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
1509         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1510
1511         if (unlikely(extra_len == 4))
1512                 skb->no_fcs = 1;
1513
1514         skb_probe_transport_header(skb, 0);
1515
1516         dev_queue_xmit(skb);
1517         rcu_read_unlock();
1518         return len;
1519
1520 out_unlock:
1521         rcu_read_unlock();
1522 out_free:
1523         kfree_skb(skb);
1524         return err;
1525 }
1526
1527 static unsigned int run_filter(const struct sk_buff *skb,
1528                                       const struct sock *sk,
1529                                       unsigned int res)
1530 {
1531         struct sk_filter *filter;
1532
1533         rcu_read_lock();
1534         filter = rcu_dereference(sk->sk_filter);
1535         if (filter != NULL)
1536                 res = SK_RUN_FILTER(filter, skb);
1537         rcu_read_unlock();
1538
1539         return res;
1540 }
1541
1542 /*
1543  * This function makes lazy skb cloning in hope that most of packets
1544  * are discarded by BPF.
1545  *
1546  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1547  * and skb->cb are mangled. It works because (and until) packets
1548  * falling here are owned by current CPU. Output packets are cloned
1549  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1550  * sequencially, so that if we return skb to original state on exit,
1551  * we will not harm anyone.
1552  */
1553
1554 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1555                       struct packet_type *pt, struct net_device *orig_dev)
1556 {
1557         struct sock *sk;
1558         struct sockaddr_ll *sll;
1559         struct packet_sock *po;
1560         u8 *skb_head = skb->data;
1561         int skb_len = skb->len;
1562         unsigned int snaplen, res;
1563
1564         if (skb->pkt_type == PACKET_LOOPBACK)
1565                 goto drop;
1566
1567         sk = pt->af_packet_priv;
1568         po = pkt_sk(sk);
1569
1570         if (!net_eq(dev_net(dev), sock_net(sk)))
1571                 goto drop;
1572
1573         skb->dev = dev;
1574
1575         if (dev->header_ops) {
1576                 /* The device has an explicit notion of ll header,
1577                  * exported to higher levels.
1578                  *
1579                  * Otherwise, the device hides details of its frame
1580                  * structure, so that corresponding packet head is
1581                  * never delivered to user.
1582                  */
1583                 if (sk->sk_type != SOCK_DGRAM)
1584                         skb_push(skb, skb->data - skb_mac_header(skb));
1585                 else if (skb->pkt_type == PACKET_OUTGOING) {
1586                         /* Special case: outgoing packets have ll header at head */
1587                         skb_pull(skb, skb_network_offset(skb));
1588                 }
1589         }
1590
1591         snaplen = skb->len;
1592
1593         res = run_filter(skb, sk, snaplen);
1594         if (!res)
1595                 goto drop_n_restore;
1596         if (snaplen > res)
1597                 snaplen = res;
1598
1599         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1600                 goto drop_n_acct;
1601
1602         if (skb_shared(skb)) {
1603                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1604                 if (nskb == NULL)
1605                         goto drop_n_acct;
1606
1607                 if (skb_head != skb->data) {
1608                         skb->data = skb_head;
1609                         skb->len = skb_len;
1610                 }
1611                 consume_skb(skb);
1612                 skb = nskb;
1613         }
1614
1615         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1616                      sizeof(skb->cb));
1617
1618         sll = &PACKET_SKB_CB(skb)->sa.ll;
1619         sll->sll_family = AF_PACKET;
1620         sll->sll_hatype = dev->type;
1621         sll->sll_protocol = skb->protocol;
1622         sll->sll_pkttype = skb->pkt_type;
1623         if (unlikely(po->origdev))
1624                 sll->sll_ifindex = orig_dev->ifindex;
1625         else
1626                 sll->sll_ifindex = dev->ifindex;
1627
1628         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1629
1630         PACKET_SKB_CB(skb)->origlen = skb->len;
1631
1632         if (pskb_trim(skb, snaplen))
1633                 goto drop_n_acct;
1634
1635         skb_set_owner_r(skb, sk);
1636         skb->dev = NULL;
1637         skb_dst_drop(skb);
1638
1639         /* drop conntrack reference */
1640         nf_reset(skb);
1641
1642         spin_lock(&sk->sk_receive_queue.lock);
1643         po->stats.tp_packets++;
1644         skb->dropcount = atomic_read(&sk->sk_drops);
1645         __skb_queue_tail(&sk->sk_receive_queue, skb);
1646         spin_unlock(&sk->sk_receive_queue.lock);
1647         sk->sk_data_ready(sk, skb->len);
1648         return 0;
1649
1650 drop_n_acct:
1651         spin_lock(&sk->sk_receive_queue.lock);
1652         po->stats.tp_drops++;
1653         atomic_inc(&sk->sk_drops);
1654         spin_unlock(&sk->sk_receive_queue.lock);
1655
1656 drop_n_restore:
1657         if (skb_head != skb->data && skb_shared(skb)) {
1658                 skb->data = skb_head;
1659                 skb->len = skb_len;
1660         }
1661 drop:
1662         consume_skb(skb);
1663         return 0;
1664 }
1665
1666 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1667                        struct packet_type *pt, struct net_device *orig_dev)
1668 {
1669         struct sock *sk;
1670         struct packet_sock *po;
1671         struct sockaddr_ll *sll;
1672         union {
1673                 struct tpacket_hdr *h1;
1674                 struct tpacket2_hdr *h2;
1675                 struct tpacket3_hdr *h3;
1676                 void *raw;
1677         } h;
1678         u8 *skb_head = skb->data;
1679         int skb_len = skb->len;
1680         unsigned int snaplen, res;
1681         unsigned long status = TP_STATUS_USER;
1682         unsigned short macoff, netoff, hdrlen;
1683         struct sk_buff *copy_skb = NULL;
1684         struct timeval tv;
1685         struct timespec ts;
1686         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1687
1688         if (skb->pkt_type == PACKET_LOOPBACK)
1689                 goto drop;
1690
1691         sk = pt->af_packet_priv;
1692         po = pkt_sk(sk);
1693
1694         if (!net_eq(dev_net(dev), sock_net(sk)))
1695                 goto drop;
1696
1697         if (dev->header_ops) {
1698                 if (sk->sk_type != SOCK_DGRAM)
1699                         skb_push(skb, skb->data - skb_mac_header(skb));
1700                 else if (skb->pkt_type == PACKET_OUTGOING) {
1701                         /* Special case: outgoing packets have ll header at head */
1702                         skb_pull(skb, skb_network_offset(skb));
1703                 }
1704         }
1705
1706         if (skb->ip_summed == CHECKSUM_PARTIAL)
1707                 status |= TP_STATUS_CSUMNOTREADY;
1708
1709         snaplen = skb->len;
1710
1711         res = run_filter(skb, sk, snaplen);
1712         if (!res)
1713                 goto drop_n_restore;
1714         if (snaplen > res)
1715                 snaplen = res;
1716
1717         if (sk->sk_type == SOCK_DGRAM) {
1718                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1719                                   po->tp_reserve;
1720         } else {
1721                 unsigned int maclen = skb_network_offset(skb);
1722                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1723                                        (maclen < 16 ? 16 : maclen)) +
1724                         po->tp_reserve;
1725                 macoff = netoff - maclen;
1726         }
1727         if (po->tp_version <= TPACKET_V2) {
1728                 if (macoff + snaplen > po->rx_ring.frame_size) {
1729                         if (po->copy_thresh &&
1730                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1731                                 if (skb_shared(skb)) {
1732                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1733                                 } else {
1734                                         copy_skb = skb_get(skb);
1735                                         skb_head = skb->data;
1736                                 }
1737                                 if (copy_skb)
1738                                         skb_set_owner_r(copy_skb, sk);
1739                         }
1740                         snaplen = po->rx_ring.frame_size - macoff;
1741                         if ((int)snaplen < 0)
1742                                 snaplen = 0;
1743                 }
1744         }
1745         spin_lock(&sk->sk_receive_queue.lock);
1746         h.raw = packet_current_rx_frame(po, skb,
1747                                         TP_STATUS_KERNEL, (macoff+snaplen));
1748         if (!h.raw)
1749                 goto ring_is_full;
1750         if (po->tp_version <= TPACKET_V2) {
1751                 packet_increment_rx_head(po, &po->rx_ring);
1752         /*
1753          * LOSING will be reported till you read the stats,
1754          * because it's COR - Clear On Read.
1755          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1756          * at packet level.
