net/packet/af_packet.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              PACKET - implements raw packet sockets.
   7  *
   8  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  11  *
  12  * Fixes:
  13  *              Alan Cox        :       verify_area() now used correctly
  14  *              Alan Cox        :       new skbuff lists, look ma no backlogs!
  15  *              Alan Cox        :       tidied skbuff lists.
  16  *              Alan Cox        :       Now uses generic datagram routines I
  17  *                                      added. Also fixed the peek/read crash
  18  *                                      from all old Linux datagram code.
  19  *              Alan Cox        :       Uses the improved datagram code.
  20  *              Alan Cox        :       Added NULL's for socket options.
  21  *              Alan Cox        :       Re-commented the code.
  22  *              Alan Cox        :       Use new kernel side addressing
  23  *              Rob Janssen     :       Correct MTU usage.
  24  *              Dave Platt      :       Counter leaks caused by incorrect
  25  *                                      interrupt locking and some slightly
  26  *                                      dubious gcc output. Can you read
  27  *                                      compiler: it said _VOLATILE_
  28  *      Richard Kooijman        :       Timestamp fixes.
  29  *              Alan Cox        :       New buffers. Use sk->mac.raw.
  30  *              Alan Cox        :       sendmsg/recvmsg support.
  31  *              Alan Cox        :       Protocol setting support
  32  *      Alexey Kuznetsov        :       Untied from IPv4 stack.
  33  *      Cyrus Durgin            :       Fixed kerneld for kmod.
  34  *      Michal Ostrowski        :       Module initialization cleanup.
  35  *         Ulises Alonso        :       Frame number limit removal and
  36  *                                      packet_set_ring memory leak.
  37  *              Eric Biederman  :       Allow for > 8 byte hardware addresses.
  38  *                                      The convention is that longer addresses
  39  *                                      will simply extend the hardware address
  40  *                                      byte arrays at the end of sockaddr_ll
  41  *                                      and packet_mreq.
  42  *              Johann Baudy    :       Added TX RING.
  43  *              Chetan Loke     :       Implemented TPACKET_V3 block abstraction
  44  *                                      layer.
  45  *                                      Copyright (C) 2011, <lokec@ccs.neu.edu>
  46  *
  47  *
  48  *              This program is free software; you can redistribute it and/or
  49  *              modify it under the terms of the GNU General Public License
  50  *              as published by the Free Software Foundation; either version
  51  *              2 of the License, or (at your option) any later version.
  52  *
  53  */
  54
  55 #include <linux/types.h>
  56 #include <linux/mm.h>
  57 #include <linux/capability.h>
  58 #include <linux/fcntl.h>
  59 #include <linux/socket.h>
  60 #include <linux/in.h>
  61 #include <linux/inet.h>
  62 #include <linux/netdevice.h>
  63 #include <linux/if_packet.h>
  64 #include <linux/wireless.h>
  65 #include <linux/kernel.h>
  66 #include <linux/kmod.h>
  67 #include <linux/slab.h>
  68 #include <linux/vmalloc.h>
  69 #include <net/net_namespace.h>
  70 #include <net/ip.h>
  71 #include <net/protocol.h>
  72 #include <linux/skbuff.h>
  73 #include <net/sock.h>
  74 #include <linux/errno.h>
  75 #include <linux/timer.h>
  76 #include <asm/uaccess.h>
  77 #include <asm/ioctls.h>
  78 #include <asm/page.h>
  79 #include <asm/cacheflush.h>
  80 #include <asm/io.h>
  81 #include <linux/proc_fs.h>
  82 #include <linux/seq_file.h>
  83 #include <linux/poll.h>
  84 #include <linux/module.h>
  85 #include <linux/init.h>
  86 #include <linux/mutex.h>
  87 #include <linux/if_vlan.h>
  88 #include <linux/virtio_net.h>
  89 #include <linux/errqueue.h>
  90 #include <linux/net_tstamp.h>
  91
  92 #ifdef CONFIG_INET
  93 #include <net/inet_common.h>
  94 #endif
  95
  96 #include "internal.h"
  97
  98 /*
  99    Assumptions:
 100    - if device has no dev->hard_header routine, it adds and removes ll header
 101      inside itself. In this case ll header is invisible outside of device,
 102      but higher levels still should reserve dev->hard_header_len.
 103      Some devices are enough clever to reallocate skb, when header
 104      will not fit to reserved space (tunnel), another ones are silly
 105      (PPP).
 106    - packet socket receives packets with pulled ll header,
 107      so that SOCK_RAW should push it back.
 108
 109 On receive:
 110 -----------
 111
 112 Incoming, dev->hard_header!=NULL
 113    mac_header -> ll header
 114    data       -> data
 115
 116 Outgoing, dev->hard_header!=NULL
 117    mac_header -> ll header
 118    data       -> ll header
 119
 120 Incoming, dev->hard_header==NULL
 121    mac_header -> UNKNOWN position. It is very likely, that it points to ll
 122                  header.  PPP makes it, that is wrong, because introduce
 123                  assymetry between rx and tx paths.
 124    data       -> data
 125
 126 Outgoing, dev->hard_header==NULL
 127    mac_header -> data. ll header is still not built!
 128    data       -> data
 129
 130 Resume
 131   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 132
 133
 134 On transmit:
 135 ------------
 136
 137 dev->hard_header != NULL
 138    mac_header -> ll header
 139    data       -> ll header
 140
 141 dev->hard_header == NULL (ll header is added by device, we cannot control it)
 142    mac_header -> data
 143    data       -> data
 144
 145    We should set nh.raw on output to correct posistion,
 146    packet classifier depends on it.
 147  */
 148
 149 /* Private packet socket structures. */
 150
 151 /* identical to struct packet_mreq except it has
 152  * a longer address field.
 153  */
 154 struct packet_mreq_max {
 155         int             mr_ifindex;
 156         unsigned short  mr_type;
 157         unsigned short  mr_alen;
 158         unsigned char   mr_address[MAX_ADDR_LEN];
 159 };
 160
 161 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 162                 int closing, int tx_ring);
 163
 164
 165 #define V3_ALIGNMENT    (8)
 166
 167 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 168
 169 #define BLK_PLUS_PRIV(sz_of_priv) \
 170         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 171
 172 #define PGV_FROM_VMALLOC 1
 173
 174 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
 175 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
 176 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
 177 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
 178 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
 179 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
 180 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
 181
 182 struct packet_sock;
 183 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
 184 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 185                        struct packet_type *pt, struct net_device *orig_dev);
 186
 187 static void *packet_previous_frame(struct packet_sock *po,
 188                 struct packet_ring_buffer *rb,
 189                 int status);
 190 static void packet_increment_head(struct packet_ring_buffer *buff);
 191 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
 192                         struct tpacket_block_desc *);
 193 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 194                         struct packet_sock *);
 195 static void prb_retire_current_block(struct tpacket_kbdq_core *,
 196                 struct packet_sock *, unsigned int status);
 197 static int prb_queue_frozen(struct tpacket_kbdq_core *);
 198 static void prb_open_block(struct tpacket_kbdq_core *,
 199                 struct tpacket_block_desc *);
 200 static void prb_retire_rx_blk_timer_expired(unsigned long);
 201 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 202 static void prb_init_blk_timer(struct packet_sock *,
 203                 struct tpacket_kbdq_core *,
 204                 void (*func) (unsigned long));
 205 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 206 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 207                 struct tpacket3_hdr *);
 208 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 209                 struct tpacket3_hdr *);
 210 static void packet_flush_mclist(struct sock *sk);
 211
 212 struct packet_skb_cb {
 213         unsigned int origlen;
 214         union {
 215                 struct sockaddr_pkt pkt;
 216                 struct sockaddr_ll ll;
 217         } sa;
 218 };
 219
 220 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 221
 222 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 223 #define GET_PBLOCK_DESC(x, bid) \
 224         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 225 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 226         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 227 #define GET_NEXT_PRB_BLK_NUM(x) \
 228         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 229         ((x)->kactive_blk_num+1) : 0)
 230
 231 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 232 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 233
 234 /* register_prot_hook must be invoked with the po->bind_lock held,
 235  * or from a context in which asynchronous accesses to the packet
 236  * socket is not possible (packet_create()).
 237  */
 238 static void register_prot_hook(struct sock *sk)
 239 {
 240         struct packet_sock *po = pkt_sk(sk);
 241         if (!po->running) {
 242                 if (po->fanout)
 243                         __fanout_link(sk, po);
 244                 else
 245                         dev_add_pack(&po->prot_hook);
 246                 sock_hold(sk);
 247                 po->running = 1;
 248         }
 249 }
 250
 251 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
 252  * held.   If the sync parameter is true, we will temporarily drop
 253  * the po->bind_lock and do a synchronize_net to make sure no
 254  * asynchronous packet processing paths still refer to the elements
 255  * of po->prot_hook.  If the sync parameter is false, it is the
 256  * callers responsibility to take care of this.
 257  */
 258 static void __unregister_prot_hook(struct sock *sk, bool sync)
 259 {
 260         struct packet_sock *po = pkt_sk(sk);
 261
 262         po->running = 0;
 263         if (po->fanout)
 264                 __fanout_unlink(sk, po);
 265         else
 266                 __dev_remove_pack(&po->prot_hook);
 267         __sock_put(sk);
 268
 269         if (sync) {
 270                 spin_unlock(&po->bind_lock);
 271                 synchronize_net();
 272                 spin_lock(&po->bind_lock);
 273         }
 274 }
 275
 276 static void unregister_prot_hook(struct sock *sk, bool sync)
 277 {
 278         struct packet_sock *po = pkt_sk(sk);
 279
 280         if (po->running)
 281                 __unregister_prot_hook(sk, sync);
 282 }
 283
 284 static inline __pure struct page *pgv_to_page(void *addr)
 285 {
 286         if (is_vmalloc_addr(addr))
 287                 return vmalloc_to_page(addr);
 288         return virt_to_page(addr);
 289 }
 290
 291 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 292 {
 293         union {
 294                 struct tpacket_hdr *h1;
 295                 struct tpacket2_hdr *h2;
 296                 void *raw;
 297         } h;
 298
 299         h.raw = frame;
 300         switch (po->tp_version) {
 301         case TPACKET_V1:
 302                 h.h1->tp_status = status;
 303                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 304                 break;
 305         case TPACKET_V2:
 306                 h.h2->tp_status = status;
 307                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 308                 break;
 309         case TPACKET_V3:
 310         default:
 311                 WARN(1, "TPACKET version not supported.\n");
 312                 BUG();
 313         }
 314
 315         smp_wmb();
 316 }
 317
 318 static int __packet_get_status(struct packet_sock *po, void *frame)
 319 {
 320         union {
 321                 struct tpacket_hdr *h1;
 322                 struct tpacket2_hdr *h2;
 323                 void *raw;
 324         } h;
 325
 326         smp_rmb();
 327
 328         h.raw = frame;
 329         switch (po->tp_version) {
 330         case TPACKET_V1:
 331                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 332                 return h.h1->tp_status;
 333         case TPACKET_V2:
 334                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 335                 return h.h2->tp_status;
 336         case TPACKET_V3:
 337         default:
 338                 WARN(1, "TPACKET version not supported.\n");
 339                 BUG();
 340                 return 0;
 341         }
 342 }
 343
 344 static void *packet_lookup_frame(struct packet_sock *po,
 345                 struct packet_ring_buffer *rb,
 346                 unsigned int position,
 347                 int status)
 348 {
 349         unsigned int pg_vec_pos, frame_offset;
 350         union {
 351                 struct tpacket_hdr *h1;
 352                 struct tpacket2_hdr *h2;
 353                 void *raw;
 354         } h;
 355
 356         pg_vec_pos = position / rb->frames_per_block;
 357         frame_offset = position % rb->frames_per_block;
 358
 359         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 360                 (frame_offset * rb->frame_size);
 361
 362         if (status != __packet_get_status(po, h.raw))
 363                 return NULL;
 364
 365         return h.raw;
 366 }
 367
 368 static void *packet_current_frame(struct packet_sock *po,
 369                 struct packet_ring_buffer *rb,
 370                 int status)
 371 {
 372         return packet_lookup_frame(po, rb, rb->head, status);
 373 }
 374
 375 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 376 {
 377         del_timer_sync(&pkc->retire_blk_timer);
 378 }
 379
 380 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 381                 int tx_ring,
 382                 struct sk_buff_head *rb_queue)
 383 {
 384         struct tpacket_kbdq_core *pkc;
 385
 386         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 387
 388         spin_lock(&rb_queue->lock);
 389         pkc->delete_blk_timer = 1;
 390         spin_unlock(&rb_queue->lock);
 391
 392         prb_del_retire_blk_timer(pkc);
 393 }
 394
 395 static void prb_init_blk_timer(struct packet_sock *po,
 396                 struct tpacket_kbdq_core *pkc,
 397                 void (*func) (unsigned long))
 398 {
 399         init_timer(&pkc->retire_blk_timer);
 400         pkc->retire_blk_timer.data = (long)po;
 401         pkc->retire_blk_timer.function = func;
 402         pkc->retire_blk_timer.expires = jiffies;
 403 }
 404
 405 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
 406 {
 407         struct tpacket_kbdq_core *pkc;
 408
 409         if (tx_ring)
 410                 BUG();
 411
 412         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 413         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
 414 }
 415
 416 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 417                                 int blk_size_in_bytes)
 418 {
 419         struct net_device *dev;
 420         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 421         struct ethtool_cmd ecmd;
 422         int err;
 423         u32 speed;
 424
 425         rtnl_lock();
 426         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 427         if (unlikely(!dev)) {
 428                 rtnl_unlock();
 429                 return DEFAULT_PRB_RETIRE_TOV;
 430         }
 431         err = __ethtool_get_settings(dev, &ecmd);
 432         speed = ethtool_cmd_speed(&ecmd);
 433         rtnl_unlock();
 434         if (!err) {
 435                 /*
 436                  * If the link speed is so slow you don't really
 437                  * need to worry about perf anyways
 438                  */
 439                 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
 440                         return DEFAULT_PRB_RETIRE_TOV;
 441                 } else {
 442                         msec = 1;
 443                         div = speed / 1000;
 444                 }
 445         }
 446
 447         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 448
 449         if (div)
 450                 mbits /= div;
 451
 452         tmo = mbits * msec;
 453
 454         if (div)
 455                 return tmo+1;
 456         return tmo;
 457 }
 458
 459 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 460                         union tpacket_req_u *req_u)
 461 {
 462         p1->feature_req_word = req_u->req3.tp_feature_req_word;
 463 }
 464
 465 static void init_prb_bdqc(struct packet_sock *po,
 466                         struct packet_ring_buffer *rb,
 467                         struct pgv *pg_vec,
 468                         union tpacket_req_u *req_u, int tx_ring)
 469 {
 470         struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
 471         struct tpacket_block_desc *pbd;
 472
 473         memset(p1, 0x0, sizeof(*p1));
 474
 475         p1->knxt_seq_num = 1;
 476         p1->pkbdq = pg_vec;
 477         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 478         p1->pkblk_start = pg_vec[0].buffer;
 479         p1->kblk_size = req_u->req3.tp_block_size;
 480         p1->knum_blocks = req_u->req3.tp_block_nr;
 481         p1->hdrlen = po->tp_hdrlen;
 482         p1->version = po->tp_version;
 483         p1->last_kactive_blk_num = 0;
 484         po->stats_u.stats3.tp_freeze_q_cnt = 0;
 485         if (req_u->req3.tp_retire_blk_tov)
 486                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 487         else
 488                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 489                                                 req_u->req3.tp_block_size);
 490         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 491         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 492
 493         prb_init_ft_ops(p1, req_u);
 494         prb_setup_retire_blk_timer(po, tx_ring);
 495         prb_open_block(p1, pbd);
 496 }
 497
 498 /*  Do NOT update the last_blk_num first.
