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 #include <linux/reciprocal_div.h>
  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 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
 241 {
 242         struct net_device *dev;
 243
 244         rcu_read_lock();
 245         dev = rcu_dereference(po->cached_dev);
 246         if (likely(dev))
 247                 dev_hold(dev);
 248         rcu_read_unlock();
 249
 250         return dev;
 251 }
 252
 253 static void packet_cached_dev_assign(struct packet_sock *po,
 254                                      struct net_device *dev)
 255 {
 256         rcu_assign_pointer(po->cached_dev, dev);
 257 }
 258
 259 static void packet_cached_dev_reset(struct packet_sock *po)
 260 {
 261         RCU_INIT_POINTER(po->cached_dev, NULL);
 262 }
 263
 264 /* register_prot_hook must be invoked with the po->bind_lock held,
 265  * or from a context in which asynchronous accesses to the packet
 266  * socket is not possible (packet_create()).
 267  */
 268 static void register_prot_hook(struct sock *sk)
 269 {
 270         struct packet_sock *po = pkt_sk(sk);
 271
 272         if (!po->running) {
 273                 if (po->fanout)
 274                         __fanout_link(sk, po);
 275                 else
 276                         dev_add_pack(&po->prot_hook);
 277
 278                 sock_hold(sk);
 279                 po->running = 1;
 280         }
 281 }
 282
 283 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
 284  * held.   If the sync parameter is true, we will temporarily drop
 285  * the po->bind_lock and do a synchronize_net to make sure no
 286  * asynchronous packet processing paths still refer to the elements
 287  * of po->prot_hook.  If the sync parameter is false, it is the
 288  * callers responsibility to take care of this.
 289  */
 290 static void __unregister_prot_hook(struct sock *sk, bool sync)
 291 {
 292         struct packet_sock *po = pkt_sk(sk);
 293
 294         po->running = 0;
 295
 296         if (po->fanout)
 297                 __fanout_unlink(sk, po);
 298         else
 299                 __dev_remove_pack(&po->prot_hook);
 300
 301         __sock_put(sk);
 302
 303         if (sync) {
 304                 spin_unlock(&po->bind_lock);
 305                 synchronize_net();
 306                 spin_lock(&po->bind_lock);
 307         }
 308 }
 309
 310 static void unregister_prot_hook(struct sock *sk, bool sync)
 311 {
 312         struct packet_sock *po = pkt_sk(sk);
 313
 314         if (po->running)
 315                 __unregister_prot_hook(sk, sync);
 316 }
 317
 318 static inline __pure struct page *pgv_to_page(void *addr)
 319 {
 320         if (is_vmalloc_addr(addr))
 321                 return vmalloc_to_page(addr);
 322         return virt_to_page(addr);
 323 }
 324
 325 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 326 {
 327         union tpacket_uhdr h;
 328
 329         h.raw = frame;
 330         switch (po->tp_version) {
 331         case TPACKET_V1:
 332                 h.h1->tp_status = status;
 333                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 334                 break;
 335         case TPACKET_V2:
 336                 h.h2->tp_status = status;
 337                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 338                 break;
 339         case TPACKET_V3:
 340         default:
 341                 WARN(1, "TPACKET version not supported.\n");
 342                 BUG();
 343         }
 344
 345         smp_wmb();
 346 }
 347
 348 static int __packet_get_status(struct packet_sock *po, void *frame)
 349 {
 350         union tpacket_uhdr h;
 351
 352         smp_rmb();
 353
 354         h.raw = frame;
 355         switch (po->tp_version) {
 356         case TPACKET_V1:
 357                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 358                 return h.h1->tp_status;
 359         case TPACKET_V2:
 360                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 361                 return h.h2->tp_status;
 362         case TPACKET_V3:
 363         default:
 364                 WARN(1, "TPACKET version not supported.\n");
 365                 BUG();
 366                 return 0;
 367         }
 368 }
 369
 370 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
 371                                    unsigned int flags)
 372 {
 373         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 374
 375         if (shhwtstamps) {
 376                 if ((flags & SOF_TIMESTAMPING_SYS_HARDWARE) &&
 377                     ktime_to_timespec_cond(shhwtstamps->syststamp, ts))
 378                         return TP_STATUS_TS_SYS_HARDWARE;
 379                 if ((flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
 380                     ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
 381                         return TP_STATUS_TS_RAW_HARDWARE;
 382         }
 383
 384         if (ktime_to_timespec_cond(skb->tstamp, ts))
 385                 return TP_STATUS_TS_SOFTWARE;
 386
 387         return 0;
 388 }
 389
 390 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
 391                                     struct sk_buff *skb)
 392 {
 393         union tpacket_uhdr h;
 394         struct timespec ts;
 395         __u32 ts_status;
 396
 397         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
 398                 return 0;
 399
 400         h.raw = frame;
 401         switch (po->tp_version) {
 402         case TPACKET_V1:
 403                 h.h1->tp_sec = ts.tv_sec;
 404                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
 405                 break;
 406         case TPACKET_V2:
 407                 h.h2->tp_sec = ts.tv_sec;
 408                 h.h2->tp_nsec = ts.tv_nsec;
 409                 break;
 410         case TPACKET_V3:
 411         default:
 412                 WARN(1, "TPACKET version not supported.\n");
 413                 BUG();
 414         }
 415
 416         /* one flush is safe, as both fields always lie on the same cacheline */
 417         flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
 418         smp_wmb();
 419
 420         return ts_status;
 421 }
 422
 423 static void *packet_lookup_frame(struct packet_sock *po,
 424                 struct packet_ring_buffer *rb,
 425                 unsigned int position,
 426                 int status)
 427 {
 428         unsigned int pg_vec_pos, frame_offset;
 429         union tpacket_uhdr h;
 430
 431         pg_vec_pos = position / rb->frames_per_block;
 432         frame_offset = position % rb->frames_per_block;
 433
 434         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 435                 (frame_offset * rb->frame_size);
 436
 437         if (status != __packet_get_status(po, h.raw))
 438                 return NULL;
 439
 440         return h.raw;
 441 }
 442
 443 static void *packet_current_frame(struct packet_sock *po,
 444                 struct packet_ring_buffer *rb,
 445                 int status)
 446 {
 447         return packet_lookup_frame(po, rb, rb->head, status);
 448 }
 449
 450 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 451 {
 452         del_timer_sync(&pkc->retire_blk_timer);
 453 }
 454
 455 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 456                 int tx_ring,
 457                 struct sk_buff_head *rb_queue)
 458 {
 459         struct tpacket_kbdq_core *pkc;
 460
 461         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 462
 463         spin_lock_bh(&rb_queue->lock);
 464         pkc->delete_blk_timer = 1;
 465         spin_unlock_bh(&rb_queue->lock);
 466
 467         prb_del_retire_blk_timer(pkc);
 468 }
 469
 470 static void prb_init_blk_timer(struct packet_sock *po,
 471                 struct tpacket_kbdq_core *pkc,
 472                 void (*func) (unsigned long))
 473 {
 474         init_timer(&pkc->retire_blk_timer);
 475         pkc->retire_blk_timer.data = (long)po;
 476         pkc->retire_blk_timer.function = func;
 477         pkc->retire_blk_timer.expires = jiffies;
 478 }
 479
 480 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
 481 {
 482         struct tpacket_kbdq_core *pkc;
 483
 484         if (tx_ring)
 485                 BUG();
 486
 487         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 488         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
 489 }
 490
 491 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 492                                 int blk_size_in_bytes)
 493 {
 494         struct net_device *dev;
 495         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 496         struct ethtool_cmd ecmd;
 497         int err;
 498         u32 speed;
 499
 500         rtnl_lock();
 501         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 502         if (unlikely(!dev)) {
 503                 rtnl_unlock();
 504                 return DEFAULT_PRB_RETIRE_TOV;
 505         }
 506         err = __ethtool_get_settings(dev, &ecmd);
 507         speed = ethtool_cmd_speed(&ecmd);
 508         rtnl_unlock();
 509         if (!err) {
 510                 /*
 511                  * If the link speed is so slow you don't really
 512                  * need to worry about perf anyways
 513                  */
 514                 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
 515                         return DEFAULT_PRB_RETIRE_TOV;
 516                 } else {
 517                         msec = 1;
 518                         div = speed / 1000;
 519                 }
 520         }
 521
 522         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 523
 524         if (div)
 525                 mbits /= div;
 526
 527         tmo = mbits * msec;
 528
 529         if (div)
 530                 return tmo+1;
 531         return tmo;
 532 }
 533
 534 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 535                         union tpacket_req_u *req_u)
 536 {
 537         p1->feature_req_word = req_u->req3.tp_feature_req_word;
 538 }
 539
 540 static void init_prb_bdqc(struct packet_sock *po,
 541                         struct packet_ring_buffer *rb,
 542                         struct pgv *pg_vec,
 543                         union tpacket_req_u *req_u, int tx_ring)
 544 {
 545         struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
 546         struct tpacket_block_desc *pbd;
 547
 548         memset(p1, 0x0, sizeof(*p1));
 549
 550         p1->knxt_seq_num = 1;
 551         p1->pkbdq = pg_vec;
 552         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 553         p1->pkblk_start = pg_vec[0].buffer;
 554         p1->kblk_size = req_u->req3.tp_block_size;
 555         p1->knum_blocks = req_u->req3.tp_block_nr;
 556         p1->hdrlen = po->tp_hdrlen;
 557         p1->version = po->tp_version;
 558         p1->last_kactive_blk_num = 0;
 559         po->stats.stats3.tp_freeze_q_cnt = 0;
 560         if (req_u->req3.tp_retire_blk_tov)
 561                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 562         else
 563                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 564                                                 req_u->req3.tp_block_size);
 565         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 566         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 567
 568         prb_init_ft_ops(p1, req_u);
 569         prb_setup_retire_blk_timer(po, tx_ring);
 570         prb_open_block(p1, pbd);
 571 }
 572
 573 /*  Do NOT update the last_blk_num first.
 574  *  Assumes sk_buff_head lock is held.
 575  */
 576 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 577 {
 578         mod_timer(&pkc->retire_blk_timer,
 579                         jiffies + pkc->tov_in_jiffies);
 580         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 581 }
 582
 583 /*
 584  * Timer logic:
 585  * 1) We refresh the timer only when we open a block.
 586  *    By doing this we don't waste cycles refreshing the timer
 587  *        on packet-by-packet basis.
 588  *
 589  * With a 1MB block-size, on a 1Gbps line, it will take
 590  * i) ~8 ms to fill a block + ii) memcpy etc.
 591  * In this cut we are not accounting for the memcpy time.
 592  *
 593  * So, if the user sets the 'tmo' to 10ms then the timer
 594  * will never fire while the block is still getting filled
 595  * (which is what we want). However, the user could choose
 596  * to close a block early and that's fine.
 597  *
 598  * But when the timer does fire, we check whether or not to refresh it.
 599  * Since the tmo granularity is in msecs, it is not too expensive
 600  * to refresh the timer, lets say every '8' msecs.
 601  * Either the user can set the 'tmo' or we can derive it based on
 602  * a) line-speed and b) block-size.
 603  * prb_calc_retire_blk_tmo() calculates the tmo.
 604  *
 605  */
 606 static void prb_retire_rx_blk_timer_expired(unsigned long data)
 607 {
 608         struct packet_sock *po = (struct packet_sock *)data;
 609         struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
 610         unsigned int frozen;
 611         struct tpacket_block_desc *pbd;
 612
 613         spin_lock(&po->sk.sk_receive_queue.lock);
 614
 615         frozen = prb_queue_frozen(pkc);
 616         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 617
 618         if (unlikely(pkc->delete_blk_timer))
 619                 goto out;
 620
 621         /* We only need to plug the race when the block is partially filled.
 622          * tpacket_rcv:
 623          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 624          *              copy_bits() is in progress ...
 625          *              timer fires on other cpu:
 626          *              we can't retire the current block because copy_bits
 627          *              is in progress.
 628          *
 629          */
 630         if (BLOCK_NUM_PKTS(pbd)) {
 631                 while (atomic_read(&pkc->blk_fill_in_prog)) {
 632                         /* Waiting for skb_copy_bits to finish... */
 633                         cpu_relax();
 634                 }
 635         }
 636
 637         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 638                 if (!frozen) {
 639                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 640                         if (!prb_dispatch_next_block(pkc, po))
 641                                 goto refresh_timer;
 642                         else
 643                                 goto out;
 644                 } else {
 645                         /* Case 1. Queue was frozen because user-space was
 646                          *         lagging behind.
