
3045                 if (len > sizeof(mreq))
3046                         len = sizeof(mreq);
3047                 if (copy_from_user(&mreq, optval, len))
3048                         return -EFAULT;
3049                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3050                         return -EINVAL;
3051                 if (optname == PACKET_ADD_MEMBERSHIP)
3052                         ret = packet_mc_add(sk, &mreq);
3053                 else
3054                         ret = packet_mc_drop(sk, &mreq);
3055                 return ret;
3056         }
3057
3058         case PACKET_RX_RING:
3059         case PACKET_TX_RING:
3060         {
3061                 union tpacket_req_u req_u;
3062                 int len;
3063
3064                 switch (po->tp_version) {
3065                 case TPACKET_V1:
3066                 case TPACKET_V2:
3067                         len = sizeof(req_u.req);
3068                         break;
3069                 case TPACKET_V3:
3070                 default:
3071                         len = sizeof(req_u.req3);
3072                         break;
3073                 }
3074                 if (optlen < len)
3075                         return -EINVAL;
3076                 if (pkt_sk(sk)->has_vnet_hdr)
3077                         return -EINVAL;
3078                 if (copy_from_user(&req_u.req, optval, len))
3079                         return -EFAULT;
3080                 return packet_set_ring(sk, &req_u, 0,
3081                         optname == PACKET_TX_RING);
3082         }
3083         case PACKET_COPY_THRESH:
3084         {
3085                 int val;
3086
3087                 if (optlen != sizeof(val))
3088                         return -EINVAL;
3089                 if (copy_from_user(&val, optval, sizeof(val)))
3090                         return -EFAULT;
3091
3092                 pkt_sk(sk)->copy_thresh = val;
3093                 return 0;
3094         }
3095         case PACKET_VERSION:
3096         {
3097                 int val;
3098
3099                 if (optlen != sizeof(val))
3100                         return -EINVAL;
3101                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3102                         return -EBUSY;
3103                 if (copy_from_user(&val, optval, sizeof(val)))
3104                         return -EFAULT;
3105                 switch (val) {
3106                 case TPACKET_V1:
3107                 case TPACKET_V2:
3108                 case TPACKET_V3:
3109                         po->tp_version = val;
3110                         return 0;
3111                 default:
3112                         return -EINVAL;
3113                 }
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         case PACKET_TX_HAS_OFF:
3205         {
3206                 unsigned int val;
3207
3208                 if (optlen != sizeof(val))
3209                         return -EINVAL;
3210                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3211                         return -EBUSY;
3212                 if (copy_from_user(&val, optval, sizeof(val)))
3213                         return -EFAULT;
3214                 po->tp_tx_has_off = !!val;
3215                 return 0;
3216         }
3217         default:
3218                 return -ENOPROTOOPT;
3219         }
3220 }
3221
3222 static int packet_getsockopt(struct socket *sock, int level, int optname,
3223                              char __user *optval, int __user *optlen)
3224 {
3225         int len;
3226         int val, lv = sizeof(val);
3227         struct sock *sk = sock->sk;
3228         struct packet_sock *po = pkt_sk(sk);
3229         void *data = &val;
3230         struct tpacket_stats st;
3231         union tpacket_stats_u st_u;
3232
3233         if (level != SOL_PACKET)
3234                 return -ENOPROTOOPT;
3235
3236         if (get_user(len, optlen))
3237                 return -EFAULT;
3238
3239         if (len < 0)
3240                 return -EINVAL;
3241
3242         switch (optname) {
3243         case PACKET_STATISTICS:
3244                 spin_lock_bh(&sk->sk_receive_queue.lock);
3245                 if (po->tp_version == TPACKET_V3) {
3246                         lv = sizeof(struct tpacket_stats_v3);
3247                         memcpy(&st_u.stats3, &po->stats,
3248                                sizeof(struct tpacket_stats));
3249                         st_u.stats3.tp_freeze_q_cnt =
3250                                         po->stats_u.stats3.tp_freeze_q_cnt;
3251                         st_u.stats3.tp_packets += po->stats.tp_drops;
3252                         data = &st_u.stats3;
3253                 } else {
3254                         lv = sizeof(struct tpacket_stats);
3255                         st = po->stats;
3256                         st.tp_packets += st.tp_drops;
3257                         data = &st;