 499  *  Assumes sk_buff_head lock is held.
 500  */
 501 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 502 {
 503         mod_timer(&pkc->retire_blk_timer,
 504                         jiffies + pkc->tov_in_jiffies);
 505         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 506 }
 507
 508 /*
 509  * Timer logic:
 510  * 1) We refresh the timer only when we open a block.
 511  *    By doing this we don't waste cycles refreshing the timer
 512  *        on packet-by-packet basis.
 513  *
 514  * With a 1MB block-size, on a 1Gbps line, it will take
 515  * i) ~8 ms to fill a block + ii) memcpy etc.
 516  * In this cut we are not accounting for the memcpy time.
 517  *
 518  * So, if the user sets the 'tmo' to 10ms then the timer
 519  * will never fire while the block is still getting filled
 520  * (which is what we want). However, the user could choose
 521  * to close a block early and that's fine.
 522  *
 523  * But when the timer does fire, we check whether or not to refresh it.
 524  * Since the tmo granularity is in msecs, it is not too expensive
 525  * to refresh the timer, lets say every '8' msecs.
 526  * Either the user can set the 'tmo' or we can derive it based on
 527  * a) line-speed and b) block-size.
 528  * prb_calc_retire_blk_tmo() calculates the tmo.
 529  *
 530  */
 531 static void prb_retire_rx_blk_timer_expired(unsigned long data)
 532 {
 533         struct packet_sock *po = (struct packet_sock *)data;
 534         struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
 535         unsigned int frozen;
 536         struct tpacket_block_desc *pbd;
 537
 538         spin_lock(&po->sk.sk_receive_queue.lock);
 539
 540         frozen = prb_queue_frozen(pkc);
 541         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 542
 543         if (unlikely(pkc->delete_blk_timer))
 544                 goto out;
 545
 546         /* We only need to plug the race when the block is partially filled.
 547          * tpacket_rcv:
 548          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 549          *              copy_bits() is in progress ...
 550          *              timer fires on other cpu:
 551          *              we can't retire the current block because copy_bits
 552          *              is in progress.
 553          *
 554          */
 555         if (BLOCK_NUM_PKTS(pbd)) {
 556                 while (atomic_read(&pkc->blk_fill_in_prog)) {
 557                         /* Waiting for skb_copy_bits to finish... */
 558                         cpu_relax();
 559                 }
 560         }
 561
 562         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 563                 if (!frozen) {
 564                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 565                         if (!prb_dispatch_next_block(pkc, po))
 566                                 goto refresh_timer;
 567                         else
 568                                 goto out;
 569                 } else {
 570                         /* Case 1. Queue was frozen because user-space was
 571                          *         lagging behind.
 572                          */
 573                         if (prb_curr_blk_in_use(pkc, pbd)) {
 574                                 /*
 575                                  * Ok, user-space is still behind.
 576                                  * So just refresh the timer.
 577                                  */
 578                                 goto refresh_timer;
 579                         } else {
 580                                /* Case 2. queue was frozen,user-space caught up,
 581                                 * now the link went idle && the timer fired.
 582                                 * We don't have a block to close.So we open this
 583                                 * block and restart the timer.
 584                                 * opening a block thaws the queue,restarts timer
 585                                 * Thawing/timer-refresh is a side effect.
 586                                 */
 587                                 prb_open_block(pkc, pbd);
 588                                 goto out;
 589                         }
 590                 }
 591         }
 592
 593 refresh_timer:
 594         _prb_refresh_rx_retire_blk_timer(pkc);
 595
 596 out:
 597         spin_unlock(&po->sk.sk_receive_queue.lock);
 598 }
 599
 600 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 601                 struct tpacket_block_desc *pbd1, __u32 status)
 602 {
 603         /* Flush everything minus the block header */
 604
 605 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 606         u8 *start, *end;
 607
 608         start = (u8 *)pbd1;
 609
 610         /* Skip the block header(we know header WILL fit in 4K) */
 611         start += PAGE_SIZE;
 612
 613         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 614         for (; start < end; start += PAGE_SIZE)
 615                 flush_dcache_page(pgv_to_page(start));
 616
 617         smp_wmb();
 618 #endif
 619
 620         /* Now update the block status. */
 621
 622         BLOCK_STATUS(pbd1) = status;
 623
 624         /* Flush the block header */
 625
 626 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 627         start = (u8 *)pbd1;
 628         flush_dcache_page(pgv_to_page(start));
 629
 630         smp_wmb();
 631 #endif
 632 }
 633
 634 /*
 635  * Side effect:
 636  *
 637  * 1) flush the block
 638  * 2) Increment active_blk_num
 639  *
 640  * Note:We DONT refresh the timer on purpose.
 641  *      Because almost always the next block will be opened.
 642  */
 643 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 644                 struct tpacket_block_desc *pbd1,
 645                 struct packet_sock *po, unsigned int stat)
 646 {
 647         __u32 status = TP_STATUS_USER | stat;
 648
 649         struct tpacket3_hdr *last_pkt;
 650         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 651
 652         if (po->stats.tp_drops)
 653                 status |= TP_STATUS_LOSING;
 654
 655         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 656         last_pkt->tp_next_offset = 0;
 657
 658         /* Get the ts of the last pkt */
 659         if (BLOCK_NUM_PKTS(pbd1)) {
 660                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 661                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 662         } else {
 663                 /* Ok, we tmo'd - so get the current time */
 664                 struct timespec ts;
 665                 getnstimeofday(&ts);
 666                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 667                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 668         }
 669
 670         smp_wmb();
 671
 672         /* Flush the block */
 673         prb_flush_block(pkc1, pbd1, status);
 674
 675         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 676 }
 677
 678 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 679 {
 680         pkc->reset_pending_on_curr_blk = 0;
 681 }
 682
 683 /*
 684  * Side effect of opening a block:
 685  *
 686  * 1) prb_queue is thawed.
 687  * 2) retire_blk_timer is refreshed.
 688  *
 689  */
 690 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 691         struct tpacket_block_desc *pbd1)
 692 {
 693         struct timespec ts;
 694         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 695
 696         smp_rmb();
 697
 698         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd1))) {
 699
 700                 /* We could have just memset this but we will lose the
 701                  * flexibility of making the priv area sticky
 702                  */
 703                 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 704                 BLOCK_NUM_PKTS(pbd1) = 0;
 705                 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 706                 getnstimeofday(&ts);
 707                 h1->ts_first_pkt.ts_sec = ts.tv_sec;
 708                 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 709                 pkc1->pkblk_start = (char *)pbd1;
 710                 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 711                 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 712                 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 713                 pbd1->version = pkc1->version;
 714                 pkc1->prev = pkc1->nxt_offset;
 715                 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 716                 prb_thaw_queue(pkc1);
 717                 _prb_refresh_rx_retire_blk_timer(pkc1);
 718
 719                 smp_wmb();
 720
 721                 return;
 722         }
 723
 724         WARN(1, "ERROR block:%p is NOT FREE status:%d kactive_blk_num:%d\n",
 725                 pbd1, BLOCK_STATUS(pbd1), pkc1->kactive_blk_num);
 726         dump_stack();
 727         BUG();
 728 }
 729
 730 /*
 731  * Queue freeze logic:
 732  * 1) Assume tp_block_nr = 8 blocks.
 733  * 2) At time 't0', user opens Rx ring.
 734  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 735  * 4) user-space is either sleeping or processing block '0'.
 736  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 737  *    it will close block-7,loop around and try to fill block '0'.
 738  *    call-flow:
 739  *    __packet_lookup_frame_in_block
 740  *      prb_retire_current_block()
 741  *      prb_dispatch_next_block()
 742  *        |->(BLOCK_STATUS == USER) evaluates to true
 743  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 744  * 6) Now there are two cases:
 745  *    6.1) Link goes idle right after the queue is frozen.
 746  *         But remember, the last open_block() refreshed the timer.
 747  *         When this timer expires,it will refresh itself so that we can
 748  *         re-open block-0 in near future.
 749  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 750  *         case and __packet_lookup_frame_in_block will check if block-0
 751  *         is free and can now be re-used.
 752  */
 753 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 754                                   struct packet_sock *po)
 755 {
 756         pkc->reset_pending_on_curr_blk = 1;
 757         po->stats_u.stats3.tp_freeze_q_cnt++;
 758 }
 759
 760 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 761
 762 /*
 763  * If the next block is free then we will dispatch it
 764  * and return a good offset.
 765  * Else, we will freeze the queue.
 766  * So, caller must check the return value.
 767  */
 768 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 769                 struct packet_sock *po)
 770 {
 771         struct tpacket_block_desc *pbd;
 772
 773         smp_rmb();
 774
 775         /* 1. Get current block num */
 776         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 777
 778         /* 2. If this block is currently in_use then freeze the queue */
 779         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 780                 prb_freeze_queue(pkc, po);
 781                 return NULL;
 782         }
 783
 784         /*
 785          * 3.
 786          * open this block and return the offset where the first packet
 787          * needs to get stored.
 788          */
 789         prb_open_block(pkc, pbd);
 790         return (void *)pkc->nxt_offset;
 791 }
 792
 793 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 794                 struct packet_sock *po, unsigned int status)
 795 {
 796         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 797
 798         /* retire/close the current block */
 799         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 800                 /*
 801                  * Plug the case where copy_bits() is in progress on
 802                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 803                  * have space to copy the pkt in the current block and
 804                  * called prb_retire_current_block()
 805                  *
 806                  * We don't need to worry about the TMO case because
 807                  * the timer-handler already handled this case.
 808                  */
 809                 if (!(status & TP_STATUS_BLK_TMO)) {
 810                         while (atomic_read(&pkc->blk_fill_in_prog)) {
 811                                 /* Waiting for skb_copy_bits to finish... */
 812                                 cpu_relax();
 813                         }
 814                 }
 815                 prb_close_block(pkc, pbd, po, status);
 816                 return;
 817         }
 818
 819         WARN(1, "ERROR-pbd[%d]:%p\n", pkc->kactive_blk_num, pbd);
 820         dump_stack();
 821         BUG();
 822 }
 823
 824 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
 825                                       struct tpacket_block_desc *pbd)
 826 {
 827         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 828 }
 829
 830 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 831 {
 832         return pkc->reset_pending_on_curr_blk;
 833 }
 834
 835 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 836 {
 837         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 838         atomic_dec(&pkc->blk_fill_in_prog);
 839 }
 840
 841 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 842                         struct tpacket3_hdr *ppd)
 843 {
 844         ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
 845 }
 846
 847 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 848                         struct tpacket3_hdr *ppd)
 849 {
 850         ppd->hv1.tp_rxhash = 0;
 851 }
 852
 853 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
 854                         struct tpacket3_hdr *ppd)
 855 {
 856         if (vlan_tx_tag_present(pkc->skb)) {
 857                 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
 858                 ppd->tp_status = TP_STATUS_VLAN_VALID;
 859         } else {
 860                 ppd->hv1.tp_vlan_tci = 0;
 861                 ppd->tp_status = TP_STATUS_AVAILABLE;
 862         }
 863 }
 864
 865 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
 866                         struct tpacket3_hdr *ppd)
 867 {
 868         prb_fill_vlan_info(pkc, ppd);
 869
 870         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
 871                 prb_fill_rxhash(pkc, ppd);
 872         else
 873                 prb_clear_rxhash(pkc, ppd);
 874 }
 875
 876 static void prb_fill_curr_block(char *curr,
 877                                 struct tpacket_kbdq_core *pkc,
 878                                 struct tpacket_block_desc *pbd,
 879                                 unsigned int len)
 880 {
 881         struct tpacket3_hdr *ppd;
 882
 883         ppd  = (struct tpacket3_hdr *)curr;
 884         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
 885         pkc->prev = curr;
 886         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
 887         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
 888         BLOCK_NUM_PKTS(pbd) += 1;
 889         atomic_inc(&pkc->blk_fill_in_prog);
 890         prb_run_all_ft_ops(pkc, ppd);
 891 }
 892
 893 /* Assumes caller has the sk->rx_queue.lock */
 894 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
 895                                             struct sk_buff *skb,
 896                                                 int status,
 897                                             unsigned int len
 898                                             )
 899 {
 900         struct tpacket_kbdq_core *pkc;
 901         struct tpacket_block_desc *pbd;
 902         char *curr, *end;
 903
 904         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 905         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 906
 907         /* Queue is frozen when user space is lagging behind */
 908         if (prb_queue_frozen(pkc)) {
 909                 /*
 910                  * Check if that last block which caused the queue to freeze,
 911                  * is still in_use by user-space.
 912                  */
 913                 if (prb_curr_blk_in_use(pkc, pbd)) {
 914                         /* Can't record this packet */
 915                         return NULL;
 916                 } else {
 917                         /*
 918                          * Ok, the block was released by user-space.
 919                          * Now let's open that block.
 920                          * opening a block also thaws the queue.
 921                          * Thawing is a side effect.
 922                          */
 923                         prb_open_block(pkc, pbd);
 924                 }
 925         }
 926
 927         smp_mb();
 928         curr = pkc->nxt_offset;
 929         pkc->skb = skb;
 930         end = (char *)pbd + pkc->kblk_size;
 931
 932         /* first try the current block */
 933         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
 934                 prb_fill_curr_block(curr, pkc, pbd, len);
 935                 return (void *)curr;
 936         }
 937
 938         /* Ok, close the current block */
 939         prb_retire_current_block(pkc, po, 0);
 940
 941         /* Now, try to dispatch the next block */
 942         curr = (char *)prb_dispatch_next_block(pkc, po);
 943         if (curr) {
 944                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 945                 prb_fill_curr_block(curr, pkc, pbd, len);
 946                 return (void *)curr;
 947         }
 948
 949         /*
 950          * No free blocks are available.user_space hasn't caught up yet.
 951          * Queue was just frozen and now this packet will get dropped.