1757          */
1758                 if (po->stats.tp_drops)
1759                         status |= TP_STATUS_LOSING;
1760         }
1761         po->stats.tp_packets++;
1762         if (copy_skb) {
1763                 status |= TP_STATUS_COPY;
1764                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1765         }
1766         spin_unlock(&sk->sk_receive_queue.lock);
1767
1768         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1769
1770         switch (po->tp_version) {
1771         case TPACKET_V1:
1772                 h.h1->tp_len = skb->len;
1773                 h.h1->tp_snaplen = snaplen;
1774                 h.h1->tp_mac = macoff;
1775                 h.h1->tp_net = netoff;
1776                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1777                                 && shhwtstamps->syststamp.tv64)
1778                         tv = ktime_to_timeval(shhwtstamps->syststamp);
1779                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1780                                 && shhwtstamps->hwtstamp.tv64)
1781                         tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1782                 else if (skb->tstamp.tv64)
1783                         tv = ktime_to_timeval(skb->tstamp);
1784                 else
1785                         do_gettimeofday(&tv);
1786                 h.h1->tp_sec = tv.tv_sec;
1787                 h.h1->tp_usec = tv.tv_usec;
1788                 hdrlen = sizeof(*h.h1);
1789                 break;
1790         case TPACKET_V2:
1791                 h.h2->tp_len = skb->len;
1792                 h.h2->tp_snaplen = snaplen;
1793                 h.h2->tp_mac = macoff;
1794                 h.h2->tp_net = netoff;
1795                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1796                                 && shhwtstamps->syststamp.tv64)
1797                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1798                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1799                                 && shhwtstamps->hwtstamp.tv64)
1800                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1801                 else if (skb->tstamp.tv64)
1802                         ts = ktime_to_timespec(skb->tstamp);
1803                 else
1804                         getnstimeofday(&ts);
1805                 h.h2->tp_sec = ts.tv_sec;
1806                 h.h2->tp_nsec = ts.tv_nsec;
1807                 if (vlan_tx_tag_present(skb)) {
1808                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1809                         status |= TP_STATUS_VLAN_VALID;
1810                 } else {
1811                         h.h2->tp_vlan_tci = 0;
1812                 }
1813                 h.h2->tp_padding = 0;
1814                 hdrlen = sizeof(*h.h2);
1815                 break;
1816         case TPACKET_V3:
1817                 /* tp_nxt_offset,vlan are already populated above.
1818                  * So DONT clear those fields here
1819                  */
1820                 h.h3->tp_status |= status;
1821                 h.h3->tp_len = skb->len;
1822                 h.h3->tp_snaplen = snaplen;
1823                 h.h3->tp_mac = macoff;
1824                 h.h3->tp_net = netoff;
1825                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1826                                 && shhwtstamps->syststamp.tv64)
1827                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1828                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1829                                 && shhwtstamps->hwtstamp.tv64)
1830                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1831                 else if (skb->tstamp.tv64)
1832                         ts = ktime_to_timespec(skb->tstamp);
1833                 else
1834                         getnstimeofday(&ts);
1835                 h.h3->tp_sec  = ts.tv_sec;
1836                 h.h3->tp_nsec = ts.tv_nsec;
1837                 hdrlen = sizeof(*h.h3);
1838                 break;
1839         default:
1840                 BUG();
1841         }
1842
1843         sll = h.raw + TPACKET_ALIGN(hdrlen);
1844         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1845         sll->sll_family = AF_PACKET;
1846         sll->sll_hatype = dev->type;
1847         sll->sll_protocol = skb->protocol;
1848         sll->sll_pkttype = skb->pkt_type;
1849         if (unlikely(po->origdev))
1850                 sll->sll_ifindex = orig_dev->ifindex;
1851         else
1852                 sll->sll_ifindex = dev->ifindex;
1853
1854         smp_mb();
1855 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1856         {
1857                 u8 *start, *end;
1858
1859                 if (po->tp_version <= TPACKET_V2) {
1860                         end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1861                                 + macoff + snaplen);
1862                         for (start = h.raw; start < end; start += PAGE_SIZE)
1863                                 flush_dcache_page(pgv_to_page(start));
1864                 }
1865                 smp_wmb();
1866         }
1867 #endif
1868         if (po->tp_version <= TPACKET_V2)
1869                 __packet_set_status(po, h.raw, status);
1870         else
1871                 prb_clear_blk_fill_status(&po->rx_ring);
1872
1873         sk->sk_data_ready(sk, 0);
1874
1875 drop_n_restore:
1876         if (skb_head != skb->data && skb_shared(skb)) {
1877                 skb->data = skb_head;
1878                 skb->len = skb_len;
1879         }
1880 drop:
1881         kfree_skb(skb);
1882         return 0;
1883
1884 ring_is_full:
1885         po->stats.tp_drops++;
1886         spin_unlock(&sk->sk_receive_queue.lock);
1887
1888         sk->sk_data_ready(sk, 0);
1889         kfree_skb(copy_skb);
1890         goto drop_n_restore;
1891 }
1892
1893 static void tpacket_destruct_skb(struct sk_buff *skb)
1894 {
1895         struct packet_sock *po = pkt_sk(skb->sk);
1896         void *ph;
1897
1898         if (likely(po->tx_ring.pg_vec)) {
1899                 ph = skb_shinfo(skb)->destructor_arg;
1900                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1901                 atomic_dec(&po->tx_ring.pending);
1902                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1903         }
1904
1905         sock_wfree(skb);
1906 }
1907
1908 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1909                 void *frame, struct net_device *dev, int size_max,
1910                 __be16 proto, unsigned char *addr, int hlen)
1911 {
1912         union {
1913                 struct tpacket_hdr *h1;
1914                 struct tpacket2_hdr *h2;
1915                 void *raw;
1916         } ph;
1917         int to_write, offset, len, tp_len, nr_frags, len_max;
1918         struct socket *sock = po->sk.sk_socket;
1919         struct page *page;
1920         void *data;
1921         int err;
1922
1923         ph.raw = frame;
1924
1925         skb->protocol = proto;
1926         skb->dev = dev;
1927         skb->priority = po->sk.sk_priority;
1928         skb->mark = po->sk.sk_mark;
1929         skb_shinfo(skb)->destructor_arg = ph.raw;
1930
1931         switch (po->tp_version) {
1932         case TPACKET_V2:
1933                 tp_len = ph.h2->tp_len;
1934                 break;
1935         default:
1936                 tp_len = ph.h1->tp_len;
1937                 break;
1938         }
1939         if (unlikely(tp_len > size_max)) {
1940                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1941                 return -EMSGSIZE;
1942         }
1943
1944         skb_reserve(skb, hlen);
1945         skb_reset_network_header(skb);
1946         skb_probe_transport_header(skb, 0);
1947
1948         if (po->tp_tx_has_off) {
1949                 int off_min, off_max, off;
1950                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
1951                 off_max = po->tx_ring.frame_size - tp_len;
1952                 if (sock->type == SOCK_DGRAM) {
1953                         switch (po->tp_version) {
1954                         case TPACKET_V2:
1955                                 off = ph.h2->tp_net;
1956                                 break;
1957                         default:
1958                                 off = ph.h1->tp_net;
1959                                 break;
1960                         }
1961                 } else {
1962                         switch (po->tp_version) {
1963                         case TPACKET_V2:
1964                                 off = ph.h2->tp_mac;
1965                                 break;
1966                         default:
1967                                 off = ph.h1->tp_mac;
1968                                 break;
1969                         }
1970                 }
1971                 if (unlikely((off < off_min) || (off_max < off)))
1972                         return -EINVAL;
1973                 data = ph.raw + off;
1974         } else {
1975                 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1976         }
1977         to_write = tp_len;
1978
1979         if (sock->type == SOCK_DGRAM) {
1980                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
1981                                 NULL, tp_len);
1982                 if (unlikely(err < 0))
1983                         return -EINVAL;
1984         } else if (dev->hard_header_len) {
1985                 /* net device doesn't like empty head */
1986                 if (unlikely(tp_len <= dev->hard_header_len)) {
1987                         pr_err("packet size is too short (%d < %d)\n",
1988                                tp_len, dev->hard_header_len);
1989                         return -EINVAL;
1990                 }
1991
1992                 skb_push(skb, dev->hard_header_len);
1993                 err = skb_store_bits(skb, 0, data,
1994                                 dev->hard_header_len);
1995                 if (unlikely(err))
1996                         return err;
1997
1998                 data += dev->hard_header_len;
1999                 to_write -= dev->hard_header_len;
2000         }
2001
2002         offset = offset_in_page(data);
2003         len_max = PAGE_SIZE - offset;
2004         len = ((to_write > len_max) ? len_max : to_write);
2005
2006         skb->data_len = to_write;
2007         skb->len += to_write;
2008         skb->truesize += to_write;
2009         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2010
2011         while (likely(to_write)) {
2012                 nr_frags = skb_shinfo(skb)->nr_frags;
2013
2014                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2015                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2016                                MAX_SKB_FRAGS);
2017                         return -EFAULT;
2018                 }
2019
2020                 page = pgv_to_page(data);
2021                 data += len;
2022                 flush_dcache_page(page);
2023                 get_page(page);
2024                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2025                 to_write -= len;
2026                 offset = 0;
2027                 len_max = PAGE_SIZE;
2028                 len = ((to_write > len_max) ? len_max : to_write);
2029         }
2030
2031         return tp_len;
2032 }
2033
2034 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2035 {
2036         struct sk_buff *skb;
2037         struct net_device *dev;
2038         __be16 proto;
2039         bool need_rls_dev = false;
2040         int err, reserve = 0;
2041         void *ph;
2042         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2043         int tp_len, size_max;
2044         unsigned char *addr;
2045         int len_sum = 0;
2046         int status = TP_STATUS_AVAILABLE;
2047         int hlen, tlen;
2048
2049         mutex_lock(&po->pg_vec_lock);
2050
2051         if (saddr == NULL) {
2052                 dev = po->prot_hook.dev;
2053                 proto   = po->num;
2054                 addr    = NULL;
2055         } else {
2056                 err = -EINVAL;
2057                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2058                         goto out;
2059                 if (msg->msg_namelen < (saddr->sll_halen
2060                                         + offsetof(struct sockaddr_ll,
2061                                                 sll_addr)))
2062                         goto out;
2063                 proto   = saddr->sll_protocol;
