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 union tpacket_uhdr {
 162         struct tpacket_hdr  *h1;
 163         struct tpacket2_hdr *h2;
 164         struct tpacket3_hdr *h3;
 165         void *raw;
 166 };
 167
 168 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 169                 int closing, int tx_ring);
 170
 171 #define V3_ALIGNMENT    (8)
 172
 173 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 174
 175 #define BLK_PLUS_PRIV(sz_of_priv) \
 176         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 177
 178 #define PGV_FROM_VMALLOC 1
 179
 180 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
 181 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
 182 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
 183 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
 184 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
 185 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
 186 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
 187
 188 struct packet_sock;
 189 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
 190 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 191                        struct packet_type *pt, struct net_device *orig_dev);
 192
 193 static void *packet_previous_frame(struct packet_sock *po,
 194                 struct packet_ring_buffer *rb,
 195                 int status);
 196 static void packet_increment_head(struct packet_ring_buffer *buff);
 197 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
 198                         struct tpacket_block_desc *);
 199 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 200                         struct packet_sock *);
 201 static void prb_retire_current_block(struct tpacket_kbdq_core *,
 202                 struct packet_sock *, unsigned int status);
 203 static int prb_queue_frozen(struct tpacket_kbdq_core *);
 204 static void prb_open_block(struct tpacket_kbdq_core *,
 205                 struct tpacket_block_desc *);
 206 static void prb_retire_rx_blk_timer_expired(unsigned long);
 207 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 208 static void prb_init_blk_timer(struct packet_sock *,
 209                 struct tpacket_kbdq_core *,
 210                 void (*func) (unsigned long));
 211 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 212 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 213                 struct tpacket3_hdr *);
 214 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 215                 struct tpacket3_hdr *);
 216 static void packet_flush_mclist(struct sock *sk);
 217
 218 struct packet_skb_cb {
 219         unsigned int origlen;
 220         union {
 221                 struct sockaddr_pkt pkt;
 222                 struct sockaddr_ll ll;
 223         } sa;
 224 };
 225
 226 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 227
 228 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 229 #define GET_PBLOCK_DESC(x, bid) \
 230         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 231 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 232         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 233 #define GET_NEXT_PRB_BLK_NUM(x) \
 234         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 235         ((x)->kactive_blk_num+1) : 0)
 236
 237 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 238 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 239
 240 /* register_prot_hook must be invoked with the po->bind_lock held,
 241  * or from a context in which asynchronous accesses to the packet
 242  * socket is not possible (packet_create()).
 243  */
 244 static void register_prot_hook(struct sock *sk)
 245 {
 246         struct packet_sock *po = pkt_sk(sk);
 247         if (!po->running) {
 248                 if (po->fanout)
 249                         __fanout_link(sk, po);
 250                 else
 251                         dev_add_pack(&po->prot_hook);
 252                 sock_hold(sk);
 253                 po->running = 1;
 254         }
 255 }
 256
 257 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
 258  * held.   If the sync parameter is true, we will temporarily drop
 259  * the po->bind_lock and do a synchronize_net to make sure no
 260  * asynchronous packet processing paths still refer to the elements
 261  * of po->prot_hook.  If the sync parameter is false, it is the
 262  * callers responsibility to take care of this.
 263  */
 264 static void __unregister_prot_hook(struct sock *sk, bool sync)
 265 {
 266         struct packet_sock *po = pkt_sk(sk);
 267
 268         po->running = 0;
 269         if (po->fanout)
 270                 __fanout_unlink(sk, po);
 271         else
 272                 __dev_remove_pack(&po->prot_hook);
 273         __sock_put(sk);
 274
 275         if (sync) {
 276                 spin_unlock(&po->bind_lock);
 277                 synchronize_net();
 278                 spin_lock(&po->bind_lock);
 279         }
 280 }
 281
 282 static void unregister_prot_hook(struct sock *sk, bool sync)
 283 {
 284         struct packet_sock *po = pkt_sk(sk);
 285
 286         if (po->running)
 287                 __unregister_prot_hook(sk, sync);
 288 }
 289
 290 static inline __pure struct page *pgv_to_page(void *addr)
 291 {
 292         if (is_vmalloc_addr(addr))
 293                 return vmalloc_to_page(addr);
 294         return virt_to_page(addr);
 295 }
 296
 297 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 298 {
 299         union tpacket_uhdr h;
 300
 301         h.raw = frame;
 302         switch (po->tp_version) {
 303         case TPACKET_V1:
 304                 h.h1->tp_status = status;
 305                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 306                 break;
 307         case TPACKET_V2:
 308                 h.h2->tp_status = status;
 309                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 310                 break;
 311         case TPACKET_V3:
 312         default:
 313                 WARN(1, "TPACKET version not supported.\n");
 314                 BUG();
 315         }
 316
 317         smp_wmb();
 318 }
 319
 320 static int __packet_get_status(struct packet_sock *po, void *frame)
 321 {
 322         union tpacket_uhdr 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 __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
 343                                    unsigned int flags)
 344 {
 345         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 346
 347         if (shhwtstamps) {
 348                 if ((flags & SOF_TIMESTAMPING_SYS_HARDWARE) &&
 349                     ktime_to_timespec_cond(shhwtstamps->syststamp, ts))
 350                         return TP_STATUS_TS_SYS_HARDWARE;
 351                 if ((flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
 352                     ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
 353                         return TP_STATUS_TS_RAW_HARDWARE;
 354         }
 355
 356         if (ktime_to_timespec_cond(skb->tstamp, ts))
 357                 return TP_STATUS_TS_SOFTWARE;
 358
 359         return 0;
 360 }
 361
 362 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
 363                                     struct sk_buff *skb)
 364 {
 365         union tpacket_uhdr h;
 366         struct timespec ts;
 367         __u32 ts_status;
 368
 369         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
 370                 return 0;
 371
 372         h.raw = frame;
 373         switch (po->tp_version) {
 374         case TPACKET_V1:
 375                 h.h1->tp_sec = ts.tv_sec;
 376                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
 377                 break;
 378         case TPACKET_V2:
 379                 h.h2->tp_sec = ts.tv_sec;
 380                 h.h2->tp_nsec = ts.tv_nsec;
 381                 break;
 382         case TPACKET_V3:
 383         default:
 384                 WARN(1, "TPACKET version not supported.\n");
 385                 BUG();
 386         }
 387
 388         /* one flush is safe, as both fields always lie on the same cacheline */
 389         flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
 390         smp_wmb();
 391
 392         return ts_status;
 393 }
 394
 395 static void *packet_lookup_frame(struct packet_sock *po,
 396                 struct packet_ring_buffer *rb,
 397                 unsigned int position,
 398                 int status)
 399 {
 400         unsigned int pg_vec_pos, frame_offset;
 401         union tpacket_uhdr h;
 402
 403         pg_vec_pos = position / rb->frames_per_block;
 404         frame_offset = position % rb->frames_per_block;
 405
 406         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 407                 (frame_offset * rb->frame_size);
 408
 409         if (status != __packet_get_status(po, h.raw))
 410                 return NULL;
 411
 412         return h.raw;
 413 }
 414
 415 static void *packet_current_frame(struct packet_sock *po,
 416                 struct packet_ring_buffer *rb,
 417                 int status)
 418 {
 419         return packet_lookup_frame(po, rb, rb->head, status);
 420 }
 421
 422 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 423 {
 424         del_timer_sync(&pkc->retire_blk_timer);
 425 }
 426
 427 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 428                 int tx_ring,
 429                 struct sk_buff_head *rb_queue)
 430 {
 431         struct tpacket_kbdq_core *pkc;
 432
 433         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 434
 435         spin_lock(&rb_queue->lock);
 436         pkc->delete_blk_timer = 1;
 437         spin_unlock(&rb_queue->lock);
 438
 439         prb_del_retire_blk_timer(pkc);
 440 }
 441
 442 static void prb_init_blk_timer(struct packet_sock *po,
 443                 struct tpacket_kbdq_core *pkc,
 444                 void (*func) (unsigned long))
 445 {
 446         init_timer(&pkc->retire_blk_timer);
 447         pkc->retire_blk_timer.data = (long)po;
 448         pkc->retire_blk_timer.function = func;
 449         pkc->retire_blk_timer.expires = jiffies;
 450 }
 451
 452 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
 453 {
 454         struct tpacket_kbdq_core *pkc;
 455
 456         if (tx_ring)
 457                 BUG();
 458
 459         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 460         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
 461 }
 462
 463 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 464                                 int blk_size_in_bytes)
 465 {
 466         struct net_device *dev;
 467         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 468         struct ethtool_cmd ecmd;
 469         int err;
 470         u32 speed;
 471
 472         rtnl_lock();
 473         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 474         if (unlikely(!dev)) {
 475                 rtnl_unlock();
 476                 return DEFAULT_PRB_RETIRE_TOV;
 477         }
 478         err = __ethtool_get_settings(dev, &ecmd);
 479         speed = ethtool_cmd_speed(&ecmd);
 480         rtnl_unlock();
 481         if (!err) {
 482                 /*
 483                  * If the link speed is so slow you don't really
 484                  * need to worry about perf anyways
 485                  */
 486                 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
 487                         return DEFAULT_PRB_RETIRE_TOV;
 488                 } else {
 489                         msec = 1;
 490                         div = speed / 1000;
 491                 }
 492         }
 493
 494         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 495
 496         if (div)
 497                 mbits /= div;
 498
 499         tmo = mbits * msec;
 500
 501         if (div)
 502                 return tmo+1;
 503         return tmo;
 504 }
 505
 506 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 507                         union tpacket_req_u *req_u)
 508 {
 509         p1->feature_req_word = req_u->req3.tp_feature_req_word;
 510 }
 511
 512 static void init_prb_bdqc(struct packet_sock *po,
 513                         struct packet_ring_buffer *rb,
 514                         struct pgv *pg_vec,
 515                         union tpacket_req_u *req_u, int tx_ring)
 516 {
 517         struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
 518         struct tpacket_block_desc *pbd;
 519
 520         memset(p1, 0x0, sizeof(*p1));
 521
 522         p1->knxt_seq_num = 1;
 523         p1->pkbdq = pg_vec;
 524         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 525         p1->pkblk_start = pg_vec[0].buffer;
 526         p1->kblk_size = req_u->req3.tp_block_size;
 527         p1->knum_blocks = req_u->req3.tp_block_nr;
 528         p1->hdrlen = po->tp_hdrlen;
 529         p1->version = po->tp_version;
 530         p1->last_kactive_blk_num = 0;
 531         po->stats.stats3.tp_freeze_q_cnt = 0;
 532         if (req_u->req3.tp_retire_blk_tov)
 533                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 534         else
 535                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 536                                                 req_u->req3.tp_block_size);
 537         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 538         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 539
 540         prb_init_ft_ops(p1, req_u);
 541         prb_setup_retire_blk_timer(po, tx_ring);
 542         prb_open_block(p1, pbd);
 543 }
 544
 545 /*  Do NOT update the last_blk_num first.
 546  *  Assumes sk_buff_head lock is held.
 547  */
 548 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 549 {
 550         mod_timer(&pkc->retire_blk_timer,
 551                         jiffies + pkc->tov_in_jiffies);
 552         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 553 }
 554
 555 /*
 556  * Timer logic:
 557  * 1) We refresh the timer only when we open a block.
 558  *    By doing this we don't waste cycles refreshing the timer
 559  *        on packet-by-packet basis.
 560  *
 561  * With a 1MB block-size, on a 1Gbps line, it will take
 562  * i) ~8 ms to fill a block + ii) memcpy etc.
 563  * In this cut we are not accounting for the memcpy time.
 564  *
 565  * So, if the user sets the 'tmo' to 10ms then the timer
 566  * will never fire while the block is still getting filled
 567  * (which is what we want). However, the user could choose
 568  * to close a block early and that's fine.
 569  *
 570  * But when the timer does fire, we check whether or not to refresh it.
 571  * Since the tmo granularity is in msecs, it is not too expensive
 572  * to refresh the timer, lets say every '8' msecs.
 573  * Either the user can set the 'tmo' or we can derive it based on
 574  * a) line-speed and b) block-size.
 575  * prb_calc_retire_blk_tmo() calculates the tmo.
 576  *
 577  */
 578 static void prb_retire_rx_blk_timer_expired(unsigned long data)
 579 {
 580         struct packet_sock *po = (struct packet_sock *)data;
 581         struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
 582         unsigned int frozen;
 583         struct tpacket_block_desc *pbd;
 584
 585         spin_lock(&po->sk.sk_receive_queue.lock);
 586
 587         frozen = prb_queue_frozen(pkc);
 588         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 589
 590         if (unlikely(pkc->delete_blk_timer))
 591                 goto out;
 592
 593         /* We only need to plug the race when the block is partially filled.
 594          * tpacket_rcv:
 595          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 596          *              copy_bits() is in progress ...
 597          *              timer fires on other cpu:
 598          *              we can't retire the current block because copy_bits
 599          *              is in progress.
 600          *
 601          */
 602         if (BLOCK_NUM_PKTS(pbd)) {
 603                 while (atomic_read(&pkc->blk_fill_in_prog)) {
 604                         /* Waiting for skb_copy_bits to finish... */
 605                         cpu_relax();
 606                 }
 607         }
 608
 609         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 610                 if (!frozen) {
 611                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 612                         if (!prb_dispatch_next_block(pkc, po))
 613                                 goto refresh_timer;
 614                         else
 615                                 goto out;
 616                 } else {
 617                         /* Case 1. Queue was frozen because user-space was
 618                          *         lagging behind.
 619                          */
 620                         if (prb_curr_blk_in_use(pkc, pbd)) {
 621                                 /*
 622                                  * Ok, user-space is still behind.
 623                                  * So just refresh the timer.
 624                                  */
 625                                 goto refresh_timer;
 626                         } else {
 627                                /* Case 2. queue was frozen,user-space caught up,
 628                                 * now the link went idle && the timer fired.
 629                                 * We don't have a block to close.So we open this
 630                                 * block and restart the timer.
 631                                 * opening a block thaws the queue,restarts timer
 632                                 * Thawing/timer-refresh is a side effect.