 952          */
 953         return NULL;
 954 }
 955
 956 static void *packet_current_rx_frame(struct packet_sock *po,
 957                                             struct sk_buff *skb,
 958                                             int status, unsigned int len)
 959 {
 960         char *curr = NULL;
 961         switch (po->tp_version) {
 962         case TPACKET_V1:
 963         case TPACKET_V2:
 964                 curr = packet_lookup_frame(po, &po->rx_ring,
 965                                         po->rx_ring.head, status);
 966                 return curr;
 967         case TPACKET_V3:
 968                 return __packet_lookup_frame_in_block(po, skb, status, len);
 969         default:
 970                 WARN(1, "TPACKET version not supported\n");
 971                 BUG();
 972                 return NULL;
 973         }
 974 }
 975
 976 static void *prb_lookup_block(struct packet_sock *po,
 977                                      struct packet_ring_buffer *rb,
 978                                      unsigned int idx,
 979                                      int status)
 980 {
 981         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 982         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
 983
 984         if (status != BLOCK_STATUS(pbd))
 985                 return NULL;
 986         return pbd;
 987 }
 988
 989 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
 990 {
 991         unsigned int prev;
 992         if (rb->prb_bdqc.kactive_blk_num)
 993                 prev = rb->prb_bdqc.kactive_blk_num-1;
 994         else
 995                 prev = rb->prb_bdqc.knum_blocks-1;
 996         return prev;
 997 }
 998
 999 /* Assumes caller has held the rx_queue.lock */
1000 static void *__prb_previous_block(struct packet_sock *po,
1001                                          struct packet_ring_buffer *rb,
1002                                          int status)
1003 {
1004         unsigned int previous = prb_previous_blk_num(rb);
1005         return prb_lookup_block(po, rb, previous, status);
1006 }
1007
1008 static void *packet_previous_rx_frame(struct packet_sock *po,
1009                                              struct packet_ring_buffer *rb,
1010                                              int status)
1011 {
1012         if (po->tp_version <= TPACKET_V2)
1013                 return packet_previous_frame(po, rb, status);
1014
1015         return __prb_previous_block(po, rb, status);
1016 }
1017
1018 static void packet_increment_rx_head(struct packet_sock *po,
1019                                             struct packet_ring_buffer *rb)
1020 {
1021         switch (po->tp_version) {
1022         case TPACKET_V1:
1023         case TPACKET_V2:
1024                 return packet_increment_head(rb);
1025         case TPACKET_V3:
1026         default:
1027                 WARN(1, "TPACKET version not supported.\n");
1028                 BUG();
1029                 return;
1030         }
1031 }
1032
1033 static void *packet_previous_frame(struct packet_sock *po,
1034                 struct packet_ring_buffer *rb,
1035                 int status)
1036 {
1037         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1038         return packet_lookup_frame(po, rb, previous, status);
1039 }
1040
1041 static void packet_increment_head(struct packet_ring_buffer *buff)
1042 {
1043         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1044 }
1045
1046 static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1047 {
1048         struct sock *sk = &po->sk;
1049         bool has_room;
1050
1051         if (po->prot_hook.func != tpacket_rcv)
1052                 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1053                         <= sk->sk_rcvbuf;
1054
1055         spin_lock(&sk->sk_receive_queue.lock);
1056         if (po->tp_version == TPACKET_V3)
1057                 has_room = prb_lookup_block(po, &po->rx_ring,
1058                                             po->rx_ring.prb_bdqc.kactive_blk_num,
1059                                             TP_STATUS_KERNEL);
1060         else
1061                 has_room = packet_lookup_frame(po, &po->rx_ring,
1062                                                po->rx_ring.head,
1063                                                TP_STATUS_KERNEL);
1064         spin_unlock(&sk->sk_receive_queue.lock);
1065
1066         return has_room;
1067 }
1068
1069 static void packet_sock_destruct(struct sock *sk)
1070 {
1071         skb_queue_purge(&sk->sk_error_queue);
1072
1073         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1074         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1075
1076         if (!sock_flag(sk, SOCK_DEAD)) {
1077                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1078                 return;
1079         }
1080
1081         sk_refcnt_debug_dec(sk);
1082 }
1083
1084 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1085 {
1086         int x = atomic_read(&f->rr_cur) + 1;
1087
1088         if (x >= num)
1089                 x = 0;
1090
1091         return x;
1092 }
1093
1094 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1095                                       struct sk_buff *skb,
1096                                       unsigned int num)
1097 {
1098         return (((u64)skb->rxhash) * num) >> 32;
1099 }
1100
1101 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1102                                     struct sk_buff *skb,
1103                                     unsigned int num)
1104 {
1105         int cur, old;
1106
1107         cur = atomic_read(&f->rr_cur);
1108         while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1109                                      fanout_rr_next(f, num))) != cur)
1110                 cur = old;
1111         return cur;
1112 }
1113
1114 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1115                                      struct sk_buff *skb,
1116                                      unsigned int num)
1117 {
1118         return smp_processor_id() % num;
1119 }
1120
1121 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1122                                           struct sk_buff *skb,
1123                                           unsigned int idx, unsigned int skip,
1124                                           unsigned int num)
1125 {
1126         unsigned int i, j;
1127
1128         i = j = min_t(int, f->next[idx], num - 1);
1129         do {
1130                 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1131                         if (i != j)
1132                                 f->next[idx] = i;
1133                         return i;
1134                 }
1135                 if (++i == num)
1136                         i = 0;
1137         } while (i != j);
1138
1139         return idx;
1140 }
1141
1142 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1143 {
1144         return f->flags & (flag >> 8);
1145 }
1146
1147 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1148                              struct packet_type *pt, struct net_device *orig_dev)
1149 {
1150         struct packet_fanout *f = pt->af_packet_priv;
1151         unsigned int num = f->num_members;
1152         struct packet_sock *po;
1153         unsigned int idx;
1154
1155         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1156             !num) {
1157                 kfree_skb(skb);
1158                 return 0;
1159         }
1160
1161         switch (f->type) {
1162         case PACKET_FANOUT_HASH:
1163         default:
1164                 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1165                         skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1166                         if (!skb)
1167                                 return 0;
1168                 }
1169                 skb_get_rxhash(skb);
1170                 idx = fanout_demux_hash(f, skb, num);
1171                 break;
1172         case PACKET_FANOUT_LB:
1173                 idx = fanout_demux_lb(f, skb, num);
1174                 break;
1175         case PACKET_FANOUT_CPU:
1176                 idx = fanout_demux_cpu(f, skb, num);
1177                 break;
1178         case PACKET_FANOUT_ROLLOVER:
1179                 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1180                 break;
1181         }
1182
1183         po = pkt_sk(f->arr[idx]);
1184         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1185             unlikely(!packet_rcv_has_room(po, skb))) {
1186                 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1187                 po = pkt_sk(f->arr[idx]);
1188         }
1189
1190         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1191 }
1192
1193 DEFINE_MUTEX(fanout_mutex);
1194 EXPORT_SYMBOL_GPL(fanout_mutex);
1195 static LIST_HEAD(fanout_list);
1196
1197 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1198 {
1199         struct packet_fanout *f = po->fanout;
1200
1201         spin_lock(&f->lock);
1202         f->arr[f->num_members] = sk;
1203         smp_wmb();
1204         f->num_members++;
1205         spin_unlock(&f->lock);
1206 }
1207
1208 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1209 {
1210         struct packet_fanout *f = po->fanout;
1211         int i;
1212
1213         spin_lock(&f->lock);
1214         for (i = 0; i < f->num_members; i++) {
1215                 if (f->arr[i] == sk)
1216                         break;
1217         }
1218         BUG_ON(i >= f->num_members);
1219         f->arr[i] = f->arr[f->num_members - 1];
1220         f->num_members--;
1221         spin_unlock(&f->lock);
1222 }
1223
1224 static bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1225 {
1226         if (ptype->af_packet_priv == (void*)((struct packet_sock *)sk)->fanout)
1227                 return true;
1228
1229         return false;
1230 }
1231
1232 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1233 {
1234         struct packet_sock *po = pkt_sk(sk);
1235         struct packet_fanout *f, *match;
1236         u8 type = type_flags & 0xff;
1237         u8 flags = type_flags >> 8;
1238         int err;
1239
1240         switch (type) {
1241         case PACKET_FANOUT_ROLLOVER:
1242                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1243                         return -EINVAL;
1244         case PACKET_FANOUT_HASH:
1245         case PACKET_FANOUT_LB:
1246         case PACKET_FANOUT_CPU:
1247                 break;
1248         default:
1249                 return -EINVAL;
1250         }
1251
1252         if (!po->running)
1253                 return -EINVAL;
1254
1255         if (po->fanout)
1256                 return -EALREADY;
1257
1258         mutex_lock(&fanout_mutex);
1259         match = NULL;
1260         list_for_each_entry(f, &fanout_list, list) {
1261                 if (f->id == id &&
1262                     read_pnet(&f->net) == sock_net(sk)) {
1263                         match = f;
1264                         break;
1265                 }
1266         }
1267         err = -EINVAL;
1268         if (match && match->flags != flags)
1269                 goto out;
1270         if (!match) {
1271                 err = -ENOMEM;
1272                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1273                 if (!match)
1274                         goto out;
1275                 write_pnet(&match->net, sock_net(sk));
1276                 match->id = id;
1277                 match->type = type;
1278                 match->flags = flags;
1279                 atomic_set(&match->rr_cur, 0);
1280                 INIT_LIST_HEAD(&match->list);
1281                 spin_lock_init(&match->lock);
1282                 atomic_set(&match->sk_ref, 0);
1283                 match->prot_hook.type = po->prot_hook.type;
1284                 match->prot_hook.dev = po->prot_hook.dev;
1285                 match->prot_hook.func = packet_rcv_fanout;
1286                 match->prot_hook.af_packet_priv = match;
1287                 match->prot_hook.id_match = match_fanout_group;
1288                 dev_add_pack(&match->prot_hook);
1289                 list_add(&match->list, &fanout_list);
1290         }
1291         err = -EINVAL;
1292         if (match->type == type &&
1293             match->prot_hook.type == po->prot_hook.type &&
1294             match->prot_hook.dev == po->prot_hook.dev) {
1295                 err = -ENOSPC;
1296                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1297                         __dev_remove_pack(&po->prot_hook);
1298                         po->fanout = match;
1299                         atomic_inc(&match->sk_ref);
1300                         __fanout_link(sk, po);
1301                         err = 0;
1302                 }
1303         }
1304 out:
1305         mutex_unlock(&fanout_mutex);
1306         return err;
1307 }
1308
1309 static void fanout_release(struct sock *sk)
1310 {
1311         struct packet_sock *po = pkt_sk(sk);
1312         struct packet_fanout *f;
1313
1314         f = po->fanout;
1315         if (!f)
1316                 return;
1317
1318         mutex_lock(&fanout_mutex);
1319         po->fanout = NULL;
1320
1321         if (atomic_dec_and_test(&f->sk_ref)) {
1322                 list_del(&f->list);
1323                 dev_remove_pack(&f->prot_hook);
1324                 kfree(f);
1325         }
1326         mutex_unlock(&fanout_mutex);
1327 }
1328
1329 static const struct proto_ops packet_ops;
1330
1331 static const struct proto_ops packet_ops_spkt;
1332
1333 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1334                            struct packet_type *pt, struct net_device *orig_dev)
1335 {
1336         struct sock *sk;
1337         struct sockaddr_pkt *spkt;
1338
1339         /*
1340          *      When we registered the protocol we saved the socket in the data
1341          *      field for just this event.
1342          */
1343
1344         sk = pt->af_packet_priv;
1345
1346         /*
1347          *      Yank back the headers [hope the device set this
1348          *      right or kerboom...]
1349          *
1350          *      Incoming packets have ll header pulled,
1351          *      push it back.
1352          *
1353          *      For outgoing ones skb->data == skb_mac_header(skb)
1354          *      so that this procedure is noop.
1355          */
1356
1357         if (skb->pkt_type == PACKET_LOOPBACK)
1358                 goto out;
1359
1360         if (!net_eq(dev_net(dev), sock_net(sk)))
1361                 goto out;
1362
1363         skb = skb_share_check(skb, GFP_ATOMIC);
1364         if (skb == NULL)
1365                 goto oom;
1366
1367         /* drop any routing info */
1368         skb_dst_drop(skb);
1369
1370         /* drop conntrack reference */
1371         nf_reset(skb);
1372
1373         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1374
1375         skb_push(skb, skb->data - skb_mac_header(skb));
1376
1377         /*
1378          *      The SOCK_PACKET socket receives _all_ frames.
1379          */
1380
1381         spkt->spkt_family = dev->type;
1382         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1383         spkt->spkt_protocol = skb->protocol;
1384
1385         /*
1386          *      Charge the memory to the socket. This is done specifically
1387          *      to prevent sockets using all the memory up.
1388          */
1389
1390         if (sock_queue_rcv_skb(sk, skb) == 0)
1391                 return 0;
1392
1393 out:
1394         kfree_skb(skb);
1395 oom:
1396         return 0;
1397 }
1398
1399
1400 /*
1401  *      Output a raw packet to a device layer. This bypasses all the other
1402  *      protocol layers and you must therefore supply it with a complete frame
1403  */
1404
1405 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1406                                struct msghdr *msg, size_t len)
1407 {
1408         struct sock *sk = sock->sk;
1409         struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1410         struct sk_buff *skb = NULL;
1411         struct net_device *dev;
1412         __be16 proto = 0;
1413         int err;
1414         int extra_len = 0;
1415
1416         /*
1417          *      Get and verify the address.
1418          */
1419
1420         if (saddr) {
1421                 if (msg->msg_namelen < sizeof(struct sockaddr))
1422                         return -EINVAL;
1423                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1424                         proto = saddr->spkt_protocol;
1425         } else
1426                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1427
1428         /*
1429          *      Find the device first to size check it
1430          */
1431
1432         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1433 retry:
1434         rcu_read_lock();
1435         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1436         err = -ENODEV;
1437         if (dev == NULL)
1438                 goto out_unlock;
1439
1440         err = -ENETDOWN;
1441         if (!(dev->flags & IFF_UP))
1442                 goto out_unlock;
1443
1444         /*
1445          * You may not queue a frame bigger than the mtu. This is the lowest level
1446          * raw protocol and you must do your own fragmentation at this level.
1447          */
1448
1449         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1450                 if (!netif_supports_nofcs(dev)) {
1451                         err = -EPROTONOSUPPORT;
1452                         goto out_unlock;
1453                 }
1454                 extra_len = 4; /* We're doing our own CRC */
1455         }
1456
1457         err = -EMSGSIZE;
1458         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1459                 goto out_unlock;
1460
1461         if (!skb) {
1462                 size_t reserved = LL_RESERVED_SPACE(dev);
1463                 int tlen = dev->needed_tailroom;
1464                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1465
1466                 rcu_read_unlock();
1467                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1468                 if (skb == NULL)
1469                         return -ENOBUFS;
1470                 /* FIXME: Save some space for broken drivers that write a hard
1471                  * header at transmission time by themselves. PPP is the notable
1472                  * one here. This should really be fixed at the driver level.
1473                  */
1474                 skb_reserve(skb, reserved);
1475                 skb_reset_network_header(skb);
1476
1477                 /* Try to align data part correctly */
1478                 if (hhlen) {
1479                         skb->data -= hhlen;
1480                         skb->tail -= hhlen;
1481                         if (len < hhlen)
1482                                 skb_reset_network_header(skb);
1483                 }
1484                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1485                 if (err)
1486                         goto out_free;
1487                 goto retry;
1488         }
1489
1490         if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1491                 /* Earlier code assumed this would be a VLAN pkt,
1492                  * double-check this now that we have the actual
1493                  * packet in hand.