2064                 addr    = saddr->sll_addr;
2065                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2066                 need_rls_dev = true;
2067         }
2068
2069         err = -ENXIO;
2070         if (unlikely(dev == NULL))
2071                 goto out;
2072
2073         reserve = dev->hard_header_len;
2074
2075         err = -ENETDOWN;
2076         if (unlikely(!(dev->flags & IFF_UP)))
2077                 goto out_put;
2078
2079         size_max = po->tx_ring.frame_size
2080                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2081
2082         if (size_max > dev->mtu + reserve)
2083                 size_max = dev->mtu + reserve;
2084
2085         do {
2086                 ph = packet_current_frame(po, &po->tx_ring,
2087                                 TP_STATUS_SEND_REQUEST);
2088
2089                 if (unlikely(ph == NULL)) {
2090                         schedule();
2091                         continue;
2092                 }
2093
2094                 status = TP_STATUS_SEND_REQUEST;
2095                 hlen = LL_RESERVED_SPACE(dev);
2096                 tlen = dev->needed_tailroom;
2097                 skb = sock_alloc_send_skb(&po->sk,
2098                                 hlen + tlen + sizeof(struct sockaddr_ll),
2099                                 0, &err);
2100
2101                 if (unlikely(skb == NULL))
2102                         goto out_status;
2103
2104                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2105                                 addr, hlen);
2106
2107                 if (unlikely(tp_len < 0)) {
2108                         if (po->tp_loss) {
2109                                 __packet_set_status(po, ph,
2110                                                 TP_STATUS_AVAILABLE);
2111                                 packet_increment_head(&po->tx_ring);
2112                                 kfree_skb(skb);
2113                                 continue;
2114                         } else {
2115                                 status = TP_STATUS_WRONG_FORMAT;
2116                                 err = tp_len;
2117                                 goto out_status;
2118                         }
2119                 }
2120
2121                 skb->destructor = tpacket_destruct_skb;
2122                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2123                 atomic_inc(&po->tx_ring.pending);
2124
2125                 status = TP_STATUS_SEND_REQUEST;
2126                 err = dev_queue_xmit(skb);
2127                 if (unlikely(err > 0)) {
2128                         err = net_xmit_errno(err);
2129                         if (err && __packet_get_status(po, ph) ==
2130                                    TP_STATUS_AVAILABLE) {
2131                                 /* skb was destructed already */
2132                                 skb = NULL;
2133                                 goto out_status;
2134                         }
2135                         /*
2136                          * skb was dropped but not destructed yet;
2137                          * let's treat it like congestion or err < 0
2138                          */
2139                         err = 0;
2140                 }
2141                 packet_increment_head(&po->tx_ring);
2142                 len_sum += tp_len;
2143         } while (likely((ph != NULL) ||
2144                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2145                          (atomic_read(&po->tx_ring.pending))))
2146                 );
2147
2148         err = len_sum;
2149         goto out_put;
2150
2151 out_status:
2152         __packet_set_status(po, ph, status);
2153         kfree_skb(skb);
2154 out_put:
2155         if (need_rls_dev)
2156                 dev_put(dev);
2157 out:
2158         mutex_unlock(&po->pg_vec_lock);
2159         return err;
2160 }
2161
2162 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2163                                         size_t reserve, size_t len,
2164                                         size_t linear, int noblock,
2165                                         int *err)
2166 {
2167         struct sk_buff *skb;
2168
2169         /* Under a page?  Don't bother with paged skb. */
2170         if (prepad + len < PAGE_SIZE || !linear)
2171                 linear = len;
2172
2173         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2174                                    err);
2175         if (!skb)
2176                 return NULL;
2177
2178         skb_reserve(skb, reserve);
2179         skb_put(skb, linear);
2180         skb->data_len = len - linear;
2181         skb->len += len - linear;
2182
2183         return skb;
2184 }
2185
2186 static int packet_snd(struct socket *sock,
2187                           struct msghdr *msg, size_t len)
2188 {
2189         struct sock *sk = sock->sk;
2190         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2191         struct sk_buff *skb;
2192         struct net_device *dev;
2193         __be16 proto;
2194         bool need_rls_dev = false;
2195         unsigned char *addr;
2196         int err, reserve = 0;
2197         struct virtio_net_hdr vnet_hdr = { 0 };
2198         int offset = 0;
2199         int vnet_hdr_len;
2200         struct packet_sock *po = pkt_sk(sk);
2201         unsigned short gso_type = 0;
2202         int hlen, tlen;
2203         int extra_len = 0;
2204
2205         /*
2206          *      Get and verify the address.
2207          */
2208
2209         if (saddr == NULL) {
2210                 dev = po->prot_hook.dev;
2211                 proto   = po->num;
2212                 addr    = NULL;
2213         } else {
2214                 err = -EINVAL;
2215                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2216                         goto out;
2217                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2218                         goto out;
2219                 proto   = saddr->sll_protocol;
2220                 addr    = saddr->sll_addr;
2221                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2222                 need_rls_dev = true;
2223         }
2224
2225         err = -ENXIO;
2226         if (dev == NULL)
2227                 goto out_unlock;
2228         if (sock->type == SOCK_RAW)
2229                 reserve = dev->hard_header_len;
2230
2231         err = -ENETDOWN;
2232         if (!(dev->flags & IFF_UP))
2233                 goto out_unlock;
2234
2235         if (po->has_vnet_hdr) {
2236                 vnet_hdr_len = sizeof(vnet_hdr);
2237
2238                 err = -EINVAL;
2239                 if (len < vnet_hdr_len)
2240                         goto out_unlock;
2241
2242                 len -= vnet_hdr_len;
2243
2244                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2245                                        vnet_hdr_len);
2246                 if (err < 0)
2247                         goto out_unlock;
2248
2249                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2250                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2251                       vnet_hdr.hdr_len))
2252                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
2253                                                  vnet_hdr.csum_offset + 2;
2254
2255                 err = -EINVAL;
2256                 if (vnet_hdr.hdr_len > len)
2257                         goto out_unlock;
2258
2259                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2260                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2261                         case VIRTIO_NET_HDR_GSO_TCPV4:
2262                                 gso_type = SKB_GSO_TCPV4;
2263                                 break;
2264                         case VIRTIO_NET_HDR_GSO_TCPV6:
2265                                 gso_type = SKB_GSO_TCPV6;
2266                                 break;
2267                         case VIRTIO_NET_HDR_GSO_UDP:
2268                                 gso_type = SKB_GSO_UDP;
2269                                 break;
2270                         default:
2271                                 goto out_unlock;
2272                         }
2273
2274                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2275                                 gso_type |= SKB_GSO_TCP_ECN;
2276
2277                         if (vnet_hdr.gso_size == 0)
2278                                 goto out_unlock;
2279
2280                 }
2281         }
2282
2283         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2284                 if (!netif_supports_nofcs(dev)) {
2285                         err = -EPROTONOSUPPORT;
2286                         goto out_unlock;
2287                 }
2288                 extra_len = 4; /* We're doing our own CRC */
2289         }
2290
2291         err = -EMSGSIZE;
2292         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2293                 goto out_unlock;
2294
2295         err = -ENOBUFS;
2296         hlen = LL_RESERVED_SPACE(dev);
2297         tlen = dev->needed_tailroom;
2298         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2299                                msg->msg_flags & MSG_DONTWAIT, &err);
2300         if (skb == NULL)
2301                 goto out_unlock;
2302
2303         skb_set_network_header(skb, reserve);
2304
2305         err = -EINVAL;
2306         if (sock->type == SOCK_DGRAM &&
2307             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2308                 goto out_free;
2309
2310         /* Returns -EFAULT on error */
2311         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2312         if (err)
2313                 goto out_free;
2314
2315         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2316
2317         if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2318                 /* Earlier code assumed this would be a VLAN pkt,
2319                  * double-check this now that we have the actual
2320                  * packet in hand.
2321                  */
2322                 struct ethhdr *ehdr;
2323                 skb_reset_mac_header(skb);
2324                 ehdr = eth_hdr(skb);
2325                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2326                         err = -EMSGSIZE;
2327                         goto out_free;
2328                 }
2329         }
2330
2331         skb->protocol = proto;
2332         skb->dev = dev;
2333         skb->priority = sk->sk_priority;
2334         skb->mark = sk->sk_mark;
2335
2336         if (po->has_vnet_hdr) {
2337                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2338                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2339                                                   vnet_hdr.csum_offset)) {
2340                                 err = -EINVAL;
2341                                 goto out_free;
2342                         }
2343                 }
2344
2345                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2346                 skb_shinfo(skb)->gso_type = gso_type;
2347
2348                 /* Header must be checked, and gso_segs computed. */
2349                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2350                 skb_shinfo(skb)->gso_segs = 0;
2351
2352                 len += vnet_hdr_len;
2353         }
2354
2355         skb_probe_transport_header(skb, reserve);
2356
2357         if (unlikely(extra_len == 4))
2358                 skb->no_fcs = 1;
2359
2360         /*
2361          *      Now send it
2362          */
2363
2364         err = dev_queue_xmit(skb);
2365         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2366                 goto out_unlock;
2367
2368         if (need_rls_dev)
2369                 dev_put(dev);
2370
2371         return len;
2372
2373 out_free:
2374         kfree_skb(skb);
2375 out_unlock:
2376         if (dev && need_rls_dev)
2377                 dev_put(dev);
2378 out:
2379         return err;
2380 }
2381
2382 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2383                 struct msghdr *msg, size_t len)
2384 {
2385         struct sock *sk = sock->sk;
2386         struct packet_sock *po = pkt_sk(sk);
2387         if (po->tx_ring.pg_vec)
2388                 return tpacket_snd(po, msg);
2389         else
2390                 return packet_snd(sock, msg, len);
2391 }
2392
2393 /*
2394  *      Close a PACKET socket. This is fairly simple. We immediately go
2395  *      to 'closed' state and remove our protocol entry in the device list.
