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