1494                  */
1495                 struct ethhdr *ehdr;
1496                 skb_reset_mac_header(skb);
1497                 ehdr = eth_hdr(skb);
1498                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1499                         err = -EMSGSIZE;
1500                         goto out_unlock;
1501                 }
1502         }
1503
1504         skb->protocol = proto;
1505         skb->dev = dev;
1506         skb->priority = sk->sk_priority;
1507         skb->mark = sk->sk_mark;
1508         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1509         if (err < 0)
1510                 goto out_unlock;
1511
1512         if (unlikely(extra_len == 4))
1513                 skb->no_fcs = 1;
1514
1515         skb_probe_transport_header(skb, 0);
1516
1517         dev_queue_xmit(skb);
1518         rcu_read_unlock();
1519         return len;
1520
1521 out_unlock:
1522         rcu_read_unlock();
1523 out_free:
1524         kfree_skb(skb);
1525         return err;
1526 }
1527
1528 static unsigned int run_filter(const struct sk_buff *skb,
1529                                       const struct sock *sk,
1530                                       unsigned int res)
1531 {
1532         struct sk_filter *filter;
1533
1534         rcu_read_lock();
1535         filter = rcu_dereference(sk->sk_filter);
1536         if (filter != NULL)
1537                 res = SK_RUN_FILTER(filter, skb);
1538         rcu_read_unlock();
1539
1540         return res;
1541 }
1542
1543 /*
1544  * This function makes lazy skb cloning in hope that most of packets
1545  * are discarded by BPF.
1546  *
1547  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1548  * and skb->cb are mangled. It works because (and until) packets
1549  * falling here are owned by current CPU. Output packets are cloned
1550  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1551  * sequencially, so that if we return skb to original state on exit,
1552  * we will not harm anyone.
1553  */
1554
1555 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1556                       struct packet_type *pt, struct net_device *orig_dev)
1557 {
1558         struct sock *sk;
1559         struct sockaddr_ll *sll;
1560         struct packet_sock *po;
1561         u8 *skb_head = skb->data;
1562         int skb_len = skb->len;
1563         unsigned int snaplen, res;
1564
1565         if (skb->pkt_type == PACKET_LOOPBACK)
1566                 goto drop;
1567
1568         sk = pt->af_packet_priv;
1569         po = pkt_sk(sk);
1570
1571         if (!net_eq(dev_net(dev), sock_net(sk)))
1572                 goto drop;
1573
1574         skb->dev = dev;
1575
1576         if (dev->header_ops) {
1577                 /* The device has an explicit notion of ll header,
1578                  * exported to higher levels.
1579                  *
1580                  * Otherwise, the device hides details of its frame
1581                  * structure, so that corresponding packet head is
1582                  * never delivered to user.
1583                  */
1584                 if (sk->sk_type != SOCK_DGRAM)
1585                         skb_push(skb, skb->data - skb_mac_header(skb));
1586                 else if (skb->pkt_type == PACKET_OUTGOING) {
1587                         /* Special case: outgoing packets have ll header at head */
1588                         skb_pull(skb, skb_network_offset(skb));
1589                 }
1590         }
1591
1592         snaplen = skb->len;
1593
1594         res = run_filter(skb, sk, snaplen);
1595         if (!res)
1596                 goto drop_n_restore;
1597         if (snaplen > res)
1598                 snaplen = res;
1599
1600         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1601                 goto drop_n_acct;
1602
1603         if (skb_shared(skb)) {
1604                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1605                 if (nskb == NULL)
1606                         goto drop_n_acct;
1607
1608                 if (skb_head != skb->data) {
1609                         skb->data = skb_head;
1610                         skb->len = skb_len;
1611                 }
1612                 consume_skb(skb);
1613                 skb = nskb;
1614         }
1615
1616         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1617                      sizeof(skb->cb));
1618
1619         sll = &PACKET_SKB_CB(skb)->sa.ll;
1620         sll->sll_family = AF_PACKET;
1621         sll->sll_hatype = dev->type;
1622         sll->sll_protocol = skb->protocol;
1623         sll->sll_pkttype = skb->pkt_type;
1624         if (unlikely(po->origdev))
1625                 sll->sll_ifindex = orig_dev->ifindex;
1626         else
1627                 sll->sll_ifindex = dev->ifindex;
1628
1629         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1630
1631         PACKET_SKB_CB(skb)->origlen = skb->len;
1632
1633         if (pskb_trim(skb, snaplen))
1634                 goto drop_n_acct;
1635
1636         skb_set_owner_r(skb, sk);
1637         skb->dev = NULL;
1638         skb_dst_drop(skb);
1639
1640         /* drop conntrack reference */
1641         nf_reset(skb);
1642
1643         spin_lock(&sk->sk_receive_queue.lock);
1644         po->stats.tp_packets++;
1645         skb->dropcount = atomic_read(&sk->sk_drops);
1646         __skb_queue_tail(&sk->sk_receive_queue, skb);
1647         spin_unlock(&sk->sk_receive_queue.lock);
1648         sk->sk_data_ready(sk, skb->len);
1649         return 0;
1650
1651 drop_n_acct:
1652         spin_lock(&sk->sk_receive_queue.lock);
1653         po->stats.tp_drops++;
1654         atomic_inc(&sk->sk_drops);
1655         spin_unlock(&sk->sk_receive_queue.lock);
1656
1657 drop_n_restore:
1658         if (skb_head != skb->data && skb_shared(skb)) {
1659                 skb->data = skb_head;
1660                 skb->len = skb_len;
1661         }
1662 drop:
1663         consume_skb(skb);
1664         return 0;
1665 }
1666
1667 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1668                        struct packet_type *pt, struct net_device *orig_dev)
1669 {
1670         struct sock *sk;
1671         struct packet_sock *po;
1672         struct sockaddr_ll *sll;
1673         union {
1674                 struct tpacket_hdr *h1;
1675                 struct tpacket2_hdr *h2;
1676                 struct tpacket3_hdr *h3;
1677                 void *raw;
1678         } h;
1679         u8 *skb_head = skb->data;
1680         int skb_len = skb->len;
1681         unsigned int snaplen, res;
1682         unsigned long status = TP_STATUS_USER;
1683         unsigned short macoff, netoff, hdrlen;
1684         struct sk_buff *copy_skb = NULL;
1685         struct timeval tv;
1686         struct timespec ts;
1687         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1688
1689         if (skb->pkt_type == PACKET_LOOPBACK)
1690                 goto drop;
1691
1692         sk = pt->af_packet_priv;
1693         po = pkt_sk(sk);
1694
1695         if (!net_eq(dev_net(dev), sock_net(sk)))
1696                 goto drop;
1697
1698         if (dev->header_ops) {
1699                 if (sk->sk_type != SOCK_DGRAM)
1700                         skb_push(skb, skb->data - skb_mac_header(skb));
1701                 else if (skb->pkt_type == PACKET_OUTGOING) {
1702                         /* Special case: outgoing packets have ll header at head */
1703                         skb_pull(skb, skb_network_offset(skb));
1704                 }
1705         }
1706
1707         if (skb->ip_summed == CHECKSUM_PARTIAL)
1708                 status |= TP_STATUS_CSUMNOTREADY;
1709
1710         snaplen = skb->len;
1711
1712         res = run_filter(skb, sk, snaplen);
1713         if (!res)
1714                 goto drop_n_restore;
1715         if (snaplen > res)
1716                 snaplen = res;
1717
1718         if (sk->sk_type == SOCK_DGRAM) {
1719                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1720                                   po->tp_reserve;
1721         } else {
1722                 unsigned int maclen = skb_network_offset(skb);
1723                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1724                                        (maclen < 16 ? 16 : maclen)) +
1725                         po->tp_reserve;
1726                 macoff = netoff - maclen;
1727         }
1728         if (po->tp_version <= TPACKET_V2) {
1729                 if (macoff + snaplen > po->rx_ring.frame_size) {
1730                         if (po->copy_thresh &&
1731                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1732                                 if (skb_shared(skb)) {
1733                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1734                                 } else {
1735                                         copy_skb = skb_get(skb);
1736                                         skb_head = skb->data;
1737                                 }
1738                                 if (copy_skb)
1739                                         skb_set_owner_r(copy_skb, sk);
1740                         }
1741                         snaplen = po->rx_ring.frame_size - macoff;
1742                         if ((int)snaplen < 0)
1743                                 snaplen = 0;
1744                 }
1745         }
1746         spin_lock(&sk->sk_receive_queue.lock);
1747         h.raw = packet_current_rx_frame(po, skb,
1748                                         TP_STATUS_KERNEL, (macoff+snaplen));
1749         if (!h.raw)
1750                 goto ring_is_full;
1751         if (po->tp_version <= TPACKET_V2) {
1752                 packet_increment_rx_head(po, &po->rx_ring);
1753         /*
1754          * LOSING will be reported till you read the stats,
1755          * because it's COR - Clear On Read.
1756          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1757          * at packet level.
1758          */
1759                 if (po->stats.tp_drops)
1760                         status |= TP_STATUS_LOSING;
1761         }
1762         po->stats.tp_packets++;
1763         if (copy_skb) {
1764                 status |= TP_STATUS_COPY;
1765                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1766         }
1767         spin_unlock(&sk->sk_receive_queue.lock);
1768
1769         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1770
1771         switch (po->tp_version) {
1772         case TPACKET_V1:
1773                 h.h1->tp_len = skb->len;
1774                 h.h1->tp_snaplen = snaplen;
1775                 h.h1->tp_mac = macoff;
1776                 h.h1->tp_net = netoff;
1777                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1778                                 && shhwtstamps->syststamp.tv64)
1779                         tv = ktime_to_timeval(shhwtstamps->syststamp);
1780                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1781                                 && shhwtstamps->hwtstamp.tv64)
1782                         tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1783                 else if (skb->tstamp.tv64)
1784                         tv = ktime_to_timeval(skb->tstamp);
1785                 else
1786                         do_gettimeofday(&tv);
1787                 h.h1->tp_sec = tv.tv_sec;
1788                 h.h1->tp_usec = tv.tv_usec;
1789                 hdrlen = sizeof(*h.h1);
1790                 break;
1791         case TPACKET_V2:
1792                 h.h2->tp_len = skb->len;
1793                 h.h2->tp_snaplen = snaplen;
1794                 h.h2->tp_mac = macoff;
1795                 h.h2->tp_net = netoff;
1796                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1797                                 && shhwtstamps->syststamp.tv64)
1798                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1799                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1800                                 && shhwtstamps->hwtstamp.tv64)
1801                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1802                 else if (skb->tstamp.tv64)
1803                         ts = ktime_to_timespec(skb->tstamp);
1804                 else
1805                         getnstimeofday(&ts);
1806                 h.h2->tp_sec = ts.tv_sec;
1807                 h.h2->tp_nsec = ts.tv_nsec;
1808                 if (vlan_tx_tag_present(skb)) {
1809                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1810                         status |= TP_STATUS_VLAN_VALID;
1811                 } else {
1812                         h.h2->tp_vlan_tci = 0;
1813                 }
1814                 h.h2->tp_padding = 0;
1815                 hdrlen = sizeof(*h.h2);
1816                 break;
1817         case TPACKET_V3:
1818                 /* tp_nxt_offset,vlan are already populated above.