2396  */
2397
2398 static int packet_release(struct socket *sock)
2399 {
2400         struct sock *sk = sock->sk;
2401         struct packet_sock *po;
2402         struct net *net;
2403         union tpacket_req_u req_u;
2404
2405         if (!sk)
2406                 return 0;
2407
2408         net = sock_net(sk);
2409         po = pkt_sk(sk);
2410
2411         mutex_lock(&net->packet.sklist_lock);
2412         sk_del_node_init_rcu(sk);
2413         mutex_unlock(&net->packet.sklist_lock);
2414
2415         preempt_disable();
2416         sock_prot_inuse_add(net, sk->sk_prot, -1);
2417         preempt_enable();
2418
2419         spin_lock(&po->bind_lock);
2420         unregister_prot_hook(sk, false);
2421         if (po->prot_hook.dev) {
2422                 dev_put(po->prot_hook.dev);
2423                 po->prot_hook.dev = NULL;
2424         }
2425         spin_unlock(&po->bind_lock);
2426
2427         packet_flush_mclist(sk);
2428
2429         if (po->rx_ring.pg_vec) {
2430                 memset(&req_u, 0, sizeof(req_u));
2431                 packet_set_ring(sk, &req_u, 1, 0);
2432         }
2433
2434         if (po->tx_ring.pg_vec) {
2435                 memset(&req_u, 0, sizeof(req_u));
2436                 packet_set_ring(sk, &req_u, 1, 1);
2437         }
2438
2439         fanout_release(sk);
2440
2441         synchronize_net();
2442         /*
2443          *      Now the socket is dead. No more input will appear.
2444          */
2445         sock_orphan(sk);
2446         sock->sk = NULL;
2447
2448         /* Purge queues */
2449
2450         skb_queue_purge(&sk->sk_receive_queue);
2451         sk_refcnt_debug_release(sk);
2452
2453         sock_put(sk);
2454         return 0;
2455 }
2456
2457 /*
2458  *      Attach a packet hook.
2459  */
2460
2461 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2462 {
2463         struct packet_sock *po = pkt_sk(sk);
2464
2465         if (po->fanout) {
2466                 if (dev)
2467                         dev_put(dev);
2468
2469                 return -EINVAL;
2470         }
2471
2472         lock_sock(sk);
2473
2474         spin_lock(&po->bind_lock);
2475         unregister_prot_hook(sk, true);
2476         po->num = protocol;
2477         po->prot_hook.type = protocol;
2478         if (po->prot_hook.dev)
2479                 dev_put(po->prot_hook.dev);
2480         po->prot_hook.dev = dev;
2481
2482         po->ifindex = dev ? dev->ifindex : 0;
2483
2484         if (protocol == 0)
2485                 goto out_unlock;
2486
2487         if (!dev || (dev->flags & IFF_UP)) {
2488                 register_prot_hook(sk);
2489         } else {
2490                 sk->sk_err = ENETDOWN;
2491                 if (!sock_flag(sk, SOCK_DEAD))
2492                         sk->sk_error_report(sk);
2493         }
2494
2495 out_unlock:
2496         spin_unlock(&po->bind_lock);
2497         release_sock(sk);
2498         return 0;
2499 }
2500
2501 /*
2502  *      Bind a packet socket to a device
2503  */
2504
2505 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2506                             int addr_len)
2507 {
2508         struct sock *sk = sock->sk;
2509         char name[15];
2510         struct net_device *dev;
2511         int err = -ENODEV;
2512
2513         /*
2514          *      Check legality
2515          */
2516
2517         if (addr_len != sizeof(struct sockaddr))
2518                 return -EINVAL;
2519         strlcpy(name, uaddr->sa_data, sizeof(name));
2520
2521         dev = dev_get_by_name(sock_net(sk), name);
2522         if (dev)
2523                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2524         return err;
2525 }
2526
2527 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2528 {
2529         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2530         struct sock *sk = sock->sk;
2531         struct net_device *dev = NULL;
2532         int err;
2533
2534
2535         /*
2536          *      Check legality
2537          */
2538
2539         if (addr_len < sizeof(struct sockaddr_ll))
2540                 return -EINVAL;
2541         if (sll->sll_family != AF_PACKET)
2542                 return -EINVAL;
2543
2544         if (sll->sll_ifindex) {
2545                 err = -ENODEV;
2546                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2547                 if (dev == NULL)
2548                         goto out;
2549         }
2550         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2551
2552 out:
2553         return err;
2554 }
2555
2556 static struct proto packet_proto = {
2557         .name     = "PACKET",
2558         .owner    = THIS_MODULE,
2559         .obj_size = sizeof(struct packet_sock),
2560 };
2561
2562 /*
2563  *      Create a packet of type SOCK_PACKET.
2564  */
2565
2566 static int packet_create(struct net *net, struct socket *sock, int protocol,
2567                          int kern)
2568 {
2569         struct sock *sk;
2570         struct packet_sock *po;
2571         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2572         int err;
2573
2574         if (!ns_capable(net->user_ns, CAP_NET_RAW))
2575                 return -EPERM;
2576         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2577             sock->type != SOCK_PACKET)
2578                 return -ESOCKTNOSUPPORT;
2579
2580         sock->state = SS_UNCONNECTED;
2581
2582         err = -ENOBUFS;
2583         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2584         if (sk == NULL)
2585                 goto out;
2586
2587         sock->ops = &packet_ops;
2588         if (sock->type == SOCK_PACKET)
2589                 sock->ops = &packet_ops_spkt;
2590
2591         sock_init_data(sock, sk);
2592
2593         po = pkt_sk(sk);
2594         sk->sk_family = PF_PACKET;
2595         po->num = proto;
2596
2597         sk->sk_destruct = packet_sock_destruct;
2598         sk_refcnt_debug_inc(sk);
2599
2600         /*
2601          *      Attach a protocol block
2602          */
2603
2604         spin_lock_init(&po->bind_lock);
2605         mutex_init(&po->pg_vec_lock);
2606         po->prot_hook.func = packet_rcv;
2607
2608         if (sock->type == SOCK_PACKET)
2609                 po->prot_hook.func = packet_rcv_spkt;
2610
2611         po->prot_hook.af_packet_priv = sk;
2612
2613         if (proto) {
2614                 po->prot_hook.type = proto;
2615                 register_prot_hook(sk);
2616         }
2617
2618         mutex_lock(&net->packet.sklist_lock);
2619         sk_add_node_rcu(sk, &net->packet.sklist);
2620         mutex_unlock(&net->packet.sklist_lock);
2621
2622         preempt_disable();
2623         sock_prot_inuse_add(net, &packet_proto, 1);
2624         preempt_enable();
2625
2626         return 0;
2627 out:
2628         return err;
2629 }
2630
2631 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2632 {
2633         struct sock_exterr_skb *serr;
2634         struct sk_buff *skb, *skb2;
2635         int copied, err;
2636
2637         err = -EAGAIN;
2638         skb = skb_dequeue(&sk->sk_error_queue);
2639         if (skb == NULL)
2640                 goto out;
2641
2642         copied = skb->len;
2643         if (copied > len) {
2644                 msg->msg_flags |= MSG_TRUNC;
2645                 copied = len;
2646         }
2647         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2648         if (err)
2649                 goto out_free_skb;
2650
2651         sock_recv_timestamp(msg, sk, skb);
2652
2653         serr = SKB_EXT_ERR(skb);
2654         put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2655                  sizeof(serr->ee), &serr->ee);
2656
2657         msg->msg_flags |= MSG_ERRQUEUE;
2658         err = copied;
2659
2660         /* Reset and regenerate socket error */
2661         spin_lock_bh(&sk->sk_error_queue.lock);
2662         sk->sk_err = 0;
2663         if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2664                 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2665                 spin_unlock_bh(&sk->sk_error_queue.lock);
2666                 sk->sk_error_report(sk);
2667         } else
2668                 spin_unlock_bh(&sk->sk_error_queue.lock);
2669
2670 out_free_skb:
2671         kfree_skb(skb);
2672 out:
2673         return err;
2674 }
2675
2676 /*
2677  *      Pull a packet from our receive queue and hand it to the user.
2678  *      If necessary we block.
2679  */
2680
2681 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2682                           struct msghdr *msg, size_t len, int flags)
2683 {
2684         struct sock *sk = sock->sk;
2685         struct sk_buff *skb;
2686         int copied, err;
2687         struct sockaddr_ll *sll;
2688         int vnet_hdr_len = 0;
2689
2690         err = -EINVAL;
2691         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2692                 goto out;
2693
2694 #if 0
2695         /* What error should we return now? EUNATTACH? */
2696         if (pkt_sk(sk)->ifindex < 0)
2697                 return -ENODEV;
2698 #endif
2699
2700         if (flags & MSG_ERRQUEUE) {
2701                 err = packet_recv_error(sk, msg, len);
2702                 goto out;
2703         }
2704
2705         /*
2706          *      Call the generic datagram receiver. This handles all sorts
2707          *      of horrible races and re-entrancy so we can forget about it
2708          *      in the protocol layers.
2709          *
2710          *      Now it will return ENETDOWN, if device have just gone down,
2711          *      but then it will block.