 633                                 */
 634                                 prb_open_block(pkc, pbd);
 635                                 goto out;
 636                         }
 637                 }
 638         }
 639
 640 refresh_timer:
 641         _prb_refresh_rx_retire_blk_timer(pkc);
 642
 643 out:
 644         spin_unlock(&po->sk.sk_receive_queue.lock);
 645 }
 646
 647 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 648                 struct tpacket_block_desc *pbd1, __u32 status)
 649 {
 650         /* Flush everything minus the block header */
 651
 652 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 653         u8 *start, *end;
 654
 655         start = (u8 *)pbd1;
 656
 657         /* Skip the block header(we know header WILL fit in 4K) */
 658         start += PAGE_SIZE;
 659
 660         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 661         for (; start < end; start += PAGE_SIZE)
 662                 flush_dcache_page(pgv_to_page(start));
 663
 664         smp_wmb();
 665 #endif
 666
 667         /* Now update the block status. */
 668
 669         BLOCK_STATUS(pbd1) = status;
 670
 671         /* Flush the block header */
 672
 673 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 674         start = (u8 *)pbd1;
 675         flush_dcache_page(pgv_to_page(start));
 676
 677         smp_wmb();
 678 #endif
 679 }
 680
 681 /*
 682  * Side effect:
 683  *
 684  * 1) flush the block
 685  * 2) Increment active_blk_num
 686  *
 687  * Note:We DONT refresh the timer on purpose.
 688  *      Because almost always the next block will be opened.
 689  */
 690 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 691                 struct tpacket_block_desc *pbd1,
 692                 struct packet_sock *po, unsigned int stat)
 693 {
 694         __u32 status = TP_STATUS_USER | stat;
 695
 696         struct tpacket3_hdr *last_pkt;
 697         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 698
 699         if (po->stats.stats3.tp_drops)
 700                 status |= TP_STATUS_LOSING;
 701
 702         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 703         last_pkt->tp_next_offset = 0;
 704
 705         /* Get the ts of the last pkt */
 706         if (BLOCK_NUM_PKTS(pbd1)) {
 707                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 708                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 709         } else {
 710                 /* Ok, we tmo'd - so get the current time */
 711                 struct timespec ts;
 712                 getnstimeofday(&ts);
 713                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 714                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 715         }
 716
 717         smp_wmb();
 718
 719         /* Flush the block */
 720         prb_flush_block(pkc1, pbd1, status);
 721
 722         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 723 }
 724
 725 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 726 {
 727         pkc->reset_pending_on_curr_blk = 0;
 728 }
 729
 730 /*
 731  * Side effect of opening a block:
 732  *
 733  * 1) prb_queue is thawed.
 734  * 2) retire_blk_timer is refreshed.
 735  *
 736  */
 737 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 738         struct tpacket_block_desc *pbd1)
 739 {
 740         struct timespec ts;
 741         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 742
 743         smp_rmb();
 744
 745         /* We could have just memset this but we will lose the
 746          * flexibility of making the priv area sticky
 747          */
 748
 749         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 750         BLOCK_NUM_PKTS(pbd1) = 0;
 751         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 752
 753         getnstimeofday(&ts);
 754
 755         h1->ts_first_pkt.ts_sec = ts.tv_sec;
 756         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 757
 758         pkc1->pkblk_start = (char *)pbd1;
 759         pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 760
 761         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 762         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 763
 764         pbd1->version = pkc1->version;
 765         pkc1->prev = pkc1->nxt_offset;
 766         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 767
 768         prb_thaw_queue(pkc1);
 769         _prb_refresh_rx_retire_blk_timer(pkc1);
 770
 771         smp_wmb();
 772 }
 773
 774 /*
 775  * Queue freeze logic:
 776  * 1) Assume tp_block_nr = 8 blocks.
 777  * 2) At time 't0', user opens Rx ring.
 778  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 779  * 4) user-space is either sleeping or processing block '0'.
 780  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 781  *    it will close block-7,loop around and try to fill block '0'.
 782  *    call-flow:
 783  *    __packet_lookup_frame_in_block
 784  *      prb_retire_current_block()
 785  *      prb_dispatch_next_block()
 786  *        |->(BLOCK_STATUS == USER) evaluates to true
 787  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 788  * 6) Now there are two cases:
 789  *    6.1) Link goes idle right after the queue is frozen.
 790  *         But remember, the last open_block() refreshed the timer.
 791  *         When this timer expires,it will refresh itself so that we can
 792  *         re-open block-0 in near future.
 793  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 794  *         case and __packet_lookup_frame_in_block will check if block-0
 795  *         is free and can now be re-used.
 796  */
 797 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 798                                   struct packet_sock *po)
 799 {
 800         pkc->reset_pending_on_curr_blk = 1;
 801         po->stats.stats3.tp_freeze_q_cnt++;
 802 }
 803
 804 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 805
 806 /*
 807  * If the next block is free then we will dispatch it
 808  * and return a good offset.
 809  * Else, we will freeze the queue.
 810  * So, caller must check the return value.
 811  */
 812 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 813                 struct packet_sock *po)
 814 {
 815         struct tpacket_block_desc *pbd;
 816
 817         smp_rmb();
 818
 819         /* 1. Get current block num */
 820         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 821
 822         /* 2. If this block is currently in_use then freeze the queue */
 823         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 824                 prb_freeze_queue(pkc, po);
 825                 return NULL;
 826         }
 827
 828         /*
 829          * 3.
 830          * open this block and return the offset where the first packet
 831          * needs to get stored.
 832          */
 833         prb_open_block(pkc, pbd);
 834         return (void *)pkc->nxt_offset;
 835 }
 836
 837 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 838                 struct packet_sock *po, unsigned int status)
 839 {
 840         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 841
 842         /* retire/close the current block */
 843         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 844                 /*
 845                  * Plug the case where copy_bits() is in progress on
 846                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 847                  * have space to copy the pkt in the current block and
 848                  * called prb_retire_current_block()
 849                  *
 850                  * We don't need to worry about the TMO case because
 851                  * the timer-handler already handled this case.
 852                  */
 853                 if (!(status & TP_STATUS_BLK_TMO)) {
 854                         while (atomic_read(&pkc->blk_fill_in_prog)) {
 855                                 /* Waiting for skb_copy_bits to finish... */
 856                                 cpu_relax();
 857                         }
 858                 }
 859                 prb_close_block(pkc, pbd, po, status);
 860                 return;
 861         }
 862 }
 863
 864 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
 865                                       struct tpacket_block_desc *pbd)
 866 {
 867         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 868 }
 869
 870 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 871 {
 872         return pkc->reset_pending_on_curr_blk;
 873 }
 874
 875 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 876 {
 877         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 878         atomic_dec(&pkc->blk_fill_in_prog);
 879 }
 880
 881 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 882                         struct tpacket3_hdr *ppd)
 883 {
 884         ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
 885 }
 886
 887 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 888                         struct tpacket3_hdr *ppd)
 889 {
 890         ppd->hv1.tp_rxhash = 0;
 891 }
 892
 893 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
 894                         struct tpacket3_hdr *ppd)
 895 {
 896         if (vlan_tx_tag_present(pkc->skb)) {
 897                 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
 898                 ppd->tp_status = TP_STATUS_VLAN_VALID;
 899         } else {
 900                 ppd->hv1.tp_vlan_tci = 0;
 901                 ppd->tp_status = TP_STATUS_AVAILABLE;
 902         }
 903 }
 904
 905 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
 906                         struct tpacket3_hdr *ppd)
 907 {
 908         prb_fill_vlan_info(pkc, ppd);
 909
 910         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
 911                 prb_fill_rxhash(pkc, ppd);
 912         else
 913                 prb_clear_rxhash(pkc, ppd);
 914 }
 915
 916 static void prb_fill_curr_block(char *curr,
 917                                 struct tpacket_kbdq_core *pkc,
 918                                 struct tpacket_block_desc *pbd,
 919                                 unsigned int len)
 920 {
 921         struct tpacket3_hdr *ppd;
 922
 923         ppd  = (struct tpacket3_hdr *)curr;
 924         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
 925         pkc->prev = curr;
 926         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
 927         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
 928         BLOCK_NUM_PKTS(pbd) += 1;
 929         atomic_inc(&pkc->blk_fill_in_prog);
 930         prb_run_all_ft_ops(pkc, ppd);
 931 }
 932
 933 /* Assumes caller has the sk->rx_queue.lock */
 934 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
 935                                             struct sk_buff *skb,
 936                                                 int status,
 937                                             unsigned int len
 938                                             )
 939 {
 940         struct tpacket_kbdq_core *pkc;
 941         struct tpacket_block_desc *pbd;
 942         char *curr, *end;
 943
 944         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 945         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 946
 947         /* Queue is frozen when user space is lagging behind */
 948         if (prb_queue_frozen(pkc)) {
 949                 /*
 950                  * Check if that last block which caused the queue to freeze,
 951                  * is still in_use by user-space.
 952                  */
 953                 if (prb_curr_blk_in_use(pkc, pbd)) {
 954                         /* Can't record this packet */
 955                         return NULL;
 956                 } else {
 957                         /*
 958                          * Ok, the block was released by user-space.
 959                          * Now let's open that block.
 960                          * opening a block also thaws the queue.
 961                          * Thawing is a side effect.
 962                          */
 963                         prb_open_block(pkc, pbd);
 964                 }
 965         }
 966
 967         smp_mb();
 968         curr = pkc->nxt_offset;
 969         pkc->skb = skb;
 970         end = (char *)pbd + pkc->kblk_size;
 971
 972         /* first try the current block */
 973         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
 974                 prb_fill_curr_block(curr, pkc, pbd, len);
 975                 return (void *)curr;
 976         }
 977
 978         /* Ok, close the current block */
 979         prb_retire_current_block(pkc, po, 0);
 980
 981         /* Now, try to dispatch the next block */
 982         curr = (char *)prb_dispatch_next_block(pkc, po);
 983         if (curr) {
 984                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 985                 prb_fill_curr_block(curr, pkc, pbd, len);
 986                 return (void *)curr;
 987         }
 988
 989         /*
 990          * No free blocks are available.user_space hasn't caught up yet.
 991          * Queue was just frozen and now this packet will get dropped.