3258                 }
3259                 memset(&po->stats, 0, sizeof(st));
3260                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3261                 break;
3262         case PACKET_AUXDATA:
3263                 val = po->auxdata;
3264                 break;
3265         case PACKET_ORIGDEV:
3266                 val = po->origdev;
3267                 break;
3268         case PACKET_VNET_HDR:
3269                 val = po->has_vnet_hdr;
3270                 break;
3271         case PACKET_VERSION:
3272                 val = po->tp_version;
3273                 break;
3274         case PACKET_HDRLEN:
3275                 if (len > sizeof(int))
3276                         len = sizeof(int);
3277                 if (copy_from_user(&val, optval, len))
3278                         return -EFAULT;
3279                 switch (val) {
3280                 case TPACKET_V1:
3281                         val = sizeof(struct tpacket_hdr);
3282                         break;
3283                 case TPACKET_V2:
3284                         val = sizeof(struct tpacket2_hdr);
3285                         break;
3286                 case TPACKET_V3:
3287                         val = sizeof(struct tpacket3_hdr);
3288                         break;
3289                 default:
3290                         return -EINVAL;
3291                 }
3292                 break;
3293         case PACKET_RESERVE:
3294                 val = po->tp_reserve;
3295                 break;
3296         case PACKET_LOSS:
3297                 val = po->tp_loss;
3298                 break;
3299         case PACKET_TIMESTAMP:
3300                 val = po->tp_tstamp;
3301                 break;
3302         case PACKET_FANOUT:
3303                 val = (po->fanout ?
3304                        ((u32)po->fanout->id |
3305                         ((u32)po->fanout->type << 16) |
3306                         ((u32)po->fanout->flags << 24)) :
3307                        0);
3308                 break;
3309         case PACKET_TX_HAS_OFF:
3310                 val = po->tp_tx_has_off;
3311                 break;
3312         default:
3313                 return -ENOPROTOOPT;
3314         }
3315
3316         if (len > lv)
3317                 len = lv;
3318         if (put_user(len, optlen))
3319                 return -EFAULT;
3320         if (copy_to_user(optval, data, len))
3321                 return -EFAULT;
3322         return 0;
3323 }
3324
3325
3326 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3327 {
3328         struct sock *sk;
3329         struct net_device *dev = data;
3330         struct net *net = dev_net(dev);
3331
3332         rcu_read_lock();
3333         sk_for_each_rcu(sk, &net->packet.sklist) {
3334                 struct packet_sock *po = pkt_sk(sk);
3335
3336                 switch (msg) {
3337                 case NETDEV_UNREGISTER:
3338                         if (po->mclist)
3339                                 packet_dev_mclist(dev, po->mclist, -1);
3340                         /* fallthrough */
3341
3342                 case NETDEV_DOWN:
3343                         if (dev->ifindex == po->ifindex) {
3344                                 spin_lock(&po->bind_lock);
3345                                 if (po->running) {
3346                                         __unregister_prot_hook(sk, false);
3347                                         sk->sk_err = ENETDOWN;
3348                                         if (!sock_flag(sk, SOCK_DEAD))
3349                                                 sk->sk_error_report(sk);
3350                                 }
3351                                 if (msg == NETDEV_UNREGISTER) {
3352                                         po->ifindex = -1;
3353                                         if (po->prot_hook.dev)
3354                                                 dev_put(po->prot_hook.dev);
3355                                         po->prot_hook.dev = NULL;
3356                                 }
3357                                 spin_unlock(&po->bind_lock);
3358                         }
3359                         break;
3360                 case NETDEV_UP:
3361                         if (dev->ifindex == po->ifindex) {
3362                                 spin_lock(&po->bind_lock);
3363                                 if (po->num)
3364                                         register_prot_hook(sk);
3365                                 spin_unlock(&po->bind_lock);
3366                         }
3367                         break;
3368                 }
3369         }
3370         rcu_read_unlock();
3371         return NOTIFY_DONE;
3372 }
3373
3374
3375 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3376                         unsigned long arg)
3377 {
3378         struct sock *sk = sock->sk;
3379
3380         switch (cmd) {
3381         case SIOCOUTQ:
3382         {
3383                 int amount = sk_wmem_alloc_get(sk);
3384
3385                 return put_user(amount, (int __user *)arg);
3386         }
3387         case SIOCINQ:
3388         {
3389                 struct sk_buff *skb;
3390                 int amount = 0;
3391
3392                 spin_lock_bh(&sk->sk_receive_queue.lock);
3393                 skb = skb_peek(&sk->sk_receive_queue);
3394                 if (skb)
3395                         amount = skb->len;
3396                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3397                 return put_user(amount, (int __user *)arg);
3398         }
3399         case SIOCGSTAMP:
3400                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3401         case SIOCGSTAMPNS:
3402                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3403
3404 #ifdef CONFIG_INET
3405         case SIOCADDRT:
3406         case SIOCDELRT:
3407         case SIOCDARP:
3408         case SIOCGARP:
3409         case SIOCSARP:
3410         case SIOCGIFADDR:
3411         case SIOCSIFADDR:
3412         case SIOCGIFBRDADDR:
3413         case SIOCSIFBRDADDR:
3414         case SIOCGIFNETMASK:
3415         case SIOCSIFNETMASK:
3416         case SIOCGIFDSTADDR:
3417         case SIOCSIFDSTADDR:
3418         case SIOCSIFFLAGS:
3419                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3420 #endif
3421
3422         default:
3423                 return -ENOIOCTLCMD;
3424         }
3425         return 0;
3426 }
3427
3428 static unsigned int packet_poll(struct file *file, struct socket *sock,
3429                                 poll_table *wait)
3430 {
3431         struct sock *sk = sock->sk;
3432         struct packet_sock *po = pkt_sk(sk);
3433         unsigned int mask = datagram_poll(file, sock, wait);
3434
3435         spin_lock_bh(&sk->sk_receive_queue.lock);
3436         if (po->rx_ring.pg_vec) {
3437                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3438                         TP_STATUS_KERNEL))
3439                         mask |= POLLIN | POLLRDNORM;
3440         }
3441         spin_unlock_bh(&sk->sk_receive_queue.lock);
3442         spin_lock_bh(&sk->sk_write_queue.lock);
3443         if (po->tx_ring.pg_vec) {
3444                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3445                         mask |= POLLOUT | POLLWRNORM;
3446         }
3447         spin_unlock_bh(&sk->sk_write_queue.lock);
3448         return mask;
3449 }
3450
3451
3452 /* Dirty? Well, I still did not learn better way to account
3453  * for user mmaps.
3454  */
3455
3456 static void packet_mm_open(struct vm_area_struct *vma)
3457 {
3458         struct file *file = vma->vm_file;
3459         struct socket *sock = file->private_data;
3460         struct sock *sk = sock->sk;
3461
3462         if (sk)
3463                 atomic_inc(&pkt_sk(sk)->mapped);
3464 }
3465
3466 static void packet_mm_close(struct vm_area_struct *vma)
3467 {
3468         struct file *file = vma->vm_file;
3469         struct socket *sock = file->private_data;
3470         struct sock *sk = sock->sk;
3471
3472         if (sk)
3473                 atomic_dec(&pkt_sk(sk)->mapped);
3474 }
3475
3476 static const struct vm_operations_struct packet_mmap_ops = {
3477         .open   =       packet_mm_open,
3478         .close  =       packet_mm_close,
3479 };
3480
3481 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3482                         unsigned int len)
3483 {
3484         int i;
3485
3486         for (i = 0; i < len; i++) {
3487                 if (likely(pg_vec[i].buffer)) {
3488                         if (is_vmalloc_addr(pg_vec[i].buffer))
3489                                 vfree(pg_vec[i].buffer);
3490                         else
3491                                 free_pages((unsigned long)pg_vec[i].buffer,
3492                                            order);
3493                         pg_vec[i].buffer = NULL;
3494                 }
3495         }
3496         kfree(pg_vec);
3497 }
3498
3499 static char *alloc_one_pg_vec_page(unsigned long order)
3500 {
3501         char *buffer = NULL;
3502         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3503                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3504
3505         buffer = (char *) __get_free_pages(gfp_flags, order);
3506
3507         if (buffer)
3508                 return buffer;
3509
3510         /*
3511          * __get_free_pages failed, fall back to vmalloc
3512          */
3513         buffer = vzalloc((1 << order) * PAGE_SIZE);
3514
3515         if (buffer)
3516                 return buffer;
3517
3518         /*
3519          * vmalloc failed, lets dig into swap here
3520          */
3521         gfp_flags &= ~__GFP_NORETRY;
3522         buffer = (char *)__get_free_pages(gfp_flags, order);
3523         if (buffer)
3524                 return buffer;
3525