2712          */
2713
2714         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2715
2716         /*
2717          *      An error occurred so return it. Because skb_recv_datagram()
2718          *      handles the blocking we don't see and worry about blocking
2719          *      retries.
2720          */
2721
2722         if (skb == NULL)
2723                 goto out;
2724
2725         if (pkt_sk(sk)->has_vnet_hdr) {
2726                 struct virtio_net_hdr vnet_hdr = { 0 };
2727
2728                 err = -EINVAL;
2729                 vnet_hdr_len = sizeof(vnet_hdr);
2730                 if (len < vnet_hdr_len)
2731                         goto out_free;
2732
2733                 len -= vnet_hdr_len;
2734
2735                 if (skb_is_gso(skb)) {
2736                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2737
2738                         /* This is a hint as to how much should be linear. */
2739                         vnet_hdr.hdr_len = skb_headlen(skb);
2740                         vnet_hdr.gso_size = sinfo->gso_size;
2741                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2742                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2743                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2744                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2745                         else if (sinfo->gso_type & SKB_GSO_UDP)
2746                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2747                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2748                                 goto out_free;
2749                         else
2750                                 BUG();
2751                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2752                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2753                 } else
2754                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2755
2756                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2757                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2758                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2759                         vnet_hdr.csum_offset = skb->csum_offset;
2760                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2761                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2762                 } /* else everything is zero */
2763
2764                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2765                                      vnet_hdr_len);
2766                 if (err < 0)
2767                         goto out_free;
2768         }
2769
2770         /*
2771          *      If the address length field is there to be filled in, we fill
2772          *      it in now.
2773          */
2774
2775         sll = &PACKET_SKB_CB(skb)->sa.ll;
2776         if (sock->type == SOCK_PACKET)
2777                 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2778         else
2779                 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2780
2781         /*
2782          *      You lose any data beyond the buffer you gave. If it worries a
2783          *      user program they can ask the device for its MTU anyway.
2784          */
2785
2786         copied = skb->len;
2787         if (copied > len) {
2788                 copied = len;
2789                 msg->msg_flags |= MSG_TRUNC;
2790         }
2791
2792         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2793         if (err)
2794                 goto out_free;
2795
2796         sock_recv_ts_and_drops(msg, sk, skb);
2797
2798         if (msg->msg_name)
2799                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2800                        msg->msg_namelen);
2801
2802         if (pkt_sk(sk)->auxdata) {
2803                 struct tpacket_auxdata aux;
2804
2805                 aux.tp_status = TP_STATUS_USER;
2806                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2807                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2808                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2809                 aux.tp_snaplen = skb->len;
2810                 aux.tp_mac = 0;
2811                 aux.tp_net = skb_network_offset(skb);
2812                 if (vlan_tx_tag_present(skb)) {
2813                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2814                         aux.tp_status |= TP_STATUS_VLAN_VALID;
2815                 } else {
2816                         aux.tp_vlan_tci = 0;
2817                 }
2818                 aux.tp_padding = 0;
2819                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2820         }
2821
2822         /*
2823          *      Free or return the buffer as appropriate. Again this
2824          *      hides all the races and re-entrancy issues from us.
2825          */
2826         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2827
2828 out_free:
2829         skb_free_datagram(sk, skb);
2830 out:
2831         return err;
2832 }
2833
2834 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2835                                int *uaddr_len, int peer)
2836 {
2837         struct net_device *dev;
2838         struct sock *sk = sock->sk;
2839
2840         if (peer)
2841                 return -EOPNOTSUPP;
2842
2843         uaddr->sa_family = AF_PACKET;
2844         rcu_read_lock();
2845         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2846         if (dev)
2847                 strncpy(uaddr->sa_data, dev->name, 14);
2848         else
2849                 memset(uaddr->sa_data, 0, 14);
2850         rcu_read_unlock();
2851         *uaddr_len = sizeof(*uaddr);
2852
2853         return 0;
2854 }
2855
2856 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2857                           int *uaddr_len, int peer)
2858 {
2859         struct net_device *dev;
2860         struct sock *sk = sock->sk;
2861         struct packet_sock *po = pkt_sk(sk);
2862         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2863
2864         if (peer)
2865                 return -EOPNOTSUPP;
2866
2867         sll->sll_family = AF_PACKET;
2868         sll->sll_ifindex = po->ifindex;
2869         sll->sll_protocol = po->num;
2870         sll->sll_pkttype = 0;
2871         rcu_read_lock();
2872         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2873         if (dev) {
2874                 sll->sll_hatype = dev->type;
2875                 sll->sll_halen = dev->addr_len;
2876                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2877         } else {
2878                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
2879                 sll->sll_halen = 0;
2880         }
2881         rcu_read_unlock();
2882         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2883
2884         return 0;
2885 }
2886
2887 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2888                          int what)
2889 {
2890         switch (i->type) {
2891         case PACKET_MR_MULTICAST:
2892                 if (i->alen != dev->addr_len)
2893                         return -EINVAL;
2894                 if (what > 0)
2895                         return dev_mc_add(dev, i->addr);
2896                 else
2897                         return dev_mc_del(dev, i->addr);
2898                 break;
2899         case PACKET_MR_PROMISC:
2900                 return dev_set_promiscuity(dev, what);
2901                 break;
2902         case PACKET_MR_ALLMULTI:
2903                 return dev_set_allmulti(dev, what);
2904                 break;
2905         case PACKET_MR_UNICAST:
2906                 if (i->alen != dev->addr_len)
2907                         return -EINVAL;
2908                 if (what > 0)
2909                         return dev_uc_add(dev, i->addr);
2910                 else
2911                         return dev_uc_del(dev, i->addr);
2912                 break;
2913         default:
2914                 break;
2915         }
2916         return 0;
2917 }
2918
2919 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2920 {
2921         for ( ; i; i = i->next) {
2922                 if (i->ifindex == dev->ifindex)
2923                         packet_dev_mc(dev, i, what);
2924         }
2925 }
2926
2927 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2928 {
2929         struct packet_sock *po = pkt_sk(sk);
2930         struct packet_mclist *ml, *i;
2931         struct net_device *dev;
2932         int err;
2933
2934         rtnl_lock();
2935
2936         err = -ENODEV;
2937         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2938         if (!dev)
2939                 goto done;
2940
2941         err = -EINVAL;
2942         if (mreq->mr_alen > dev->addr_len)
2943                 goto done;
2944
2945         err = -ENOBUFS;
2946         i = kmalloc(sizeof(*i), GFP_KERNEL);
2947         if (i == NULL)
2948                 goto done;
2949
2950         err = 0;
2951         for (ml = po->mclist; ml; ml = ml->next) {
2952                 if (ml->ifindex == mreq->mr_ifindex &&
2953                     ml->type == mreq->mr_type &&
2954                     ml->alen == mreq->mr_alen &&
2955                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2956                         ml->count++;
2957                         /* Free the new element ... */
2958                         kfree(i);
2959                         goto done;
2960                 }
2961         }
2962
2963         i->type = mreq->mr_type;
2964         i->ifindex = mreq->mr_ifindex;
2965         i->alen = mreq->mr_alen;
2966         memcpy(i->addr, mreq->mr_address, i->alen);
2967         i->count = 1;
2968         i->next = po->mclist;
2969         po->mclist = i;
2970         err = packet_dev_mc(dev, i, 1);
2971         if (err) {
2972                 po->mclist = i->next;
2973                 kfree(i);
2974         }
2975
2976 done:
2977         rtnl_unlock();
2978         return err;
2979 }
2980
2981 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2982 {
2983         struct packet_mclist *ml, **mlp;
2984
2985         rtnl_lock();
2986
2987         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
2988                 if (ml->ifindex == mreq->mr_ifindex &&
2989                     ml->type == mreq->mr_type &&
2990                     ml->alen == mreq->mr_alen &&
2991                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2992                         if (--ml->count == 0) {
2993                                 struct net_device *dev;
2994                                 *mlp = ml->next;
2995                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2996                                 if (dev)
2997                                         packet_dev_mc(dev, ml, -1);
2998                                 kfree(ml);
2999                         }
3000                         rtnl_unlock();
3001                         return 0;
3002                 }
3003         }
3004         rtnl_unlock();
3005         return -EADDRNOTAVAIL;