 992          */
 993         return NULL;
 994 }
 995
 996 static void *packet_current_rx_frame(struct packet_sock *po,
 997                                             struct sk_buff *skb,
 998                                             int status, unsigned int len)
 999 {
1000         char *curr = NULL;
1001         switch (po->tp_version) {
1002         case TPACKET_V1:
1003         case TPACKET_V2:
1004                 curr = packet_lookup_frame(po, &po->rx_ring,
1005                                         po->rx_ring.head, status);
1006                 return curr;
1007         case TPACKET_V3:
1008                 return __packet_lookup_frame_in_block(po, skb, status, len);
1009         default:
1010                 WARN(1, "TPACKET version not supported\n");
1011                 BUG();
1012                 return NULL;
1013         }
1014 }
1015
1016 static void *prb_lookup_block(struct packet_sock *po,
1017                                      struct packet_ring_buffer *rb,
1018                                      unsigned int idx,
1019                                      int status)
1020 {
1021         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1022         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1023
1024         if (status != BLOCK_STATUS(pbd))
1025                 return NULL;
1026         return pbd;
1027 }
1028
1029 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1030 {
1031         unsigned int prev;
1032         if (rb->prb_bdqc.kactive_blk_num)
1033                 prev = rb->prb_bdqc.kactive_blk_num-1;
1034         else
1035                 prev = rb->prb_bdqc.knum_blocks-1;
1036         return prev;
1037 }
1038
1039 /* Assumes caller has held the rx_queue.lock */
1040 static void *__prb_previous_block(struct packet_sock *po,
1041                                          struct packet_ring_buffer *rb,
1042                                          int status)
1043 {
1044         unsigned int previous = prb_previous_blk_num(rb);
1045         return prb_lookup_block(po, rb, previous, status);
1046 }
1047
1048 static void *packet_previous_rx_frame(struct packet_sock *po,
1049                                              struct packet_ring_buffer *rb,
1050                                              int status)
1051 {
1052         if (po->tp_version <= TPACKET_V2)
1053                 return packet_previous_frame(po, rb, status);
1054
1055         return __prb_previous_block(po, rb, status);
1056 }
1057
1058 static void packet_increment_rx_head(struct packet_sock *po,
1059                                             struct packet_ring_buffer *rb)
1060 {
1061         switch (po->tp_version) {
1062         case TPACKET_V1:
1063         case TPACKET_V2:
1064                 return packet_increment_head(rb);
1065         case TPACKET_V3:
1066         default:
1067                 WARN(1, "TPACKET version not supported.\n");
1068                 BUG();
1069                 return;
1070         }
1071 }
1072
1073 static void *packet_previous_frame(struct packet_sock *po,
1074                 struct packet_ring_buffer *rb,
1075                 int status)
1076 {
1077         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1078         return packet_lookup_frame(po, rb, previous, status);
1079 }
1080
1081 static void packet_increment_head(struct packet_ring_buffer *buff)
1082 {
1083         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1084 }
1085
1086 static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1087 {
1088         struct sock *sk = &po->sk;
1089         bool has_room;
1090
1091         if (po->prot_hook.func != tpacket_rcv)
1092                 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1093                         <= sk->sk_rcvbuf;
1094
1095         spin_lock(&sk->sk_receive_queue.lock);
1096         if (po->tp_version == TPACKET_V3)
1097                 has_room = prb_lookup_block(po, &po->rx_ring,
1098                                             po->rx_ring.prb_bdqc.kactive_blk_num,
1099                                             TP_STATUS_KERNEL);
1100         else
1101                 has_room = packet_lookup_frame(po, &po->rx_ring,
1102                                                po->rx_ring.head,
1103                                                TP_STATUS_KERNEL);
1104         spin_unlock(&sk->sk_receive_queue.lock);
1105
1106         return has_room;
1107 }
1108
1109 static void packet_sock_destruct(struct sock *sk)
1110 {
1111         skb_queue_purge(&sk->sk_error_queue);
1112
1113         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1114         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1115
1116         if (!sock_flag(sk, SOCK_DEAD)) {
1117                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1118                 return;
1119         }
1120
1121         sk_refcnt_debug_dec(sk);
1122 }
1123
1124 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1125 {
1126         int x = atomic_read(&f->rr_cur) + 1;
1127
1128         if (x >= num)
1129                 x = 0;
1130
1131         return x;
1132 }
1133
1134 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1135                                       struct sk_buff *skb,
1136                                       unsigned int num)
1137 {
1138         return (((u64)skb->rxhash) * num) >> 32;
1139 }
1140
1141 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1142                                     struct sk_buff *skb,
1143                                     unsigned int num)
1144 {
1145         int cur, old;
1146
1147         cur = atomic_read(&f->rr_cur);
1148         while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1149                                      fanout_rr_next(f, num))) != cur)
1150                 cur = old;
1151         return cur;
1152 }
1153
1154 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1155                                      struct sk_buff *skb,
1156                                      unsigned int num)
1157 {
1158         return smp_processor_id() % num;
1159 }
1160
1161 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1162                                           struct sk_buff *skb,
1163                                           unsigned int idx, unsigned int skip,
1164                                           unsigned int num)
1165 {
1166         unsigned int i, j;
1167
1168         i = j = min_t(int, f->next[idx], num - 1);
1169         do {
1170                 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1171                         if (i != j)
1172                                 f->next[idx] = i;
1173                         return i;
1174                 }
1175                 if (++i == num)
1176                         i = 0;
1177         } while (i != j);
1178
1179         return idx;
1180 }
1181
1182 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1183 {
1184         return f->flags & (flag >> 8);
1185 }
1186
1187 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1188                              struct packet_type *pt, struct net_device *orig_dev)
1189 {
1190         struct packet_fanout *f = pt->af_packet_priv;
1191         unsigned int num = f->num_members;
1192         struct packet_sock *po;
1193         unsigned int idx;
1194
1195         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1196             !num) {
1197                 kfree_skb(skb);
1198                 return 0;
1199         }
1200
1201         switch (f->type) {
1202         case PACKET_FANOUT_HASH:
1203         default:
1204                 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1205                         skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1206                         if (!skb)
1207                                 return 0;
1208                 }
1209                 skb_get_rxhash(skb);
1210                 idx = fanout_demux_hash(f, skb, num);
1211                 break;
1212         case PACKET_FANOUT_LB:
1213                 idx = fanout_demux_lb(f, skb, num);
1214                 break;
1215         case PACKET_FANOUT_CPU:
1216                 idx = fanout_demux_cpu(f, skb, num);
1217                 break;
1218         case PACKET_FANOUT_ROLLOVER:
1219                 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1220                 break;
1221         }
1222
1223         po = pkt_sk(f->arr[idx]);
1224         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1225             unlikely(!packet_rcv_has_room(po, skb))) {
1226                 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1227                 po = pkt_sk(f->arr[idx]);
1228         }
1229
1230         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1231 }
1232
1233 DEFINE_MUTEX(fanout_mutex);
1234 EXPORT_SYMBOL_GPL(fanout_mutex);
1235 static LIST_HEAD(fanout_list);
1236
1237 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1238 {
1239         struct packet_fanout *f = po->fanout;
1240
1241         spin_lock(&f->lock);
1242         f->arr[f->num_members] = sk;
1243         smp_wmb();
1244         f->num_members++;
1245         spin_unlock(&f->lock);
1246 }
1247
1248 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1249 {
1250         struct packet_fanout *f = po->fanout;
1251         int i;
1252
1253         spin_lock(&f->lock);
1254         for (i = 0; i < f->num_members; i++) {
1255                 if (f->arr[i] == sk)
1256                         break;
1257         }
1258         BUG_ON(i >= f->num_members);
1259         f->arr[i] = f->arr[f->num_members - 1];
1260         f->num_members--;
1261         spin_unlock(&f->lock);
1262 }
1263
1264 static bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1265 {
1266         if (ptype->af_packet_priv == (void*)((struct packet_sock *)sk)->fanout)
1267                 return true;
1268
1269         return false;
1270 }
1271
1272 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1273 {
1274         struct packet_sock *po = pkt_sk(sk);
1275         struct packet_fanout *f, *match;
1276         u8 type = type_flags & 0xff;
1277         u8 flags = type_flags >> 8;
1278         int err;
1279
1280         switch (type) {
1281         case PACKET_FANOUT_ROLLOVER:
1282                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1283                         return -EINVAL;
1284         case PACKET_FANOUT_HASH:
1285         case PACKET_FANOUT_LB:
1286         case PACKET_FANOUT_CPU:
1287                 break;
1288         default:
1289                 return -EINVAL;
1290         }
1291
1292         if (!po->running)
1293                 return -EINVAL;
1294
1295         if (po->fanout)
1296                 return -EALREADY;
1297
1298         mutex_lock(&fanout_mutex);
1299         match = NULL;
1300         list_for_each_entry(f, &fanout_list, list) {
1301                 if (f->id == id &&
1302                     read_pnet(&f->net) == sock_net(sk)) {
1303                         match = f;
1304                         break;
1305                 }
1306         }
1307         err = -EINVAL;
1308         if (match && match->flags != flags)
1309                 goto out;
1310         if (!match) {
1311                 err = -ENOMEM;
1312                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1313                 if (!match)
1314                         goto out;
1315                 write_pnet(&match->net, sock_net(sk));
1316                 match->id = id;
1317                 match->type = type;
1318                 match->flags = flags;
1319                 atomic_set(&match->rr_cur, 0);
1320                 INIT_LIST_HEAD(&match->list);
1321                 spin_lock_init(&match->lock);
1322                 atomic_set(&match->sk_ref, 0);
1323                 match->prot_hook.type = po->prot_hook.type;
1324                 match->prot_hook.dev = po->prot_hook.dev;
1325                 match->prot_hook.func = packet_rcv_fanout;
1326                 match->prot_hook.af_packet_priv = match;
1327                 match->prot_hook.id_match = match_fanout_group;
1328                 dev_add_pack(&match->prot_hook);
1329                 list_add(&match->list, &fanout_list);
1330         }
1331         err = -EINVAL;
1332         if (match->type == type &&
1333             match->prot_hook.type == po->prot_hook.type &&
1334             match->prot_hook.dev == po->prot_hook.dev) {
1335                 err = -ENOSPC;
1336                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1337                         __dev_remove_pack(&po->prot_hook);
1338                         po->fanout = match;
1339                         atomic_inc(&match->sk_ref);
1340                         __fanout_link(sk, po);
1341                         err = 0;
1342                 }
1343         }
1344 out:
1345         mutex_unlock(&fanout_mutex);
1346         return err;
1347 }
1348
1349 static void fanout_release(struct sock *sk)
1350 {
1351         struct packet_sock *po = pkt_sk(sk);
1352         struct packet_fanout *f;
1353
1354         f = po->fanout;
1355         if (!f)
1356                 return;
1357
1358         mutex_lock(&fanout_mutex);
1359         po->fanout = NULL;
1360
1361         if (atomic_dec_and_test(&f->sk_ref)) {
1362                 list_del(&f->list);
1363                 dev_remove_pack(&f->prot_hook);
1364                 kfree(f);
1365         }
1366         mutex_unlock(&fanout_mutex);
1367 }
1368
1369 static const struct proto_ops packet_ops;
1370
1371 static const struct proto_ops packet_ops_spkt;
1372
1373 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1374                            struct packet_type *pt, struct net_device *orig_dev)
1375 {
1376         struct sock *sk;
1377         struct sockaddr_pkt *spkt;
1378
1379         /*
1380          *      When we registered the protocol we saved the socket in the data
1381          *      field for just this event.
1382          */
1383
1384         sk = pt->af_packet_priv;
1385
1386         /*
1387          *      Yank back the headers [hope the device set this
1388          *      right or kerboom...]
1389          *
1390          *      Incoming packets have ll header pulled,
1391          *      push it back.
1392          *
1393          *      For outgoing ones skb->data == skb_mac_header(skb)
1394          *      so that this procedure is noop.
1395          */
1396
1397         if (skb->pkt_type == PACKET_LOOPBACK)
1398                 goto out;
1399
1400         if (!net_eq(dev_net(dev), sock_net(sk)))
1401                 goto out;
1402
1403         skb = skb_share_check(skb, GFP_ATOMIC);
1404         if (skb == NULL)
1405                 goto oom;
1406
1407         /* drop any routing info */
1408         skb_dst_drop(skb);
1409
1410         /* drop conntrack reference */
1411         nf_reset(skb);
1412
1413         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1414
1415         skb_push(skb, skb->data - skb_mac_header(skb));
1416
1417         /*
1418          *      The SOCK_PACKET socket receives _all_ frames.
1419          */
1420
1421         spkt->spkt_family = dev->type;
1422         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1423         spkt->spkt_protocol = skb->protocol;
1424
1425         /*
1426          *      Charge the memory to the socket. This is done specifically
1427          *      to prevent sockets using all the memory up.
1428          */
1429
1430         if (sock_queue_rcv_skb(sk, skb) == 0)
1431                 return 0;
1432
1433 out:
1434         kfree_skb(skb);
1435 oom:
1436         return 0;
1437 }
1438
1439
1440 /*
1441  *      Output a raw packet to a device layer. This bypasses all the other
1442  *      protocol layers and you must therefore supply it with a complete frame
1443  */
1444
1445 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1446                                struct msghdr *msg, size_t len)
1447 {
1448         struct sock *sk = sock->sk;
1449         struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1450         struct sk_buff *skb = NULL;
1451         struct net_device *dev;
1452         __be16 proto = 0;
1453         int err;
1454         int extra_len = 0;
1455
1456         /*
1457          *      Get and verify the address.
1458          */
1459
1460         if (saddr) {
1461                 if (msg->msg_namelen < sizeof(struct sockaddr))
1462                         return -EINVAL;
1463                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1464                         proto = saddr->spkt_protocol;
1465         } else
1466                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1467
1468         /*
1469          *      Find the device first to size check it
1470          */
1471
1472         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1473 retry:
1474         rcu_read_lock();
1475         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1476         err = -ENODEV;
1477         if (dev == NULL)
1478                 goto out_unlock;
1479
1480         err = -ENETDOWN;
1481         if (!(dev->flags & IFF_UP))
1482                 goto out_unlock;
1483
1484         /*
1485          * You may not queue a frame bigger than the mtu. This is the lowest level
1486          * raw protocol and you must do your own fragmentation at this level.
1487          */
1488
1489         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1490                 if (!netif_supports_nofcs(dev)) {
1491                         err = -EPROTONOSUPPORT;
1492                         goto out_unlock;
1493                 }
1494                 extra_len = 4; /* We're doing our own CRC */
1495         }
1496
1497         err = -EMSGSIZE;
1498         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1499                 goto out_unlock;
1500
1501         if (!skb) {
1502                 size_t reserved = LL_RESERVED_SPACE(dev);
1503                 int tlen = dev->needed_tailroom;
1504                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1505
1506                 rcu_read_unlock();
1507                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1508                 if (skb == NULL)
1509                         return -ENOBUFS;
1510                 /* FIXME: Save some space for broken drivers that write a hard
1511                  * header at transmission time by themselves. PPP is the notable
1512                  * one here. This should really be fixed at the driver level.
1513                  */
1514                 skb_reserve(skb, reserved);
1515                 skb_reset_network_header(skb);
1516
1517                 /* Try to align data part correctly */
1518                 if (hhlen) {
1519                         skb->data -= hhlen;
1520                         skb->tail -= hhlen;
1521                         if (len < hhlen)
1522                                 skb_reset_network_header(skb);
1523                 }
1524                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1525                 if (err)
1526                         goto out_free;
1527                 goto retry;
1528         }
1529
1530         if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1531                 /* Earlier code assumed this would be a VLAN pkt,
1532                  * double-check this now that we have the actual
1533                  * packet in hand.
1534                  */
1535                 struct ethhdr *ehdr;
1536                 skb_reset_mac_header(skb);
1537                 ehdr = eth_hdr(skb);
1538                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1539                         err = -EMSGSIZE;
1540                         goto out_unlock;
1541                 }
1542         }
1543
1544         skb->protocol = proto;
1545         skb->dev = dev;
1546         skb->priority = sk->sk_priority;
1547         skb->mark = sk->sk_mark;
1548
1549         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1550
1551         if (unlikely(extra_len == 4))
1552                 skb->no_fcs = 1;
1553
1554         skb_probe_transport_header(skb, 0);
1555
1556         dev_queue_xmit(skb);
1557         rcu_read_unlock();
1558         return len;
1559
1560 out_unlock:
1561         rcu_read_unlock();
1562 out_free:
1563         kfree_skb(skb);
1564         return err;
1565 }
1566
1567 static unsigned int run_filter(const struct sk_buff *skb,
1568                                       const struct sock *sk,
1569                                       unsigned int res)
1570 {
1571         struct sk_filter *filter;
1572
1573         rcu_read_lock();
1574         filter = rcu_dereference(sk->sk_filter);
1575         if (filter != NULL)
1576                 res = SK_RUN_FILTER(filter, skb);
1577         rcu_read_unlock();
1578
1579         return res;
1580 }
1581
1582 /*
1583  * This function makes lazy skb cloning in hope that most of packets
1584  * are discarded by BPF.
1585  *
1586  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1587  * and skb->cb are mangled. It works because (and until) packets
1588  * falling here are owned by current CPU. Output packets are cloned
1589  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1590  * sequencially, so that if we return skb to original state on exit,
1591  * we will not harm anyone.
1592  */
1593
1594 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1595                       struct packet_type *pt, struct net_device *orig_dev)
1596 {
1597         struct sock *sk;
1598         struct sockaddr_ll *sll;
1599         struct packet_sock *po;
1600         u8 *skb_head = skb->data;
1601         int skb_len = skb->len;
1602         unsigned int snaplen, res;
1603
1604         if (skb->pkt_type == PACKET_LOOPBACK)
1605                 goto drop;
1606
1607         sk = pt->af_packet_priv;
1608         po = pkt_sk(sk);
1609
1610         if (!net_eq(dev_net(dev), sock_net(sk)))
1611                 goto drop;
1612
1613         skb->dev = dev;
1614
1615         if (dev->header_ops) {
1616                 /* The device has an explicit notion of ll header,
1617                  * exported to higher levels.