3526         /*
3527          * complete and utter failure
3528          */
3529         return NULL;
3530 }
3531
3532 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3533 {
3534         unsigned int block_nr = req->tp_block_nr;
3535         struct pgv *pg_vec;
3536         int i;
3537
3538         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3539         if (unlikely(!pg_vec))
3540                 goto out;
3541
3542         for (i = 0; i < block_nr; i++) {
3543                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3544                 if (unlikely(!pg_vec[i].buffer))
3545                         goto out_free_pgvec;
3546         }
3547
3548 out:
3549         return pg_vec;
3550
3551 out_free_pgvec:
3552         free_pg_vec(pg_vec, order, block_nr);
3553         pg_vec = NULL;
3554         goto out;
3555 }
3556
3557 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3558                 int closing, int tx_ring)
3559 {
3560         struct pgv *pg_vec = NULL;
3561         struct packet_sock *po = pkt_sk(sk);
3562         int was_running, order = 0;
3563         struct packet_ring_buffer *rb;
3564         struct sk_buff_head *rb_queue;
3565         __be16 num;
3566         int err = -EINVAL;
3567         /* Added to avoid minimal code churn */
3568         struct tpacket_req *req = &req_u->req;
3569
3570         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3571         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3572                 WARN(1, "Tx-ring is not supported.\n");
3573                 goto out;
3574         }
3575
3576         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3577         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3578
3579         err = -EBUSY;
3580         if (!closing) {
3581                 if (atomic_read(&po->mapped))
3582                         goto out;
3583                 if (atomic_read(&rb->pending))
3584                         goto out;
3585         }
3586
3587         if (req->tp_block_nr) {
3588                 /* Sanity tests and some calculations */
3589                 err = -EBUSY;
3590                 if (unlikely(rb->pg_vec))
3591                         goto out;
3592
3593                 switch (po->tp_version) {
3594                 case TPACKET_V1:
3595                         po->tp_hdrlen = TPACKET_HDRLEN;
3596                         break;
3597                 case TPACKET_V2:
3598                         po->tp_hdrlen = TPACKET2_HDRLEN;
3599                         break;
3600                 case TPACKET_V3:
3601                         po->tp_hdrlen = TPACKET3_HDRLEN;
3602                         break;
3603                 }
3604
3605                 err = -EINVAL;
3606                 if (unlikely((int)req->tp_block_size <= 0))
3607                         goto out;
3608                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3609                         goto out;
3610                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3611                                         po->tp_reserve))
3612                         goto out;
3613                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3614                         goto out;
3615
3616                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3617                 if (unlikely(rb->frames_per_block <= 0))
3618                         goto out;
3619                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3620                                         req->tp_frame_nr))
3621                         goto out;
3622
3623                 err = -ENOMEM;
3624                 order = get_order(req->tp_block_size);
3625                 pg_vec = alloc_pg_vec(req, order);
3626                 if (unlikely(!pg_vec))
3627                         goto out;
3628                 switch (po->tp_version) {
3629                 case TPACKET_V3:
3630                 /* Transmit path is not supported. We checked
3631                  * it above but just being paranoid
3632                  */
3633                         if (!tx_ring)
3634                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3635                                 break;
3636                 default:
3637                         break;
3638                 }
3639         }
3640         /* Done */
3641         else {
3642                 err = -EINVAL;
3643                 if (unlikely(req->tp_frame_nr))
3644                         goto out;
3645         }
3646
3647         lock_sock(sk);
3648
3649         /* Detach socket from network */
3650         spin_lock(&po->bind_lock);
3651         was_running = po->running;
3652         num = po->num;
3653         if (was_running) {
3654                 po->num = 0;
3655                 __unregister_prot_hook(sk, false);
3656         }
3657         spin_unlock(&po->bind_lock);
3658
3659         synchronize_net();