 647                          */
 648                         if (prb_curr_blk_in_use(pkc, pbd)) {
 649                                 /*
 650                                  * Ok, user-space is still behind.
 651                                  * So just refresh the timer.
 652                                  */
 653                                 goto refresh_timer;
 654                         } else {
 655                                /* Case 2. queue was frozen,user-space caught up,
 656                                 * now the link went idle && the timer fired.
 657                                 * We don't have a block to close.So we open this
 658                                 * block and restart the timer.
 659                                 * opening a block thaws the queue,restarts timer
 660                                 * Thawing/timer-refresh is a side effect.
 661                                 */
 662                                 prb_open_block(pkc, pbd);
 663                                 goto out;
 664                         }
 665                 }
 666         }
 667
 668 refresh_timer:
 669         _prb_refresh_rx_retire_blk_timer(pkc);
 670
 671 out:
 672         spin_unlock(&po->sk.sk_receive_queue.lock);
 673 }
 674
 675 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 676                 struct tpacket_block_desc *pbd1, __u32 status)
 677 {
 678         /* Flush everything minus the block header */
 679
 680 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 681         u8 *start, *end;
 682
 683         start = (u8 *)pbd1;
 684
 685         /* Skip the block header(we know header WILL fit in 4K) */
 686         start += PAGE_SIZE;
 687
 688         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 689         for (; start < end; start += PAGE_SIZE)
 690                 flush_dcache_page(pgv_to_page(start));
 691
 692         smp_wmb();
 693 #endif
 694
 695         /* Now update the block status. */
 696
 697         BLOCK_STATUS(pbd1) = status;
 698
 699         /* Flush the block header */
 700
 701 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 702         start = (u8 *)pbd1;
 703         flush_dcache_page(pgv_to_page(start));
 704
 705         smp_wmb();
 706 #endif
 707 }
 708
 709 /*
 710  * Side effect:
 711  *
 712  * 1) flush the block
 713  * 2) Increment active_blk_num
 714  *
 715  * Note:We DONT refresh the timer on purpose.
 716  *      Because almost always the next block will be opened.
 717  */
 718 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 719                 struct tpacket_block_desc *pbd1,
 720                 struct packet_sock *po, unsigned int stat)
 721 {
 722         __u32 status = TP_STATUS_USER | stat;
 723
 724         struct tpacket3_hdr *last_pkt;
 725         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 726
 727         if (po->stats.stats3.tp_drops)
 728                 status |= TP_STATUS_LOSING;
 729
 730         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 731         last_pkt->tp_next_offset = 0;
 732
 733         /* Get the ts of the last pkt */
 734         if (BLOCK_NUM_PKTS(pbd1)) {
 735                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 736                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 737         } else {
 738                 /* Ok, we tmo'd - so get the current time */
 739                 struct timespec ts;
 740                 getnstimeofday(&ts);
 741                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 742                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 743         }
 744
 745         smp_wmb();
 746
 747         /* Flush the block */
 748         prb_flush_block(pkc1, pbd1, status);
 749
 750         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 751 }
 752
 753 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 754 {
 755         pkc->reset_pending_on_curr_blk = 0;
 756 }
 757
 758 /*
 759  * Side effect of opening a block:
 760  *
 761  * 1) prb_queue is thawed.
 762  * 2) retire_blk_timer is refreshed.
 763  *
 764  */
 765 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 766         struct tpacket_block_desc *pbd1)
 767 {
 768         struct timespec ts;
 769         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 770
 771         smp_rmb();
 772
 773         /* We could have just memset this but we will lose the
 774          * flexibility of making the priv area sticky
 775          */
 776
 777         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 778         BLOCK_NUM_PKTS(pbd1) = 0;
 779         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 780
 781         getnstimeofday(&ts);
 782
 783         h1->ts_first_pkt.ts_sec = ts.tv_sec;
 784         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 785
 786         pkc1->pkblk_start = (char *)pbd1;
 787         pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 788
 789         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 790         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 791
 792         pbd1->version = pkc1->version;
 793         pkc1->prev = pkc1->nxt_offset;
 794         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 795
 796         prb_thaw_queue(pkc1);
 797         _prb_refresh_rx_retire_blk_timer(pkc1);
 798
 799         smp_wmb();
 800 }
 801
 802 /*
 803  * Queue freeze logic:
 804  * 1) Assume tp_block_nr = 8 blocks.
 805  * 2) At time 't0', user opens Rx ring.
 806  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 807  * 4) user-space is either sleeping or processing block '0'.
 808  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 809  *    it will close block-7,loop around and try to fill block '0'.
 810  *    call-flow:
 811  *    __packet_lookup_frame_in_block
 812  *      prb_retire_current_block()
 813  *      prb_dispatch_next_block()
 814  *        |->(BLOCK_STATUS == USER) evaluates to true
 815  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 816  * 6) Now there are two cases:
 817  *    6.1) Link goes idle right after the queue is frozen.
 818  *         But remember, the last open_block() refreshed the timer.
 819  *         When this timer expires,it will refresh itself so that we can
 820  *         re-open block-0 in near future.
 821  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 822  *         case and __packet_lookup_frame_in_block will check if block-0
 823  *         is free and can now be re-used.
 824  */
 825 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 826                                   struct packet_sock *po)
 827 {
 828         pkc->reset_pending_on_curr_blk = 1;
 829         po->stats.stats3.tp_freeze_q_cnt++;
 830 }
 831
 832 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 833
 834 /*
 835  * If the next block is free then we will dispatch it
 836  * and return a good offset.
 837  * Else, we will freeze the queue.
 838  * So, caller must check the return value.
 839  */
 840 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 841                 struct packet_sock *po)
 842 {
 843         struct tpacket_block_desc *pbd;
 844
 845         smp_rmb();
 846
 847         /* 1. Get current block num */
 848         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 849
 850         /* 2. If this block is currently in_use then freeze the queue */
 851         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 852                 prb_freeze_queue(pkc, po);
 853                 return NULL;
 854         }
 855
 856         /*
 857          * 3.
 858          * open this block and return the offset where the first packet
 859          * needs to get stored.
 860          */
 861         prb_open_block(pkc, pbd);
 862         return (void *)pkc->nxt_offset;
 863 }
 864
 865 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 866                 struct packet_sock *po, unsigned int status)
 867 {
 868         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 869
 870         /* retire/close the current block */
 871         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 872                 /*
 873                  * Plug the case where copy_bits() is in progress on
 874                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 875                  * have space to copy the pkt in the current block and
 876                  * called prb_retire_current_block()
 877                  *
 878                  * We don't need to worry about the TMO case because
 879                  * the timer-handler already handled this case.
 880                  */
 881                 if (!(status & TP_STATUS_BLK_TMO)) {
 882                         while (atomic_read(&pkc->blk_fill_in_prog)) {
 883                                 /* Waiting for skb_copy_bits to finish... */
 884                                 cpu_relax();
 885                         }
 886                 }
 887                 prb_close_block(pkc, pbd, po, status);
 888                 return;
 889         }
 890 }
 891
 892 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
 893                                       struct tpacket_block_desc *pbd)
 894 {
 895         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 896 }
 897
 898 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 899 {
 900         return pkc->reset_pending_on_curr_blk;
 901 }
 902
 903 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 904 {
 905         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 906         atomic_dec(&pkc->blk_fill_in_prog);
 907 }
 908
 909 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 910                         struct tpacket3_hdr *ppd)
 911 {
 912         ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
 913 }
 914
 915 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 916                         struct tpacket3_hdr *ppd)
 917 {
 918         ppd->hv1.tp_rxhash = 0;
 919 }
 920
 921 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
 922                         struct tpacket3_hdr *ppd)
 923 {
 924         if (vlan_tx_tag_present(pkc->skb)) {
 925                 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
 926                 ppd->tp_status = TP_STATUS_VLAN_VALID;
 927         } else {
 928                 ppd->hv1.tp_vlan_tci = 0;
 929                 ppd->tp_status = TP_STATUS_AVAILABLE;
 930         }
 931 }
 932
 933 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
 934                         struct tpacket3_hdr *ppd)
 935 {
 936         prb_fill_vlan_info(pkc, ppd);
 937
 938         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
 939                 prb_fill_rxhash(pkc, ppd);
 940         else
 941                 prb_clear_rxhash(pkc, ppd);
 942 }
 943
 944 static void prb_fill_curr_block(char *curr,
 945                                 struct tpacket_kbdq_core *pkc,
 946                                 struct tpacket_block_desc *pbd,
 947                                 unsigned int len)
 948 {
 949         struct tpacket3_hdr *ppd;
 950
 951         ppd  = (struct tpacket3_hdr *)curr;
 952         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
 953         pkc->prev = curr;
 954         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
 955         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
 956         BLOCK_NUM_PKTS(pbd) += 1;
 957         atomic_inc(&pkc->blk_fill_in_prog);
 958         prb_run_all_ft_ops(pkc, ppd);
 959 }
 960
 961 /* Assumes caller has the sk->rx_queue.lock */
 962 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
 963                                             struct sk_buff *skb,
 964                                                 int status,
 965                                             unsigned int len
 966                                             )
 967 {
 968         struct tpacket_kbdq_core *pkc;
 969         struct tpacket_block_desc *pbd;
 970         char *curr, *end;
 971
 972         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 973         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 974
 975         /* Queue is frozen when user space is lagging behind */
 976         if (prb_queue_frozen(pkc)) {
 977                 /*
 978                  * Check if that last block which caused the queue to freeze,
 979                  * is still in_use by user-space.
 980                  */
 981                 if (prb_curr_blk_in_use(pkc, pbd)) {
 982                         /* Can't record this packet */
 983                         return NULL;
 984                 } else {
 985                         /*
 986                          * Ok, the block was released by user-space.
 987                          * Now let's open that block.
 988                          * opening a block also thaws the queue.
 989                          * Thawing is a side effect.
 990                          */
 991                         prb_open_block(pkc, pbd);
 992                 }
 993         }
 994
 995         smp_mb();
 996         curr = pkc->nxt_offset;
 997         pkc->skb = skb;
 998         end = (char *)pbd + pkc->kblk_size;
 999
1000         /* first try the current block */
1001         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1002                 prb_fill_curr_block(curr, pkc, pbd, len);
1003                 return (void *)curr;
1004         }
1005
1006         /* Ok, close the current block */
1007         prb_retire_current_block(pkc, po, 0);
1008
1009         /* Now, try to dispatch the next block */
1010         curr = (char *)prb_dispatch_next_block(pkc, po);
1011         if (curr) {
1012                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1013                 prb_fill_curr_block(curr, pkc, pbd, len);
1014                 return (void *)curr;
1015         }
1016
1017         /*
1018          * No free blocks are available.user_space hasn't caught up yet.
1019          * Queue was just frozen and now this packet will get dropped.