1618                  *
1619                  * Otherwise, the device hides details of its frame
1620                  * structure, so that corresponding packet head is
1621                  * never delivered to user.
1622                  */
1623                 if (sk->sk_type != SOCK_DGRAM)
1624                         skb_push(skb, skb->data - skb_mac_header(skb));
1625                 else if (skb->pkt_type == PACKET_OUTGOING) {
1626                         /* Special case: outgoing packets have ll header at head */
1627                         skb_pull(skb, skb_network_offset(skb));
1628                 }
1629         }
1630
1631         snaplen = skb->len;
1632
1633         res = run_filter(skb, sk, snaplen);
1634         if (!res)
1635                 goto drop_n_restore;
1636         if (snaplen > res)
1637                 snaplen = res;
1638
1639         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1640                 goto drop_n_acct;
1641
1642         if (skb_shared(skb)) {
1643                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1644                 if (nskb == NULL)
1645                         goto drop_n_acct;
1646
1647                 if (skb_head != skb->data) {
1648                         skb->data = skb_head;
1649                         skb->len = skb_len;
1650                 }
1651                 consume_skb(skb);
1652                 skb = nskb;
1653         }
1654
1655         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1656                      sizeof(skb->cb));
1657
1658         sll = &PACKET_SKB_CB(skb)->sa.ll;
1659         sll->sll_family = AF_PACKET;
1660         sll->sll_hatype = dev->type;
1661         sll->sll_protocol = skb->protocol;
1662         sll->sll_pkttype = skb->pkt_type;
1663         if (unlikely(po->origdev))
1664                 sll->sll_ifindex = orig_dev->ifindex;
1665         else
1666                 sll->sll_ifindex = dev->ifindex;
1667
1668         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1669
1670         PACKET_SKB_CB(skb)->origlen = skb->len;
1671
1672         if (pskb_trim(skb, snaplen))
1673                 goto drop_n_acct;
1674
1675         skb_set_owner_r(skb, sk);
1676         skb->dev = NULL;
1677         skb_dst_drop(skb);
1678
1679         /* drop conntrack reference */
1680         nf_reset(skb);
1681
1682         spin_lock(&sk->sk_receive_queue.lock);
1683         po->stats.stats1.tp_packets++;
1684         skb->dropcount = atomic_read(&sk->sk_drops);
1685         __skb_queue_tail(&sk->sk_receive_queue, skb);
1686         spin_unlock(&sk->sk_receive_queue.lock);
1687         sk->sk_data_ready(sk, skb->len);
1688         return 0;
1689
1690 drop_n_acct:
1691         spin_lock(&sk->sk_receive_queue.lock);
1692         po->stats.stats1.tp_drops++;
1693         atomic_inc(&sk->sk_drops);
1694         spin_unlock(&sk->sk_receive_queue.lock);
1695
1696 drop_n_restore:
1697         if (skb_head != skb->data && skb_shared(skb)) {
1698                 skb->data = skb_head;
1699                 skb->len = skb_len;
1700         }
1701 drop:
1702         consume_skb(skb);
1703         return 0;
1704 }
1705
1706 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1707                        struct packet_type *pt, struct net_device *orig_dev)
1708 {
1709         struct sock *sk;
1710         struct packet_sock *po;
1711         struct sockaddr_ll *sll;
1712         union tpacket_uhdr h;
1713         u8 *skb_head = skb->data;
1714         int skb_len = skb->len;
1715         unsigned int snaplen, res;
1716         unsigned long status = TP_STATUS_USER;
1717         unsigned short macoff, netoff, hdrlen;
1718         struct sk_buff *copy_skb = NULL;
1719         struct timespec ts;
1720         __u32 ts_status;
1721
1722         if (skb->pkt_type == PACKET_LOOPBACK)
1723                 goto drop;
1724
1725         sk = pt->af_packet_priv;
1726         po = pkt_sk(sk);
1727
1728         if (!net_eq(dev_net(dev), sock_net(sk)))
1729                 goto drop;
1730
1731         if (dev->header_ops) {
1732                 if (sk->sk_type != SOCK_DGRAM)
1733                         skb_push(skb, skb->data - skb_mac_header(skb));
1734                 else if (skb->pkt_type == PACKET_OUTGOING) {
1735                         /* Special case: outgoing packets have ll header at head */
1736                         skb_pull(skb, skb_network_offset(skb));
1737                 }
1738         }
1739
1740         if (skb->ip_summed == CHECKSUM_PARTIAL)
1741                 status |= TP_STATUS_CSUMNOTREADY;
1742
1743         snaplen = skb->len;
1744
1745         res = run_filter(skb, sk, snaplen);
1746         if (!res)
1747                 goto drop_n_restore;
1748         if (snaplen > res)
1749                 snaplen = res;
1750
1751         if (sk->sk_type == SOCK_DGRAM) {
1752                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1753                                   po->tp_reserve;
1754         } else {
1755                 unsigned int maclen = skb_network_offset(skb);
1756                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1757                                        (maclen < 16 ? 16 : maclen)) +
1758                         po->tp_reserve;
1759                 macoff = netoff - maclen;
1760         }
1761         if (po->tp_version <= TPACKET_V2) {
1762                 if (macoff + snaplen > po->rx_ring.frame_size) {
1763                         if (po->copy_thresh &&
1764                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1765                                 if (skb_shared(skb)) {
1766                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1767                                 } else {
1768                                         copy_skb = skb_get(skb);
1769                                         skb_head = skb->data;
1770                                 }
1771                                 if (copy_skb)
1772                                         skb_set_owner_r(copy_skb, sk);
1773                         }
1774                         snaplen = po->rx_ring.frame_size - macoff;
1775                         if ((int)snaplen < 0)
1776                                 snaplen = 0;
1777                 }
1778         }
1779         spin_lock(&sk->sk_receive_queue.lock);
1780         h.raw = packet_current_rx_frame(po, skb,
1781                                         TP_STATUS_KERNEL, (macoff+snaplen));
1782         if (!h.raw)
1783                 goto ring_is_full;
1784         if (po->tp_version <= TPACKET_V2) {
1785                 packet_increment_rx_head(po, &po->rx_ring);
1786         /*
1787          * LOSING will be reported till you read the stats,
1788          * because it's COR - Clear On Read.
1789          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1790          * at packet level.
1791          */
1792                 if (po->stats.stats1.tp_drops)
1793                         status |= TP_STATUS_LOSING;
1794         }
1795         po->stats.stats1.tp_packets++;
1796         if (copy_skb) {
1797                 status |= TP_STATUS_COPY;
1798                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1799         }
1800         spin_unlock(&sk->sk_receive_queue.lock);
1801
1802         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1803
1804         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
1805                 getnstimeofday(&ts);
1806
1807         status |= ts_status;
1808
1809         switch (po->tp_version) {
1810         case TPACKET_V1:
1811                 h.h1->tp_len = skb->len;
1812                 h.h1->tp_snaplen = snaplen;
1813                 h.h1->tp_mac = macoff;
1814                 h.h1->tp_net = netoff;
1815                 h.h1->tp_sec = ts.tv_sec;
1816                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
1817                 hdrlen = sizeof(*h.h1);
1818                 break;
1819         case TPACKET_V2:
1820                 h.h2->tp_len = skb->len;
1821                 h.h2->tp_snaplen = snaplen;
1822                 h.h2->tp_mac = macoff;
1823                 h.h2->tp_net = netoff;
1824                 h.h2->tp_sec = ts.tv_sec;
1825                 h.h2->tp_nsec = ts.tv_nsec;
1826                 if (vlan_tx_tag_present(skb)) {
1827                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1828                         status |= TP_STATUS_VLAN_VALID;
1829                 } else {
1830                         h.h2->tp_vlan_tci = 0;
1831                 }
1832                 h.h2->tp_padding = 0;
1833                 hdrlen = sizeof(*h.h2);
1834                 break;
1835         case TPACKET_V3:
1836                 /* tp_nxt_offset,vlan are already populated above.
1837                  * So DONT clear those fields here
1838                  */
1839                 h.h3->tp_status |= status;
1840                 h.h3->tp_len = skb->len;
1841                 h.h3->tp_snaplen = snaplen;
1842                 h.h3->tp_mac = macoff;
1843                 h.h3->tp_net = netoff;
1844                 h.h3->tp_sec  = ts.tv_sec;
1845                 h.h3->tp_nsec = ts.tv_nsec;
1846                 hdrlen = sizeof(*h.h3);
1847                 break;
1848         default:
1849                 BUG();
1850         }
1851
1852         sll = h.raw + TPACKET_ALIGN(hdrlen);
1853         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1854         sll->sll_family = AF_PACKET;
1855         sll->sll_hatype = dev->type;
1856         sll->sll_protocol = skb->protocol;
1857         sll->sll_pkttype = skb->pkt_type;
1858         if (unlikely(po->origdev))
1859                 sll->sll_ifindex = orig_dev->ifindex;
1860         else
1861                 sll->sll_ifindex = dev->ifindex;
1862
1863         smp_mb();
1864 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1865         {
1866                 u8 *start, *end;
1867
1868                 if (po->tp_version <= TPACKET_V2) {
1869                         end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1870                                 + macoff + snaplen);
1871                         for (start = h.raw; start < end; start += PAGE_SIZE)
1872                                 flush_dcache_page(pgv_to_page(start));
1873                 }
1874                 smp_wmb();
1875         }
1876 #endif
1877         if (po->tp_version <= TPACKET_V2)
1878                 __packet_set_status(po, h.raw, status);
1879         else
1880                 prb_clear_blk_fill_status(&po->rx_ring);
1881
1882         sk->sk_data_ready(sk, 0);
1883
1884 drop_n_restore:
1885         if (skb_head != skb->data && skb_shared(skb)) {
1886                 skb->data = skb_head;
1887                 skb->len = skb_len;
1888         }
1889 drop:
1890         kfree_skb(skb);
1891         return 0;
1892
1893 ring_is_full:
1894         po->stats.stats1.tp_drops++;
1895         spin_unlock(&sk->sk_receive_queue.lock);
1896
1897         sk->sk_data_ready(sk, 0);
1898         kfree_skb(copy_skb);
1899         goto drop_n_restore;
1900 }
1901
1902 static void tpacket_destruct_skb(struct sk_buff *skb)
1903 {
1904         struct packet_sock *po = pkt_sk(skb->sk);
1905         void *ph;
1906
1907         if (likely(po->tx_ring.pg_vec)) {
1908                 __u32 ts;
1909
1910                 ph = skb_shinfo(skb)->destructor_arg;
1911                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1912                 atomic_dec(&po->tx_ring.pending);
1913
1914                 ts = __packet_set_timestamp(po, ph, skb);
1915                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
1916         }
1917
1918         sock_wfree(skb);
1919 }
1920
1921 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1922                 void *frame, struct net_device *dev, int size_max,
1923                 __be16 proto, unsigned char *addr, int hlen)
1924 {
1925         union tpacket_uhdr ph;
1926         int to_write, offset, len, tp_len, nr_frags, len_max;
1927         struct socket *sock = po->sk.sk_socket;
1928         struct page *page;
1929         void *data;
1930         int err;
1931
1932         ph.raw = frame;
1933
1934         skb->protocol = proto;
1935         skb->dev = dev;
1936         skb->priority = po->sk.sk_priority;
1937         skb->mark = po->sk.sk_mark;
1938         sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
1939         skb_shinfo(skb)->destructor_arg = ph.raw;
1940
1941         switch (po->tp_version) {
1942         case TPACKET_V2:
1943                 tp_len = ph.h2->tp_len;
1944                 break;
1945         default:
1946                 tp_len = ph.h1->tp_len;
1947                 break;
1948         }
1949         if (unlikely(tp_len > size_max)) {
1950                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1951                 return -EMSGSIZE;
1952         }
1953
1954         skb_reserve(skb, hlen);
1955         skb_reset_network_header(skb);
1956         skb_probe_transport_header(skb, 0);
1957
1958         if (po->tp_tx_has_off) {
1959                 int off_min, off_max, off;
1960                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
1961                 off_max = po->tx_ring.frame_size - tp_len;
1962                 if (sock->type == SOCK_DGRAM) {
1963                         switch (po->tp_version) {
1964                         case TPACKET_V2:
1965                                 off = ph.h2->tp_net;
1966                                 break;
1967                         default:
1968                                 off = ph.h1->tp_net;
1969                                 break;
1970                         }
1971                 } else {
1972                         switch (po->tp_version) {
1973                         case TPACKET_V2:
1974                                 off = ph.h2->tp_mac;
1975                                 break;
1976                         default:
1977                                 off = ph.h1->tp_mac;
1978                                 break;
1979                         }
1980                 }
1981                 if (unlikely((off < off_min) || (off_max < off)))
1982                         return -EINVAL;
1983                 data = ph.raw + off;
1984         } else {
1985                 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1986         }
1987         to_write = tp_len;
1988
1989         if (sock->type == SOCK_DGRAM) {
1990                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
1991                                 NULL, tp_len);
1992                 if (unlikely(err < 0))
1993                         return -EINVAL;
1994         } else if (dev->hard_header_len) {
1995                 /* net device doesn't like empty head */
1996                 if (unlikely(tp_len <= dev->hard_header_len)) {
1997                         pr_err("packet size is too short (%d < %d)\n",
1998                                tp_len, dev->hard_header_len);
1999                         return -EINVAL;
2000                 }
2001
2002                 skb_push(skb, dev->hard_header_len);
2003                 err = skb_store_bits(skb, 0, data,
2004                                 dev->hard_header_len);
2005                 if (unlikely(err))
2006                         return err;
2007
2008                 data += dev->hard_header_len;
2009                 to_write -= dev->hard_header_len;
2010         }
2011
2012         offset = offset_in_page(data);
2013         len_max = PAGE_SIZE - offset;
2014         len = ((to_write > len_max) ? len_max : to_write);
2015
2016         skb->data_len = to_write;
2017         skb->len += to_write;
2018         skb->truesize += to_write;
2019         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2020
2021         while (likely(to_write)) {
2022                 nr_frags = skb_shinfo(skb)->nr_frags;
2023
2024                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2025                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2026                                MAX_SKB_FRAGS);