3006 }
3007
3008 static void packet_flush_mclist(struct sock *sk)
3009 {
3010         struct packet_sock *po = pkt_sk(sk);
3011         struct packet_mclist *ml;
3012
3013         if (!po->mclist)
3014                 return;
3015
3016         rtnl_lock();
3017         while ((ml = po->mclist) != NULL) {
3018                 struct net_device *dev;
3019
3020                 po->mclist = ml->next;
3021                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3022                 if (dev != NULL)
3023                         packet_dev_mc(dev, ml, -1);
3024                 kfree(ml);
3025         }
3026         rtnl_unlock();
3027 }
3028
3029 static int
3030 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3031 {
3032         struct sock *sk = sock->sk;
3033         struct packet_sock *po = pkt_sk(sk);
3034         int ret;
3035
3036         if (level != SOL_PACKET)
3037                 return -ENOPROTOOPT;
3038
3039         switch (optname) {
3040         case PACKET_ADD_MEMBERSHIP:
3041         case PACKET_DROP_MEMBERSHIP:
3042         {
3043                 struct packet_mreq_max mreq;
3044                 int len = optlen;
3045                 memset(&mreq, 0, sizeof(mreq));
3046                 if (len < sizeof(struct packet_mreq))
3047                         return -EINVAL;
3048                 if (len > sizeof(mreq))
3049                         len = sizeof(mreq);
3050                 if (copy_from_user(&mreq, optval, len))
3051                         return -EFAULT;
3052                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3053                         return -EINVAL;
3054                 if (optname == PACKET_ADD_MEMBERSHIP)
3055                         ret = packet_mc_add(sk, &mreq);
3056                 else
3057                         ret = packet_mc_drop(sk, &mreq);
3058                 return ret;
3059         }
3060
3061         case PACKET_RX_RING:
3062         case PACKET_TX_RING:
3063         {
3064                 union tpacket_req_u req_u;
3065                 int len;
3066
3067                 switch (po->tp_version) {
3068                 case TPACKET_V1:
3069                 case TPACKET_V2:
3070                         len = sizeof(req_u.req);
3071                         break;
3072                 case TPACKET_V3:
3073                 default:
3074                         len = sizeof(req_u.req3);
3075                         break;
3076                 }
3077                 if (optlen < len)
3078                         return -EINVAL;
3079                 if (pkt_sk(sk)->has_vnet_hdr)
3080                         return -EINVAL;
3081                 if (copy_from_user(&req_u.req, optval, len))
3082                         return -EFAULT;
3083                 return packet_set_ring(sk, &req_u, 0,
3084                         optname == PACKET_TX_RING);
3085         }
3086         case PACKET_COPY_THRESH:
3087         {
3088                 int val;
3089
3090                 if (optlen != sizeof(val))
3091                         return -EINVAL;
3092                 if (copy_from_user(&val, optval, sizeof(val)))
3093                         return -EFAULT;
3094
3095                 pkt_sk(sk)->copy_thresh = val;
3096                 return 0;
3097         }
3098         case PACKET_VERSION:
3099         {
3100                 int val;
3101
3102                 if (optlen != sizeof(val))
3103                         return -EINVAL;
3104                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3105                         return -EBUSY;
3106                 if (copy_from_user(&val, optval, sizeof(val)))
3107                         return -EFAULT;
3108                 switch (val) {
3109                 case TPACKET_V1:
3110                 case TPACKET_V2:
3111                 case TPACKET_V3:
3112                         po->tp_version = val;
3113                         return 0;
3114                 default:
3115                         return -EINVAL;
3116                 }
3117         }
3118         case PACKET_RESERVE:
3119         {
3120                 unsigned int val;
3121
3122                 if (optlen != sizeof(val))
3123                         return -EINVAL;
3124                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3125                         return -EBUSY;
3126                 if (copy_from_user(&val, optval, sizeof(val)))
3127                         return -EFAULT;
3128                 po->tp_reserve = val;
3129                 return 0;
3130         }
3131         case PACKET_LOSS:
3132         {
3133                 unsigned int val;
3134
3135                 if (optlen != sizeof(val))
3136                         return -EINVAL;
3137                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3138                         return -EBUSY;
3139                 if (copy_from_user(&val, optval, sizeof(val)))
3140                         return -EFAULT;
3141                 po->tp_loss = !!val;
3142                 return 0;
3143         }
3144         case PACKET_AUXDATA:
3145         {
3146                 int val;
3147
3148                 if (optlen < sizeof(val))
3149                         return -EINVAL;
3150                 if (copy_from_user(&val, optval, sizeof(val)))
3151                         return -EFAULT;
3152
3153                 po->auxdata = !!val;
3154                 return 0;
3155         }
3156         case PACKET_ORIGDEV:
3157         {
3158                 int val;
3159
3160                 if (optlen < sizeof(val))
3161                         return -EINVAL;
3162                 if (copy_from_user(&val, optval, sizeof(val)))
3163                         return -EFAULT;
3164
3165                 po->origdev = !!val;
3166                 return 0;
3167         }
3168         case PACKET_VNET_HDR:
3169         {
3170                 int val;
3171
3172                 if (sock->type != SOCK_RAW)
3173                         return -EINVAL;
3174                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3175                         return -EBUSY;
3176                 if (optlen < sizeof(val))
3177                         return -EINVAL;
3178                 if (copy_from_user(&val, optval, sizeof(val)))
3179                         return -EFAULT;
3180
3181                 po->has_vnet_hdr = !!val;
3182                 return 0;
3183         }
3184         case PACKET_TIMESTAMP:
3185         {
3186                 int val;
3187
3188                 if (optlen != sizeof(val))
3189                         return -EINVAL;
3190                 if (copy_from_user(&val, optval, sizeof(val)))
3191                         return -EFAULT;
3192
3193                 po->tp_tstamp = val;
3194                 return 0;
3195         }
3196         case PACKET_FANOUT:
3197         {
3198                 int val;
3199
3200                 if (optlen != sizeof(val))
3201                         return -EINVAL;
3202                 if (copy_from_user(&val, optval, sizeof(val)))
3203                         return -EFAULT;
3204
3205                 return fanout_add(sk, val & 0xffff, val >> 16);
3206         }
3207         case PACKET_TX_HAS_OFF:
3208         {
3209                 unsigned int val;
3210
3211                 if (optlen != sizeof(val))
3212                         return -EINVAL;
3213                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3214                         return -EBUSY;
3215                 if (copy_from_user(&val, optval, sizeof(val)))
3216                         return -EFAULT;
3217                 po->tp_tx_has_off = !!val;
3218                 return 0;
3219         }
3220         default:
3221                 return -ENOPROTOOPT;
3222         }
3223 }
3224
3225 static int packet_getsockopt(struct socket *sock, int level, int optname,
3226                              char __user *optval, int __user *optlen)
3227 {
3228         int len;
3229         int val, lv = sizeof(val);
3230         struct sock *sk = sock->sk;
3231         struct packet_sock *po = pkt_sk(sk);
3232         void *data = &val;
3233         struct tpacket_stats st;
3234         union tpacket_stats_u st_u;
3235
3236         if (level != SOL_PACKET)
3237                 return -ENOPROTOOPT;
3238
3239         if (get_user(len, optlen))
3240                 return -EFAULT;
3241
3242         if (len < 0)
3243                 return -EINVAL;
3244
3245         switch (optname) {
3246         case PACKET_STATISTICS:
3247                 spin_lock_bh(&sk->sk_receive_queue.lock);
3248                 if (po->tp_version == TPACKET_V3) {
3249                         lv = sizeof(struct tpacket_stats_v3);
3250                         memcpy(&st_u.stats3, &po->stats,
3251                                sizeof(struct tpacket_stats));
3252                         st_u.stats3.tp_freeze_q_cnt =
3253                                         po->stats_u.stats3.tp_freeze_q_cnt;
3254                         st_u.stats3.tp_packets += po->stats.tp_drops;
3255                         data = &st_u.stats3;
3256                 } else {
3257                         lv = sizeof(struct tpacket_stats);
3258                         st = po->stats;
3259                         st.tp_packets += st.tp_drops;
3260                         data = &st;
3261                 }
3262                 memset(&po->stats, 0, sizeof(st));
3263                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3264                 break;
3265         case PACKET_AUXDATA:
3266                 val = po->auxdata;
3267                 break;
3268         case PACKET_ORIGDEV:
3269                 val = po->origdev;
3270                 break;
3271         case PACKET_VNET_HDR:
3272                 val = po->has_vnet_hdr;
3273                 break;
3274         case PACKET_VERSION:
3275                 val = po->tp_version;
3276                 break;
3277         case PACKET_HDRLEN:
3278                 if (len > sizeof(int))
3279                         len = sizeof(int);
3280                 if (copy_from_user(&val, optval, len))
3281                         return -EFAULT;
3282                 switch (val) {
3283                 case TPACKET_V1:
3284                         val = sizeof(struct tpacket_hdr);
3285                         break;
3286                 case TPACKET_V2:
3287                         val = sizeof(struct tpacket2_hdr);
3288                         break;
3289                 case TPACKET_V3:
3290                         val = sizeof(struct tpacket3_hdr);
3291                         break;
3292                 default:
3293                         return -EINVAL;
3294                 }
3295                 break;
3296         case PACKET_RESERVE:
3297                 val = po->tp_reserve;
3298                 break;
3299         case PACKET_LOSS:
3300                 val = po->tp_loss;
3301                 break;
3302         case PACKET_TIMESTAMP:
3303                 val = po->tp_tstamp;
3304                 break;
3305         case PACKET_FANOUT:
3306                 val = (po->fanout ?