1020          */
1021         return NULL;
1022 }
1023
1024 static void *packet_current_rx_frame(struct packet_sock *po,
1025                                             struct sk_buff *skb,
1026                                             int status, unsigned int len)
1027 {
1028         char *curr = NULL;
1029         switch (po->tp_version) {
1030         case TPACKET_V1:
1031         case TPACKET_V2:
1032                 curr = packet_lookup_frame(po, &po->rx_ring,
1033                                         po->rx_ring.head, status);
1034                 return curr;
1035         case TPACKET_V3:
1036                 return __packet_lookup_frame_in_block(po, skb, status, len);
1037         default:
1038                 WARN(1, "TPACKET version not supported\n");
1039                 BUG();
1040                 return NULL;
1041         }
1042 }
1043
1044 static void *prb_lookup_block(struct packet_sock *po,
1045                                      struct packet_ring_buffer *rb,
1046                                      unsigned int idx,
1047                                      int status)
1048 {
1049         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1050         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1051
1052         if (status != BLOCK_STATUS(pbd))
1053                 return NULL;
1054         return pbd;
1055 }
1056
1057 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1058 {
1059         unsigned int prev;
1060         if (rb->prb_bdqc.kactive_blk_num)
1061                 prev = rb->prb_bdqc.kactive_blk_num-1;
1062         else
1063                 prev = rb->prb_bdqc.knum_blocks-1;
1064         return prev;
1065 }
1066
1067 /* Assumes caller has held the rx_queue.lock */
1068 static void *__prb_previous_block(struct packet_sock *po,
1069                                          struct packet_ring_buffer *rb,
1070                                          int status)
1071 {
1072         unsigned int previous = prb_previous_blk_num(rb);
1073         return prb_lookup_block(po, rb, previous, status);
1074 }
1075
1076 static void *packet_previous_rx_frame(struct packet_sock *po,
1077                                              struct packet_ring_buffer *rb,
1078                                              int status)
1079 {
1080         if (po->tp_version <= TPACKET_V2)
1081                 return packet_previous_frame(po, rb, status);
1082
1083         return __prb_previous_block(po, rb, status);
1084 }
1085
1086 static void packet_increment_rx_head(struct packet_sock *po,
1087                                             struct packet_ring_buffer *rb)
1088 {
1089         switch (po->tp_version) {
1090         case TPACKET_V1:
1091         case TPACKET_V2:
1092                 return packet_increment_head(rb);
1093         case TPACKET_V3:
1094         default:
1095                 WARN(1, "TPACKET version not supported.\n");
1096                 BUG();
1097                 return;
1098         }
1099 }
1100
1101 static void *packet_previous_frame(struct packet_sock *po,
1102                 struct packet_ring_buffer *rb,
1103                 int status)
1104 {
1105         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1106         return packet_lookup_frame(po, rb, previous, status);
1107 }
1108
1109 static void packet_increment_head(struct packet_ring_buffer *buff)
1110 {
1111         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1112 }
1113
1114 static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1115 {
1116         struct sock *sk = &po->sk;
1117         bool has_room;
1118
1119         if (po->prot_hook.func != tpacket_rcv)
1120                 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1121                         <= sk->sk_rcvbuf;
1122
1123         spin_lock(&sk->sk_receive_queue.lock);
1124         if (po->tp_version == TPACKET_V3)
1125                 has_room = prb_lookup_block(po, &po->rx_ring,
1126                                             po->rx_ring.prb_bdqc.kactive_blk_num,
1127                                             TP_STATUS_KERNEL);
1128         else
1129                 has_room = packet_lookup_frame(po, &po->rx_ring,
1130                                                po->rx_ring.head,
1131                                                TP_STATUS_KERNEL);
1132         spin_unlock(&sk->sk_receive_queue.lock);
1133
1134         return has_room;
1135 }
1136
1137 static void packet_sock_destruct(struct sock *sk)
1138 {
1139         skb_queue_purge(&sk->sk_error_queue);
1140
1141         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1142         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1143
1144         if (!sock_flag(sk, SOCK_DEAD)) {
1145                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1146                 return;
1147         }
1148
1149         sk_refcnt_debug_dec(sk);
1150 }
1151
1152 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1153 {
1154         int x = atomic_read(&f->rr_cur) + 1;
1155
1156         if (x >= num)
1157                 x = 0;
1158
1159         return x;
1160 }
1161
1162 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1163                                       struct sk_buff *skb,
1164                                       unsigned int num)
1165 {
1166         return reciprocal_divide(skb->rxhash, num);
1167 }
1168
1169 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1170                                     struct sk_buff *skb,
1171                                     unsigned int num)
1172 {
1173         int cur, old;
1174
1175         cur = atomic_read(&f->rr_cur);
1176         while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1177                                      fanout_rr_next(f, num))) != cur)
1178                 cur = old;
1179         return cur;
1180 }
1181
1182 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1183                                      struct sk_buff *skb,
1184                                      unsigned int num)
1185 {
1186         return smp_processor_id() % num;
1187 }
1188
1189 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1190                                      struct sk_buff *skb,
1191                                      unsigned int num)
1192 {
1193         return reciprocal_divide(prandom_u32(), num);
1194 }
1195
1196 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1197                                           struct sk_buff *skb,
1198                                           unsigned int idx, unsigned int skip,
1199                                           unsigned int num)
1200 {
1201         unsigned int i, j;
1202
1203         i = j = min_t(int, f->next[idx], num - 1);
1204         do {
1205                 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1206                         if (i != j)
1207                                 f->next[idx] = i;
1208                         return i;
1209                 }
1210                 if (++i == num)
1211                         i = 0;
1212         } while (i != j);
1213
1214         return idx;
1215 }
1216
1217 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1218 {
1219         return f->flags & (flag >> 8);
1220 }
1221
1222 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1223                              struct packet_type *pt, struct net_device *orig_dev)
1224 {
1225         struct packet_fanout *f = pt->af_packet_priv;
1226         unsigned int num = f->num_members;
1227         struct packet_sock *po;
1228         unsigned int idx;
1229
1230         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1231             !num) {
1232                 kfree_skb(skb);
1233                 return 0;
1234         }
1235
1236         switch (f->type) {
1237         case PACKET_FANOUT_HASH:
1238         default:
1239                 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1240                         skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1241                         if (!skb)
1242                                 return 0;
1243                 }
1244                 skb_get_rxhash(skb);
1245                 idx = fanout_demux_hash(f, skb, num);
1246                 break;
1247         case PACKET_FANOUT_LB:
1248                 idx = fanout_demux_lb(f, skb, num);
1249                 break;
1250         case PACKET_FANOUT_CPU:
1251                 idx = fanout_demux_cpu(f, skb, num);
1252                 break;
1253         case PACKET_FANOUT_RND:
1254                 idx = fanout_demux_rnd(f, skb, num);
1255                 break;
1256         case PACKET_FANOUT_ROLLOVER:
1257                 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1258                 break;
1259         }
1260
1261         po = pkt_sk(f->arr[idx]);
1262         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1263             unlikely(!packet_rcv_has_room(po, skb))) {
1264                 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1265                 po = pkt_sk(f->arr[idx]);
1266         }
1267
1268         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1269 }
1270
1271 DEFINE_MUTEX(fanout_mutex);
1272 EXPORT_SYMBOL_GPL(fanout_mutex);
1273 static LIST_HEAD(fanout_list);
1274
1275 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1276 {
1277         struct packet_fanout *f = po->fanout;
1278
1279         spin_lock(&f->lock);
1280         f->arr[f->num_members] = sk;
1281         smp_wmb();
1282         f->num_members++;
1283         spin_unlock(&f->lock);
1284 }
1285
1286 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1287 {
1288         struct packet_fanout *f = po->fanout;
1289         int i;
1290
1291         spin_lock(&f->lock);
1292         for (i = 0; i < f->num_members; i++) {
1293                 if (f->arr[i] == sk)
1294                         break;
1295         }
1296         BUG_ON(i >= f->num_members);
1297         f->arr[i] = f->arr[f->num_members - 1];
1298         f->num_members--;
1299         spin_unlock(&f->lock);
1300 }
1301
1302 static bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1303 {
1304         if (ptype->af_packet_priv == (void*)((struct packet_sock *)sk)->fanout)
1305                 return true;
1306
1307         return false;
1308 }
1309
1310 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1311 {
1312         struct packet_sock *po = pkt_sk(sk);
1313         struct packet_fanout *f, *match;
1314         u8 type = type_flags & 0xff;
1315         u8 flags = type_flags >> 8;
1316         int err;
1317
1318         switch (type) {
1319         case PACKET_FANOUT_ROLLOVER:
1320                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1321                         return -EINVAL;
1322         case PACKET_FANOUT_HASH:
1323         case PACKET_FANOUT_LB:
1324         case PACKET_FANOUT_CPU:
1325         case PACKET_FANOUT_RND:
1326                 break;
1327         default:
1328                 return -EINVAL;
1329         }
1330
1331         if (!po->running)
1332                 return -EINVAL;
1333
1334         if (po->fanout)
1335                 return -EALREADY;
1336
1337         mutex_lock(&fanout_mutex);
1338         match = NULL;
1339         list_for_each_entry(f, &fanout_list, list) {
1340                 if (f->id == id &&
1341                     read_pnet(&f->net) == sock_net(sk)) {
1342                         match = f;
1343                         break;
1344                 }
1345         }
1346         err = -EINVAL;
1347         if (match && match->flags != flags)
1348                 goto out;
1349         if (!match) {
1350                 err = -ENOMEM;
1351                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1352                 if (!match)
1353                         goto out;
1354                 write_pnet(&match->net, sock_net(sk));
1355                 match->id = id;
1356                 match->type = type;
1357                 match->flags = flags;
1358                 atomic_set(&match->rr_cur, 0);
1359                 INIT_LIST_HEAD(&match->list);
1360                 spin_lock_init(&match->lock);
1361                 atomic_set(&match->sk_ref, 0);
1362                 match->prot_hook.type = po->prot_hook.type;
1363                 match->prot_hook.dev = po->prot_hook.dev;
1364                 match->prot_hook.func = packet_rcv_fanout;
1365                 match->prot_hook.af_packet_priv = match;
1366                 match->prot_hook.id_match = match_fanout_group;
1367                 dev_add_pack(&match->prot_hook);
1368                 list_add(&match->list, &fanout_list);
1369         }
1370         err = -EINVAL;
1371         if (match->type == type &&
1372             match->prot_hook.type == po->prot_hook.type &&
1373             match->prot_hook.dev == po->prot_hook.dev) {
1374                 err = -ENOSPC;
1375                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1376                         __dev_remove_pack(&po->prot_hook);
1377                         po->fanout = match;
1378                         atomic_inc(&match->sk_ref);
1379                         __fanout_link(sk, po);
1380                         err = 0;
1381                 }
1382         }
1383 out:
1384         mutex_unlock(&fanout_mutex);
1385         return err;
1386 }
1387
1388 static void fanout_release(struct sock *sk)
1389 {
1390         struct packet_sock *po = pkt_sk(sk);
1391         struct packet_fanout *f;
1392
1393         f = po->fanout;
1394         if (!f)
1395                 return;
1396
1397         mutex_lock(&fanout_mutex);
1398         po->fanout = NULL;
1399
1400         if (atomic_dec_and_test(&f->sk_ref)) {
1401                 list_del(&f->list);
1402                 dev_remove_pack(&f->prot_hook);
1403                 kfree(f);
1404         }
1405         mutex_unlock(&fanout_mutex);
1406 }
1407
1408 static const struct proto_ops packet_ops;
1409
1410 static const struct proto_ops packet_ops_spkt;
1411
1412 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1413                            struct packet_type *pt, struct net_device *orig_dev)
1414 {
1415         struct sock *sk;
1416         struct sockaddr_pkt *spkt;
1417
1418         /*
1419          *      When we registered the protocol we saved the socket in the data
1420          *      field for just this event.
1421          */
1422
1423         sk = pt->af_packet_priv;
1424
1425         /*
1426          *      Yank back the headers [hope the device set this
1427          *      right or kerboom...]
1428          *
1429          *      Incoming packets have ll header pulled,
1430          *      push it back.
1431          *
1432          *      For outgoing ones skb->data == skb_mac_header(skb)
1433          *      so that this procedure is noop.
1434          */
1435
1436         if (skb->pkt_type == PACKET_LOOPBACK)
1437                 goto out;
1438
1439         if (!net_eq(dev_net(dev), sock_net(sk)))
1440                 goto out;
1441
1442         skb = skb_share_check(skb, GFP_ATOMIC);
1443         if (skb == NULL)
1444                 goto oom;
1445
1446         /* drop any routing info */
1447         skb_dst_drop(skb);
1448
1449         /* drop conntrack reference */
1450         nf_reset(skb);
1451
1452         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1453
1454         skb_push(skb, skb->data - skb_mac_header(skb));
1455
1456         /*
1457          *      The SOCK_PACKET socket receives _all_ frames.
1458          */
1459
1460         spkt->spkt_family = dev->type;
1461         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1462         spkt->spkt_protocol = skb->protocol;
1463
1464         /*
1465          *      Charge the memory to the socket. This is done specifically
1466          *      to prevent sockets using all the memory up.
1467          */
1468
1469         if (sock_queue_rcv_skb(sk, skb) == 0)
1470                 return 0;
1471
1472 out:
1473         kfree_skb(skb);
1474 oom:
1475         return 0;
1476 }
1477
1478
1479 /*
1480  *      Output a raw packet to a device layer. This bypasses all the other
1481  *      protocol layers and you must therefore supply it with a complete frame
1482  */
1483
1484 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1485                                struct msghdr *msg, size_t len)
1486 {
1487         struct sock *sk = sock->sk;
1488         struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1489         struct sk_buff *skb = NULL;
1490         struct net_device *dev;
1491         __be16 proto = 0;
1492         int err;
1493         int extra_len = 0;
1494
1495         /*
1496          *      Get and verify the address.