3660
3661         err = -EBUSY;
3662         mutex_lock(&po->pg_vec_lock);
3663         if (closing || atomic_read(&po->mapped) == 0) {
3664                 err = 0;
3665                 spin_lock_bh(&rb_queue->lock);
3666                 swap(rb->pg_vec, pg_vec);
3667                 rb->frame_max = (req->tp_frame_nr - 1);
3668                 rb->head = 0;
3669                 rb->frame_size = req->tp_frame_size;
3670                 spin_unlock_bh(&rb_queue->lock);
3671
3672                 swap(rb->pg_vec_order, order);
3673                 swap(rb->pg_vec_len, req->tp_block_nr);
3674
3675                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3676                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3677                                                 tpacket_rcv : packet_rcv;
3678                 skb_queue_purge(rb_queue);
3679                 if (atomic_read(&po->mapped))
3680                         pr_err("packet_mmap: vma is busy: %d\n",
3681                                atomic_read(&po->mapped));
3682         }
3683         mutex_unlock(&po->pg_vec_lock);
3684
3685         spin_lock(&po->bind_lock);
3686         if (was_running) {
3687                 po->num = num;
3688                 register_prot_hook(sk);
3689         }
3690         spin_unlock(&po->bind_lock);
3691         if (closing && (po->tp_version > TPACKET_V2)) {
3692                 /* Because we don't support block-based V3 on tx-ring */
3693                 if (!tx_ring)
3694                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3695         }
3696         release_sock(sk);
3697
3698         if (pg_vec)
3699                 free_pg_vec(pg_vec, order, req->tp_block_nr);
3700 out:
3701         return err;
3702 }
3703
3704 static int packet_mmap(struct file *file, struct socket *sock,
3705                 struct vm_area_struct *vma)
3706 {
3707         struct sock *sk = sock->sk;
3708         struct packet_sock *po = pkt_sk(sk);
3709         unsigned long size, expected_size;
3710         struct packet_ring_buffer *rb;
3711         unsigned long start;
3712         int err = -EINVAL;
3713         int i;
3714
3715         if (vma->vm_pgoff)
3716                 return -EINVAL;
3717
3718         mutex_lock(&po->pg_vec_lock);
3719
3720         expected_size = 0;
3721         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3722                 if (rb->pg_vec) {
3723                         expected_size += rb->pg_vec_len
3724                                                 * rb->pg_vec_pages
3725                                                 * PAGE_SIZE;
3726                 }
3727         }
3728
3729         if (expected_size == 0)
3730                 goto out;
3731
3732         size = vma->vm_end - vma->vm_start;
3733         if (size != expected_size)
3734                 goto out;
3735
3736         start = vma->vm_start;
3737         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3738                 if (rb->pg_vec == NULL)
3739                         continue;
3740
3741                 for (i = 0; i < rb->pg_vec_len; i++) {
3742                         struct page *page;
3743                         void *kaddr = rb->pg_vec[i].buffer;
3744                         int pg_num;
3745
3746                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3747                                 page = pgv_to_page(kaddr);
3748                                 err = vm_insert_page(vma, start, page);
3749                                 if (unlikely(err))
3750                                         goto out;
3751                                 start += PAGE_SIZE;
3752                                 kaddr += PAGE_SIZE;
3753                         }
3754                 }
3755         }
3756
3757         atomic_inc(&po->mapped);
3758         vma->vm_ops = &packet_mmap_ops;
3759         err = 0;
3760
3761 out:
3762         mutex_unlock(&po->pg_vec_lock);
3763         return err;
3764 }
3765
3766 static const struct proto_ops packet_ops_spkt = {
3767         .family =       PF_PACKET,
3768         .owner =        THIS_MODULE,
3769         .release =      packet_release,
3770         .bind =         packet_bind_spkt,
3771         .connect =      sock_no_connect,
3772         .socketpair =   sock_no_socketpair,
3773         .accept =       sock_no_accept,
3774         .getname =      packet_getname_spkt,
3775         .poll =         datagram_poll,
3776         .ioctl =        packet_ioctl,
3777         .listen =       sock_no_listen,
3778         .shutdown =     sock_no_shutdown,
3779         .setsockopt =   sock_no_setsockopt,
3780         .getsockopt =   sock_no_getsockopt,
3781         .sendmsg =      packet_sendmsg_spkt,
3782         .recvmsg =      packet_recvmsg,
3783         .mmap =         sock_no_mmap,
3784         .sendpage =     sock_no_sendpage,
3785 };
3786