2027                         return -EFAULT;
2028                 }
2029
2030                 page = pgv_to_page(data);
2031                 data += len;
2032                 flush_dcache_page(page);
2033                 get_page(page);
2034                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2035                 to_write -= len;
2036                 offset = 0;
2037                 len_max = PAGE_SIZE;
2038                 len = ((to_write > len_max) ? len_max : to_write);
2039         }
2040
2041         return tp_len;
2042 }
2043
2044 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2045 {
2046         struct sk_buff *skb;
2047         struct net_device *dev;
2048         __be16 proto;
2049         bool need_rls_dev = false;
2050         int err, reserve = 0;
2051         void *ph;
2052         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2053         int tp_len, size_max;
2054         unsigned char *addr;
2055         int len_sum = 0;
2056         int status = TP_STATUS_AVAILABLE;
2057         int hlen, tlen;
2058
2059         mutex_lock(&po->pg_vec_lock);
2060
2061         if (saddr == NULL) {
2062                 dev = po->prot_hook.dev;
2063                 proto   = po->num;
2064                 addr    = NULL;
2065         } else {
2066                 err = -EINVAL;
2067                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2068                         goto out;
2069                 if (msg->msg_namelen < (saddr->sll_halen
2070                                         + offsetof(struct sockaddr_ll,
2071                                                 sll_addr)))
2072                         goto out;
2073                 proto   = saddr->sll_protocol;
2074                 addr    = saddr->sll_addr;
2075                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2076                 need_rls_dev = true;
2077         }
2078
2079         err = -ENXIO;
2080         if (unlikely(dev == NULL))
2081                 goto out;
2082
2083         reserve = dev->hard_header_len;
2084
2085         err = -ENETDOWN;
2086         if (unlikely(!(dev->flags & IFF_UP)))
2087                 goto out_put;
2088
2089         size_max = po->tx_ring.frame_size
2090                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2091
2092         if (size_max > dev->mtu + reserve)
2093                 size_max = dev->mtu + reserve;
2094
2095         do {
2096                 ph = packet_current_frame(po, &po->tx_ring,
2097                                 TP_STATUS_SEND_REQUEST);
2098
2099                 if (unlikely(ph == NULL)) {
2100                         schedule();
2101                         continue;
2102                 }
2103
2104                 status = TP_STATUS_SEND_REQUEST;
2105                 hlen = LL_RESERVED_SPACE(dev);
2106                 tlen = dev->needed_tailroom;
2107                 skb = sock_alloc_send_skb(&po->sk,
2108                                 hlen + tlen + sizeof(struct sockaddr_ll),
2109                                 0, &err);
2110
2111                 if (unlikely(skb == NULL))
2112                         goto out_status;
2113
2114                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2115                                 addr, hlen);
2116
2117                 if (unlikely(tp_len < 0)) {
2118                         if (po->tp_loss) {
2119                                 __packet_set_status(po, ph,
2120                                                 TP_STATUS_AVAILABLE);
2121                                 packet_increment_head(&po->tx_ring);
2122                                 kfree_skb(skb);
2123                                 continue;
2124                         } else {
2125                                 status = TP_STATUS_WRONG_FORMAT;
2126                                 err = tp_len;
2127                                 goto out_status;
2128                         }
2129                 }
2130
2131                 skb->destructor = tpacket_destruct_skb;
2132                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2133                 atomic_inc(&po->tx_ring.pending);
2134
2135                 status = TP_STATUS_SEND_REQUEST;
2136                 err = dev_queue_xmit(skb);
2137                 if (unlikely(err > 0)) {
2138                         err = net_xmit_errno(err);
2139                         if (err && __packet_get_status(po, ph) ==
2140                                    TP_STATUS_AVAILABLE) {
2141                                 /* skb was destructed already */
2142                                 skb = NULL;
2143                                 goto out_status;
2144                         }
2145                         /*
2146                          * skb was dropped but not destructed yet;
2147                          * let's treat it like congestion or err < 0
2148                          */
2149                         err = 0;
2150                 }
2151                 packet_increment_head(&po->tx_ring);
2152                 len_sum += tp_len;
2153         } while (likely((ph != NULL) ||
2154                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2155                          (atomic_read(&po->tx_ring.pending))))
2156                 );
2157
2158         err = len_sum;
2159         goto out_put;
2160
2161 out_status:
2162         __packet_set_status(po, ph, status);
2163         kfree_skb(skb);
2164 out_put:
2165         if (need_rls_dev)
2166                 dev_put(dev);
2167 out:
2168         mutex_unlock(&po->pg_vec_lock);
2169         return err;
2170 }
2171
2172 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2173                                         size_t reserve, size_t len,
2174                                         size_t linear, int noblock,
2175                                         int *err)
2176 {
2177         struct sk_buff *skb;
2178
2179         /* Under a page?  Don't bother with paged skb. */
2180         if (prepad + len < PAGE_SIZE || !linear)
2181                 linear = len;
2182
2183         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2184                                    err, 0);
2185         if (!skb)
2186                 return NULL;
2187
2188         skb_reserve(skb, reserve);
2189         skb_put(skb, linear);
2190         skb->data_len = len - linear;
2191         skb->len += len - linear;
2192
2193         return skb;
2194 }
2195
2196 static int packet_snd(struct socket *sock,
2197                           struct msghdr *msg, size_t len)
2198 {
2199         struct sock *sk = sock->sk;
2200         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2201         struct sk_buff *skb;
2202         struct net_device *dev;
2203         __be16 proto;
2204         bool need_rls_dev = false;
2205         unsigned char *addr;
2206         int err, reserve = 0;
2207         struct virtio_net_hdr vnet_hdr = { 0 };
2208         int offset = 0;
2209         int vnet_hdr_len;
2210         struct packet_sock *po = pkt_sk(sk);
2211         unsigned short gso_type = 0;
2212         int hlen, tlen;
2213         int extra_len = 0;
2214
2215         /*
2216          *      Get and verify the address.
2217          */
2218
2219         if (saddr == NULL) {
2220                 dev = po->prot_hook.dev;
2221                 proto   = po->num;
2222                 addr    = NULL;
2223         } else {
2224                 err = -EINVAL;
2225                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2226                         goto out;
2227                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2228                         goto out;
2229                 proto   = saddr->sll_protocol;
2230                 addr    = saddr->sll_addr;
2231                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2232                 need_rls_dev = true;
2233         }
2234
2235         err = -ENXIO;
2236         if (dev == NULL)
2237                 goto out_unlock;
2238         if (sock->type == SOCK_RAW)
2239                 reserve = dev->hard_header_len;
2240
2241         err = -ENETDOWN;
2242         if (!(dev->flags & IFF_UP))
2243                 goto out_unlock;
2244
2245         if (po->has_vnet_hdr) {
2246                 vnet_hdr_len = sizeof(vnet_hdr);
2247
2248                 err = -EINVAL;
2249                 if (len < vnet_hdr_len)
2250                         goto out_unlock;
2251
2252                 len -= vnet_hdr_len;
2253
2254                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2255                                        vnet_hdr_len);
2256                 if (err < 0)
2257                         goto out_unlock;
2258
2259                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2260                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2261                       vnet_hdr.hdr_len))
2262                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
2263                                                  vnet_hdr.csum_offset + 2;
2264
2265                 err = -EINVAL;
2266                 if (vnet_hdr.hdr_len > len)
2267                         goto out_unlock;
2268
2269                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2270                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2271                         case VIRTIO_NET_HDR_GSO_TCPV4:
2272                                 gso_type = SKB_GSO_TCPV4;
2273                                 break;
2274                         case VIRTIO_NET_HDR_GSO_TCPV6:
2275                                 gso_type = SKB_GSO_TCPV6;
2276                                 break;
2277                         case VIRTIO_NET_HDR_GSO_UDP:
2278                                 gso_type = SKB_GSO_UDP;
2279                                 break;
2280                         default:
2281                                 goto out_unlock;
2282                         }
2283
2284                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2285                                 gso_type |= SKB_GSO_TCP_ECN;
2286
2287                         if (vnet_hdr.gso_size == 0)
2288                                 goto out_unlock;
2289
2290                 }
2291         }
2292
2293         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2294                 if (!netif_supports_nofcs(dev)) {
2295                         err = -EPROTONOSUPPORT;
2296                         goto out_unlock;
2297                 }
2298                 extra_len = 4; /* We're doing our own CRC */
2299         }
2300
2301         err = -EMSGSIZE;
2302         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2303                 goto out_unlock;
2304
2305         err = -ENOBUFS;
2306         hlen = LL_RESERVED_SPACE(dev);
2307         tlen = dev->needed_tailroom;
2308         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2309                                msg->msg_flags & MSG_DONTWAIT, &err);
2310         if (skb == NULL)
2311                 goto out_unlock;
2312
2313         skb_set_network_header(skb, reserve);
2314
2315         err = -EINVAL;
2316         if (sock->type == SOCK_DGRAM &&
2317             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2318                 goto out_free;
2319
2320         /* Returns -EFAULT on error */
2321         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2322         if (err)
2323                 goto out_free;
2324
2325         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2326
2327         if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2328                 /* Earlier code assumed this would be a VLAN pkt,
2329                  * double-check this now that we have the actual
2330                  * packet in hand.
2331                  */
2332                 struct ethhdr *ehdr;
2333                 skb_reset_mac_header(skb);
2334                 ehdr = eth_hdr(skb);
2335                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2336                         err = -EMSGSIZE;
2337                         goto out_free;
2338                 }
2339         }
2340
2341         skb->protocol = proto;
2342         skb->dev = dev;
2343         skb->priority = sk->sk_priority;
2344         skb->mark = sk->sk_mark;
2345
2346         if (po->has_vnet_hdr) {
2347                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2348                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2349                                                   vnet_hdr.csum_offset)) {
2350                                 err = -EINVAL;
2351                                 goto out_free;
2352                         }
2353                 }
2354
2355                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2356                 skb_shinfo(skb)->gso_type = gso_type;
2357
2358                 /* Header must be checked, and gso_segs computed. */
2359                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2360                 skb_shinfo(skb)->gso_segs = 0;
2361
2362                 len += vnet_hdr_len;
2363         }
2364
2365         skb_probe_transport_header(skb, reserve);
2366
2367         if (unlikely(extra_len == 4))
2368                 skb->no_fcs = 1;
2369
2370         /*
2371          *      Now send it
2372          */
2373
2374         err = dev_queue_xmit(skb);
2375         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2376                 goto out_unlock;
2377
2378         if (need_rls_dev)
2379                 dev_put(dev);
2380
2381         return len;
2382
2383 out_free:
2384         kfree_skb(skb);
2385 out_unlock:
2386         if (dev && need_rls_dev)
2387                 dev_put(dev);
2388 out:
2389         return err;
2390 }
2391
2392 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2393                 struct msghdr *msg, size_t len)
2394 {
2395         struct sock *sk = sock->sk;
2396         struct packet_sock *po = pkt_sk(sk);
2397         if (po->tx_ring.pg_vec)
2398                 return tpacket_snd(po, msg);
2399         else
2400                 return packet_snd(sock, msg, len);
2401 }
2402
2403 /*
2404  *      Close a PACKET socket. This is fairly simple. We immediately go
2405  *      to 'closed' state and remove our protocol entry in the device list.
2406  */
2407
2408 static int packet_release(struct socket *sock)
2409 {
2410         struct sock *sk = sock->sk;
2411         struct packet_sock *po;
2412         struct net *net;
2413         union tpacket_req_u req_u;
2414
2415         if (!sk)
2416                 return 0;
2417
2418         net = sock_net(sk);
2419         po = pkt_sk(sk);
2420
2421         mutex_lock(&net->packet.sklist_lock);
2422         sk_del_node_init_rcu(sk);
2423         mutex_unlock(&net->packet.sklist_lock);
2424
2425         preempt_disable();
2426         sock_prot_inuse_add(net, sk->sk_prot, -1);
2427         preempt_enable();
2428
2429         spin_lock(&po->bind_lock);
2430         unregister_prot_hook(sk, false);
2431         if (po->prot_hook.dev) {
2432                 dev_put(po->prot_hook.dev);
2433                 po->prot_hook.dev = NULL;
2434         }
2435         spin_unlock(&po->bind_lock);
2436
2437         packet_flush_mclist(sk);
2438
2439         if (po->rx_ring.pg_vec) {
2440                 memset(&req_u, 0, sizeof(req_u));
2441                 packet_set_ring(sk, &req_u, 1, 0);
2442         }
2443
2444         if (po->tx_ring.pg_vec) {
2445                 memset(&req_u, 0, sizeof(req_u));
2446                 packet_set_ring(sk, &req_u, 1, 1);
2447         }
2448
2449         fanout_release(sk);
2450
2451         synchronize_net();
2452         /*
2453          *      Now the socket is dead. No more input will appear.
2454          */
2455         sock_orphan(sk);
2456         sock->sk = NULL;
2457
2458         /* Purge queues */
2459
2460         skb_queue_purge(&sk->sk_receive_queue);
2461         sk_refcnt_debug_release(sk);
2462
2463         sock_put(sk);
2464         return 0;
2465 }
2466
2467 /*
2468  *      Attach a packet hook.