1497          */
1498
1499         if (saddr) {
1500                 if (msg->msg_namelen < sizeof(struct sockaddr))
1501                         return -EINVAL;
1502                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1503                         proto = saddr->spkt_protocol;
1504         } else
1505                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1506
1507         /*
1508          *      Find the device first to size check it
1509          */
1510
1511         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1512 retry:
1513         rcu_read_lock();
1514         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1515         err = -ENODEV;
1516         if (dev == NULL)
1517                 goto out_unlock;
1518
1519         err = -ENETDOWN;
1520         if (!(dev->flags & IFF_UP))
1521                 goto out_unlock;
1522
1523         /*
1524          * You may not queue a frame bigger than the mtu. This is the lowest level
1525          * raw protocol and you must do your own fragmentation at this level.
1526          */
1527
1528         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1529                 if (!netif_supports_nofcs(dev)) {
1530                         err = -EPROTONOSUPPORT;
1531                         goto out_unlock;
1532                 }
1533                 extra_len = 4; /* We're doing our own CRC */
1534         }
1535
1536         err = -EMSGSIZE;
1537         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1538                 goto out_unlock;
1539
1540         if (!skb) {
1541                 size_t reserved = LL_RESERVED_SPACE(dev);
1542                 int tlen = dev->needed_tailroom;
1543                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1544
1545                 rcu_read_unlock();
1546                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1547                 if (skb == NULL)
1548                         return -ENOBUFS;
1549                 /* FIXME: Save some space for broken drivers that write a hard
1550                  * header at transmission time by themselves. PPP is the notable
1551                  * one here. This should really be fixed at the driver level.
1552                  */
1553                 skb_reserve(skb, reserved);
1554                 skb_reset_network_header(skb);
1555
1556                 /* Try to align data part correctly */
1557                 if (hhlen) {
1558                         skb->data -= hhlen;
1559                         skb->tail -= hhlen;
1560                         if (len < hhlen)
1561                                 skb_reset_network_header(skb);
1562                 }
1563                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1564                 if (err)
1565                         goto out_free;
1566                 goto retry;
1567         }
1568
1569         if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1570                 /* Earlier code assumed this would be a VLAN pkt,
1571                  * double-check this now that we have the actual
1572                  * packet in hand.
1573                  */
1574                 struct ethhdr *ehdr;
1575                 skb_reset_mac_header(skb);
1576                 ehdr = eth_hdr(skb);
1577                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1578                         err = -EMSGSIZE;
1579                         goto out_unlock;
1580                 }
1581         }
1582
1583         skb->protocol = proto;
1584         skb->dev = dev;
1585         skb->priority = sk->sk_priority;
1586         skb->mark = sk->sk_mark;
1587
1588         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1589
1590         if (unlikely(extra_len == 4))
1591                 skb->no_fcs = 1;
1592
1593         skb_probe_transport_header(skb, 0);
1594
1595         dev_queue_xmit(skb);
1596         rcu_read_unlock();
1597         return len;
1598
1599 out_unlock:
1600         rcu_read_unlock();
1601 out_free:
1602         kfree_skb(skb);
1603         return err;
1604 }
1605
1606 static unsigned int run_filter(const struct sk_buff *skb,
1607                                       const struct sock *sk,
1608                                       unsigned int res)
1609 {
1610         struct sk_filter *filter;
1611
1612         rcu_read_lock();
1613         filter = rcu_dereference(sk->sk_filter);
1614         if (filter != NULL)
1615                 res = SK_RUN_FILTER(filter, skb);
1616         rcu_read_unlock();
1617
1618         return res;
1619 }
1620
1621 /*
1622  * This function makes lazy skb cloning in hope that most of packets
1623  * are discarded by BPF.
1624  *
1625  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1626  * and skb->cb are mangled. It works because (and until) packets
1627  * falling here are owned by current CPU. Output packets are cloned
1628  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1629  * sequencially, so that if we return skb to original state on exit,
1630  * we will not harm anyone.
1631  */
1632
1633 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1634                       struct packet_type *pt, struct net_device *orig_dev)
1635 {
1636         struct sock *sk;
1637         struct sockaddr_ll *sll;
1638         struct packet_sock *po;
1639         u8 *skb_head = skb->data;
1640         int skb_len = skb->len;
1641         unsigned int snaplen, res;
1642
1643         if (skb->pkt_type == PACKET_LOOPBACK)
1644                 goto drop;
1645
1646         sk = pt->af_packet_priv;
1647         po = pkt_sk(sk);
1648
1649         if (!net_eq(dev_net(dev), sock_net(sk)))
1650                 goto drop;
1651
1652         skb->dev = dev;
1653
1654         if (dev->header_ops) {
1655                 /* The device has an explicit notion of ll header,
1656                  * exported to higher levels.
1657                  *
1658                  * Otherwise, the device hides details of its frame
1659                  * structure, so that corresponding packet head is
1660                  * never delivered to user.
1661                  */
1662                 if (sk->sk_type != SOCK_DGRAM)
1663                         skb_push(skb, skb->data - skb_mac_header(skb));
1664                 else if (skb->pkt_type == PACKET_OUTGOING) {
1665                         /* Special case: outgoing packets have ll header at head */
1666                         skb_pull(skb, skb_network_offset(skb));
1667                 }
1668         }
1669
1670         snaplen = skb->len;
1671
1672         res = run_filter(skb, sk, snaplen);
1673         if (!res)
1674                 goto drop_n_restore;
1675         if (snaplen > res)
1676                 snaplen = res;
1677
1678         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1679                 goto drop_n_acct;
1680
1681         if (skb_shared(skb)) {
1682                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1683                 if (nskb == NULL)
1684                         goto drop_n_acct;
1685
1686                 if (skb_head != skb->data) {
1687                         skb->data = skb_head;
1688                         skb->len = skb_len;
1689                 }
1690                 consume_skb(skb);
1691                 skb = nskb;
1692         }
1693
1694         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1695                      sizeof(skb->cb));
1696
1697         sll = &PACKET_SKB_CB(skb)->sa.ll;
1698         sll->sll_family = AF_PACKET;
1699         sll->sll_hatype = dev->type;
1700         sll->sll_protocol = skb->protocol;
1701         sll->sll_pkttype = skb->pkt_type;
1702         if (unlikely(po->origdev))
1703                 sll->sll_ifindex = orig_dev->ifindex;
1704         else
1705                 sll->sll_ifindex = dev->ifindex;
1706
1707         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1708
1709         PACKET_SKB_CB(skb)->origlen = skb->len;
1710
1711         if (pskb_trim(skb, snaplen))
1712                 goto drop_n_acct;
1713
1714         skb_set_owner_r(skb, sk);
1715         skb->dev = NULL;
1716         skb_dst_drop(skb);
1717
1718         /* drop conntrack reference */
1719         nf_reset(skb);
1720
1721         spin_lock(&sk->sk_receive_queue.lock);
1722         po->stats.stats1.tp_packets++;
1723         skb->dropcount = atomic_read(&sk->sk_drops);
1724         __skb_queue_tail(&sk->sk_receive_queue, skb);
1725         spin_unlock(&sk->sk_receive_queue.lock);
1726         sk->sk_data_ready(sk, skb->len);
1727         return 0;
1728
1729 drop_n_acct:
1730         spin_lock(&sk->sk_receive_queue.lock);
1731         po->stats.stats1.tp_drops++;
1732         atomic_inc(&sk->sk_drops);
1733         spin_unlock(&sk->sk_receive_queue.lock);
1734
1735 drop_n_restore:
1736         if (skb_head != skb->data && skb_shared(skb)) {
1737                 skb->data = skb_head;
1738                 skb->len = skb_len;
1739         }
1740 drop:
1741         consume_skb(skb);
1742         return 0;
1743 }
1744
1745 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1746                        struct packet_type *pt, struct net_device *orig_dev)
1747 {
1748         struct sock *sk;
1749         struct packet_sock *po;
1750         struct sockaddr_ll *sll;
1751         union tpacket_uhdr h;
1752         u8 *skb_head = skb->data;
1753         int skb_len = skb->len;
1754         unsigned int snaplen, res;
1755         unsigned long status = TP_STATUS_USER;
1756         unsigned short macoff, netoff, hdrlen;
1757         struct sk_buff *copy_skb = NULL;
1758         struct timespec ts;
1759         __u32 ts_status;
1760
1761         if (skb->pkt_type == PACKET_LOOPBACK)
1762                 goto drop;
1763
1764         sk = pt->af_packet_priv;
1765         po = pkt_sk(sk);
1766
1767         if (!net_eq(dev_net(dev), sock_net(sk)))
1768                 goto drop;
1769
1770         if (dev->header_ops) {
1771                 if (sk->sk_type != SOCK_DGRAM)
1772                         skb_push(skb, skb->data - skb_mac_header(skb));
1773                 else if (skb->pkt_type == PACKET_OUTGOING) {
1774                         /* Special case: outgoing packets have ll header at head */
1775                         skb_pull(skb, skb_network_offset(skb));
1776                 }
1777         }
1778
1779         if (skb->ip_summed == CHECKSUM_PARTIAL)
1780                 status |= TP_STATUS_CSUMNOTREADY;
1781
1782         snaplen = skb->len;
1783
1784         res = run_filter(skb, sk, snaplen);
1785         if (!res)
1786                 goto drop_n_restore;
1787         if (snaplen > res)
1788                 snaplen = res;
1789
1790         if (sk->sk_type == SOCK_DGRAM) {
1791                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1792                                   po->tp_reserve;
1793         } else {
1794                 unsigned int maclen = skb_network_offset(skb);
1795                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1796                                        (maclen < 16 ? 16 : maclen)) +
1797                         po->tp_reserve;
1798                 macoff = netoff - maclen;
1799         }
1800         if (po->tp_version <= TPACKET_V2) {
1801                 if (macoff + snaplen > po->rx_ring.frame_size) {
1802                         if (po->copy_thresh &&
1803                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1804                                 if (skb_shared(skb)) {
1805                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1806                                 } else {
1807                                         copy_skb = skb_get(skb);
1808                                         skb_head = skb->data;
1809                                 }
1810                                 if (copy_skb)
1811                                         skb_set_owner_r(copy_skb, sk);
1812                         }
1813                         snaplen = po->rx_ring.frame_size - macoff;
1814                         if ((int)snaplen < 0)
1815                                 snaplen = 0;
1816                 }
1817         }
1818         spin_lock(&sk->sk_receive_queue.lock);
1819         h.raw = packet_current_rx_frame(po, skb,
1820                                         TP_STATUS_KERNEL, (macoff+snaplen));
1821         if (!h.raw)
1822                 goto ring_is_full;
1823         if (po->tp_version <= TPACKET_V2) {
1824                 packet_increment_rx_head(po, &po->rx_ring);
1825         /*
1826          * LOSING will be reported till you read the stats,
1827          * because it's COR - Clear On Read.
1828          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1829          * at packet level.
1830          */
1831                 if (po->stats.stats1.tp_drops)
1832                         status |= TP_STATUS_LOSING;
1833         }
1834         po->stats.stats1.tp_packets++;
1835         if (copy_skb) {
1836                 status |= TP_STATUS_COPY;
1837                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1838         }
1839         spin_unlock(&sk->sk_receive_queue.lock);
1840
1841         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1842
1843         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
1844                 getnstimeofday(&ts);
1845
1846         status |= ts_status;
1847
1848         switch (po->tp_version) {
1849         case TPACKET_V1:
1850                 h.h1->tp_len = skb->len;
1851                 h.h1->tp_snaplen = snaplen;
1852                 h.h1->tp_mac = macoff;
1853                 h.h1->tp_net = netoff;
1854                 h.h1->tp_sec = ts.tv_sec;
1855                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
1856                 hdrlen = sizeof(*h.h1);
1857                 break;
1858         case TPACKET_V2:
1859                 h.h2->tp_len = skb->len;
1860                 h.h2->tp_snaplen = snaplen;
1861                 h.h2->tp_mac = macoff;
1862                 h.h2->tp_net = netoff;
1863                 h.h2->tp_sec = ts.tv_sec;
1864                 h.h2->tp_nsec = ts.tv_nsec;
1865                 if (vlan_tx_tag_present(skb)) {
1866                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1867                         status |= TP_STATUS_VLAN_VALID;
1868                 } else {
1869                         h.h2->tp_vlan_tci = 0;
1870                 }
1871                 h.h2->tp_padding = 0;
1872                 hdrlen = sizeof(*h.h2);
1873                 break;
1874         case TPACKET_V3:
1875                 /* tp_nxt_offset,vlan are already populated above.