3787 static const struct proto_ops packet_ops = {
3788         .family =       PF_PACKET,
3789         .owner =        THIS_MODULE,
3790         .release =      packet_release,
3791         .bind =         packet_bind,
3792         .connect =      sock_no_connect,
3793         .socketpair =   sock_no_socketpair,
3794         .accept =       sock_no_accept,
3795         .getname =      packet_getname,
3796         .poll =         packet_poll,
3797         .ioctl =        packet_ioctl,
3798         .listen =       sock_no_listen,
3799         .shutdown =     sock_no_shutdown,
3800         .setsockopt =   packet_setsockopt,
3801         .getsockopt =   packet_getsockopt,
3802         .sendmsg =      packet_sendmsg,
3803         .recvmsg =      packet_recvmsg,
3804         .mmap =         packet_mmap,
3805         .sendpage =     sock_no_sendpage,
3806 };
3807
3808 static const struct net_proto_family packet_family_ops = {
3809         .family =       PF_PACKET,
3810         .create =       packet_create,
3811         .owner  =       THIS_MODULE,
3812 };
3813
3814 static struct notifier_block packet_netdev_notifier = {
3815         .notifier_call =        packet_notifier,
3816 };
3817
3818 #ifdef CONFIG_PROC_FS
3819
3820 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3821         __acquires(RCU)
3822 {
3823         struct net *net = seq_file_net(seq);
3824
3825         rcu_read_lock();
3826         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3827 }
3828
3829 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3830 {
3831         struct net *net = seq_file_net(seq);
3832         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3833 }
3834
3835 static void packet_seq_stop(struct seq_file *seq, void *v)
3836         __releases(RCU)
3837 {
3838         rcu_read_unlock();
3839 }
3840
3841 static int packet_seq_show(struct seq_file *seq, void *v)
3842 {
3843         if (v == SEQ_START_TOKEN)
3844                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
3845         else {
3846                 struct sock *s = sk_entry(v);
3847                 const struct packet_sock *po = pkt_sk(s);
3848
3849                 seq_printf(seq,
3850                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
3851                            s,
3852                            atomic_read(&s->sk_refcnt),
3853                            s->sk_type,
3854                            ntohs(po->num),
3855                            po->ifindex,
3856                            po->running,
3857                            atomic_read(&s->sk_rmem_alloc),
3858                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
3859                            sock_i_ino(s));
3860         }
3861
3862         return 0;
3863 }
3864
3865 static const struct seq_operations packet_seq_ops = {
3866         .start  = packet_seq_start,
3867         .next   = packet_seq_next,
3868         .stop   = packet_seq_stop,
3869         .show   = packet_seq_show,
3870 };
3871
3872 static int packet_seq_open(struct inode *inode, struct file *file)
3873 {
3874         return seq_open_net(inode, file, &packet_seq_ops,
3875                             sizeof(struct seq_net_private));
3876 }
3877
3878 static const struct file_operations packet_seq_fops = {
3879         .owner          = THIS_MODULE,
3880         .open           = packet_seq_open,
3881         .read           = seq_read,
3882         .llseek         = seq_lseek,
3883         .release        = seq_release_net,
3884 };
3885
3886 #endif
3887
3888 static int __net_init packet_net_init(struct net *net)
3889 {
3890         mutex_init(&net->packet.sklist_lock);
3891         INIT_HLIST_HEAD(&net->packet.sklist);
3892
3893         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
3894                 return -ENOMEM;
3895
3896         return 0;
3897 }
3898
3899 static void __net_exit packet_net_exit(struct net *net)
3900 {
3901         remove_proc_entry("packet", net->proc_net);
3902 }
3903
3904 static struct pernet_operations packet_net_ops = {
3905         .init = packet_net_init,
3906         .exit = packet_net_exit,
3907 };
3908
3909
3910 static void __exit packet_exit(void)
3911 {
3912         unregister_netdevice_notifier(&packet_netdev_notifier);
3913         unregister_pernet_subsys(&packet_net_ops);
3914         sock_unregister(PF_PACKET);
3915         proto_unregister(&packet_proto);
3916 }
3917
3918 static int __init packet_init(void)
3919 {
3920         int rc = proto_register(&packet_proto, 0);
3921
3922         if (rc != 0)
3923                 goto out;
3924
3925         sock_register(&packet_family_ops);
3926         register_pernet_subsys(&packet_net_ops);
3927         register_netdevice_notifier(&packet_netdev_notifier);
3928 out:
3929         return rc;
3930 }
3931
3932 module_init(packet_init);
3933 module_exit(packet_exit);
3934 MODULE_LICENSE("GPL");
3935 MODULE_ALIAS_NETPROTO(PF_PACKET);