2469  */
2470
2471 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2472 {
2473         struct packet_sock *po = pkt_sk(sk);
2474
2475         if (po->fanout) {
2476                 if (dev)
2477                         dev_put(dev);
2478
2479                 return -EINVAL;
2480         }
2481
2482         lock_sock(sk);
2483
2484         spin_lock(&po->bind_lock);
2485         unregister_prot_hook(sk, true);
2486         po->num = protocol;
2487         po->prot_hook.type = protocol;
2488         if (po->prot_hook.dev)
2489                 dev_put(po->prot_hook.dev);
2490         po->prot_hook.dev = dev;
2491
2492         po->ifindex = dev ? dev->ifindex : 0;
2493
2494         if (protocol == 0)
2495                 goto out_unlock;
2496
2497         if (!dev || (dev->flags & IFF_UP)) {
2498                 register_prot_hook(sk);
2499         } else {
2500                 sk->sk_err = ENETDOWN;
2501                 if (!sock_flag(sk, SOCK_DEAD))
2502                         sk->sk_error_report(sk);
2503         }
2504
2505 out_unlock:
2506         spin_unlock(&po->bind_lock);
2507         release_sock(sk);
2508         return 0;
2509 }
2510
2511 /*
2512  *      Bind a packet socket to a device
2513  */
2514
2515 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2516                             int addr_len)
2517 {
2518         struct sock *sk = sock->sk;
2519         char name[15];
2520         struct net_device *dev;
2521         int err = -ENODEV;
2522
2523         /*
2524          *      Check legality
2525          */
2526
2527         if (addr_len != sizeof(struct sockaddr))
2528                 return -EINVAL;
2529         strlcpy(name, uaddr->sa_data, sizeof(name));
2530
2531         dev = dev_get_by_name(sock_net(sk), name);
2532         if (dev)
2533                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2534         return err;
2535 }
2536
2537 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2538 {
2539         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2540         struct sock *sk = sock->sk;
2541         struct net_device *dev = NULL;
2542         int err;
2543
2544
2545         /*
2546          *      Check legality
2547          */
2548
2549         if (addr_len < sizeof(struct sockaddr_ll))
2550                 return -EINVAL;
2551         if (sll->sll_family != AF_PACKET)
2552                 return -EINVAL;
2553
2554         if (sll->sll_ifindex) {
2555                 err = -ENODEV;
2556                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2557                 if (dev == NULL)
2558                         goto out;
2559         }
2560         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2561
2562 out:
2563         return err;
2564 }
2565
2566 static struct proto packet_proto = {
2567         .name     = "PACKET",
2568         .owner    = THIS_MODULE,
2569         .obj_size = sizeof(struct packet_sock),
2570 };
2571
2572 /*
2573  *      Create a packet of type SOCK_PACKET.
2574  */
2575
2576 static int packet_create(struct net *net, struct socket *sock, int protocol,
2577                          int kern)
2578 {
2579         struct sock *sk;
2580         struct packet_sock *po;
2581         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2582         int err;
2583
2584         if (!ns_capable(net->user_ns, CAP_NET_RAW))
2585                 return -EPERM;
2586         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2587             sock->type != SOCK_PACKET)
2588                 return -ESOCKTNOSUPPORT;
2589
2590         sock->state = SS_UNCONNECTED;
2591
2592         err = -ENOBUFS;
2593         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2594         if (sk == NULL)
2595                 goto out;
2596
2597         sock->ops = &packet_ops;
2598         if (sock->type == SOCK_PACKET)
2599                 sock->ops = &packet_ops_spkt;
2600
2601         sock_init_data(sock, sk);
2602
2603         po = pkt_sk(sk);
2604         sk->sk_family = PF_PACKET;
2605         po->num = proto;
2606
2607         sk->sk_destruct = packet_sock_destruct;
2608         sk_refcnt_debug_inc(sk);
2609
2610         /*
2611          *      Attach a protocol block
2612          */
2613
2614         spin_lock_init(&po->bind_lock);
2615         mutex_init(&po->pg_vec_lock);
2616         po->prot_hook.func = packet_rcv;
2617
2618         if (sock->type == SOCK_PACKET)
2619                 po->prot_hook.func = packet_rcv_spkt;
2620
2621         po->prot_hook.af_packet_priv = sk;
2622
2623         if (proto) {
2624                 po->prot_hook.type = proto;
2625                 register_prot_hook(sk);
2626         }
2627
2628         mutex_lock(&net->packet.sklist_lock);
2629         sk_add_node_rcu(sk, &net->packet.sklist);
2630         mutex_unlock(&net->packet.sklist_lock);
2631
2632         preempt_disable();
2633         sock_prot_inuse_add(net, &packet_proto, 1);
2634         preempt_enable();
2635
2636         return 0;
2637 out:
2638         return err;
2639 }
2640
2641 /*
2642  *      Pull a packet from our receive queue and hand it to the user.
2643  *      If necessary we block.
2644  */
2645
2646 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2647                           struct msghdr *msg, size_t len, int flags)
2648 {
2649         struct sock *sk = sock->sk;
2650         struct sk_buff *skb;
2651         int copied, err;
2652         struct sockaddr_ll *sll;
2653         int vnet_hdr_len = 0;
2654
2655         err = -EINVAL;
2656         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2657                 goto out;
2658
2659 #if 0
2660         /* What error should we return now? EUNATTACH? */
2661         if (pkt_sk(sk)->ifindex < 0)
2662                 return -ENODEV;
2663 #endif
2664
2665         if (flags & MSG_ERRQUEUE) {
2666                 err = sock_recv_errqueue(sk, msg, len,
2667                                          SOL_PACKET, PACKET_TX_TIMESTAMP);
2668                 goto out;
2669         }
2670
2671         /*
2672          *      Call the generic datagram receiver. This handles all sorts
2673          *      of horrible races and re-entrancy so we can forget about it
2674          *      in the protocol layers.
2675          *
2676          *      Now it will return ENETDOWN, if device have just gone down,
2677          *      but then it will block.
2678          */
2679
2680         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2681
2682         /*
2683          *      An error occurred so return it. Because skb_recv_datagram()
2684          *      handles the blocking we don't see and worry about blocking
2685          *      retries.
2686          */
2687
2688         if (skb == NULL)
2689                 goto out;
2690
2691         if (pkt_sk(sk)->has_vnet_hdr) {
2692                 struct virtio_net_hdr vnet_hdr = { 0 };
2693
2694                 err = -EINVAL;
2695                 vnet_hdr_len = sizeof(vnet_hdr);
2696                 if (len < vnet_hdr_len)
2697                         goto out_free;
2698
2699                 len -= vnet_hdr_len;
2700
2701                 if (skb_is_gso(skb)) {
2702                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2703
2704                         /* This is a hint as to how much should be linear. */
2705                         vnet_hdr.hdr_len = skb_headlen(skb);
2706                         vnet_hdr.gso_size = sinfo->gso_size;
2707                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2708                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2709                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2710                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2711                         else if (sinfo->gso_type & SKB_GSO_UDP)
2712                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2713                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2714                                 goto out_free;
2715                         else
2716                                 BUG();
2717                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2718                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2719                 } else
2720                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2721
2722                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2723                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2724                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2725                         vnet_hdr.csum_offset = skb->csum_offset;
2726                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2727                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2728                 } /* else everything is zero */
2729
2730                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2731                                      vnet_hdr_len);
2732                 if (err < 0)
2733                         goto out_free;
2734         }
2735
2736         /*
2737          *      If the address length field is there to be filled in, we fill
2738          *      it in now.
2739          */
2740
2741         sll = &PACKET_SKB_CB(skb)->sa.ll;
2742         if (sock->type == SOCK_PACKET)
2743                 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2744         else
2745                 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2746
2747         /*
2748          *      You lose any data beyond the buffer you gave. If it worries a
2749          *      user program they can ask the device for its MTU anyway.
2750          */
2751
2752         copied = skb->len;
2753         if (copied > len) {
2754                 copied = len;
2755                 msg->msg_flags |= MSG_TRUNC;
2756         }
2757
2758         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2759         if (err)
2760                 goto out_free;
2761
2762         sock_recv_ts_and_drops(msg, sk, skb);
2763
2764         if (msg->msg_name)
2765                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2766                        msg->msg_namelen);
2767
2768         if (pkt_sk(sk)->auxdata) {
2769                 struct tpacket_auxdata aux;
2770
2771                 aux.tp_status = TP_STATUS_USER;
2772                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2773                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2774                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2775                 aux.tp_snaplen = skb->len;
2776                 aux.tp_mac = 0;
2777                 aux.tp_net = skb_network_offset(skb);
2778                 if (vlan_tx_tag_present(skb)) {
2779                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2780                         aux.tp_status |= TP_STATUS_VLAN_VALID;
2781                 } else {
2782                         aux.tp_vlan_tci = 0;
2783                 }
2784                 aux.tp_padding = 0;
2785                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2786         }
2787
2788         /*
2789          *      Free or return the buffer as appropriate. Again this
2790          *      hides all the races and re-entrancy issues from us.
2791          */
2792         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2793
2794 out_free:
2795         skb_free_datagram(sk, skb);
2796 out:
2797         return err;
2798 }
2799
2800 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2801                                int *uaddr_len, int peer)
2802 {
2803         struct net_device *dev;
2804         struct sock *sk = sock->sk;
2805
2806         if (peer)
2807                 return -EOPNOTSUPP;
2808
2809         uaddr->sa_family = AF_PACKET;
2810         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
2811         rcu_read_lock();
2812         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2813         if (dev)
2814                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
2815         rcu_read_unlock();
2816         *uaddr_len = sizeof(*uaddr);
2817
2818         return 0;
2819 }
2820
2821 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2822                           int *uaddr_len, int peer)
2823 {
2824         struct net_device *dev;
2825         struct sock *sk = sock->sk;
2826         struct packet_sock *po = pkt_sk(sk);
2827         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2828
2829         if (peer)
2830                 return -EOPNOTSUPP;
2831
2832         sll->sll_family = AF_PACKET;
2833         sll->sll_ifindex = po->ifindex;
2834         sll->sll_protocol = po->num;
2835         sll->sll_pkttype = 0;
2836         rcu_read_lock();
2837         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2838         if (dev) {
2839                 sll->sll_hatype = dev->type;
2840                 sll->sll_halen = dev->addr_len;
2841                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2842         } else {
2843                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
2844                 sll->sll_halen = 0;
2845         }
2846         rcu_read_unlock();
2847         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2848
2849         return 0;
2850 }
2851
2852 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2853                          int what)
2854 {
2855         switch (i->type) {
2856         case PACKET_MR_MULTICAST:
2857                 if (i->alen != dev->addr_len)
2858                         return -EINVAL;
2859                 if (what > 0)
2860                         return dev_mc_add(dev, i->addr);
2861                 else
2862                         return dev_mc_del(dev, i->addr);
2863                 break;
2864         case PACKET_MR_PROMISC:
2865                 return dev_set_promiscuity(dev, what);
2866                 break;
2867         case PACKET_MR_ALLMULTI:
2868                 return dev_set_allmulti(dev, what);
2869                 break;
2870         case PACKET_MR_UNICAST:
2871                 if (i->alen != dev->addr_len)
2872                         return -EINVAL;
2873                 if (what > 0)
2874                         return dev_uc_add(dev, i->addr);
2875                 else
2876                         return dev_uc_del(dev, i->addr);
2877                 break;
2878         default:
2879                 break;
2880         }
2881         return 0;
2882 }
2883
2884 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2885 {
2886         for ( ; i; i = i->next) {
2887                 if (i->ifindex == dev->ifindex)
2888                         packet_dev_mc(dev, i, what);
2889         }
2890 }
2891
2892 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2893 {
2894         struct packet_sock *po = pkt_sk(sk);
2895         struct packet_mclist *ml, *i;
2896         struct net_device *dev;
2897         int err;
2898
2899         rtnl_lock();
2900
2901         err = -ENODEV;
2902         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2903         if (!dev)
2904                 goto done;
2905
2906         err = -EINVAL;
2907         if (mreq->mr_alen > dev->addr_len)
2908                 goto done;
2909
2910         err = -ENOBUFS;
2911         i = kmalloc(sizeof(*i), GFP_KERNEL);
2912         if (i == NULL)
2913                 goto done;
2914
2915         err = 0;
2916         for (ml = po->mclist; ml; ml = ml->next) {
2917                 if (ml->ifindex == mreq->mr_ifindex &&
2918                     ml->type == mreq->mr_type &&
2919                     ml->alen == mreq->mr_alen &&
2920                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2921                         ml->count++;
2922                         /* Free the new element ... */
2923                         kfree(i);
2924                         goto done;
2925                 }
2926         }
2927
2928         i->type = mreq->mr_type;
2929         i->ifindex = mreq->mr_ifindex;
2930         i->alen = mreq->mr_alen;
2931         memcpy(i->addr, mreq->mr_address, i->alen);
2932         i->count = 1;
2933         i->next = po->mclist;
2934         po->mclist = i;
2935         err = packet_dev_mc(dev, i, 1);
2936         if (err) {
2937                 po->mclist = i->next;
2938                 kfree(i);
2939         }
2940
2941 done:
2942         rtnl_unlock();
2943         return err;
2944 }
2945
2946 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2947 {
2948         struct packet_mclist *ml, **mlp;
2949
2950         rtnl_lock();
2951
2952         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
2953                 if (ml->ifindex == mreq->mr_ifindex &&
2954                     ml->type == mreq->mr_type &&
2955                     ml->alen == mreq->mr_alen &&
2956                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2957                         if (--ml->count == 0) {
2958                                 struct net_device *dev;
2959                                 *mlp = ml->next;