1876                  * So DONT clear those fields here
1877                  */
1878                 h.h3->tp_status |= status;
1879                 h.h3->tp_len = skb->len;
1880                 h.h3->tp_snaplen = snaplen;
1881                 h.h3->tp_mac = macoff;
1882                 h.h3->tp_net = netoff;
1883                 h.h3->tp_sec  = ts.tv_sec;
1884                 h.h3->tp_nsec = ts.tv_nsec;
1885                 hdrlen = sizeof(*h.h3);
1886                 break;
1887         default:
1888                 BUG();
1889         }
1890
1891         sll = h.raw + TPACKET_ALIGN(hdrlen);
1892         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1893         sll->sll_family = AF_PACKET;
1894         sll->sll_hatype = dev->type;
1895         sll->sll_protocol = skb->protocol;
1896         sll->sll_pkttype = skb->pkt_type;
1897         if (unlikely(po->origdev))
1898                 sll->sll_ifindex = orig_dev->ifindex;
1899         else
1900                 sll->sll_ifindex = dev->ifindex;
1901
1902         smp_mb();
1903 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1904         {
1905                 u8 *start, *end;
1906
1907                 if (po->tp_version <= TPACKET_V2) {
1908                         end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1909                                 + macoff + snaplen);
1910                         for (start = h.raw; start < end; start += PAGE_SIZE)
1911                                 flush_dcache_page(pgv_to_page(start));
1912                 }
1913                 smp_wmb();
1914         }
1915 #endif
1916         if (po->tp_version <= TPACKET_V2)
1917                 __packet_set_status(po, h.raw, status);
1918         else
1919                 prb_clear_blk_fill_status(&po->rx_ring);
1920
1921         sk->sk_data_ready(sk, 0);
1922
1923 drop_n_restore:
1924         if (skb_head != skb->data && skb_shared(skb)) {
1925                 skb->data = skb_head;
1926                 skb->len = skb_len;
1927         }
1928 drop:
1929         kfree_skb(skb);
1930         return 0;
1931
1932 ring_is_full:
1933         po->stats.stats1.tp_drops++;
1934         spin_unlock(&sk->sk_receive_queue.lock);
1935
1936         sk->sk_data_ready(sk, 0);
1937         kfree_skb(copy_skb);
1938         goto drop_n_restore;
1939 }
1940
1941 static void tpacket_destruct_skb(struct sk_buff *skb)
1942 {
1943         struct packet_sock *po = pkt_sk(skb->sk);
1944         void *ph;
1945
1946         if (likely(po->tx_ring.pg_vec)) {
1947                 __u32 ts;
1948
1949                 ph = skb_shinfo(skb)->destructor_arg;
1950                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1951                 atomic_dec(&po->tx_ring.pending);
1952
1953                 ts = __packet_set_timestamp(po, ph, skb);
1954                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
1955         }
1956
1957         sock_wfree(skb);
1958 }
1959
1960 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1961                 void *frame, struct net_device *dev, int size_max,
1962                 __be16 proto, unsigned char *addr, int hlen)
1963 {
1964         union tpacket_uhdr ph;
1965         int to_write, offset, len, tp_len, nr_frags, len_max;
1966         struct socket *sock = po->sk.sk_socket;
1967         struct page *page;
1968         void *data;
1969         int err;
1970
1971         ph.raw = frame;
1972
1973         skb->protocol = proto;
1974         skb->dev = dev;
1975         skb->priority = po->sk.sk_priority;
1976         skb->mark = po->sk.sk_mark;
1977         sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
1978         skb_shinfo(skb)->destructor_arg = ph.raw;
1979
1980         switch (po->tp_version) {
1981         case TPACKET_V2:
1982                 tp_len = ph.h2->tp_len;
1983                 break;
1984         default:
1985                 tp_len = ph.h1->tp_len;
1986                 break;
1987         }
1988         if (unlikely(tp_len > size_max)) {
1989                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1990                 return -EMSGSIZE;
1991         }
1992
1993         skb_reserve(skb, hlen);
1994         skb_reset_network_header(skb);
1995         skb_probe_transport_header(skb, 0);
1996
1997         if (po->tp_tx_has_off) {
1998                 int off_min, off_max, off;
1999                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2000                 off_max = po->tx_ring.frame_size - tp_len;
2001                 if (sock->type == SOCK_DGRAM) {
2002                         switch (po->tp_version) {
2003                         case TPACKET_V2:
2004                                 off = ph.h2->tp_net;
2005                                 break;
2006                         default:
2007                                 off = ph.h1->tp_net;
2008                                 break;
2009                         }
2010                 } else {
2011                         switch (po->tp_version) {
2012                         case TPACKET_V2:
2013                                 off = ph.h2->tp_mac;
2014                                 break;
2015                         default:
2016                                 off = ph.h1->tp_mac;
2017                                 break;
2018                         }
2019                 }
2020                 if (unlikely((off < off_min) || (off_max < off)))
2021                         return -EINVAL;
2022                 data = ph.raw + off;
2023         } else {
2024                 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2025         }
2026         to_write = tp_len;
2027
2028         if (sock->type == SOCK_DGRAM) {
2029                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2030                                 NULL, tp_len);
2031                 if (unlikely(err < 0))
2032                         return -EINVAL;
2033         } else if (dev->hard_header_len) {
2034                 /* net device doesn't like empty head */
2035                 if (unlikely(tp_len <= dev->hard_header_len)) {
2036                         pr_err("packet size is too short (%d < %d)\n",
2037                                tp_len, dev->hard_header_len);
2038                         return -EINVAL;
2039                 }
2040
2041                 skb_push(skb, dev->hard_header_len);
2042                 err = skb_store_bits(skb, 0, data,
2043                                 dev->hard_header_len);
2044                 if (unlikely(err))
2045                         return err;
2046
2047                 data += dev->hard_header_len;
2048                 to_write -= dev->hard_header_len;
2049         }
2050
2051         offset = offset_in_page(data);
2052         len_max = PAGE_SIZE - offset;
2053         len = ((to_write > len_max) ? len_max : to_write);
2054
2055         skb->data_len = to_write;
2056         skb->len += to_write;
2057         skb->truesize += to_write;
2058         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2059
2060         while (likely(to_write)) {
2061                 nr_frags = skb_shinfo(skb)->nr_frags;
2062
2063                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2064                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2065                                MAX_SKB_FRAGS);
2066                         return -EFAULT;
2067                 }
2068
2069                 page = pgv_to_page(data);
2070                 data += len;
2071                 flush_dcache_page(page);
2072                 get_page(page);
2073                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2074                 to_write -= len;
2075                 offset = 0;
2076                 len_max = PAGE_SIZE;
2077                 len = ((to_write > len_max) ? len_max : to_write);
2078         }
2079
2080         return tp_len;
2081 }
2082
2083 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2084 {
2085         struct sk_buff *skb;
2086         struct net_device *dev;
2087         __be16 proto;
2088         int err, reserve = 0;
2089         void *ph;
2090         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2091         int tp_len, size_max;
2092         unsigned char *addr;
2093         int len_sum = 0;
2094         int status = TP_STATUS_AVAILABLE;
2095         int hlen, tlen;
2096
2097         mutex_lock(&po->pg_vec_lock);
2098
2099         if (likely(saddr == NULL)) {
2100                 dev     = packet_cached_dev_get(po);
2101                 proto   = po->num;
2102                 addr    = NULL;
2103         } else {
2104                 err = -EINVAL;
2105                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2106                         goto out;
2107                 if (msg->msg_namelen < (saddr->sll_halen
2108                                         + offsetof(struct sockaddr_ll,
2109                                                 sll_addr)))
2110                         goto out;
2111                 proto   = saddr->sll_protocol;
2112                 addr    = saddr->sll_addr;
2113                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2114         }
2115
2116         err = -ENXIO;
2117         if (unlikely(dev == NULL))
2118                 goto out;
2119         err = -ENETDOWN;
2120         if (unlikely(!(dev->flags & IFF_UP)))
2121                 goto out_put;
2122
2123         reserve = dev->hard_header_len;
2124
2125         size_max = po->tx_ring.frame_size
2126                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2127
2128         if (size_max > dev->mtu + reserve)
2129                 size_max = dev->mtu + reserve;
2130
2131         do {
2132                 ph = packet_current_frame(po, &po->tx_ring,
2133                                 TP_STATUS_SEND_REQUEST);
2134
2135                 if (unlikely(ph == NULL)) {
2136                         schedule();
2137                         continue;
2138                 }
2139
2140                 status = TP_STATUS_SEND_REQUEST;
2141                 hlen = LL_RESERVED_SPACE(dev);
2142                 tlen = dev->needed_tailroom;
2143                 skb = sock_alloc_send_skb(&po->sk,
2144                                 hlen + tlen + sizeof(struct sockaddr_ll),
2145                                 0, &err);
2146
2147                 if (unlikely(skb == NULL))
2148                         goto out_status;
2149
2150                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2151                                 addr, hlen);
2152
2153                 if (unlikely(tp_len < 0)) {
2154                         if (po->tp_loss) {
2155                                 __packet_set_status(po, ph,
2156                                                 TP_STATUS_AVAILABLE);
2157                                 packet_increment_head(&po->tx_ring);
2158                                 kfree_skb(skb);
2159                                 continue;
2160                         } else {
2161                                 status = TP_STATUS_WRONG_FORMAT;
2162                                 err = tp_len;
2163                                 goto out_status;
2164                         }
2165                 }
2166
2167                 skb->destructor = tpacket_destruct_skb;
2168                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2169                 atomic_inc(&po->tx_ring.pending);
2170
2171                 status = TP_STATUS_SEND_REQUEST;
2172                 err = dev_queue_xmit(skb);
2173                 if (unlikely(err > 0)) {
2174                         err = net_xmit_errno(err);
2175                         if (err && __packet_get_status(po, ph) ==
2176                                    TP_STATUS_AVAILABLE) {
2177                                 /* skb was destructed already */
2178                                 skb = NULL;
2179                                 goto out_status;
2180                         }
2181                         /*
2182                          * skb was dropped but not destructed yet;
2183                          * let's treat it like congestion or err < 0
2184                          */
2185                         err = 0;
2186                 }
2187                 packet_increment_head(&po->tx_ring);
2188                 len_sum += tp_len;
2189         } while (likely((ph != NULL) ||
2190                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2191                          (atomic_read(&po->tx_ring.pending))))
2192                 );
2193
2194         err = len_sum;
2195         goto out_put;
2196
2197 out_status:
2198         __packet_set_status(po, ph, status);
2199         kfree_skb(skb);
2200 out_put:
2201         dev_put(dev);
2202 out:
2203         mutex_unlock(&po->pg_vec_lock);
2204         return err;
2205 }
2206
2207 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2208                                         size_t reserve, size_t len,
2209                                         size_t linear, int noblock,
2210                                         int *err)
2211 {
2212         struct sk_buff *skb;
2213
2214         /* Under a page?  Don't bother with paged skb. */
2215         if (prepad + len < PAGE_SIZE || !linear)
2216                 linear = len;
2217
2218         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2219                                    err, 0);
2220         if (!skb)
2221                 return NULL;
2222
2223         skb_reserve(skb, reserve);
2224         skb_put(skb, linear);
2225         skb->data_len = len - linear;
2226         skb->len += len - linear;
2227
2228         return skb;
2229 }
2230
2231 static int packet_snd(struct socket *sock,
2232                           struct msghdr *msg, size_t len)
2233 {
2234         struct sock *sk = sock->sk;
2235         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2236         struct sk_buff *skb;
2237         struct net_device *dev;
2238         __be16 proto;
2239         unsigned char *addr;
2240         int err, reserve = 0;
2241         struct virtio_net_hdr vnet_hdr = { 0 };
2242         int offset = 0;
2243         int vnet_hdr_len;
2244         struct packet_sock *po = pkt_sk(sk);
2245         unsigned short gso_type = 0;
2246         int hlen, tlen;
2247         int extra_len = 0;
2248
2249         /*
2250          *      Get and verify the address.