1819                  * So DONT clear those fields here
1820                  */
1821                 h.h3->tp_status |= status;
1822                 h.h3->tp_len = skb->len;
1823                 h.h3->tp_snaplen = snaplen;
1824                 h.h3->tp_mac = macoff;
1825                 h.h3->tp_net = netoff;
1826                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1827                                 && shhwtstamps->syststamp.tv64)
1828                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1829                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1830                                 && shhwtstamps->hwtstamp.tv64)
1831                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1832                 else if (skb->tstamp.tv64)
1833                         ts = ktime_to_timespec(skb->tstamp);
1834                 else
1835                         getnstimeofday(&ts);
1836                 h.h3->tp_sec  = ts.tv_sec;
1837                 h.h3->tp_nsec = ts.tv_nsec;
1838                 hdrlen = sizeof(*h.h3);
1839                 break;
1840         default:
1841                 BUG();
1842         }
1843
1844         sll = h.raw + TPACKET_ALIGN(hdrlen);
1845         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1846         sll->sll_family = AF_PACKET;
1847         sll->sll_hatype = dev->type;
1848         sll->sll_protocol = skb->protocol;
1849         sll->sll_pkttype = skb->pkt_type;
1850         if (unlikely(po->origdev))
1851                 sll->sll_ifindex = orig_dev->ifindex;
1852         else
1853                 sll->sll_ifindex = dev->ifindex;
1854
1855         smp_mb();
1856 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1857         {
1858                 u8 *start, *end;
1859
1860                 if (po->tp_version <= TPACKET_V2) {
1861                         end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1862                                 + macoff + snaplen);
1863                         for (start = h.raw; start < end; start += PAGE_SIZE)
1864                                 flush_dcache_page(pgv_to_page(start));
1865                 }
1866                 smp_wmb();
1867         }
1868 #endif
1869         if (po->tp_version <= TPACKET_V2)
1870                 __packet_set_status(po, h.raw, status);
1871         else
1872                 prb_clear_blk_fill_status(&po->rx_ring);
1873
1874         sk->sk_data_ready(sk, 0);
1875
1876 drop_n_restore:
1877         if (skb_head != skb->data && skb_shared(skb)) {
1878                 skb->data = skb_head;
1879                 skb->len = skb_len;
1880         }
1881 drop:
1882         kfree_skb(skb);
1883         return 0;
1884
1885 ring_is_full:
1886         po->stats.tp_drops++;
1887         spin_unlock(&sk->sk_receive_queue.lock);
1888
1889         sk->sk_data_ready(sk, 0);
1890         kfree_skb(copy_skb);
1891         goto drop_n_restore;
1892 }
1893
1894 static void tpacket_destruct_skb(struct sk_buff *skb)
1895 {
1896         struct packet_sock *po = pkt_sk(skb->sk);
1897         void *ph;
1898
1899         if (likely(po->tx_ring.pg_vec)) {
1900                 ph = skb_shinfo(skb)->destructor_arg;
1901                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1902                 atomic_dec(&po->tx_ring.pending);
1903                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1904         }
1905
1906         sock_wfree(skb);
1907 }
1908
1909 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1910                 void *frame, struct net_device *dev, int size_max,
1911                 __be16 proto, unsigned char *addr, int hlen)
1912 {
1913         union {
1914                 struct tpacket_hdr *h1;
1915                 struct tpacket2_hdr *h2;
1916                 void *raw;
1917         } ph;
1918         int to_write, offset, len, tp_len, nr_frags, len_max;
1919         struct socket *sock = po->sk.sk_socket;
1920         struct page *page;
1921         void *data;
1922         int err;
1923
1924         ph.raw = frame;
1925
1926         skb->protocol = proto;
1927         skb->dev = dev;
1928         skb->priority = po->sk.sk_priority;
1929         skb->mark = po->sk.sk_mark;
1930         skb_shinfo(skb)->destructor_arg = ph.raw;
1931
1932         switch (po->tp_version) {
1933         case TPACKET_V2:
1934                 tp_len = ph.h2->tp_len;
1935                 break;
1936         default:
1937                 tp_len = ph.h1->tp_len;
1938                 break;
1939         }
1940         if (unlikely(tp_len > size_max)) {
1941                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1942                 return -EMSGSIZE;
1943         }
1944
1945         skb_reserve(skb, hlen);
1946         skb_reset_network_header(skb);
1947         skb_probe_transport_header(skb, 0);
1948
1949         if (po->tp_tx_has_off) {
1950                 int off_min, off_max, off;
1951                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
1952                 off_max = po->tx_ring.frame_size - tp_len;
1953                 if (sock->type == SOCK_DGRAM) {
1954                         switch (po->tp_version) {
1955                         case TPACKET_V2:
1956                                 off = ph.h2->tp_net;
1957                                 break;
1958                         default:
1959                                 off = ph.h1->tp_net;
1960                                 break;
1961                         }
1962                 } else {
1963                         switch (po->tp_version) {
1964                         case TPACKET_V2:
1965                                 off = ph.h2->tp_mac;
1966                                 break;
1967                         default:
1968                                 off = ph.h1->tp_mac;
1969                                 break;
1970                         }
1971                 }
1972                 if (unlikely((off < off_min) || (off_max < off)))
1973                         return -EINVAL;
1974                 data = ph.raw + off;
1975         } else {
1976                 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1977         }
1978         to_write = tp_len;
1979
1980         if (sock->type == SOCK_DGRAM) {
1981                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
1982                                 NULL, tp_len);
1983                 if (unlikely(err < 0))
1984                         return -EINVAL;
1985         } else if (dev->hard_header_len) {
1986                 /* net device doesn't like empty head */
1987                 if (unlikely(tp_len <= dev->hard_header_len)) {
1988                         pr_err("packet size is too short (%d < %d)\n",
1989                                tp_len, dev->hard_header_len);
1990                         return -EINVAL;
1991                 }
1992
1993                 skb_push(skb, dev->hard_header_len);
1994                 err = skb_store_bits(skb, 0, data,
1995                                 dev->hard_header_len);
1996                 if (unlikely(err))
1997                         return err;
1998
1999                 data += dev->hard_header_len;
2000                 to_write -= dev->hard_header_len;
2001         }
2002
2003         offset = offset_in_page(data);
2004         len_max = PAGE_SIZE - offset;
2005         len = ((to_write > len_max) ? len_max : to_write);
2006
2007         skb->data_len = to_write;
2008         skb->len += to_write;
2009         skb->truesize += to_write;
2010         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2011
2012         while (likely(to_write)) {
2013                 nr_frags = skb_shinfo(skb)->nr_frags;
2014
2015                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2016                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2017                                MAX_SKB_FRAGS);
2018                         return -EFAULT;
2019                 }
2020
2021                 page = pgv_to_page(data);
2022                 data += len;
2023                 flush_dcache_page(page);
2024                 get_page(page);
2025                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2026                 to_write -= len;
2027                 offset = 0;
2028                 len_max = PAGE_SIZE;
2029                 len = ((to_write > len_max) ? len_max : to_write);
2030         }
2031
2032         return tp_len;
2033 }
2034
2035 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2036 {
2037         struct sk_buff *skb;
2038         struct net_device *dev;
2039         __be16 proto;
2040         bool need_rls_dev = false;
2041         int err, reserve = 0;
2042         void *ph;
2043         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2044         int tp_len, size_max;
2045         unsigned char *addr;
2046         int len_sum = 0;
2047         int status = TP_STATUS_AVAILABLE;
2048         int hlen, tlen;
2049
2050         mutex_lock(&po->pg_vec_lock);
2051
2052         if (saddr == NULL) {
2053                 dev = po->prot_hook.dev;
2054                 proto   = po->num;
2055                 addr    = NULL;
2056         } else {
2057                 err = -EINVAL;
2058                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2059                         goto out;
2060                 if (msg->msg_namelen < (saddr->sll_halen
2061                                         + offsetof(struct sockaddr_ll,
2062                                                 sll_addr)))
2063                         goto out;
2064                 proto   = saddr->sll_protocol;
2065                 addr    = saddr->sll_addr;
2066                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2067                 need_rls_dev = true;
2068         }
2069
2070         err = -ENXIO;
2071         if (unlikely(dev == NULL))
2072                 goto out;
2073
2074         reserve = dev->hard_header_len;
2075
2076         err = -ENETDOWN;
2077         if (unlikely(!(dev->flags & IFF_UP)))
2078                 goto out_put;
2079
2080         size_max = po->tx_ring.frame_size
2081                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2082
2083         if (size_max > dev->mtu + reserve)
2084                 size_max = dev->mtu + reserve;
2085
2086         do {
2087                 ph = packet_current_frame(po, &po->tx_ring,
2088                                 TP_STATUS_SEND_REQUEST);
2089
2090                 if (unlikely(ph == NULL)) {
2091                         schedule();
2092                         continue;
2093                 }
2094
2095                 status = TP_STATUS_SEND_REQUEST;
2096                 hlen = LL_RESERVED_SPACE(dev);
2097                 tlen = dev->needed_tailroom;
2098                 skb = sock_alloc_send_skb(&po->sk,
2099                                 hlen + tlen + sizeof(struct sockaddr_ll),
2100                                 0, &err);
2101
2102                 if (unlikely(skb == NULL))
2103                         goto out_status;
2104
2105                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2106                                 addr, hlen);
2107
2108                 if (unlikely(tp_len < 0)) {
2109                         if (po->tp_loss) {
2110                                 __packet_set_status(po, ph,
2111                                                 TP_STATUS_AVAILABLE);
2112                                 packet_increment_head(&po->tx_ring);
2113                                 kfree_skb(skb);
2114                                 continue;
2115                         } else {
2116                                 status = TP_STATUS_WRONG_FORMAT;
2117                                 err = tp_len;
2118                                 goto out_status;
2119                         }
2120                 }
2121
2122                 skb->destructor = tpacket_destruct_skb;
2123                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2124                 atomic_inc(&po->tx_ring.pending);
2125
2126                 status = TP_STATUS_SEND_REQUEST;
2127                 err = dev_queue_xmit(skb);
2128                 if (unlikely(err > 0)) {
2129                         err = net_xmit_errno(err);
2130                         if (err && __packet_get_status(po, ph) ==
2131                                    TP_STATUS_AVAILABLE) {
2132                                 /* skb was destructed already */
2133                                 skb = NULL;
2134                                 goto out_status;
2135                         }
2136                         /*
2137                          * skb was dropped but not destructed yet;
2138                          * let's treat it like congestion or err < 0
2139                          */
2140                         err = 0;
2141                 }
2142                 packet_increment_head(&po->tx_ring);
2143                 len_sum += tp_len;
2144         } while (likely((ph != NULL) ||
2145                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2146                          (atomic_read(&po->tx_ring.pending))))
2147                 );
2148
2149         err = len_sum;
2150         goto out_put;
2151
2152 out_status:
2153         __packet_set_status(po, ph, status);
2154         kfree_skb(skb);
2155 out_put:
2156         if (need_rls_dev)
2157                 dev_put(dev);
2158 out:
2159         mutex_unlock(&po->pg_vec_lock);
2160         return err;
2161 }
2162
2163 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2164                                         size_t reserve, size_t len,
2165                                         size_t linear, int noblock,
2166                                         int *err)
2167 {
2168         struct sk_buff *skb;
2169
2170         /* Under a page?  Don't bother with paged skb. */
2171         if (prepad + len < PAGE_SIZE || !linear)
2172                 linear = len;
2173
2174         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2175                                    err);
2176         if (!skb)
2177                 return NULL;
2178
2179         skb_reserve(skb, reserve);
2180         skb_put(skb, linear);
2181         skb->data_len = len - linear;
2182         skb->len += len - linear;
2183
2184         return skb;
2185 }
2186
2187 static int packet_snd(struct socket *sock,
2188                           struct msghdr *msg, size_t len)
2189 {
2190         struct sock *sk = sock->sk;
2191         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2192         struct sk_buff *skb;
2193         struct net_device *dev;
2194         __be16 proto;
2195         bool need_rls_dev = false;
2196         unsigned char *addr;
2197         int err, reserve = 0;
2198         struct virtio_net_hdr vnet_hdr = { 0 };
2199         int offset = 0;
2200         int vnet_hdr_len;
2201         struct packet_sock *po = pkt_sk(sk);
2202         unsigned short gso_type = 0;
2203         int hlen, tlen;
2204         int extra_len = 0;
2205
2206         /*
2207          *      Get and verify the address.
2208          */
2209
2210         if (saddr == NULL) {
2211                 dev = po->prot_hook.dev;
2212                 proto   = po->num;
2213                 addr    = NULL;
2214         } else {
2215                 err = -EINVAL;
2216                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2217                         goto out;
2218                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2219                         goto out;
2220                 proto   = saddr->sll_protocol;
2221                 addr    = saddr->sll_addr;
2222                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2223                 need_rls_dev = true;
2224         }
2225
2226         err = -ENXIO;
2227         if (dev == NULL)
2228                 goto out_unlock;
2229         if (sock->type == SOCK_RAW)
2230                 reserve = dev->hard_header_len;
2231
2232         err = -ENETDOWN;
2233         if (!(dev->flags & IFF_UP))
2234                 goto out_unlock;
2235
2236         if (po->has_vnet_hdr) {
2237                 vnet_hdr_len = sizeof(vnet_hdr);
2238
2239                 err = -EINVAL;
2240                 if (len < vnet_hdr_len)
2241                         goto out_unlock;
2242
2243                 len -= vnet_hdr_len;
2244
2245                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2246                                        vnet_hdr_len);
2247                 if (err < 0)
2248                         goto out_unlock;
2249
2250                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2251                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2252                       vnet_hdr.hdr_len))
2253                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
2254                                                  vnet_hdr.csum_offset + 2;
2255
2256                 err = -EINVAL;
2257                 if (vnet_hdr.hdr_len > len)
2258                         goto out_unlock;
2259
2260                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2261                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2262                         case VIRTIO_NET_HDR_GSO_TCPV4:
2263                                 gso_type = SKB_GSO_TCPV4;
2264                                 break;
2265                         case VIRTIO_NET_HDR_GSO_TCPV6:
2266                                 gso_type = SKB_GSO_TCPV6;
2267                                 break;
2268                         case VIRTIO_NET_HDR_GSO_UDP:
2269                                 gso_type = SKB_GSO_UDP;
2270                                 break;
2271                         default:
2272                                 goto out_unlock;
2273                         }
2274
2275                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2276                                 gso_type |= SKB_GSO_TCP_ECN;
2277
2278                         if (vnet_hdr.gso_size == 0)
2279                                 goto out_unlock;
2280
2281                 }
2282         }
2283
2284         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2285                 if (!netif_supports_nofcs(dev)) {
2286                         err = -EPROTONOSUPPORT;
2287                         goto out_unlock;
2288                 }
2289                 extra_len = 4; /* We're doing our own CRC */
2290         }
2291
2292         err = -EMSGSIZE;
2293         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2294                 goto out_unlock;
2295
2296         err = -ENOBUFS;
2297         hlen = LL_RESERVED_SPACE(dev);
2298         tlen = dev->needed_tailroom;
2299         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2300                                msg->msg_flags & MSG_DONTWAIT, &err);
2301         if (skb == NULL)
2302                 goto out_unlock;
2303
2304         skb_set_network_header(skb, reserve);
2305
2306         err = -EINVAL;
2307         if (sock->type == SOCK_DGRAM &&
2308             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2309                 goto out_free;
2310
2311         /* Returns -EFAULT on error */
2312         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2313         if (err)
2314                 goto out_free;
2315         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2316         if (err < 0)
2317                 goto out_free;
2318
2319         if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2320                 /* Earlier code assumed this would be a VLAN pkt,
2321                  * double-check this now that we have the actual
2322                  * packet in hand.
2323                  */
2324                 struct ethhdr *ehdr;
2325                 skb_reset_mac_header(skb);
2326                 ehdr = eth_hdr(skb);
2327                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2328                         err = -EMSGSIZE;
2329                         goto out_free;
2330                 }
2331         }
2332
2333         skb->protocol = proto;
2334         skb->dev = dev;
2335         skb->priority = sk->sk_priority;
2336         skb->mark = sk->sk_mark;
2337
2338         if (po->has_vnet_hdr) {
2339                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2340                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2341                                                   vnet_hdr.csum_offset)) {
2342                                 err = -EINVAL;
2343                                 goto out_free;
2344                         }
2345                 }
2346
2347                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2348                 skb_shinfo(skb)->gso_type = gso_type;
2349
2350                 /* Header must be checked, and gso_segs computed. */
2351                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2352                 skb_shinfo(skb)->gso_segs = 0;
2353
2354                 len += vnet_hdr_len;
2355         }
2356
2357         skb_probe_transport_header(skb, reserve);
2358
2359         if (unlikely(extra_len == 4))
2360                 skb->no_fcs = 1;
2361
2362         /*
2363          *      Now send it
2364          */
2365
2366         err = dev_queue_xmit(skb);
2367         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2368                 goto out_unlock;
2369
2370         if (need_rls_dev)
2371                 dev_put(dev);
2372
2373         return len;
2374
2375 out_free:
2376         kfree_skb(skb);
2377 out_unlock:
2378         if (dev && need_rls_dev)
2379                 dev_put(dev);
2380 out:
2381         return err;
2382 }
2383
2384 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2385                 struct msghdr *msg, size_t len)
2386 {
2387         struct sock *sk = sock->sk;
2388         struct packet_sock *po = pkt_sk(sk);
2389         if (po->tx_ring.pg_vec)
2390                 return tpacket_snd(po, msg);
2391         else
2392                 return packet_snd(sock, msg, len);
2393 }
2394
2395 /*
2396  *      Close a PACKET socket. This is fairly simple. We immediately go
2397  *      to 'closed' state and remove our protocol entry in the device list.
2398  */
2399
2400 static int packet_release(struct socket *sock)
2401 {
2402         struct sock *sk = sock->sk;
2403         struct packet_sock *po;
2404         struct net *net;
2405         union tpacket_req_u req_u;
2406
2407         if (!sk)
2408                 return 0;
2409
2410         net = sock_net(sk);
2411         po = pkt_sk(sk);
2412
2413         mutex_lock(&net->packet.sklist_lock);
2414         sk_del_node_init_rcu(sk);
2415         mutex_unlock(&net->packet.sklist_lock);
2416
2417         preempt_disable();
2418         sock_prot_inuse_add(net, sk->sk_prot, -1);
2419         preempt_enable();
2420
2421         spin_lock(&po->bind_lock);
2422         unregister_prot_hook(sk, false);
2423         if (po->prot_hook.dev) {
2424                 dev_put(po->prot_hook.dev);
2425                 po->prot_hook.dev = NULL;
2426         }
2427         spin_unlock(&po->bind_lock);
2428
2429         packet_flush_mclist(sk);
2430
2431         if (po->rx_ring.pg_vec) {
2432                 memset(&req_u, 0, sizeof(req_u));
2433                 packet_set_ring(sk, &req_u, 1, 0);
2434         }
2435
2436         if (po->tx_ring.pg_vec) {
2437                 memset(&req_u, 0, sizeof(req_u));
2438                 packet_set_ring(sk, &req_u, 1, 1);
2439         }
2440
2441         fanout_release(sk);
2442
2443         synchronize_net();
2444         /*
2445          *      Now the socket is dead. No more input will appear.