2960                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2961                                 if (dev)
2962                                         packet_dev_mc(dev, ml, -1);
2963                                 kfree(ml);
2964                         }
2965                         rtnl_unlock();
2966                         return 0;
2967                 }
2968         }
2969         rtnl_unlock();
2970         return -EADDRNOTAVAIL;
2971 }
2972
2973 static void packet_flush_mclist(struct sock *sk)
2974 {
2975         struct packet_sock *po = pkt_sk(sk);
2976         struct packet_mclist *ml;
2977
2978         if (!po->mclist)
2979                 return;
2980
2981         rtnl_lock();
2982         while ((ml = po->mclist) != NULL) {
2983                 struct net_device *dev;
2984
2985                 po->mclist = ml->next;
2986                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2987                 if (dev != NULL)
2988                         packet_dev_mc(dev, ml, -1);
2989                 kfree(ml);
2990         }
2991         rtnl_unlock();
2992 }
2993
2994 static int
2995 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
2996 {
2997         struct sock *sk = sock->sk;
2998         struct packet_sock *po = pkt_sk(sk);
2999         int ret;
3000
3001         if (level != SOL_PACKET)
3002                 return -ENOPROTOOPT;
3003
3004         switch (optname) {
3005         case PACKET_ADD_MEMBERSHIP:
3006         case PACKET_DROP_MEMBERSHIP:
3007         {
3008                 struct packet_mreq_max mreq;
3009                 int len = optlen;
3010                 memset(&mreq, 0, sizeof(mreq));
3011                 if (len < sizeof(struct packet_mreq))
3012                         return -EINVAL;
3013                 if (len > sizeof(mreq))
3014                         len = sizeof(mreq);
3015                 if (copy_from_user(&mreq, optval, len))
3016                         return -EFAULT;
3017                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3018                         return -EINVAL;
3019                 if (optname == PACKET_ADD_MEMBERSHIP)
3020                         ret = packet_mc_add(sk, &mreq);
3021                 else
3022                         ret = packet_mc_drop(sk, &mreq);
3023                 return ret;
3024         }
3025
3026         case PACKET_RX_RING:
3027         case PACKET_TX_RING:
3028         {
3029                 union tpacket_req_u req_u;
3030                 int len;
3031
3032                 switch (po->tp_version) {
3033                 case TPACKET_V1:
3034                 case TPACKET_V2:
3035                         len = sizeof(req_u.req);
3036                         break;
3037                 case TPACKET_V3:
3038                 default:
3039                         len = sizeof(req_u.req3);
3040                         break;
3041                 }
3042                 if (optlen < len)
3043                         return -EINVAL;
3044                 if (pkt_sk(sk)->has_vnet_hdr)
3045                         return -EINVAL;
3046                 if (copy_from_user(&req_u.req, optval, len))
3047                         return -EFAULT;
3048                 return packet_set_ring(sk, &req_u, 0,
3049                         optname == PACKET_TX_RING);
3050         }
3051         case PACKET_COPY_THRESH:
3052         {
3053                 int val;
3054
3055                 if (optlen != sizeof(val))
3056                         return -EINVAL;
3057                 if (copy_from_user(&val, optval, sizeof(val)))
3058                         return -EFAULT;
3059
3060                 pkt_sk(sk)->copy_thresh = val;
3061                 return 0;
3062         }
3063         case PACKET_VERSION:
3064         {
3065                 int val;
3066
3067                 if (optlen != sizeof(val))
3068                         return -EINVAL;
3069                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3070                         return -EBUSY;
3071                 if (copy_from_user(&val, optval, sizeof(val)))
3072                         return -EFAULT;
3073                 switch (val) {
3074                 case TPACKET_V1:
3075                 case TPACKET_V2:
3076                 case TPACKET_V3:
3077                         po->tp_version = val;
3078                         return 0;
3079                 default:
3080                         return -EINVAL;
3081                 }
3082         }
3083         case PACKET_RESERVE:
3084         {
3085                 unsigned int val;
3086
3087                 if (optlen != sizeof(val))
3088                         return -EINVAL;
3089                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3090                         return -EBUSY;
3091                 if (copy_from_user(&val, optval, sizeof(val)))
3092                         return -EFAULT;
3093                 po->tp_reserve = val;
3094                 return 0;
3095         }
3096         case PACKET_LOSS:
3097         {
3098                 unsigned int val;
3099
3100                 if (optlen != sizeof(val))
3101                         return -EINVAL;
3102                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3103                         return -EBUSY;
3104                 if (copy_from_user(&val, optval, sizeof(val)))
3105                         return -EFAULT;
3106                 po->tp_loss = !!val;
3107                 return 0;
3108         }
3109         case PACKET_AUXDATA:
3110         {
3111                 int val;
3112
3113                 if (optlen < sizeof(val))
3114                         return -EINVAL;
3115                 if (copy_from_user(&val, optval, sizeof(val)))
3116                         return -EFAULT;
3117
3118                 po->auxdata = !!val;
3119                 return 0;
3120         }
3121         case PACKET_ORIGDEV:
3122         {
3123                 int val;
3124
3125                 if (optlen < sizeof(val))
3126                         return -EINVAL;
3127                 if (copy_from_user(&val, optval, sizeof(val)))
3128                         return -EFAULT;
3129
3130                 po->origdev = !!val;
3131                 return 0;
3132         }
3133         case PACKET_VNET_HDR:
3134         {
3135                 int val;
3136
3137                 if (sock->type != SOCK_RAW)
3138                         return -EINVAL;
3139                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3140                         return -EBUSY;
3141                 if (optlen < sizeof(val))
3142                         return -EINVAL;
3143                 if (copy_from_user(&val, optval, sizeof(val)))
3144                         return -EFAULT;
3145
3146                 po->has_vnet_hdr = !!val;
3147                 return 0;
3148         }
3149         case PACKET_TIMESTAMP:
3150         {
3151                 int val;
3152
3153                 if (optlen != sizeof(val))
3154                         return -EINVAL;
3155                 if (copy_from_user(&val, optval, sizeof(val)))
3156                         return -EFAULT;
3157
3158                 po->tp_tstamp = val;
3159                 return 0;
3160         }
3161         case PACKET_FANOUT:
3162         {
3163                 int val;
3164
3165                 if (optlen != sizeof(val))
3166                         return -EINVAL;
3167                 if (copy_from_user(&val, optval, sizeof(val)))
3168                         return -EFAULT;
3169
3170                 return fanout_add(sk, val & 0xffff, val >> 16);
3171         }
3172         case PACKET_TX_HAS_OFF:
3173         {
3174                 unsigned int val;
3175
3176                 if (optlen != sizeof(val))
3177                         return -EINVAL;
3178                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3179                         return -EBUSY;
3180                 if (copy_from_user(&val, optval, sizeof(val)))
3181                         return -EFAULT;
3182                 po->tp_tx_has_off = !!val;
3183                 return 0;
3184         }
3185         default:
3186                 return -ENOPROTOOPT;
3187         }
3188 }
3189
3190 static int packet_getsockopt(struct socket *sock, int level, int optname,
3191                              char __user *optval, int __user *optlen)
3192 {
3193         int len;
3194         int val, lv = sizeof(val);
3195         struct sock *sk = sock->sk;
3196         struct packet_sock *po = pkt_sk(sk);
3197         void *data = &val;
3198         union tpacket_stats_u st;
3199
3200         if (level != SOL_PACKET)
3201                 return -ENOPROTOOPT;
3202
3203         if (get_user(len, optlen))
3204                 return -EFAULT;
3205
3206         if (len < 0)
3207                 return -EINVAL;
3208
3209         switch (optname) {
3210         case PACKET_STATISTICS:
3211                 spin_lock_bh(&sk->sk_receive_queue.lock);
3212                 memcpy(&st, &po->stats, sizeof(st));
3213                 memset(&po->stats, 0, sizeof(po->stats));
3214                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3215
3216                 if (po->tp_version == TPACKET_V3) {
3217                         lv = sizeof(struct tpacket_stats_v3);
3218                         st.stats3.tp_packets += st.stats3.tp_drops;
3219                         data = &st.stats3;
3220                 } else {
3221                         lv = sizeof(struct tpacket_stats);
3222                         st.stats1.tp_packets += st.stats1.tp_drops;
3223                         data = &st.stats1;
3224                 }
3225
3226                 break;
3227         case PACKET_AUXDATA:
3228                 val = po->auxdata;
3229                 break;
3230         case PACKET_ORIGDEV:
3231                 val = po->origdev;
3232                 break;
3233         case PACKET_VNET_HDR:
3234                 val = po->has_vnet_hdr;
3235                 break;
3236         case PACKET_VERSION:
3237                 val = po->tp_version;
3238                 break;
3239         case PACKET_HDRLEN:
3240                 if (len > sizeof(int))
3241                         len = sizeof(int);
3242                 if (copy_from_user(&val, optval, len))
3243                         return -EFAULT;
3244                 switch (val) {
3245                 case TPACKET_V1:
3246                         val = sizeof(struct tpacket_hdr);
3247                         break;
3248                 case TPACKET_V2:
3249                         val = sizeof(struct tpacket2_hdr);
3250                         break;
3251                 case TPACKET_V3:
3252                         val = sizeof(struct tpacket3_hdr);
3253                         break;
3254                 default:
3255                         return -EINVAL;
3256                 }
3257                 break;
3258         case PACKET_RESERVE:
3259                 val = po->tp_reserve;
3260                 break;
3261         case PACKET_LOSS:
3262                 val = po->tp_loss;
3263                 break;
3264         case PACKET_TIMESTAMP:
3265                 val = po->tp_tstamp;
3266                 break;
3267         case PACKET_FANOUT:
3268                 val = (po->fanout ?
3269                        ((u32)po->fanout->id |
3270                         ((u32)po->fanout->type << 16) |
3271                         ((u32)po->fanout->flags << 24)) :
3272                        0);
3273                 break;
3274         case PACKET_TX_HAS_OFF:
3275                 val = po->tp_tx_has_off;
3276                 break;
3277         default:
3278                 return -ENOPROTOOPT;
3279         }
3280
3281         if (len > lv)
3282                 len = lv;
3283         if (put_user(len, optlen))
3284                 return -EFAULT;
3285         if (copy_to_user(optval, data, len))
3286                 return -EFAULT;
3287         return 0;
3288 }
3289
3290
3291 static int packet_notifier(struct notifier_block *this,
3292                            unsigned long msg, void *ptr)
3293 {
3294         struct sock *sk;
3295         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3296         struct net *net = dev_net(dev);
3297
3298         rcu_read_lock();
3299         sk_for_each_rcu(sk, &net->packet.sklist) {
3300                 struct packet_sock *po = pkt_sk(sk);
3301
3302                 switch (msg) {
3303                 case NETDEV_UNREGISTER:
3304                         if (po->mclist)
3305                                 packet_dev_mclist(dev, po->mclist, -1);
3306                         /* fallthrough */
3307
3308                 case NETDEV_DOWN:
3309                         if (dev->ifindex == po->ifindex) {
3310                                 spin_lock(&po->bind_lock);
3311                                 if (po->running) {
3312                                         __unregister_prot_hook(sk, false);
3313                                         sk->sk_err = ENETDOWN;
3314                                         if (!sock_flag(sk, SOCK_DEAD))
3315                                                 sk->sk_error_report(sk);
3316                                 }
3317                                 if (msg == NETDEV_UNREGISTER) {
3318                                         po->ifindex = -1;
3319                                         if (po->prot_hook.dev)
3320                                                 dev_put(po->prot_hook.dev);
3321                                         po->prot_hook.dev = NULL;
3322                                 }
3323                                 spin_unlock(&po->bind_lock);
3324                         }
3325                         break;
3326                 case NETDEV_UP:
3327                         if (dev->ifindex == po->ifindex) {
3328                                 spin_lock(&po->bind_lock);
3329                                 if (po->num)
3330                                         register_prot_hook(sk);
3331                                 spin_unlock(&po->bind_lock);
3332                         }
3333                         break;
3334                 }
3335         }
3336         rcu_read_unlock();
3337         return NOTIFY_DONE;
3338 }
3339
3340
3341 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3342                         unsigned long arg)
3343 {
3344         struct sock *sk = sock->sk;
3345
3346         switch (cmd) {
3347         case SIOCOUTQ:
3348         {
3349                 int amount = sk_wmem_alloc_get(sk);
3350
3351                 return put_user(amount, (int __user *)arg);
3352         }
3353         case SIOCINQ:
3354         {
3355                 struct sk_buff *skb;
3356                 int amount = 0;
3357
3358                 spin_lock_bh(&sk->sk_receive_queue.lock);
3359                 skb = skb_peek(&sk->sk_receive_queue);
3360                 if (skb)
3361                         amount = skb->len;
3362                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3363                 return put_user(amount, (int __user *)arg);
3364         }
3365         case SIOCGSTAMP:
3366                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3367         case SIOCGSTAMPNS:
3368                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3369
3370 #ifdef CONFIG_INET
3371         case SIOCADDRT:
3372         case SIOCDELRT:
3373         case SIOCDARP:
3374         case SIOCGARP:
3375         case SIOCSARP:
3376         case SIOCGIFADDR:
3377         case SIOCSIFADDR:
3378         case SIOCGIFBRDADDR:
3379         case SIOCSIFBRDADDR:
3380         case SIOCGIFNETMASK:
3381         case SIOCSIFNETMASK:
3382         case SIOCGIFDSTADDR:
3383         case SIOCSIFDSTADDR:
3384         case SIOCSIFFLAGS:
3385                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3386 #endif
3387
3388         default:
3389                 return -ENOIOCTLCMD;
3390         }
3391         return 0;
3392 }
3393
3394 static unsigned int packet_poll(struct file *file, struct socket *sock,
3395                                 poll_table *wait)
3396 {
3397         struct sock *sk = sock->sk;
3398         struct packet_sock *po = pkt_sk(sk);
3399         unsigned int mask = datagram_poll(file, sock, wait);
3400
3401         spin_lock_bh(&sk->sk_receive_queue.lock);
3402         if (po->rx_ring.pg_vec) {
3403                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3404                         TP_STATUS_KERNEL))
3405                         mask |= POLLIN | POLLRDNORM;
3406         }
3407         spin_unlock_bh(&sk->sk_receive_queue.lock);
3408         spin_lock_bh(&sk->sk_write_queue.lock);
3409         if (po->tx_ring.pg_vec) {
3410                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3411                         mask |= POLLOUT | POLLWRNORM;
3412         }
3413         spin_unlock_bh(&sk->sk_write_queue.lock);
3414         return mask;
3415 }
3416
3417
3418 /* Dirty? Well, I still did not learn better way to account
3419  * for user mmaps.