2251          */
2252
2253         if (likely(saddr == NULL)) {
2254                 dev     = packet_cached_dev_get(po);
2255                 proto   = po->num;
2256                 addr    = NULL;
2257         } else {
2258                 err = -EINVAL;
2259                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2260                         goto out;
2261                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2262                         goto out;
2263                 proto   = saddr->sll_protocol;
2264                 addr    = saddr->sll_addr;
2265                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2266         }
2267
2268         err = -ENXIO;
2269         if (unlikely(dev == NULL))
2270                 goto out_unlock;
2271         err = -ENETDOWN;
2272         if (unlikely(!(dev->flags & IFF_UP)))
2273                 goto out_unlock;
2274
2275         if (sock->type == SOCK_RAW)
2276                 reserve = dev->hard_header_len;
2277         if (po->has_vnet_hdr) {
2278                 vnet_hdr_len = sizeof(vnet_hdr);
2279
2280                 err = -EINVAL;
2281                 if (len < vnet_hdr_len)
2282                         goto out_unlock;
2283
2284                 len -= vnet_hdr_len;
2285
2286                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2287                                        vnet_hdr_len);
2288                 if (err < 0)
2289                         goto out_unlock;
2290
2291                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2292                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2293                       vnet_hdr.hdr_len))
2294                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
2295                                                  vnet_hdr.csum_offset + 2;
2296
2297                 err = -EINVAL;
2298                 if (vnet_hdr.hdr_len > len)
2299                         goto out_unlock;
2300
2301                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2302                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2303                         case VIRTIO_NET_HDR_GSO_TCPV4:
2304                                 gso_type = SKB_GSO_TCPV4;
2305                                 break;
2306                         case VIRTIO_NET_HDR_GSO_TCPV6:
2307                                 gso_type = SKB_GSO_TCPV6;
2308                                 break;
2309                         case VIRTIO_NET_HDR_GSO_UDP:
2310                                 gso_type = SKB_GSO_UDP;
2311                                 break;
2312                         default:
2313                                 goto out_unlock;
2314                         }
2315
2316                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2317                                 gso_type |= SKB_GSO_TCP_ECN;
2318
2319                         if (vnet_hdr.gso_size == 0)
2320                                 goto out_unlock;
2321
2322                 }
2323         }
2324
2325         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2326                 if (!netif_supports_nofcs(dev)) {
2327                         err = -EPROTONOSUPPORT;
2328                         goto out_unlock;
2329                 }
2330                 extra_len = 4; /* We're doing our own CRC */
2331         }
2332
2333         err = -EMSGSIZE;
2334         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2335                 goto out_unlock;
2336
2337         err = -ENOBUFS;
2338         hlen = LL_RESERVED_SPACE(dev);
2339         tlen = dev->needed_tailroom;
2340         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2341                                msg->msg_flags & MSG_DONTWAIT, &err);
2342         if (skb == NULL)
2343                 goto out_unlock;
2344
2345         skb_set_network_header(skb, reserve);
2346
2347         err = -EINVAL;
2348         if (sock->type == SOCK_DGRAM &&
2349             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2350                 goto out_free;
2351
2352         /* Returns -EFAULT on error */
2353         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2354         if (err)
2355                 goto out_free;
2356
2357         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2358
2359         if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2360                 /* Earlier code assumed this would be a VLAN pkt,
2361                  * double-check this now that we have the actual
2362                  * packet in hand.
2363                  */
2364                 struct ethhdr *ehdr;
2365                 skb_reset_mac_header(skb);
2366                 ehdr = eth_hdr(skb);
2367                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2368                         err = -EMSGSIZE;
2369                         goto out_free;
2370                 }
2371         }
2372
2373         skb->protocol = proto;
2374         skb->dev = dev;
2375         skb->priority = sk->sk_priority;
2376         skb->mark = sk->sk_mark;
2377
2378         if (po->has_vnet_hdr) {
2379                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2380                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2381                                                   vnet_hdr.csum_offset)) {
2382                                 err = -EINVAL;
2383                                 goto out_free;
2384                         }
2385                 }
2386
2387                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2388                 skb_shinfo(skb)->gso_type = gso_type;
2389
2390                 /* Header must be checked, and gso_segs computed. */
2391                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2392                 skb_shinfo(skb)->gso_segs = 0;
2393
2394                 len += vnet_hdr_len;
2395         }
2396
2397         skb_probe_transport_header(skb, reserve);
2398
2399         if (unlikely(extra_len == 4))
2400                 skb->no_fcs = 1;
2401
2402         /*
2403          *      Now send it
2404          */
2405
2406         err = dev_queue_xmit(skb);
2407         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2408                 goto out_unlock;
2409
2410         dev_put(dev);
2411
2412         return len;
2413
2414 out_free:
2415         kfree_skb(skb);
2416 out_unlock:
2417         if (dev)
2418                 dev_put(dev);
2419 out:
2420         return err;
2421 }
2422
2423 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2424                 struct msghdr *msg, size_t len)
2425 {
2426         struct sock *sk = sock->sk;
2427         struct packet_sock *po = pkt_sk(sk);
2428         if (po->tx_ring.pg_vec)
2429                 return tpacket_snd(po, msg);
2430         else
2431                 return packet_snd(sock, msg, len);
2432 }
2433
2434 /*
2435  *      Close a PACKET socket. This is fairly simple. We immediately go
2436  *      to 'closed' state and remove our protocol entry in the device list.
2437  */
2438
2439 static int packet_release(struct socket *sock)
2440 {
2441         struct sock *sk = sock->sk;
2442         struct packet_sock *po;
2443         struct net *net;
2444         union tpacket_req_u req_u;
2445
2446         if (!sk)
2447                 return 0;
2448
2449         net = sock_net(sk);
2450         po = pkt_sk(sk);
2451
2452         mutex_lock(&net->packet.sklist_lock);
2453         sk_del_node_init_rcu(sk);
2454         mutex_unlock(&net->packet.sklist_lock);
2455
2456         preempt_disable();
2457         sock_prot_inuse_add(net, sk->sk_prot, -1);
2458         preempt_enable();
2459
2460         spin_lock(&po->bind_lock);
2461         unregister_prot_hook(sk, false);
2462         packet_cached_dev_reset(po);
2463
2464         if (po->prot_hook.dev) {
2465                 dev_put(po->prot_hook.dev);
2466                 po->prot_hook.dev = NULL;
2467         }
2468         spin_unlock(&po->bind_lock);
2469
2470         packet_flush_mclist(sk);
2471
2472         if (po->rx_ring.pg_vec) {
2473                 memset(&req_u, 0, sizeof(req_u));
2474                 packet_set_ring(sk, &req_u, 1, 0);
2475         }
2476
2477         if (po->tx_ring.pg_vec) {
2478                 memset(&req_u, 0, sizeof(req_u));
2479                 packet_set_ring(sk, &req_u, 1, 1);
2480         }
2481
2482         fanout_release(sk);
2483
2484         synchronize_net();
2485         /*
2486          *      Now the socket is dead. No more input will appear.
2487          */
2488         sock_orphan(sk);
2489         sock->sk = NULL;
2490
2491         /* Purge queues */
2492
2493         skb_queue_purge(&sk->sk_receive_queue);
2494         sk_refcnt_debug_release(sk);
2495
2496         sock_put(sk);
2497         return 0;
2498 }
2499
2500 /*
2501  *      Attach a packet hook.
2502  */
2503
2504 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2505 {
2506         struct packet_sock *po = pkt_sk(sk);
2507
2508         if (po->fanout) {
2509                 if (dev)
2510                         dev_put(dev);
2511
2512                 return -EINVAL;
2513         }
2514
2515         lock_sock(sk);
2516
2517         spin_lock(&po->bind_lock);
2518         unregister_prot_hook(sk, true);
2519
2520         po->num = protocol;
2521         po->prot_hook.type = protocol;
2522         if (po->prot_hook.dev)
2523                 dev_put(po->prot_hook.dev);
2524
2525         po->prot_hook.dev = dev;
2526         po->ifindex = dev ? dev->ifindex : 0;
2527
2528         packet_cached_dev_assign(po, dev);
2529
2530         if (protocol == 0)
2531                 goto out_unlock;
2532
2533         if (!dev || (dev->flags & IFF_UP)) {
2534                 register_prot_hook(sk);
2535         } else {
2536                 sk->sk_err = ENETDOWN;
2537                 if (!sock_flag(sk, SOCK_DEAD))
2538                         sk->sk_error_report(sk);
2539         }
2540
2541 out_unlock:
2542         spin_unlock(&po->bind_lock);
2543         release_sock(sk);
2544         return 0;
2545 }
2546
2547 /*
2548  *      Bind a packet socket to a device
2549  */
2550
2551 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2552                             int addr_len)
2553 {
2554         struct sock *sk = sock->sk;
2555         char name[15];
2556         struct net_device *dev;
2557         int err = -ENODEV;
2558
2559         /*
2560          *      Check legality
2561          */
2562
2563         if (addr_len != sizeof(struct sockaddr))
2564                 return -EINVAL;
2565         strlcpy(name, uaddr->sa_data, sizeof(name));
2566
2567         dev = dev_get_by_name(sock_net(sk), name);
2568         if (dev)
2569                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2570         return err;
2571 }
2572
2573 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2574 {
2575         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2576         struct sock *sk = sock->sk;
2577         struct net_device *dev = NULL;
2578         int err;
2579
2580
2581         /*
2582          *      Check legality
2583          */
2584
2585         if (addr_len < sizeof(struct sockaddr_ll))
2586                 return -EINVAL;
2587         if (sll->sll_family != AF_PACKET)
2588                 return -EINVAL;
2589
2590         if (sll->sll_ifindex) {
2591                 err = -ENODEV;
2592                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2593                 if (dev == NULL)
2594                         goto out;
2595         }
2596         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2597
2598 out:
2599         return err;
2600 }
2601
2602 static struct proto packet_proto = {
2603         .name     = "PACKET",
2604         .owner    = THIS_MODULE,
2605         .obj_size = sizeof(struct packet_sock),
2606 };
2607
2608 /*
2609  *      Create a packet of type SOCK_PACKET.
2610  */
2611
2612 static int packet_create(struct net *net, struct socket *sock, int protocol,
2613                          int kern)
2614 {
2615         struct sock *sk;
2616         struct packet_sock *po;
2617         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2618         int err;
2619
2620         if (!ns_capable(net->user_ns, CAP_NET_RAW))
2621                 return -EPERM;
2622         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2623             sock->type != SOCK_PACKET)
2624                 return -ESOCKTNOSUPPORT;
2625
2626         sock->state = SS_UNCONNECTED;
2627
2628         err = -ENOBUFS;
2629         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2630         if (sk == NULL)
2631                 goto out;
2632
2633         sock->ops = &packet_ops;
2634         if (sock->type == SOCK_PACKET)
2635                 sock->ops = &packet_ops_spkt;
2636
2637         sock_init_data(sock, sk);
2638
2639         po = pkt_sk(sk);
2640         sk->sk_family = PF_PACKET;
2641         po->num = proto;
2642
2643         packet_cached_dev_reset(po);
2644
2645         sk->sk_destruct = packet_sock_destruct;
2646         sk_refcnt_debug_inc(sk);
2647
2648         /*
2649          *      Attach a protocol block
2650          */
2651
2652         spin_lock_init(&po->bind_lock);
2653         mutex_init(&po->pg_vec_lock);
2654         po->prot_hook.func = packet_rcv;
2655
2656         if (sock->type == SOCK_PACKET)
2657                 po->prot_hook.func = packet_rcv_spkt;
2658
2659         po->prot_hook.af_packet_priv = sk;
2660
2661         if (proto) {
2662                 po->prot_hook.type = proto;
2663                 register_prot_hook(sk);
2664         }
2665
2666         mutex_lock(&net->packet.sklist_lock);
2667         sk_add_node_rcu(sk, &net->packet.sklist);
2668         mutex_unlock(&net->packet.sklist_lock);
2669
2670         preempt_disable();
2671         sock_prot_inuse_add(net, &packet_proto, 1);
2672         preempt_enable();
2673
2674         return 0;
2675 out:
2676         return err;
2677 }
2678
2679 /*
2680  *      Pull a packet from our receive queue and hand it to the user.
2681  *      If necessary we block.