3307                        ((u32)po->fanout->id |
3308                         ((u32)po->fanout->type << 16) |
3309                         ((u32)po->fanout->flags << 24)) :
3310                        0);
3311                 break;
3312         case PACKET_TX_HAS_OFF:
3313                 val = po->tp_tx_has_off;
3314                 break;
3315         default:
3316                 return -ENOPROTOOPT;
3317         }
3318
3319         if (len > lv)
3320                 len = lv;
3321         if (put_user(len, optlen))
3322                 return -EFAULT;
3323         if (copy_to_user(optval, data, len))
3324                 return -EFAULT;
3325         return 0;
3326 }
3327
3328
3329 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3330 {
3331         struct sock *sk;
3332         struct net_device *dev = data;
3333         struct net *net = dev_net(dev);
3334
3335         rcu_read_lock();
3336         sk_for_each_rcu(sk, &net->packet.sklist) {
3337                 struct packet_sock *po = pkt_sk(sk);
3338
3339                 switch (msg) {
3340                 case NETDEV_UNREGISTER:
3341                         if (po->mclist)
3342                                 packet_dev_mclist(dev, po->mclist, -1);
3343                         /* fallthrough */
3344
3345                 case NETDEV_DOWN:
3346                         if (dev->ifindex == po->ifindex) {
3347                                 spin_lock(&po->bind_lock);
3348                                 if (po->running) {
3349                                         __unregister_prot_hook(sk, false);
3350                                         sk->sk_err = ENETDOWN;
3351                                         if (!sock_flag(sk, SOCK_DEAD))
3352                                                 sk->sk_error_report(sk);
3353                                 }
3354                                 if (msg == NETDEV_UNREGISTER) {
3355                                         po->ifindex = -1;
3356                                         if (po->prot_hook.dev)
3357                                                 dev_put(po->prot_hook.dev);
3358                                         po->prot_hook.dev = NULL;
3359                                 }
3360                                 spin_unlock(&po->bind_lock);
3361                         }
3362                         break;
3363                 case NETDEV_UP:
3364                         if (dev->ifindex == po->ifindex) {
3365                                 spin_lock(&po->bind_lock);
3366                                 if (po->num)
3367                                         register_prot_hook(sk);
3368                                 spin_unlock(&po->bind_lock);
3369                         }
3370                         break;
3371                 }
3372         }
3373         rcu_read_unlock();
3374         return NOTIFY_DONE;
3375 }
3376
3377
3378 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3379                         unsigned long arg)
3380 {
3381         struct sock *sk = sock->sk;
3382
3383         switch (cmd) {
3384         case SIOCOUTQ:
3385         {
3386                 int amount = sk_wmem_alloc_get(sk);
3387
3388                 return put_user(amount, (int __user *)arg);
3389         }
3390         case SIOCINQ:
3391         {
3392                 struct sk_buff *skb;
3393                 int amount = 0;
3394
3395                 spin_lock_bh(&sk->sk_receive_queue.lock);
3396                 skb = skb_peek(&sk->sk_receive_queue);
3397                 if (skb)
3398                         amount = skb->len;
3399                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3400                 return put_user(amount, (int __user *)arg);
3401         }
3402         case SIOCGSTAMP:
3403                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3404         case SIOCGSTAMPNS:
3405                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3406
3407 #ifdef CONFIG_INET
3408         case SIOCADDRT:
3409         case SIOCDELRT:
3410         case SIOCDARP:
3411         case SIOCGARP:
3412         case SIOCSARP:
3413         case SIOCGIFADDR:
3414         case SIOCSIFADDR:
3415         case SIOCGIFBRDADDR:
3416         case SIOCSIFBRDADDR:
3417         case SIOCGIFNETMASK:
3418         case SIOCSIFNETMASK:
3419         case SIOCGIFDSTADDR:
3420         case SIOCSIFDSTADDR:
3421         case SIOCSIFFLAGS:
3422                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3423 #endif
3424
3425         default:
3426                 return -ENOIOCTLCMD;
3427         }
3428         return 0;
3429 }
3430
3431 static unsigned int packet_poll(struct file *file, struct socket *sock,
3432                                 poll_table *wait)
3433 {
3434         struct sock *sk = sock->sk;
3435         struct packet_sock *po = pkt_sk(sk);
3436         unsigned int mask = datagram_poll(file, sock, wait);
3437
3438         spin_lock_bh(&sk->sk_receive_queue.lock);
3439         if (po->rx_ring.pg_vec) {
3440                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3441                         TP_STATUS_KERNEL))
3442                         mask |= POLLIN | POLLRDNORM;
3443         }
3444         spin_unlock_bh(&sk->sk_receive_queue.lock);
3445         spin_lock_bh(&sk->sk_write_queue.lock);
3446         if (po->tx_ring.pg_vec) {
3447                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3448                         mask |= POLLOUT | POLLWRNORM;
3449         }
3450         spin_unlock_bh(&sk->sk_write_queue.lock);
3451         return mask;
3452 }
3453
3454
3455 /* Dirty? Well, I still did not learn better way to account
3456  * for user mmaps.
3457  */
3458
3459 static void packet_mm_open(struct vm_area_struct *vma)
3460 {
3461         struct file *file = vma->vm_file;
3462         struct socket *sock = file->private_data;
3463         struct sock *sk = sock->sk;
3464
3465         if (sk)
3466                 atomic_inc(&pkt_sk(sk)->mapped);
3467 }
3468
3469 static void packet_mm_close(struct vm_area_struct *vma)
3470 {
3471         struct file *file = vma->vm_file;
3472         struct socket *sock = file->private_data;
3473         struct sock *sk = sock->sk;
3474
3475         if (sk)
3476                 atomic_dec(&pkt_sk(sk)->mapped);
3477 }
3478
3479 static const struct vm_operations_struct packet_mmap_ops = {
3480         .open   =       packet_mm_open,
3481         .close  =       packet_mm_close,
3482 };
3483
3484 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3485                         unsigned int len)
3486 {
3487         int i;
3488
3489         for (i = 0; i < len; i++) {
3490                 if (likely(pg_vec[i].buffer)) {
3491                         if (is_vmalloc_addr(pg_vec[i].buffer))
3492                                 vfree(pg_vec[i].buffer);
3493                         else
3494                                 free_pages((unsigned long)pg_vec[i].buffer,
3495                                            order);
3496                         pg_vec[i].buffer = NULL;
3497                 }
3498         }
3499         kfree(pg_vec);
3500 }
3501
3502 static char *alloc_one_pg_vec_page(unsigned long order)
3503 {
3504         char *buffer = NULL;
3505         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3506                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3507
3508         buffer = (char *) __get_free_pages(gfp_flags, order);
3509
3510         if (buffer)
3511                 return buffer;
3512
3513         /*
3514          * __get_free_pages failed, fall back to vmalloc
3515          */
3516         buffer = vzalloc((1 << order) * PAGE_SIZE);
3517
3518         if (buffer)
3519                 return buffer;
3520
3521         /*
3522          * vmalloc failed, lets dig into swap here
3523          */
3524         gfp_flags &= ~__GFP_NORETRY;
3525         buffer = (char *)__get_free_pages(gfp_flags, order);
3526         if (buffer)
3527                 return buffer;
3528
3529         /*
3530          * complete and utter failure
3531          */
3532         return NULL;
3533 }
3534
3535 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3536 {
3537         unsigned int block_nr = req->tp_block_nr;
3538         struct pgv *pg_vec;
3539         int i;
3540
3541         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3542         if (unlikely(!pg_vec))
3543                 goto out;
3544
3545         for (i = 0; i < block_nr; i++) {
3546                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3547                 if (unlikely(!pg_vec[i].buffer))
3548                         goto out_free_pgvec;
3549         }
3550
3551 out:
3552         return pg_vec;
3553
3554 out_free_pgvec:
3555         free_pg_vec(pg_vec, order, block_nr);
3556         pg_vec = NULL;
3557         goto out;
3558 }
3559
3560 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3561                 int closing, int tx_ring)
3562 {
3563         struct pgv *pg_vec = NULL;
3564         struct packet_sock *po = pkt_sk(sk);
3565         int was_running, order = 0;
3566         struct packet_ring_buffer *rb;
3567         struct sk_buff_head *rb_queue;
3568         __be16 num;
3569         int err = -EINVAL;
3570         /* Added to avoid minimal code churn */
3571         struct tpacket_req *req = &req_u->req;
3572
3573         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3574         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3575                 WARN(1, "Tx-ring is not supported.\n");
3576                 goto out;
3577         }
3578
3579         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3580         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3581
3582         err = -EBUSY;
3583         if (!closing) {
3584                 if (atomic_read(&po->mapped))
3585                         goto out;
3586                 if (atomic_read(&rb->pending))
3587                         goto out;
3588         }
3589
3590         if (req->tp_block_nr) {
3591                 /* Sanity tests and some calculations */
3592                 err = -EBUSY;
3593                 if (unlikely(rb->pg_vec))
3594                         goto out;
3595
3596                 switch (po->tp_version) {
3597                 case TPACKET_V1:
3598                         po->tp_hdrlen = TPACKET_HDRLEN;
3599                         break;
3600                 case TPACKET_V2:
3601                         po->tp_hdrlen = TPACKET2_HDRLEN;
3602                         break;
3603                 case TPACKET_V3:
3604                         po->tp_hdrlen = TPACKET3_HDRLEN;
3605                         break;
3606                 }
3607
3608                 err = -EINVAL;
3609                 if (unlikely((int)req->tp_block_size <= 0))
3610                         goto out;
3611                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3612                         goto out;
3613                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3614                                         po->tp_reserve))
3615                         goto out;
3616                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3617                         goto out;
3618
3619                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3620                 if (unlikely(rb->frames_per_block <= 0))