3420  */
3421
3422 static void packet_mm_open(struct vm_area_struct *vma)
3423 {
3424         struct file *file = vma->vm_file;
3425         struct socket *sock = file->private_data;
3426         struct sock *sk = sock->sk;
3427
3428         if (sk)
3429                 atomic_inc(&pkt_sk(sk)->mapped);
3430 }
3431
3432 static void packet_mm_close(struct vm_area_struct *vma)
3433 {
3434         struct file *file = vma->vm_file;
3435         struct socket *sock = file->private_data;
3436         struct sock *sk = sock->sk;
3437
3438         if (sk)
3439                 atomic_dec(&pkt_sk(sk)->mapped);
3440 }
3441
3442 static const struct vm_operations_struct packet_mmap_ops = {
3443         .open   =       packet_mm_open,
3444         .close  =       packet_mm_close,
3445 };
3446
3447 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3448                         unsigned int len)
3449 {
3450         int i;
3451
3452         for (i = 0; i < len; i++) {
3453                 if (likely(pg_vec[i].buffer)) {
3454                         if (is_vmalloc_addr(pg_vec[i].buffer))
3455                                 vfree(pg_vec[i].buffer);
3456                         else
3457                                 free_pages((unsigned long)pg_vec[i].buffer,
3458                                            order);
3459                         pg_vec[i].buffer = NULL;
3460                 }
3461         }
3462         kfree(pg_vec);
3463 }
3464
3465 static char *alloc_one_pg_vec_page(unsigned long order)
3466 {
3467         char *buffer = NULL;
3468         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3469                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3470
3471         buffer = (char *) __get_free_pages(gfp_flags, order);
3472
3473         if (buffer)
3474                 return buffer;
3475
3476         /*
3477          * __get_free_pages failed, fall back to vmalloc
3478          */
3479         buffer = vzalloc((1 << order) * PAGE_SIZE);
3480
3481         if (buffer)
3482                 return buffer;
3483
3484         /*
3485          * vmalloc failed, lets dig into swap here
3486          */
3487         gfp_flags &= ~__GFP_NORETRY;
3488         buffer = (char *)__get_free_pages(gfp_flags, order);
3489         if (buffer)
3490                 return buffer;
3491
3492         /*
3493          * complete and utter failure
3494          */
3495         return NULL;
3496 }
3497
3498 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3499 {
3500         unsigned int block_nr = req->tp_block_nr;
3501         struct pgv *pg_vec;
3502         int i;
3503
3504         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3505         if (unlikely(!pg_vec))
3506                 goto out;
3507
3508         for (i = 0; i < block_nr; i++) {
3509                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3510                 if (unlikely(!pg_vec[i].buffer))
3511                         goto out_free_pgvec;
3512         }
3513
3514 out:
3515         return pg_vec;
3516
3517 out_free_pgvec:
3518         free_pg_vec(pg_vec, order, block_nr);
3519         pg_vec = NULL;
3520         goto out;
3521 }
3522
3523 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3524                 int closing, int tx_ring)
3525 {
3526         struct pgv *pg_vec = NULL;
3527         struct packet_sock *po = pkt_sk(sk);
3528         int was_running, order = 0;
3529         struct packet_ring_buffer *rb;
3530         struct sk_buff_head *rb_queue;
3531         __be16 num;
3532         int err = -EINVAL;
3533         /* Added to avoid minimal code churn */
3534         struct tpacket_req *req = &req_u->req;
3535
3536         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3537         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3538                 WARN(1, "Tx-ring is not supported.\n");
3539                 goto out;
3540         }
3541
3542         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3543         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3544
3545         err = -EBUSY;
3546         if (!closing) {
3547                 if (atomic_read(&po->mapped))
3548                         goto out;
3549                 if (atomic_read(&rb->pending))
3550                         goto out;
3551         }
3552
3553         if (req->tp_block_nr) {
3554                 /* Sanity tests and some calculations */
3555                 err = -EBUSY;
3556                 if (unlikely(rb->pg_vec))
3557                         goto out;
3558
3559                 switch (po->tp_version) {
3560                 case TPACKET_V1:
3561                         po->tp_hdrlen = TPACKET_HDRLEN;
3562                         break;
3563                 case TPACKET_V2:
3564                         po->tp_hdrlen = TPACKET2_HDRLEN;
3565                         break;
3566                 case TPACKET_V3:
3567                         po->tp_hdrlen = TPACKET3_HDRLEN;
3568                         break;
3569                 }
3570
3571                 err = -EINVAL;
3572                 if (unlikely((int)req->tp_block_size <= 0))
3573                         goto out;
3574                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3575                         goto out;
3576                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3577                                         po->tp_reserve))
3578                         goto out;
3579                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3580                         goto out;
3581
3582                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3583                 if (unlikely(rb->frames_per_block <= 0))
3584                         goto out;
3585                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3586                                         req->tp_frame_nr))
3587                         goto out;
3588
3589                 err = -ENOMEM;
3590                 order = get_order(req->tp_block_size);
3591                 pg_vec = alloc_pg_vec(req, order);
3592                 if (unlikely(!pg_vec))
3593                         goto out;
3594                 switch (po->tp_version) {
3595                 case TPACKET_V3:
3596                 /* Transmit path is not supported. We checked
3597                  * it above but just being paranoid
3598                  */
3599                         if (!tx_ring)
3600                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3601                                 break;
3602                 default:
3603                         break;
3604                 }
3605         }
3606         /* Done */
3607         else {
3608                 err = -EINVAL;
3609                 if (unlikely(req->tp_frame_nr))
3610                         goto out;
3611         }
3612
3613         lock_sock(sk);
3614
3615         /* Detach socket from network */
3616         spin_lock(&po->bind_lock);
3617         was_running = po->running;
3618         num = po->num;
3619         if (was_running) {
3620                 po->num = 0;
3621                 __unregister_prot_hook(sk, false);
3622         }
3623         spin_unlock(&po->bind_lock);
3624
3625         synchronize_net();
3626
3627         err = -EBUSY;
3628         mutex_lock(&po->pg_vec_lock);
3629         if (closing || atomic_read(&po->mapped) == 0) {
3630                 err = 0;
3631                 spin_lock_bh(&rb_queue->lock);
3632                 swap(rb->pg_vec, pg_vec);
3633                 rb->frame_max = (req->tp_frame_nr - 1);
3634                 rb->head = 0;
3635                 rb->frame_size = req->tp_frame_size;
3636                 spin_unlock_bh(&rb_queue->lock);
3637
3638                 swap(rb->pg_vec_order, order);
3639                 swap(rb->pg_vec_len, req->tp_block_nr);
3640
3641                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3642                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3643                                                 tpacket_rcv : packet_rcv;
3644                 skb_queue_purge(rb_queue);
3645                 if (atomic_read(&po->mapped))
3646                         pr_err("packet_mmap: vma is busy: %d\n",
3647                                atomic_read(&po->mapped));
3648         }
3649         mutex_unlock(&po->pg_vec_lock);
3650
3651         spin_lock(&po->bind_lock);
3652         if (was_running) {
3653                 po->num = num;
3654                 register_prot_hook(sk);
3655         }
3656         spin_unlock(&po->bind_lock);
3657         if (closing && (po->tp_version > TPACKET_V2)) {
3658                 /* Because we don't support block-based V3 on tx-ring */
3659                 if (!tx_ring)
3660                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3661         }
3662         release_sock(sk);
3663
3664         if (pg_vec)
3665                 free_pg_vec(pg_vec, order, req->tp_block_nr);
3666 out:
3667         return err;
3668 }
3669
3670 static int packet_mmap(struct file *file, struct socket *sock,
3671                 struct vm_area_struct *vma)
3672 {
3673         struct sock *sk = sock->sk;
3674         struct packet_sock *po = pkt_sk(sk);
3675         unsigned long size, expected_size;
3676         struct packet_ring_buffer *rb;
3677         unsigned long start;
3678         int err = -EINVAL;
3679         int i;
3680
3681         if (vma->vm_pgoff)
3682                 return -EINVAL;
3683
3684         mutex_lock(&po->pg_vec_lock);
3685
3686         expected_size = 0;
3687         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3688                 if (rb->pg_vec) {
3689                         expected_size += rb->pg_vec_len
3690                                                 * rb->pg_vec_pages
3691                                                 * PAGE_SIZE;
3692                 }
3693         }
3694
3695         if (expected_size == 0)
3696                 goto out;
3697
3698         size = vma->vm_end - vma->vm_start;
3699         if (size != expected_size)
3700                 goto out;
3701
3702         start = vma->vm_start;
3703         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3704                 if (rb->pg_vec == NULL)
3705                         continue;
3706
3707                 for (i = 0; i < rb->pg_vec_len; i++) {
3708                         struct page *page;
3709                         void *kaddr = rb->pg_vec[i].buffer;
3710                         int pg_num;
3711
3712                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3713                                 page = pgv_to_page(kaddr);
3714                                 err = vm_insert_page(vma, start, page);
3715                                 if (unlikely(err))
3716                                         goto out;
3717                                 start += PAGE_SIZE;
3718                                 kaddr += PAGE_SIZE;
3719                         }
3720                 }
3721         }
3722
3723         atomic_inc(&po->mapped);
3724         vma->vm_ops = &packet_mmap_ops;
3725         err = 0;
3726
3727 out:
3728         mutex_unlock(&po->pg_vec_lock);
3729         return err;
3730 }
3731
3732 static const struct proto_ops packet_ops_spkt = {
3733         .family =       PF_PACKET,
3734         .owner =        THIS_MODULE,
3735         .release =      packet_release,
3736         .bind =         packet_bind_spkt,
3737         .connect =      sock_no_connect,
3738         .socketpair =   sock_no_socketpair,
3739         .accept =       sock_no_accept,
3740         .getname =      packet_getname_spkt,
3741         .poll =         datagram_poll,
3742         .ioctl =        packet_ioctl,
3743         .listen =       sock_no_listen,
3744         .shutdown =     sock_no_shutdown,
3745         .setsockopt =   sock_no_setsockopt,
3746         .getsockopt =   sock_no_getsockopt,
3747         .sendmsg =      packet_sendmsg_spkt,
3748         .recvmsg =      packet_recvmsg,
3749         .mmap =         sock_no_mmap,
3750         .sendpage =     sock_no_sendpage,
3751 };
3752
3753 static const struct proto_ops packet_ops = {
3754         .family =       PF_PACKET,
3755         .owner =        THIS_MODULE,
3756         .release =      packet_release,
3757         .bind =         packet_bind,
3758         .connect =      sock_no_connect,
3759         .socketpair =   sock_no_socketpair,
3760         .accept =       sock_no_accept,
3761         .getname =      packet_getname,
3762         .poll =         packet_poll,
3763         .ioctl =        packet_ioctl,
3764         .listen =       sock_no_listen,
3765         .shutdown =     sock_no_shutdown,
3766         .setsockopt =   packet_setsockopt,
3767         .getsockopt =   packet_getsockopt,
3768         .sendmsg =      packet_sendmsg,
3769         .recvmsg =      packet_recvmsg,
3770         .mmap =         packet_mmap,
3771         .sendpage =     sock_no_sendpage,
3772 };
3773
3774 static const struct net_proto_family packet_family_ops = {
3775         .family =       PF_PACKET,
3776         .create =       packet_create,
3777         .owner  =       THIS_MODULE,
3778 };
3779
3780 static struct notifier_block packet_netdev_notifier = {
3781         .notifier_call =        packet_notifier,
3782 };
3783
3784 #ifdef CONFIG_PROC_FS
3785
3786 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3787         __acquires(RCU)
3788 {
3789         struct net *net = seq_file_net(seq);
3790
3791         rcu_read_lock();
3792         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3793 }
3794
3795 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3796 {
3797         struct net *net = seq_file_net(seq);
3798         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3799 }
3800
3801 static void packet_seq_stop(struct seq_file *seq, void *v)
3802         __releases(RCU)
3803 {
3804         rcu_read_unlock();
3805 }
3806
3807 static int packet_seq_show(struct seq_file *seq, void *v)
3808 {
3809         if (v == SEQ_START_TOKEN)
3810                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
3811         else {
3812                 struct sock *s = sk_entry(v);
3813                 const struct packet_sock *po = pkt_sk(s);
3814
3815                 seq_printf(seq,
3816                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
3817                            s,
3818                            atomic_read(&s->sk_refcnt),
3819                            s->sk_type,
3820                            ntohs(po->num),
3821                            po->ifindex,
3822                            po->running,
3823                            atomic_read(&s->sk_rmem_alloc),
3824                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
3825                            sock_i_ino(s));
3826         }
3827
3828         return 0;
3829 }
3830
3831 static const struct seq_operations packet_seq_ops = {
3832         .start  = packet_seq_start,
3833         .next   = packet_seq_next,
3834         .stop   = packet_seq_stop,
3835         .show   = packet_seq_show,
3836 };
3837
3838 static int packet_seq_open(struct inode *inode, struct file *file)
3839 {
3840         return seq_open_net(inode, file, &packet_seq_ops,
3841                             sizeof(struct seq_net_private));
3842 }
3843
3844 static const struct file_operations packet_seq_fops = {
3845         .owner          = THIS_MODULE,
3846         .open           = packet_seq_open,
3847         .read           = seq_read,
3848         .llseek         = seq_lseek,
3849         .release        = seq_release_net,
3850 };
3851
3852 #endif
3853
3854 static int __net_init packet_net_init(struct net *net)
3855 {
3856         mutex_init(&net->packet.sklist_lock);
3857         INIT_HLIST_HEAD(&net->packet.sklist);
3858
3859         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
3860                 return -ENOMEM;
3861
3862         return 0;
3863 }
3864
3865 static void __net_exit packet_net_exit(struct net *net)
3866 {
3867         remove_proc_entry("packet", net->proc_net);
3868 }
3869
3870 static struct pernet_operations packet_net_ops = {
3871         .init = packet_net_init,
3872         .exit = packet_net_exit,
3873 };
3874
3875
3876 static void __exit packet_exit(void)
3877 {
3878         unregister_netdevice_notifier(&packet_netdev_notifier);
3879         unregister_pernet_subsys(&packet_net_ops);
3880         sock_unregister(PF_PACKET);
3881         proto_unregister(&packet_proto);
3882 }
3883
3884 static int __init packet_init(void)
3885 {
3886         int rc = proto_register(&packet_proto, 0);
3887
3888         if (rc != 0)
3889                 goto out;
3890
3891         sock_register(&packet_family_ops);
3892         register_pernet_subsys(&packet_net_ops);
3893         register_netdevice_notifier(&packet_netdev_notifier);
3894 out:
3895         return rc;
3896 }
3897
3898 module_init(packet_init);
3899 module_exit(packet_exit);
3900 MODULE_LICENSE("GPL");
3901 MODULE_ALIAS_NETPROTO(PF_PACKET);