2682  */
2683
2684 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2685                           struct msghdr *msg, size_t len, int flags)
2686 {
2687         struct sock *sk = sock->sk;
2688         struct sk_buff *skb;
2689         int copied, err;
2690         int vnet_hdr_len = 0;
2691
2692         err = -EINVAL;
2693         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2694                 goto out;
2695
2696 #if 0
2697         /* What error should we return now? EUNATTACH? */
2698         if (pkt_sk(sk)->ifindex < 0)
2699                 return -ENODEV;
2700 #endif
2701
2702         if (flags & MSG_ERRQUEUE) {
2703                 err = sock_recv_errqueue(sk, msg, len,
2704                                          SOL_PACKET, PACKET_TX_TIMESTAMP);
2705                 goto out;
2706         }
2707
2708         /*
2709          *      Call the generic datagram receiver. This handles all sorts
2710          *      of horrible races and re-entrancy so we can forget about it
2711          *      in the protocol layers.
2712          *
2713          *      Now it will return ENETDOWN, if device have just gone down,
2714          *      but then it will block.
2715          */
2716
2717         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2718
2719         /*
2720          *      An error occurred so return it. Because skb_recv_datagram()
2721          *      handles the blocking we don't see and worry about blocking
2722          *      retries.
2723          */
2724
2725         if (skb == NULL)
2726                 goto out;
2727
2728         if (pkt_sk(sk)->has_vnet_hdr) {
2729                 struct virtio_net_hdr vnet_hdr = { 0 };
2730
2731                 err = -EINVAL;
2732                 vnet_hdr_len = sizeof(vnet_hdr);
2733                 if (len < vnet_hdr_len)
2734                         goto out_free;
2735
2736                 len -= vnet_hdr_len;
2737
2738                 if (skb_is_gso(skb)) {
2739                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2740
2741                         /* This is a hint as to how much should be linear. */
2742                         vnet_hdr.hdr_len = skb_headlen(skb);
2743                         vnet_hdr.gso_size = sinfo->gso_size;
2744                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2745                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2746                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2747                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2748                         else if (sinfo->gso_type & SKB_GSO_UDP)
2749                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2750                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2751                                 goto out_free;
2752                         else
2753                                 BUG();
2754                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2755                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2756                 } else
2757                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2758
2759                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2760                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2761                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2762                         vnet_hdr.csum_offset = skb->csum_offset;
2763                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2764                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2765                 } /* else everything is zero */
2766
2767                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2768                                      vnet_hdr_len);
2769                 if (err < 0)
2770                         goto out_free;
2771         }
2772
2773         /* You lose any data beyond the buffer you gave. If it worries
2774          * a user program they can ask the device for its MTU
2775          * anyway.
2776          */
2777         copied = skb->len;
2778         if (copied > len) {
2779                 copied = len;
2780                 msg->msg_flags |= MSG_TRUNC;
2781         }
2782
2783         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2784         if (err)
2785                 goto out_free;
2786
2787         sock_recv_ts_and_drops(msg, sk, skb);
2788
2789         if (msg->msg_name) {
2790                 /* If the address length field is there to be filled
2791                  * in, we fill it in now.
2792                  */
2793                 if (sock->type == SOCK_PACKET) {
2794                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
2795                 } else {
2796                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2797                         msg->msg_namelen = sll->sll_halen +
2798                                 offsetof(struct sockaddr_ll, sll_addr);
2799                 }
2800                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2801                        msg->msg_namelen);
2802         }
2803
2804         if (pkt_sk(sk)->auxdata) {
2805                 struct tpacket_auxdata aux;
2806
2807                 aux.tp_status = TP_STATUS_USER;
2808                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2809                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2810                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2811                 aux.tp_snaplen = skb->len;
2812                 aux.tp_mac = 0;
2813                 aux.tp_net = skb_network_offset(skb);
2814                 if (vlan_tx_tag_present(skb)) {
2815                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2816                         aux.tp_status |= TP_STATUS_VLAN_VALID;
2817                 } else {
2818                         aux.tp_vlan_tci = 0;
2819                 }
2820                 aux.tp_padding = 0;
2821                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2822         }
2823
2824         /*
2825          *      Free or return the buffer as appropriate. Again this
2826          *      hides all the races and re-entrancy issues from us.
2827          */
2828         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2829
2830 out_free:
2831         skb_free_datagram(sk, skb);
2832 out:
2833         return err;
2834 }
2835
2836 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2837                                int *uaddr_len, int peer)
2838 {
2839         struct net_device *dev;
2840         struct sock *sk = sock->sk;
2841
2842         if (peer)
2843                 return -EOPNOTSUPP;
2844
2845         uaddr->sa_family = AF_PACKET;
2846         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
2847         rcu_read_lock();
2848         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2849         if (dev)
2850                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
2851         rcu_read_unlock();
2852         *uaddr_len = sizeof(*uaddr);
2853
2854         return 0;
2855 }
2856
2857 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2858                           int *uaddr_len, int peer)
2859 {
2860         struct net_device *dev;
2861         struct sock *sk = sock->sk;
2862         struct packet_sock *po = pkt_sk(sk);
2863         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2864
2865         if (peer)
2866                 return -EOPNOTSUPP;
2867
2868         sll->sll_family = AF_PACKET;
2869         sll->sll_ifindex = po->ifindex;
2870         sll->sll_protocol = po->num;
2871         sll->sll_pkttype = 0;
2872         rcu_read_lock();
2873         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2874         if (dev) {
2875                 sll->sll_hatype = dev->type;
2876                 sll->sll_halen = dev->addr_len;
2877                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2878         } else {
2879                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
2880                 sll->sll_halen = 0;
2881         }
2882         rcu_read_unlock();
2883         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2884
2885         return 0;
2886 }
2887
2888 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2889                          int what)
2890 {
2891         switch (i->type) {
2892         case PACKET_MR_MULTICAST:
2893                 if (i->alen != dev->addr_len)
2894                         return -EINVAL;
2895                 if (what > 0)
2896                         return dev_mc_add(dev, i->addr);
2897                 else
2898                         return dev_mc_del(dev, i->addr);
2899                 break;
2900         case PACKET_MR_PROMISC:
2901                 return dev_set_promiscuity(dev, what);
2902                 break;
2903         case PACKET_MR_ALLMULTI:
2904                 return dev_set_allmulti(dev, what);
2905                 break;
2906         case PACKET_MR_UNICAST:
2907                 if (i->alen != dev->addr_len)
2908                         return -EINVAL;
2909                 if (what > 0)
2910                         return dev_uc_add(dev, i->addr);
2911                 else
2912                         return dev_uc_del(dev, i->addr);
2913                 break;
2914         default:
2915                 break;
2916         }
2917         return 0;
2918 }
2919
2920 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2921 {
2922         for ( ; i; i = i->next) {
2923                 if (i->ifindex == dev->ifindex)
2924                         packet_dev_mc(dev, i, what);
2925         }
2926 }
2927
2928 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2929 {
2930         struct packet_sock *po = pkt_sk(sk);
2931         struct packet_mclist *ml, *i;
2932         struct net_device *dev;
2933         int err;
2934
2935         rtnl_lock();
2936
2937         err = -ENODEV;
2938         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2939         if (!dev)
2940                 goto done;
2941
2942         err = -EINVAL;
2943         if (mreq->mr_alen > dev->addr_len)
2944                 goto done;
2945
2946         err = -ENOBUFS;
2947         i = kmalloc(sizeof(*i), GFP_KERNEL);
2948         if (i == NULL)
2949                 goto done;
2950
2951         err = 0;
2952         for (ml = po->mclist; ml; ml = 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                         ml->count++;
2958                         /* Free the new element ... */
2959                         kfree(i);
2960                         goto done;
2961                 }
2962         }
2963
2964         i->type = mreq->mr_type;
2965         i->ifindex = mreq->mr_ifindex;
2966         i->alen = mreq->mr_alen;
2967         memcpy(i->addr, mreq->mr_address, i->alen);
2968         i->count = 1;
2969         i->next = po->mclist;
2970         po->mclist = i;
2971         err = packet_dev_mc(dev, i, 1);
2972         if (err) {
2973                 po->mclist = i->next;
2974                 kfree(i);
2975         }
2976
2977 done:
2978         rtnl_unlock();
2979         return err;
2980 }
2981
2982 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2983 {
2984         struct packet_mclist *ml, **mlp;
2985
2986         rtnl_lock();
2987
2988         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
2989                 if (ml->ifindex == mreq->mr_ifindex &&
2990                     ml->type == mreq->mr_type &&
2991                     ml->alen == mreq->mr_alen &&
2992                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2993                         if (--ml->count == 0) {
2994                                 struct net_device *dev;
2995                                 *mlp = ml->next;
2996                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2997                                 if (dev)
2998                                         packet_dev_mc(dev, ml, -1);
2999                                 kfree(ml);
3000                         }
3001                         rtnl_unlock();
3002                         return 0;
3003                 }
3004         }
3005         rtnl_unlock();
3006         return -EADDRNOTAVAIL;
3007 }
3008
3009 static void packet_flush_mclist(struct sock *sk)
3010 {
3011         struct packet_sock *po = pkt_sk(sk);
3012         struct packet_mclist *ml;
3013
3014         if (!po->mclist)
3015                 return;
3016
3017         rtnl_lock();
3018         while ((ml = po->mclist) != NULL) {
3019                 struct net_device *dev;
3020
3021                 po->mclist = ml->next;
3022                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3023                 if (dev != NULL)
3024                         packet_dev_mc(dev, ml, -1);
3025                 kfree(ml);
3026         }
3027         rtnl_unlock();
3028 }
3029
3030 static int
3031 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3032 {
3033         struct sock *sk = sock->sk;
3034         struct packet_sock *po = pkt_sk(sk);
3035         int ret;
3036
3037         if (level != SOL_PACKET)
3038                 return -ENOPROTOOPT;
3039
3040         switch (optname) {
3041         case PACKET_ADD_MEMBERSHIP:
3042         case PACKET_DROP_MEMBERSHIP:
3043         {
3044                 struct packet_mreq_max mreq;
3045                 int len = optlen;
3046                 memset(&mreq, 0, sizeof(mreq));
3047                 if (len < sizeof(struct packet_mreq))
3048                         return -EINVAL;
3049                 if (len > sizeof(mreq))
3050                         len = sizeof(mreq);
3051                 if (copy_from_user(&mreq, optval, len))
3052                         return -EFAULT;
3053                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3054                         return -EINVAL;
3055                 if (optname == PACKET_ADD_MEMBERSHIP)
3056                         ret = packet_mc_add(sk, &mreq);
3057                 else
3058                         ret = packet_mc_drop(sk, &mreq);
3059                 return ret;
3060         }
3061
3062         case PACKET_RX_RING:
3063         case PACKET_TX_RING:
3064         {
3065                 union tpacket_req_u req_u;
3066                 int len;
3067
3068                 switch (po->tp_version) {
3069                 case TPACKET_V1:
3070                 case TPACKET_V2:
3071                         len = sizeof(req_u.req);
3072                         break;
3073                 case TPACKET_V3:
3074                 default:
3075                         len = sizeof(req_u.req3);
3076                         break;
3077                 }
3078                 if (optlen < len)
3079                         return -EINVAL;
3080                 if (pkt_sk(sk)->has_vnet_hdr)
3081                         return -EINVAL;
3082                 if (copy_from_user(&req_u.req, optval, len))
3083                         return -EFAULT;
3084                 return packet_set_ring(sk, &req_u, 0,
3085                         optname == PACKET_TX_RING);
3086         }
3087         case PACKET_COPY_THRESH:
3088         {
3089                 int val;
3090
3091                 if (optlen != sizeof(val))
3092                         return -EINVAL;
3093                 if (copy_from_user(&val, optval, sizeof(val)))
3094                         return -EFAULT;
3095
3096                 pkt_sk(sk)->copy_thresh = val;