3621                         goto out;
3622                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3623                                         req->tp_frame_nr))
3624                         goto out;
3625
3626                 err = -ENOMEM;
3627                 order = get_order(req->tp_block_size);
3628                 pg_vec = alloc_pg_vec(req, order);
3629                 if (unlikely(!pg_vec))
3630                         goto out;
3631                 switch (po->tp_version) {
3632                 case TPACKET_V3:
3633                 /* Transmit path is not supported. We checked
3634                  * it above but just being paranoid
3635                  */
3636                         if (!tx_ring)
3637                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3638                                 break;
3639                 default:
3640                         break;
3641                 }
3642         }
3643         /* Done */
3644         else {
3645                 err = -EINVAL;
3646                 if (unlikely(req->tp_frame_nr))
3647                         goto out;
3648         }
3649
3650         lock_sock(sk);
3651
3652         /* Detach socket from network */
3653         spin_lock(&po->bind_lock);
3654         was_running = po->running;
3655         num = po->num;
3656         if (was_running) {
3657                 po->num = 0;
3658                 __unregister_prot_hook(sk, false);
3659         }
3660         spin_unlock(&po->bind_lock);
3661
3662         synchronize_net();
3663
3664         err = -EBUSY;
3665         mutex_lock(&po->pg_vec_lock);
3666         if (closing || atomic_read(&po->mapped) == 0) {
3667                 err = 0;
3668                 spin_lock_bh(&rb_queue->lock);
3669                 swap(rb->pg_vec, pg_vec);
3670                 rb->frame_max = (req->tp_frame_nr - 1);
3671                 rb->head = 0;
3672                 rb->frame_size = req->tp_frame_size;
3673                 spin_unlock_bh(&rb_queue->lock);
3674
3675                 swap(rb->pg_vec_order, order);
3676                 swap(rb->pg_vec_len, req->tp_block_nr);
3677
3678                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3679                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3680                                                 tpacket_rcv : packet_rcv;
3681                 skb_queue_purge(rb_queue);
3682                 if (atomic_read(&po->mapped))
3683                         pr_err("packet_mmap: vma is busy: %d\n",
3684                                atomic_read(&po->mapped));
3685         }
3686         mutex_unlock(&po->pg_vec_lock);
3687
3688         spin_lock(&po->bind_lock);
3689         if (was_running) {
3690                 po->num = num;
3691                 register_prot_hook(sk);
3692         }
3693         spin_unlock(&po->bind_lock);
3694         if (closing && (po->tp_version > TPACKET_V2)) {
3695                 /* Because we don't support block-based V3 on tx-ring */
3696                 if (!tx_ring)
3697                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3698         }
3699         release_sock(sk);
3700
3701         if (pg_vec)
3702                 free_pg_vec(pg_vec, order, req->tp_block_nr);
3703 out:
3704         return err;
3705 }
3706
3707 static int packet_mmap(struct file *file, struct socket *sock,
3708                 struct vm_area_struct *vma)
3709 {
3710         struct sock *sk = sock->sk;
3711         struct packet_sock *po = pkt_sk(sk);
3712         unsigned long size, expected_size;
3713         struct packet_ring_buffer *rb;
3714         unsigned long start;
3715         int err = -EINVAL;
3716         int i;
3717
3718         if (vma->vm_pgoff)
3719                 return -EINVAL;
3720
3721         mutex_lock(&po->pg_vec_lock);
3722
3723         expected_size = 0;
3724         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3725                 if (rb->pg_vec) {
3726                         expected_size += rb->pg_vec_len
3727                                                 * rb->pg_vec_pages
3728                                                 * PAGE_SIZE;
3729                 }
3730         }
3731
3732         if (expected_size == 0)
3733                 goto out;
3734
3735         size = vma->vm_end - vma->vm_start;
3736         if (size != expected_size)
3737                 goto out;
3738
3739         start = vma->vm_start;
3740         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3741                 if (rb->pg_vec == NULL)
3742                         continue;
3743
3744                 for (i = 0; i < rb->pg_vec_len; i++) {
3745                         struct page *page;
3746                         void *kaddr = rb->pg_vec[i].buffer;
3747                         int pg_num;
3748
3749                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3750                                 page = pgv_to_page(kaddr);
3751                                 err = vm_insert_page(vma, start, page);
3752                                 if (unlikely(err))
3753                                         goto out;
3754                                 start += PAGE_SIZE;
3755                                 kaddr += PAGE_SIZE;
3756                         }
3757                 }
3758         }
3759
3760         atomic_inc(&po->mapped);
3761         vma->vm_ops = &packet_mmap_ops;
3762         err = 0;
3763
3764 out:
3765         mutex_unlock(&po->pg_vec_lock);
3766         return err;
3767 }
3768
3769 static const struct proto_ops packet_ops_spkt = {
3770         .family =       PF_PACKET,
3771         .owner =        THIS_MODULE,
3772         .release =      packet_release,
3773         .bind =         packet_bind_spkt,
3774         .connect =      sock_no_connect,
3775         .socketpair =   sock_no_socketpair,
3776         .accept =       sock_no_accept,
3777         .getname =      packet_getname_spkt,
3778         .poll =         datagram_poll,
3779         .ioctl =        packet_ioctl,
3780         .listen =       sock_no_listen,
3781         .shutdown =     sock_no_shutdown,
3782         .setsockopt =   sock_no_setsockopt,
3783         .getsockopt =   sock_no_getsockopt,
3784         .sendmsg =      packet_sendmsg_spkt,
3785         .recvmsg =      packet_recvmsg,
3786         .mmap =         sock_no_mmap,
3787         .sendpage =     sock_no_sendpage,
3788 };
3789
3790 static const struct proto_ops packet_ops = {
3791         .family =       PF_PACKET,
3792         .owner =        THIS_MODULE,
3793         .release =      packet_release,
3794         .bind =         packet_bind,
3795         .connect =      sock_no_connect,
3796         .socketpair =   sock_no_socketpair,
3797         .accept =       sock_no_accept,
3798         .getname =      packet_getname,
3799         .poll =         packet_poll,
3800         .ioctl =        packet_ioctl,
3801         .listen =       sock_no_listen,
3802         .shutdown =     sock_no_shutdown,
3803         .setsockopt =   packet_setsockopt,
3804         .getsockopt =   packet_getsockopt,
3805         .sendmsg =      packet_sendmsg,
3806         .recvmsg =      packet_recvmsg,
3807         .mmap =         packet_mmap,
3808         .sendpage =     sock_no_sendpage,
3809 };
3810
3811 static const struct net_proto_family packet_family_ops = {
3812         .family =       PF_PACKET,
3813         .create =       packet_create,
3814         .owner  =       THIS_MODULE,
3815 };
3816
3817 static struct notifier_block packet_netdev_notifier = {
3818         .notifier_call =        packet_notifier,
3819 };
3820
3821 #ifdef CONFIG_PROC_FS
3822
3823 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3824         __acquires(RCU)
3825 {
3826         struct net *net = seq_file_net(seq);
3827
3828         rcu_read_lock();
3829         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3830 }
3831
3832 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3833 {
3834         struct net *net = seq_file_net(seq);
3835         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3836 }
3837
3838 static void packet_seq_stop(struct seq_file *seq, void *v)
3839         __releases(RCU)
3840 {
3841         rcu_read_unlock();
3842 }
3843
3844 static int packet_seq_show(struct seq_file *seq, void *v)
3845 {
3846         if (v == SEQ_START_TOKEN)
3847                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
3848         else {
3849                 struct sock *s = sk_entry(v);
3850                 const struct packet_sock *po = pkt_sk(s);
3851
3852                 seq_printf(seq,
3853                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
3854                            s,
3855                            atomic_read(&s->sk_refcnt),
3856                            s->sk_type,
3857                            ntohs(po->num),
3858                            po->ifindex,
3859                            po->running,
3860                            atomic_read(&s->sk_rmem_alloc),
3861                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
3862                            sock_i_ino(s));
3863         }
3864
3865         return 0;
3866 }
3867
3868 static const struct seq_operations packet_seq_ops = {
3869         .start  = packet_seq_start,
3870         .next   = packet_seq_next,
3871         .stop   = packet_seq_stop,
3872         .show   = packet_seq_show,
3873 };
3874
3875 static int packet_seq_open(struct inode *inode, struct file *file)
3876 {
3877         return seq_open_net(inode, file, &packet_seq_ops,
3878                             sizeof(struct seq_net_private));
3879 }
3880
3881 static const struct file_operations packet_seq_fops = {
3882         .owner          = THIS_MODULE,
3883         .open           = packet_seq_open,
3884         .read           = seq_read,
3885         .llseek         = seq_lseek,
3886         .release        = seq_release_net,
3887 };
3888
3889 #endif
3890
3891 static int __net_init packet_net_init(struct net *net)
3892 {
3893         mutex_init(&net->packet.sklist_lock);
3894         INIT_HLIST_HEAD(&net->packet.sklist);
3895
3896         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
3897                 return -ENOMEM;
3898
3899         return 0;
3900 }
3901
3902 static void __net_exit packet_net_exit(struct net *net)
3903 {
3904         remove_proc_entry("packet", net->proc_net);
3905 }
3906
3907 static struct pernet_operations packet_net_ops = {
3908         .init = packet_net_init,
3909         .exit = packet_net_exit,
3910 };
3911
3912
3913 static void __exit packet_exit(void)
3914 {
3915         unregister_netdevice_notifier(&packet_netdev_notifier);
3916         unregister_pernet_subsys(&packet_net_ops);
3917         sock_unregister(PF_PACKET);
3918         proto_unregister(&packet_proto);
3919 }
3920
3921 static int __init packet_init(void)
3922 {
3923         int rc = proto_register(&packet_proto, 0);
3924
3925         if (rc != 0)
3926                 goto out;
3927
3928         sock_register(&packet_family_ops);
3929         register_pernet_subsys(&packet_net_ops);
3930         register_netdevice_notifier(&packet_netdev_notifier);
3931 out:
3932         return rc;
3933 }
3934
3935 module_init(packet_init);
3936 module_exit(packet_exit);
3937 MODULE_LICENSE("GPL");
3938 MODULE_ALIAS_NETPROTO(PF_PACKET);