2446          */
2447         sock_orphan(sk);
2448         sock->sk = NULL;
2449
2450         /* Purge queues */
2451
2452         skb_queue_purge(&sk->sk_receive_queue);
2453         sk_refcnt_debug_release(sk);
2454
2455         sock_put(sk);
2456         return 0;
2457 }
2458
2459 /*
2460  *      Attach a packet hook.
2461  */
2462
2463 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2464 {
2465         struct packet_sock *po = pkt_sk(sk);
2466
2467         if (po->fanout) {
2468                 if (dev)
2469                         dev_put(dev);
2470
2471                 return -EINVAL;
2472         }
2473
2474         lock_sock(sk);
2475
2476         spin_lock(&po->bind_lock);
2477         unregister_prot_hook(sk, true);
2478         po->num = protocol;
2479         po->prot_hook.type = protocol;
2480         if (po->prot_hook.dev)
2481                 dev_put(po->prot_hook.dev);
2482         po->prot_hook.dev = dev;
2483
2484         po->ifindex = dev ? dev->ifindex : 0;
2485
2486         if (protocol == 0)
2487                 goto out_unlock;
2488
2489         if (!dev || (dev->flags & IFF_UP)) {
2490                 register_prot_hook(sk);
2491         } else {
2492                 sk->sk_err = ENETDOWN;
2493                 if (!sock_flag(sk, SOCK_DEAD))
2494                         sk->sk_error_report(sk);
2495         }
2496
2497 out_unlock:
2498         spin_unlock(&po->bind_lock);
2499         release_sock(sk);
2500         return 0;
2501 }
2502
2503 /*
2504  *      Bind a packet socket to a device
2505  */
2506
2507 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2508                             int addr_len)
2509 {
2510         struct sock *sk = sock->sk;
2511         char name[15];
2512         struct net_device *dev;
2513         int err = -ENODEV;
2514
2515         /*
2516          *      Check legality
2517          */
2518
2519         if (addr_len != sizeof(struct sockaddr))
2520                 return -EINVAL;
2521         strlcpy(name, uaddr->sa_data, sizeof(name));
2522
2523         dev = dev_get_by_name(sock_net(sk), name);
2524         if (dev)
2525                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2526         return err;
2527 }
2528
2529 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2530 {
2531         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2532         struct sock *sk = sock->sk;
2533         struct net_device *dev = NULL;
2534         int err;
2535
2536
2537         /*
2538          *      Check legality
2539          */
2540
2541         if (addr_len < sizeof(struct sockaddr_ll))
2542                 return -EINVAL;
2543         if (sll->sll_family != AF_PACKET)
2544                 return -EINVAL;
2545
2546         if (sll->sll_ifindex) {
2547                 err = -ENODEV;
2548                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2549                 if (dev == NULL)
2550                         goto out;
2551         }
2552         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2553
2554 out:
2555         return err;
2556 }
2557
2558 static struct proto packet_proto = {
2559         .name     = "PACKET",
2560         .owner    = THIS_MODULE,
2561         .obj_size = sizeof(struct packet_sock),
2562 };
2563
2564 /*
2565  *      Create a packet of type SOCK_PACKET.
2566  */
2567
2568 static int packet_create(struct net *net, struct socket *sock, int protocol,
2569                          int kern)
2570 {
2571         struct sock *sk;
2572         struct packet_sock *po;
2573         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2574         int err;
2575
2576         if (!ns_capable(net->user_ns, CAP_NET_RAW))
2577                 return -EPERM;
2578         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2579             sock->type != SOCK_PACKET)
2580                 return -ESOCKTNOSUPPORT;
2581
2582         sock->state = SS_UNCONNECTED;
2583
2584         err = -ENOBUFS;
2585         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2586         if (sk == NULL)
2587                 goto out;
2588
2589         sock->ops = &packet_ops;
2590         if (sock->type == SOCK_PACKET)
2591                 sock->ops = &packet_ops_spkt;
2592
2593         sock_init_data(sock, sk);
2594
2595         po = pkt_sk(sk);
2596         sk->sk_family = PF_PACKET;
2597         po->num = proto;
2598
2599         sk->sk_destruct = packet_sock_destruct;
2600         sk_refcnt_debug_inc(sk);
2601
2602         /*
2603          *      Attach a protocol block
2604          */
2605
2606         spin_lock_init(&po->bind_lock);
2607         mutex_init(&po->pg_vec_lock);
2608         po->prot_hook.func = packet_rcv;
2609
2610         if (sock->type == SOCK_PACKET)
2611                 po->prot_hook.func = packet_rcv_spkt;
2612
2613         po->prot_hook.af_packet_priv = sk;
2614
2615         if (proto) {
2616                 po->prot_hook.type = proto;
2617                 register_prot_hook(sk);
2618         }
2619
2620         mutex_lock(&net->packet.sklist_lock);
2621         sk_add_node_rcu(sk, &net->packet.sklist);
2622         mutex_unlock(&net->packet.sklist_lock);
2623
2624         preempt_disable();
2625         sock_prot_inuse_add(net, &packet_proto, 1);
2626         preempt_enable();
2627
2628         return 0;
2629 out:
2630         return err;
2631 }
2632
2633 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2634 {
2635         struct sock_exterr_skb *serr;
2636         struct sk_buff *skb, *skb2;
2637         int copied, err;
2638
2639         err = -EAGAIN;
2640         skb = skb_dequeue(&sk->sk_error_queue);
2641         if (skb == NULL)
2642                 goto out;
2643
2644         copied = skb->len;
2645         if (copied > len) {
2646                 msg->msg_flags |= MSG_TRUNC;
2647                 copied = len;
2648         }
2649         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2650         if (err)
2651                 goto out_free_skb;
2652
2653         sock_recv_timestamp(msg, sk, skb);
2654
2655         serr = SKB_EXT_ERR(skb);
2656         put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2657                  sizeof(serr->ee), &serr->ee);
2658
2659         msg->msg_flags |= MSG_ERRQUEUE;
2660         err = copied;
2661
2662         /* Reset and regenerate socket error */
2663         spin_lock_bh(&sk->sk_error_queue.lock);
2664         sk->sk_err = 0;
2665         if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2666                 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2667                 spin_unlock_bh(&sk->sk_error_queue.lock);
2668                 sk->sk_error_report(sk);
2669         } else
2670                 spin_unlock_bh(&sk->sk_error_queue.lock);
2671
2672 out_free_skb:
2673         kfree_skb(skb);
2674 out:
2675         return err;
2676 }
2677
2678 /*
2679  *      Pull a packet from our receive queue and hand it to the user.
2680  *      If necessary we block.
2681  */
2682
2683 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2684                           struct msghdr *msg, size_t len, int flags)
2685 {
2686         struct sock *sk = sock->sk;
2687         struct sk_buff *skb;
2688         int copied, err;
2689         struct sockaddr_ll *sll;
2690         int vnet_hdr_len = 0;
2691
2692         err = -EINVAL;
2693         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2694                 goto out;
2695
2696 #if 0
2697         /* What error should we return now? EUNATTACH? */
2698         if (pkt_sk(sk)->ifindex < 0)
2699                 return -ENODEV;
2700 #endif
2701
2702         if (flags & MSG_ERRQUEUE) {
2703                 err = packet_recv_error(sk, msg, len);
2704                 goto out;
2705         }
2706
2707         /*
2708          *      Call the generic datagram receiver. This handles all sorts
2709          *      of horrible races and re-entrancy so we can forget about it
2710          *      in the protocol layers.
2711          *
2712          *      Now it will return ENETDOWN, if device have just gone down,
2713          *      but then it will block.
2714          */
2715
2716         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2717
2718         /*
2719          *      An error occurred so return it. Because skb_recv_datagram()
2720          *      handles the blocking we don't see and worry about blocking
2721          *      retries.
2722          */
2723
2724         if (skb == NULL)
2725                 goto out;
2726
2727         if (pkt_sk(sk)->has_vnet_hdr) {
2728                 struct virtio_net_hdr vnet_hdr = { 0 };
2729
2730                 err = -EINVAL;
2731                 vnet_hdr_len = sizeof(vnet_hdr);
2732                 if (len < vnet_hdr_len)
2733                         goto out_free;
2734
2735                 len -= vnet_hdr_len;
2736
2737                 if (skb_is_gso(skb)) {
2738                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2739
2740                         /* This is a hint as to how much should be linear. */
2741                         vnet_hdr.hdr_len = skb_headlen(skb);
2742                         vnet_hdr.gso_size = sinfo->gso_size;
2743                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2744                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2745                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2746                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2747                         else if (sinfo->gso_type & SKB_GSO_UDP)
2748                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2749                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2750                                 goto out_free;
2751                         else
2752                                 BUG();
2753                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2754                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2755                 } else
2756                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2757
2758                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2759                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2760                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2761                         vnet_hdr.csum_offset = skb->csum_offset;
2762                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2763                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2764                 } /* else everything is zero */
2765
2766                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2767                                      vnet_hdr_len);
2768                 if (err < 0)
2769                         goto out_free;
2770         }
2771
2772         /*
2773          *      If the address length field is there to be filled in, we fill
2774          *      it in now.
2775          */
2776
2777         sll = &PACKET_SKB_CB(skb)->sa.ll;
2778         if (sock->type == SOCK_PACKET)
2779                 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2780         else
2781                 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2782
2783         /*
2784          *      You lose any data beyond the buffer you gave. If it worries a
2785          *      user program they can ask the device for its MTU anyway.
2786          */
2787
2788         copied = skb->len;
2789         if (copied > len) {
2790                 copied = len;
2791                 msg->msg_flags |= MSG_TRUNC;
2792         }
2793
2794         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2795         if (err)
2796                 goto out_free;
2797
2798         sock_recv_ts_and_drops(msg, sk, skb);
2799
2800         if (msg->msg_name)
2801                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2802                        msg->msg_namelen);
2803
2804         if (pkt_sk(sk)->auxdata) {
2805                 struct tpacket_auxdata aux;
2806
2807                 aux.tp_status = TP_STATUS_USER;
2808                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2809                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2810                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2811                 aux.tp_snaplen = skb->len;
2812                 aux.tp_mac = 0;
2813                 aux.tp_net = skb_network_offset(skb);
2814                 if (vlan_tx_tag_present(skb)) {
2815                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2816                         aux.tp_status |= TP_STATUS_VLAN_VALID;
2817                 } else {
2818                         aux.tp_vlan_tci = 0;
2819                 }
2820                 aux.tp_padding = 0;
2821                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2822         }
2823
2824         /*
2825          *      Free or return the buffer as appropriate. Again this
2826          *      hides all the races and re-entrancy issues from us.
2827          */
2828         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2829
2830 out_free:
2831         skb_free_datagram(sk, skb);
2832 out:
2833         return err;
2834 }
2835
2836 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2837                                int *uaddr_len, int peer)
2838 {
2839         struct net_device *dev;
2840         struct sock *sk = sock->sk;
2841
2842         if (peer)
2843                 return -EOPNOTSUPP;
2844
2845         uaddr->sa_family = AF_PACKET;
2846         rcu_read_lock();
2847         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2848         if (dev)
2849                 strncpy(uaddr->sa_data, dev->name, 14);
2850         else
2851                 memset(uaddr->sa_data, 0, 14);
2852         rcu_read_unlock();
2853         *uaddr_len = sizeof(*uaddr);
2854
2855         return 0;
2856 }
2857
2858 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2859                           int *uaddr_len, int peer)
2860 {
2861         struct net_device *dev;
2862         struct sock *sk = sock->sk;
2863         struct packet_sock *po = pkt_sk(sk);
2864         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2865
2866         if (peer)
2867                 return -EOPNOTSUPP;
2868
2869         sll->sll_family = AF_PACKET;
2870         sll->sll_ifindex = po->ifindex;
2871         sll->sll_protocol = po->num;
2872         sll->sll_pkttype = 0;
2873         rcu_read_lock();
2874         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2875         if (dev) {
2876                 sll->sll_hatype = dev->type;
2877                 sll->sll_halen = dev->addr_len;
2878                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2879         } else {
2880                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
2881                 sll->sll_halen = 0;
2882         }
2883         rcu_read_unlock();
2884         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2885
2886         return 0;
2887 }
2888
2889 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2890                          int what)
2891 {
2892         switch (i->type) {
2893         case PACKET_MR_MULTICAST:
2894                 if (i->alen != dev->addr_len)
2895                         return -EINVAL;
2896                 if (what > 0)
2897                         return dev_mc_add(dev, i->addr);
2898                 else
2899                         return dev_mc_del(dev, i->addr);
2900                 break;
2901         case PACKET_MR_PROMISC:
2902                 return dev_set_promiscuity(dev, what);
2903                 break;
2904         case PACKET_MR_ALLMULTI:
2905                 return dev_set_allmulti(dev, what);
2906                 break;
2907         case PACKET_MR_UNICAST:
2908                 if (i->alen != dev->addr_len)
2909                         return -EINVAL;
2910                 if (what > 0)
2911                         return dev_uc_add(dev, i->addr);
2912                 else
2913                         return dev_uc_del(dev, i->addr);
2914                 break;
2915         default:
2916                 break;
2917         }
2918         return 0;
2919 }
2920
2921 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2922 {
2923         for ( ; i; i = i->next) {
2924                 if (i->ifindex == dev->ifindex)
2925                         packet_dev_mc(dev, i, what);
2926         }
2927 }
2928
2929 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2930 {
2931         struct packet_sock *po = pkt_sk(sk);
2932         struct packet_mclist *ml, *i;
2933         struct net_device *dev;
2934         int err;
2935
2936         rtnl_lock();
2937
2938         err = -ENODEV;
2939         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2940         if (!dev)
2941                 goto done;
2942
2943         err = -EINVAL;
2944         if (mreq->mr_alen > dev->addr_len)
2945                 goto done;
2946
2947         err = -ENOBUFS;
2948         i = kmalloc(sizeof(*i), GFP_KERNEL);
2949         if (i == NULL)
2950                 goto done;
2951
2952         err = 0;
2953         for (ml = po->mclist; ml; ml = ml->next) {
2954                 if (ml->ifindex == mreq->mr_ifindex &&
2955                     ml->type == mreq->mr_type &&
2956                     ml->alen == mreq->mr_alen &&
2957                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2958                         ml->count++;
2959                         /* Free the new element ... */
2960                         kfree(i);
2961                         goto done;
2962                 }
2963         }
2964
2965         i->type = mreq->mr_type;
2966         i->ifindex = mreq->mr_ifindex;
2967         i->alen = mreq->mr_alen;
2968         memcpy(i->addr, mreq->mr_address, i->alen);
2969         i->count = 1;
2970         i->next = po->mclist;
2971         po->mclist = i;
2972         err = packet_dev_mc(dev, i, 1);
2973         if (err) {
2974                 po->mclist = i->next;
2975                 kfree(i);
2976         }
2977
2978 done:
2979         rtnl_unlock();
2980         return err;
2981 }
2982
2983 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2984 {
2985         struct packet_mclist *ml, **mlp;
2986
2987         rtnl_lock();
2988
2989         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
2990                 if (ml->ifindex == mreq->mr_ifindex &&
2991                     ml->type == mreq->mr_type &&
2992                     ml->alen == mreq->mr_alen &&
2993                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2994                         if (--ml->count == 0) {
2995                                 struct net_device *dev;
2996                                 *mlp = ml->next;
2997                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2998                                 if (dev)
2999                                         packet_dev_mc(dev, ml, -1);
3000                                 kfree(ml);
3001                         }
3002                         rtnl_unlock();
3003                         return 0;
3004                 }
3005         }
3006         rtnl_unlock();
3007         return -EADDRNOTAVAIL;
3008 }
3009
3010 static void packet_flush_mclist(struct sock *sk)
3011 {
3012         struct packet_sock *po = pkt_sk(sk);
3013         struct packet_mclist *ml;
3014
3015         if (!po->mclist)
3016                 return;
3017
3018         rtnl_lock();
3019         while ((ml = po->mclist) != NULL) {
3020                 struct net_device *dev;
3021
3022                 po->mclist = ml->next;
3023                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3024                 if (dev != NULL)
3025                         packet_dev_mc(dev, ml, -1);
3026                 kfree(ml);
3027         }
3028         rtnl_unlock();
3029 }
3030
3031 static int
3032 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3033 {
3034         struct sock *sk = sock->sk;
3035         struct packet_sock *po = pkt_sk(sk);
3036         int ret;
3037
3038         if (level != SOL_PACKET)
3039                 return -ENOPROTOOPT;
3040
3041         switch (optname) {
3042         case PACKET_ADD_MEMBERSHIP:
3043         case PACKET_DROP_MEMBERSHIP:
3044         {
3045                 struct packet_mreq_max mreq;
3046                 int len = optlen;
3047                 memset(&mreq, 0, sizeof(mreq));
3048                 if (len < sizeof(struct packet_mreq))
3049                         return -EINVAL;
3050                 if (len > sizeof(mreq))
3051                         len = sizeof(mreq);
3052                 if (copy_from_user(&mreq, optval, len))
3053                         return -EFAULT;
3054                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3055                         return -EINVAL;
3056                 if (optname == PACKET_ADD_MEMBERSHIP)
3057                         ret = packet_mc_add(sk, &mreq);
3058                 else
3059                         ret = packet_mc_drop(sk, &mreq);
3060                 return ret;
3061         }
3062
3063         case PACKET_RX_RING:
3064         case PACKET_TX_RING:
3065         {
3066                 union tpacket_req_u req_u;
3067                 int len;
3068
3069                 switch (po->tp_version) {
3070                 case TPACKET_V1:
3071                 case TPACKET_V2:
3072                         len = sizeof(req_u.req);
3073                         break;
3074                 case TPACKET_V3:
3075                 default:
3076                         len = sizeof(req_u.req3);
3077                         break;
3078                 }
3079                 if (optlen < len)
3080                         return -EINVAL;
3081                 if (pkt_sk(sk)->has_vnet_hdr)
3082                         return -EINVAL;
3083                 if (copy_from_user(&req_u.req, optval, len))
3084                         return -EFAULT;
3085                 return packet_set_ring(sk, &req_u, 0,
3086                         optname == PACKET_TX_RING);
3087         }
3088         case PACKET_COPY_THRESH:
3089         {
3090                 int val;
3091
3092                 if (optlen != sizeof(val))
3093                         return -EINVAL;
3094                 if (copy_from_user(&val, optval, sizeof(val)))
3095                         return -EFAULT;
3096
3097                 pkt_sk(sk)->copy_thresh = val;
3098                 return 0;
3099         }
3100         case PACKET_VERSION:
3101         {
3102                 int val;
3103
3104                 if (optlen != sizeof(val))
3105                         return -EINVAL;
3106                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3107                         return -EBUSY;
3108                 if (copy_from_user(&val, optval, sizeof(val)))
3109                         return -EFAULT;
3110                 switch (val) {
3111                 case TPACKET_V1:
3112                 case TPACKET_V2:
3113                 case TPACKET_V3:
3114                         po->tp_version = val;
3115                         return 0;
3116                 default:
3117                         return -EINVAL;
3118                 }
3119         }
3120         case PACKET_RESERVE:
3121         {
3122                 unsigned int val;
3123
3124                 if (optlen != sizeof(val))
3125                         return -EINVAL;
3126                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3127                         return -EBUSY;
3128                 if (copy_from_user(&val, optval, sizeof(val)))
3129                         return -EFAULT;
3130                 po->tp_reserve = val;
3131                 return 0;
3132         }
3133         case PACKET_LOSS:
3134         {
3135                 unsigned int val;
3136
3137                 if (optlen != sizeof(val))
3138                         return -EINVAL;
3139                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3140                         return -EBUSY;
3141                 if (copy_from_user(&val, optval, sizeof(val)))
3142                         return -EFAULT;
3143                 po->tp_loss = !!val;
3144                 return 0;
3145         }
3146         case PACKET_AUXDATA:
3147         {
3148                 int val;
3149
3150                 if (optlen < sizeof(val))
3151                         return -EINVAL;
3152                 if (copy_from_user(&val, optval, sizeof(val)))
3153                         return -EFAULT;
3154
3155                 po->auxdata = !!val;
3156                 return 0;
3157         }
3158         case PACKET_ORIGDEV:
3159         {
3160                 int val;
3161
3162                 if (optlen < sizeof(val))
3163                         return -EINVAL;
3164                 if (copy_from_user(&val, optval, sizeof(val)))
3165                         return -EFAULT;
3166
3167                 po->origdev = !!val;
3168                 return 0;
3169         }
3170         case PACKET_VNET_HDR:
3171         {
3172                 int val;
3173
3174                 if (sock->type != SOCK_RAW)
3175                         return -EINVAL;
3176                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3177                         return -EBUSY;
3178                 if (optlen < sizeof(val))
3179                         return -EINVAL;
3180                 if (copy_from_user(&val, optval, sizeof(val)))
3181                         return -EFAULT;
3182
3183                 po->has_vnet_hdr = !!val;
3184                 return 0;
3185         }
3186         case PACKET_TIMESTAMP:
3187         {
3188                 int val;
3189
3190                 if (optlen != sizeof(val))
3191                         return -EINVAL;
3192                 if (copy_from_user(&val, optval, sizeof(val)))
3193                         return -EFAULT;
3194
3195                 po->tp_tstamp = val;
3196                 return 0;
3197         }
3198         case PACKET_FANOUT:
3199         {
3200                 int val;
3201
3202                 if (optlen != sizeof(val))
3203                         return -EINVAL;
3204                 if (copy_from_user(&val, optval, sizeof(val)))
3205                         return -EFAULT;
3206
3207                 return fanout_add(sk, val & 0xffff, val >> 16);
3208         }
3209         case PACKET_TX_HAS_OFF:
3210         {
3211                 unsigned int val;
3212
3213                 if (optlen != sizeof(val))
3214                         return -EINVAL;
3215                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3216                         return -EBUSY;
3217                 if (copy_from_user(&val, optval, sizeof(val)))
3218                         return -EFAULT;
3219                 po->tp_tx_has_off = !!val;
3220                 return 0;
3221         }
3222         default:
3223                 return -ENOPROTOOPT;
3224         }
3225 }
3226
3227 static int packet_getsockopt(struct socket *sock, int level, int optname,
3228                              char __user *optval, int __user *optlen)
3229 {
3230         int len;
3231         int val, lv = sizeof(val);
3232         struct sock *sk = sock->sk;
3233         struct packet_sock *po = pkt_sk(sk);
3234         void *data = &val;
3235         struct tpacket_stats st;
3236         union tpacket_stats_u st_u;
3237
3238         if (level != SOL_PACKET)
3239                 return -ENOPROTOOPT;
3240
3241         if (get_user(len, optlen))
3242                 return -EFAULT;
3243
3244         if (len < 0)
3245                 return -EINVAL;
3246
3247         switch (optname) {
3248         case PACKET_STATISTICS:
3249                 spin_lock_bh(&sk->sk_receive_queue.lock);
3250                 if (po->tp_version == TPACKET_V3) {
3251                         lv = sizeof(struct tpacket_stats_v3);
3252                         memcpy(&st_u.stats3, &po->stats,
3253                                sizeof(struct tpacket_stats));
3254                         st_u.stats3.tp_freeze_q_cnt =
3255                                         po->stats_u.stats3.tp_freeze_q_cnt;
3256                         st_u.stats3.tp_packets += po->stats.tp_drops;
3257                         data = &st_u.stats3;
3258                 } else {
3259                         lv = sizeof(struct tpacket_stats);
3260                         st = po->stats;
3261                         st.tp_packets += st.tp_drops;
3262                         data = &st;
3263                 }
3264                 memset(&po->stats, 0, sizeof(st));
3265                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3266                 break;
3267         case PACKET_AUXDATA:
3268                 val = po->auxdata;
3269                 break;
3270         case PACKET_ORIGDEV:
3271                 val = po->origdev;
3272                 break;
3273         case PACKET_VNET_HDR:
3274                 val = po->has_vnet_hdr;
3275                 break;
3276         case PACKET_VERSION:
3277                 val = po->tp_version;
3278                 break;
3279         case PACKET_HDRLEN:
3280                 if (len > sizeof(int))
3281                         len = sizeof(int);
3282                 if (copy_from_user(&val, optval, len))
3283                         return -EFAULT;
3284                 switch (val) {
3285                 case TPACKET_V1:
3286                         val = sizeof(struct tpacket_hdr);
3287                         break;
3288                 case TPACKET_V2:
3289                         val = sizeof(struct tpacket2_hdr);
3290                         break;
3291                 case TPACKET_V3:
3292                         val = sizeof(struct tpacket3_hdr);
3293                         break;
3294                 default:
3295                         return -EINVAL;
3296                 }
3297                 break;
3298         case PACKET_RESERVE:
3299                 val = po->tp_reserve;
3300                 break;
3301         case PACKET_LOSS:
3302                 val = po->tp_loss;
3303                 break;
3304         case PACKET_TIMESTAMP:
3305                 val = po->tp_tstamp;
3306                 break;
3307         case PACKET_FANOUT:
3308                 val = (po->fanout ?
3309                        ((u32)po->fanout->id |
3310                         ((u32)po->fanout->type << 16) |
3311                         ((u32)po->fanout->flags << 24)) :
3312                        0);
3313                 break;
3314         case PACKET_TX_HAS_OFF:
3315                 val = po->tp_tx_has_off;
3316                 break;
3317         default:
3318                 return -ENOPROTOOPT;
3319         }
3320
3321         if (len > lv)
3322                 len = lv;
3323         if (put_user(len, optlen))
3324                 return -EFAULT;
3325         if (copy_to_user(optval, data, len))
3326                 return -EFAULT;
3327         return 0;
3328 }
3329
3330
3331 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3332 {
3333         struct sock *sk;
3334         struct net_device *dev = data;
3335         struct net *net = dev_net(dev);
3336
3337         rcu_read_lock();
3338         sk_for_each_rcu(sk, &net->packet.sklist) {
3339                 struct packet_sock *po = pkt_sk(sk);
3340
3341                 switch (msg) {
3342                 case NETDEV_UNREGISTER:
3343                         if (po->mclist)
3344                                 packet_dev_mclist(dev, po->mclist, -1);
3345                         /* fallthrough */
3346
3347                 case NETDEV_DOWN:
3348                         if (dev->ifindex == po->ifindex) {
3349                                 spin_lock(&po->bind_lock);
3350                                 if (po->running) {
3351                                         __unregister_prot_hook(sk, false);
3352                                         sk->sk_err = ENETDOWN;
3353                                         if (!sock_flag(sk, SOCK_DEAD))
3354                                                 sk->sk_error_report(sk);
3355                                 }
3356                                 if (msg == NETDEV_UNREGISTER) {
3357                                         po->ifindex = -1;
3358                                         if (po->prot_hook.dev)
3359                                                 dev_put(po->prot_hook.dev);
3360                                         po->prot_hook.dev = NULL;
3361                                 }
3362                                 spin_unlock(&po->bind_lock);
3363                         }
3364                         break;
3365                 case NETDEV_UP:
3366                         if (dev->ifindex == po->ifindex) {
3367                                 spin_lock(&po->bind_lock);
3368                                 if (po->num)
3369                                         register_prot_hook(sk);
3370                                 spin_unlock(&po->bind_lock);
3371                         }
3372                         break;
3373                 }
3374         }
3375         rcu_read_unlock();
3376         return NOTIFY_DONE;
3377 }
3378
3379
3380 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3381                         unsigned long arg)
3382 {
3383         struct sock *sk = sock->sk;
3384
3385         switch (cmd) {
3386         case SIOCOUTQ:
3387         {
3388                 int amount = sk_wmem_alloc_get(sk);
3389
3390                 return put_user(amount, (int __user *)arg);
3391         }
3392         case SIOCINQ:
3393         {
3394                 struct sk_buff *skb;
3395                 int amount = 0;
3396
3397                 spin_lock_bh(&sk->sk_receive_queue.lock);
3398                 skb = skb_peek(&sk->sk_receive_queue);
3399                 if (skb)
3400                         amount = skb->len;
3401                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3402                 return put_user(amount, (int __user *)arg);
3403         }
3404         case SIOCGSTAMP:
3405                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3406         case SIOCGSTAMPNS:
3407                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3408
3409 #ifdef CONFIG_INET
3410         case SIOCADDRT:
3411         case SIOCDELRT:
3412         case SIOCDARP:
3413         case SIOCGARP:
3414         case SIOCSARP:
3415         case SIOCGIFADDR:
3416         case SIOCSIFADDR:
3417         case SIOCGIFBRDADDR:
3418         case SIOCSIFBRDADDR:
3419         case SIOCGIFNETMASK:
3420         case SIOCSIFNETMASK:
3421         case SIOCGIFDSTADDR:
3422         case SIOCSIFDSTADDR:
3423         case SIOCSIFFLAGS:
3424                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3425 #endif
3426
3427         default:
3428                 return -ENOIOCTLCMD;
3429         }
3430         return 0;
3431 }
3432
3433 static unsigned int packet_poll(struct file *file, struct socket *sock,
3434                                 poll_table *wait)
3435 {
3436         struct sock *sk = sock->sk;
3437         struct packet_sock *po = pkt_sk(sk);
3438         unsigned int mask = datagram_poll(file, sock, wait);
3439
3440         spin_lock_bh(&sk->sk_receive_queue.lock);
3441         if (po->rx_ring.pg_vec) {
3442                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3443                         TP_STATUS_KERNEL))
3444                         mask |= POLLIN | POLLRDNORM;
3445         }
3446         spin_unlock_bh(&sk->sk_receive_queue.lock);
3447         spin_lock_bh(&sk->sk_write_queue.lock);
3448         if (po->tx_ring.pg_vec) {
3449                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3450                         mask |= POLLOUT | POLLWRNORM;
3451         }
3452         spin_unlock_bh(&sk->sk_write_queue.lock);
3453         return mask;
3454 }
3455
3456
3457 /* Dirty? Well, I still did not learn better way to account
3458  * for user mmaps.
3459  */
3460
3461 static void packet_mm_open(struct vm_area_struct *vma)
3462 {
3463         struct file *file = vma->vm_file;
3464         struct socket *sock = file->private_data;
3465         struct sock *sk = sock->sk;
3466
3467         if (sk)
3468                 atomic_inc(&pkt_sk(sk)->mapped);
3469 }
3470
3471 static void packet_mm_close(struct vm_area_struct *vma)
3472 {
3473         struct file *file = vma->vm_file;
3474         struct socket *sock = file->private_data;
3475         struct sock *sk = sock->sk;
3476
3477         if (sk)
3478                 atomic_dec(&pkt_sk(sk)->mapped);
3479 }
3480
3481 static const struct vm_operations_struct packet_mmap_ops = {
3482         .open   =       packet_mm_open,
3483         .close  =       packet_mm_close,
3484 };
3485
3486 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3487                         unsigned int len)
3488 {
3489         int i;
3490
3491         for (i = 0; i < len; i++) {
3492                 if (likely(pg_vec[i].buffer)) {
3493                         if (is_vmalloc_addr(pg_vec[i].buffer))
3494                                 vfree(pg_vec[i].buffer);
3495                         else
3496                                 free_pages((unsigned long)pg_vec[i].buffer,
3497                                            order);
3498                         pg_vec[i].buffer = NULL;
3499                 }
3500         }
3501         kfree(pg_vec);
3502 }
3503
3504 static char *alloc_one_pg_vec_page(unsigned long order)
3505 {
3506         char *buffer = NULL;
3507         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3508                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3509
3510         buffer = (char *) __get_free_pages(gfp_flags, order);
3511
3512         if (buffer)
3513                 return buffer;
3514
3515         /*
3516          * __get_free_pages failed, fall back to vmalloc
3517          */
3518         buffer = vzalloc((1 << order) * PAGE_SIZE);
3519
3520         if (buffer)
3521                 return buffer;
3522
3523         /*
3524          * vmalloc failed, lets dig into swap here
3525          */
3526         gfp_flags &= ~__GFP_NORETRY;
3527         buffer = (char *)__get_free_pages(gfp_flags, order);
3528         if (buffer)
3529                 return buffer;
3530
3531         /*
3532          * complete and utter failure
3533          */
3534         return NULL;
3535 }
3536
3537 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3538 {
3539         unsigned int block_nr = req->tp_block_nr;
3540         struct pgv *pg_vec;
3541         int i;
3542
3543         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3544         if (unlikely(!pg_vec))
3545                 goto out;
3546
3547         for (i = 0; i < block_nr; i++) {
3548                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3549                 if (unlikely(!pg_vec[i].buffer))
3550                         goto out_free_pgvec;
3551         }
3552
3553 out:
3554         return pg_vec;
3555
3556 out_free_pgvec:
3557         free_pg_vec(pg_vec, order, block_nr);
3558         pg_vec = NULL;
3559         goto out;
3560 }
3561
3562 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3563                 int closing, int tx_ring)
3564 {
3565         struct pgv *pg_vec = NULL;
3566         struct packet_sock *po = pkt_sk(sk);
3567         int was_running, order = 0;
3568         struct packet_ring_buffer *rb;
3569         struct sk_buff_head *rb_queue;
3570         __be16 num;
3571         int err = -EINVAL;
3572         /* Added to avoid minimal code churn */
3573         struct tpacket_req *req = &req_u->req;
3574
3575         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3576         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3577                 WARN(1, "Tx-ring is not supported.\n");
3578                 goto out;
3579         }
3580
3581         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3582         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3583
3584         err = -EBUSY;
3585         if (!closing) {
3586                 if (atomic_read(&po->mapped))
3587                         goto out;
3588                 if (atomic_read(&rb->pending))
3589                         goto out;
3590         }
3591
3592         if (req->tp_block_nr) {
3593                 /* Sanity tests and some calculations */
3594                 err = -EBUSY;
3595                 if (unlikely(rb->pg_vec))
3596                         goto out;
3597
3598                 switch (po->tp_version) {
3599                 case TPACKET_V1:
3600                         po->tp_hdrlen = TPACKET_HDRLEN;
3601                         break;
3602                 case TPACKET_V2:
3603                         po->tp_hdrlen = TPACKET2_HDRLEN;
3604                         break;
3605                 case TPACKET_V3:
3606                         po->tp_hdrlen = TPACKET3_HDRLEN;
3607                         break;
3608                 }
3609
3610                 err = -EINVAL;
3611                 if (unlikely((int)req->tp_block_size <= 0))
3612                         goto out;
3613                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3614                         goto out;
3615                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3616                                         po->tp_reserve))
3617                         goto out;
3618                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3619                         goto out;
3620
3621                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3622                 if (unlikely(rb->frames_per_block <= 0))
3623                         goto out;
3624                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3625                                         req->tp_frame_nr))
3626                         goto out;
3627
3628                 err = -ENOMEM;
3629                 order = get_order(req->tp_block_size);
3630                 pg_vec = alloc_pg_vec(req, order);
3631                 if (unlikely(!pg_vec))
3632                         goto out;
3633                 switch (po->tp_version) {
3634                 case TPACKET_V3:
3635                 /* Transmit path is not supported. We checked
3636                  * it above but just being paranoid
3637                  */
3638                         if (!tx_ring)
3639                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3640                                 break;
3641                 default:
3642                         break;
3643                 }
3644         }
3645         /* Done */
3646         else {
3647                 err = -EINVAL;
3648                 if (unlikely(req->tp_frame_nr))
3649                         goto out;
3650         }
3651
3652         lock_sock(sk);
3653
3654         /* Detach socket from network */
3655         spin_lock(&po->bind_lock);
3656         was_running = po->running;
3657         num = po->num;
3658         if (was_running) {
3659                 po->num = 0;
3660                 __unregister_prot_hook(sk, false);
3661         }
3662         spin_unlock(&po->bind_lock);
3663
3664         synchronize_net();
3665
3666         err = -EBUSY;
3667         mutex_lock(&po->pg_vec_lock);
3668         if (closing || atomic_read(&po->mapped) == 0) {
3669                 err = 0;
3670                 spin_lock_bh(&rb_queue->lock);
3671                 swap(rb->pg_vec, pg_vec);
3672                 rb->frame_max = (req->tp_frame_nr - 1);
3673                 rb->head = 0;
3674                 rb->frame_size = req->tp_frame_size;
3675                 spin_unlock_bh(&rb_queue->lock);
3676
3677                 swap(rb->pg_vec_order, order);
3678                 swap(rb->pg_vec_len, req->tp_block_nr);
3679
3680                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3681                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3682                                                 tpacket_rcv : packet_rcv;
3683                 skb_queue_purge(rb_queue);
3684                 if (atomic_read(&po->mapped))
3685                         pr_err("packet_mmap: vma is busy: %d\n",
3686                                atomic_read(&po->mapped));
3687         }
3688         mutex_unlock(&po->pg_vec_lock);
3689
3690         spin_lock(&po->bind_lock);
3691         if (was_running) {
3692                 po->num = num;
3693                 register_prot_hook(sk);
3694         }
3695         spin_unlock(&po->bind_lock);
3696         if (closing && (po->tp_version > TPACKET_V2)) {
3697                 /* Because we don't support block-based V3 on tx-ring */
3698                 if (!tx_ring)
3699                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3700         }
3701         release_sock(sk);
3702
3703         if (pg_vec)
3704                 free_pg_vec(pg_vec, order, req->tp_block_nr);
3705 out:
3706         return err;
3707 }
3708
3709 static int packet_mmap(struct file *file, struct socket *sock,
3710                 struct vm_area_struct *vma)
3711 {
3712         struct sock *sk = sock->sk;
3713         struct packet_sock *po = pkt_sk(sk);
3714         unsigned long size, expected_size;
3715         struct packet_ring_buffer *rb;
3716         unsigned long start;
3717         int err = -EINVAL;
3718         int i;
3719
3720         if (vma->vm_pgoff)
3721                 return -EINVAL;
3722
3723         mutex_lock(&po->pg_vec_lock);
3724
3725         expected_size = 0;
3726         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3727                 if (rb->pg_vec) {
3728                         expected_size += rb->pg_vec_len
3729                                                 * rb->pg_vec_pages
3730                                                 * PAGE_SIZE;
3731                 }
3732         }
3733
3734         if (expected_size == 0)
3735                 goto out;
3736
3737         size = vma->vm_end - vma->vm_start;
3738         if (size != expected_size)
3739                 goto out;
3740
3741         start = vma->vm_start;
3742         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3743                 if (rb->pg_vec == NULL)
3744                         continue;
3745
3746                 for (i = 0; i < rb->pg_vec_len; i++) {
3747                         struct page *page;
3748                         void *kaddr = rb->pg_vec[i].buffer;
3749                         int pg_num;
3750
3751                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3752                                 page = pgv_to_page(kaddr);
3753                                 err = vm_insert_page(vma, start, page);
3754                                 if (unlikely(err))
3755                                         goto out;
3756                                 start += PAGE_SIZE;
3757                                 kaddr += PAGE_SIZE;
3758                         }
3759                 }
3760         }
3761
3762         atomic_inc(&po->mapped);
3763         vma->vm_ops = &packet_mmap_ops;
3764         err = 0;
3765
3766 out:
3767         mutex_unlock(&po->pg_vec_lock);
3768         return err;
3769 }
3770
3771 static const struct proto_ops packet_ops_spkt = {
3772         .family =       PF_PACKET,
3773         .owner =        THIS_MODULE,
3774         .release =      packet_release,
3775         .bind =         packet_bind_spkt,
3776         .connect =      sock_no_connect,
3777         .socketpair =   sock_no_socketpair,
3778         .accept =       sock_no_accept,
3779         .getname =      packet_getname_spkt,
3780         .poll =         datagram_poll,
3781         .ioctl =        packet_ioctl,
3782         .listen =       sock_no_listen,
3783         .shutdown =     sock_no_shutdown,
3784         .setsockopt =   sock_no_setsockopt,
3785         .getsockopt =   sock_no_getsockopt,
3786         .sendmsg =      packet_sendmsg_spkt,
3787         .recvmsg =      packet_recvmsg,
3788         .mmap =         sock_no_mmap,
3789         .sendpage =     sock_no_sendpage,
3790 };
3791
3792 static const struct proto_ops packet_ops = {
3793         .family =       PF_PACKET,
3794         .owner =        THIS_MODULE,
3795         .release =      packet_release,
3796         .bind =         packet_bind,
3797         .connect =      sock_no_connect,
3798         .socketpair =   sock_no_socketpair,
3799         .accept =       sock_no_accept,
3800         .getname =      packet_getname,
3801         .poll =         packet_poll,
3802         .ioctl =        packet_ioctl,
3803         .listen =       sock_no_listen,
3804         .shutdown =     sock_no_shutdown,
3805         .setsockopt =   packet_setsockopt,
3806         .getsockopt =   packet_getsockopt,
3807         .sendmsg =      packet_sendmsg,
3808         .recvmsg =      packet_recvmsg,
3809         .mmap =         packet_mmap,
3810         .sendpage =     sock_no_sendpage,
3811 };
3812
3813 static const struct net_proto_family packet_family_ops = {
3814         .family =       PF_PACKET,
3815         .create =       packet_create,
3816         .owner  =       THIS_MODULE,
3817 };
3818
3819 static struct notifier_block packet_netdev_notifier = {
3820         .notifier_call =        packet_notifier,
3821 };
3822
3823 #ifdef CONFIG_PROC_FS
3824
3825 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3826         __acquires(RCU)
3827 {
3828         struct net *net = seq_file_net(seq);
3829
3830         rcu_read_lock();
3831         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3832 }
3833
3834 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3835 {
3836         struct net *net = seq_file_net(seq);
3837         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3838 }
3839
3840 static void packet_seq_stop(struct seq_file *seq, void *v)
3841         __releases(RCU)
3842 {
3843         rcu_read_unlock();
3844 }
3845
3846 static int packet_seq_show(struct seq_file *seq, void *v)
3847 {
3848         if (v == SEQ_START_TOKEN)
3849                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
3850         else {
3851                 struct sock *s = sk_entry(v);
3852                 const struct packet_sock *po = pkt_sk(s);
3853
3854                 seq_printf(seq,
3855                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
3856                            s,
3857                            atomic_read(&s->sk_refcnt),
3858                            s->sk_type,
3859                            ntohs(po->num),
3860                            po->ifindex,
3861                            po->running,
3862                            atomic_read(&s->sk_rmem_alloc),
3863                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
3864                            sock_i_ino(s));
3865         }
3866
3867         return 0;
3868 }
3869
3870 static const struct seq_operations packet_seq_ops = {
3871         .start  = packet_seq_start,
3872         .next   = packet_seq_next,
3873         .stop   = packet_seq_stop,
3874         .show   = packet_seq_show,
3875 };
3876
3877 static int packet_seq_open(struct inode *inode, struct file *file)
3878 {
3879         return seq_open_net(inode, file, &packet_seq_ops,
3880                             sizeof(struct seq_net_private));
3881 }
3882
3883 static const struct file_operations packet_seq_fops = {
3884         .owner          = THIS_MODULE,
3885         .open           = packet_seq_open,
3886         .read           = seq_read,
3887         .llseek         = seq_lseek,
3888         .release        = seq_release_net,
3889 };
3890
3891 #endif
3892
3893 static int __net_init packet_net_init(struct net *net)
3894 {
3895         mutex_init(&net->packet.sklist_lock);
3896         INIT_HLIST_HEAD(&net->packet.sklist);
3897
3898         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
3899                 return -ENOMEM;
3900
3901         return 0;
3902 }
3903
3904 static void __net_exit packet_net_exit(struct net *net)
3905 {
3906         remove_proc_entry("packet", net->proc_net);
3907 }
3908
3909 static struct pernet_operations packet_net_ops = {
3910         .init = packet_net_init,
3911         .exit = packet_net_exit,
3912 };
3913
3914
3915 static void __exit packet_exit(void)
3916 {
3917         unregister_netdevice_notifier(&packet_netdev_notifier);
3918         unregister_pernet_subsys(&packet_net_ops);
3919         sock_unregister(PF_PACKET);
3920         proto_unregister(&packet_proto);
3921 }
3922
3923 static int __init packet_init(void)
3924 {
3925         int rc = proto_register(&packet_proto, 0);
3926
3927         if (rc != 0)
3928                 goto out;
3929
3930         sock_register(&packet_family_ops);
3931         register_pernet_subsys(&packet_net_ops);
3932         register_netdevice_notifier(&packet_netdev_notifier);
3933 out:
3934         return rc;
3935 }
3936
3937 module_init(packet_init);
3938 module_exit(packet_exit);
3939 MODULE_LICENSE("GPL");
3940 MODULE_ALIAS_NETPROTO(PF_PACKET);