3097                 return 0;
3098         }
3099         case PACKET_VERSION:
3100         {
3101                 int val;
3102
3103                 if (optlen != sizeof(val))
3104                         return -EINVAL;
3105                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3106                         return -EBUSY;
3107                 if (copy_from_user(&val, optval, sizeof(val)))
3108                         return -EFAULT;
3109                 switch (val) {
3110                 case TPACKET_V1:
3111                 case TPACKET_V2:
3112                 case TPACKET_V3:
3113                         po->tp_version = val;
3114                         return 0;
3115                 default:
3116                         return -EINVAL;
3117                 }
3118         }
3119         case PACKET_RESERVE:
3120         {
3121                 unsigned int val;
3122
3123                 if (optlen != sizeof(val))
3124                         return -EINVAL;
3125                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3126                         return -EBUSY;
3127                 if (copy_from_user(&val, optval, sizeof(val)))
3128                         return -EFAULT;
3129                 po->tp_reserve = val;
3130                 return 0;
3131         }
3132         case PACKET_LOSS:
3133         {
3134                 unsigned int val;
3135
3136                 if (optlen != sizeof(val))
3137                         return -EINVAL;
3138                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3139                         return -EBUSY;
3140                 if (copy_from_user(&val, optval, sizeof(val)))
3141                         return -EFAULT;
3142                 po->tp_loss = !!val;
3143                 return 0;
3144         }
3145         case PACKET_AUXDATA:
3146         {
3147                 int val;
3148
3149                 if (optlen < sizeof(val))
3150                         return -EINVAL;
3151                 if (copy_from_user(&val, optval, sizeof(val)))
3152                         return -EFAULT;
3153
3154                 po->auxdata = !!val;
3155                 return 0;
3156         }
3157         case PACKET_ORIGDEV:
3158         {
3159                 int val;
3160
3161                 if (optlen < sizeof(val))
3162                         return -EINVAL;
3163                 if (copy_from_user(&val, optval, sizeof(val)))
3164                         return -EFAULT;
3165
3166                 po->origdev = !!val;
3167                 return 0;
3168         }
3169         case PACKET_VNET_HDR:
3170         {
3171                 int val;
3172
3173                 if (sock->type != SOCK_RAW)
3174                         return -EINVAL;
3175                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3176                         return -EBUSY;
3177                 if (optlen < sizeof(val))
3178                         return -EINVAL;
3179                 if (copy_from_user(&val, optval, sizeof(val)))
3180                         return -EFAULT;
3181
3182                 po->has_vnet_hdr = !!val;
3183                 return 0;
3184         }
3185         case PACKET_TIMESTAMP:
3186         {
3187                 int val;
3188
3189                 if (optlen != sizeof(val))
3190                         return -EINVAL;
3191                 if (copy_from_user(&val, optval, sizeof(val)))
3192                         return -EFAULT;
3193
3194                 po->tp_tstamp = val;
3195                 return 0;
3196         }
3197         case PACKET_FANOUT:
3198         {
3199                 int val;
3200
3201                 if (optlen != sizeof(val))
3202                         return -EINVAL;
3203                 if (copy_from_user(&val, optval, sizeof(val)))
3204                         return -EFAULT;
3205
3206                 return fanout_add(sk, val & 0xffff, val >> 16);
3207         }
3208         case PACKET_TX_HAS_OFF:
3209         {
3210                 unsigned int val;
3211
3212                 if (optlen != sizeof(val))
3213                         return -EINVAL;
3214                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3215                         return -EBUSY;
3216                 if (copy_from_user(&val, optval, sizeof(val)))
3217                         return -EFAULT;
3218                 po->tp_tx_has_off = !!val;
3219                 return 0;
3220         }
3221         default:
3222                 return -ENOPROTOOPT;
3223         }
3224 }
3225
3226 static int packet_getsockopt(struct socket *sock, int level, int optname,
3227                              char __user *optval, int __user *optlen)
3228 {
3229         int len;
3230         int val, lv = sizeof(val);
3231         struct sock *sk = sock->sk;
3232         struct packet_sock *po = pkt_sk(sk);
3233         void *data = &val;
3234         union tpacket_stats_u st;
3235
3236         if (level != SOL_PACKET)
3237                 return -ENOPROTOOPT;
3238
3239         if (get_user(len, optlen))
3240                 return -EFAULT;
3241
3242         if (len < 0)
3243                 return -EINVAL;
3244
3245         switch (optname) {
3246         case PACKET_STATISTICS:
3247                 spin_lock_bh(&sk->sk_receive_queue.lock);
3248                 memcpy(&st, &po->stats, sizeof(st));
3249                 memset(&po->stats, 0, sizeof(po->stats));
3250                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3251
3252                 if (po->tp_version == TPACKET_V3) {
3253                         lv = sizeof(struct tpacket_stats_v3);
3254                         st.stats3.tp_packets += st.stats3.tp_drops;
3255                         data = &st.stats3;
3256                 } else {
3257                         lv = sizeof(struct tpacket_stats);
3258                         st.stats1.tp_packets += st.stats1.tp_drops;
3259                         data = &st.stats1;
3260                 }
3261
3262                 break;
3263         case PACKET_AUXDATA:
3264                 val = po->auxdata;
3265                 break;
3266         case PACKET_ORIGDEV:
3267                 val = po->origdev;
3268                 break;
3269         case PACKET_VNET_HDR:
3270                 val = po->has_vnet_hdr;
3271                 break;
3272         case PACKET_VERSION:
3273                 val = po->tp_version;
3274                 break;
3275         case PACKET_HDRLEN:
3276                 if (len > sizeof(int))
3277                         len = sizeof(int);
3278                 if (copy_from_user(&val, optval, len))
3279                         return -EFAULT;
3280                 switch (val) {
3281                 case TPACKET_V1:
3282                         val = sizeof(struct tpacket_hdr);
3283                         break;
3284                 case TPACKET_V2:
3285                         val = sizeof(struct tpacket2_hdr);
3286                         break;
3287                 case TPACKET_V3:
3288                         val = sizeof(struct tpacket3_hdr);
3289                         break;
3290                 default:
3291                         return -EINVAL;
3292                 }
3293                 break;
3294         case PACKET_RESERVE:
3295                 val = po->tp_reserve;
3296                 break;
3297         case PACKET_LOSS:
3298                 val = po->tp_loss;
3299                 break;
3300         case PACKET_TIMESTAMP:
3301                 val = po->tp_tstamp;
3302                 break;
3303         case PACKET_FANOUT:
3304                 val = (po->fanout ?
3305                        ((u32)po->fanout->id |
3306                         ((u32)po->fanout->type << 16) |
3307                         ((u32)po->fanout->flags << 24)) :
3308                        0);
3309                 break;
3310         case PACKET_TX_HAS_OFF:
3311                 val = po->tp_tx_has_off;
3312                 break;
3313         default:
3314                 return -ENOPROTOOPT;
3315         }
3316
3317         if (len > lv)
3318                 len = lv;
3319         if (put_user(len, optlen))
3320                 return -EFAULT;
3321         if (copy_to_user(optval, data, len))
3322                 return -EFAULT;
3323         return 0;
3324 }
3325
3326
3327 static int packet_notifier(struct notifier_block *this,
3328                            unsigned long msg, void *ptr)
3329 {
3330         struct sock *sk;
3331         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3332         struct net *net = dev_net(dev);
3333
3334         rcu_read_lock();
3335         sk_for_each_rcu(sk, &net->packet.sklist) {
3336                 struct packet_sock *po = pkt_sk(sk);
3337
3338                 switch (msg) {
3339                 case NETDEV_UNREGISTER:
3340                         if (po->mclist)
3341                                 packet_dev_mclist(dev, po->mclist, -1);
3342                         /* fallthrough */
3343
3344                 case NETDEV_DOWN:
3345                         if (dev->ifindex == po->ifindex) {
3346                                 spin_lock(&po->bind_lock);
3347                                 if (po->running) {
3348                                         __unregister_prot_hook(sk, false);
3349                                         sk->sk_err = ENETDOWN;
3350                                         if (!sock_flag(sk, SOCK_DEAD))
3351                                                 sk->sk_error_report(sk);
3352                                 }
3353                                 if (msg == NETDEV_UNREGISTER) {
3354                                         packet_cached_dev_reset(po);
3355                                         po->ifindex = -1;
3356                                         if (po->prot_hook.dev)
3357                                                 dev_put(po->prot_hook.dev);
3358                                         po->prot_hook.dev = NULL;
3359                                 }
3360                                 spin_unlock(&po->bind_lock);
3361                         }
3362                         break;
3363                 case NETDEV_UP:
3364                         if (dev->ifindex == po->ifindex) {
3365                                 spin_lock(&po->bind_lock);
3366                                 if (po->num)
3367                                         register_prot_hook(sk);
3368                                 spin_unlock(&po->bind_lock);
3369                         }
3370                         break;
3371                 }
3372         }
3373         rcu_read_unlock();
3374         return NOTIFY_DONE;
3375 }
3376
3377
3378 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3379                         unsigned long arg)
3380 {
3381         struct sock *sk = sock->sk;
3382
3383         switch (cmd) {
3384         case SIOCOUTQ:
3385         {
3386                 int amount = sk_wmem_alloc_get(sk);
3387
3388                 return put_user(amount, (int __user *)arg);
3389         }
3390         case SIOCINQ:
3391         {
3392                 struct sk_buff *skb;
3393                 int amount = 0;
3394
3395                 spin_lock_bh(&sk->sk_receive_queue.lock);
3396                 skb = skb_peek(&sk->sk_receive_queue);
3397                 if (skb)
3398                         amount = skb->len;
3399                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3400                 return put_user(amount, (int __user *)arg);
3401         }
3402         case SIOCGSTAMP:
3403                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3404         case SIOCGSTAMPNS:
3405                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3406
3407 #ifdef CONFIG_INET
3408         case SIOCADDRT:
3409         case SIOCDELRT:
3410         case SIOCDARP:
3411         case SIOCGARP:
3412         case SIOCSARP:
3413         case SIOCGIFADDR:
3414         case SIOCSIFADDR:
3415         case SIOCGIFBRDADDR:
3416         case SIOCSIFBRDADDR:
3417         case SIOCGIFNETMASK:
3418         case SIOCSIFNETMASK:
3419         case SIOCGIFDSTADDR:
3420         case SIOCSIFDSTADDR:
3421         case SIOCSIFFLAGS:
3422                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3423 #endif
3424
3425         default:
3426                 return -ENOIOCTLCMD;
3427         }
3428         return 0;
3429 }
3430
3431 static unsigned int packet_poll(struct file *file, struct socket *sock,
3432                                 poll_table *wait)
3433 {
3434         struct sock *sk = sock->sk;
3435         struct packet_sock *po = pkt_sk(sk);
3436         unsigned int mask = datagram_poll(file, sock, wait);
3437
3438         spin_lock_bh(&sk->sk_receive_queue.lock);
3439         if (po->rx_ring.pg_vec) {
3440                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3441                         TP_STATUS_KERNEL))
3442                         mask |= POLLIN | POLLRDNORM;
3443         }
3444         spin_unlock_bh(&sk->sk_receive_queue.lock);
3445         spin_lock_bh(&sk->sk_write_queue.lock);
3446         if (po->tx_ring.pg_vec) {
3447                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3448                         mask |= POLLOUT | POLLWRNORM;
3449         }
3450         spin_unlock_bh(&sk->sk_write_queue.lock);
3451         return mask;
3452 }
3453
3454
3455 /* Dirty? Well, I still did not learn better way to account
3456  * for user mmaps.
3457  */
3458
3459 static void packet_mm_open(struct vm_area_struct *vma)
3460 {
3461         struct file *file = vma->vm_file;
3462         struct socket *sock = file->private_data;
3463         struct sock *sk = sock->sk;
3464
3465         if (sk)
3466                 atomic_inc(&pkt_sk(sk)->mapped);
3467 }
3468
3469 static void packet_mm_close(struct vm_area_struct *vma)
3470 {
3471         struct file *file = vma->vm_file;
3472         struct socket *sock = file->private_data;
3473         struct sock *sk = sock->sk;
3474
3475         if (sk)
3476                 atomic_dec(&pkt_sk(sk)->mapped);
3477 }
3478
3479 static const struct vm_operations_struct packet_mmap_ops = {
3480         .open   =       packet_mm_open,
3481         .close  =       packet_mm_close,
3482 };
3483
3484 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3485                         unsigned int len)
3486 {
3487         int i;
3488
3489         for (i = 0; i < len; i++) {
3490                 if (likely(pg_vec[i].buffer)) {
3491                         if (is_vmalloc_addr(pg_vec[i].buffer))
3492                                 vfree(pg_vec[i].buffer);
3493                         else
3494                                 free_pages((unsigned long)pg_vec[i].buffer,
3495                                            order);
3496                         pg_vec[i].buffer = NULL;
3497                 }
3498         }
3499         kfree(pg_vec);
3500 }
3501
3502 static char *alloc_one_pg_vec_page(unsigned long order)
3503 {
3504         char *buffer = NULL;
3505         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3506                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3507
3508         buffer = (char *) __get_free_pages(gfp_flags, order);
3509
3510         if (buffer)
3511                 return buffer;
3512
3513         /*
3514          * __get_free_pages failed, fall back to vmalloc
3515          */
3516         buffer = vzalloc((1 << order) * PAGE_SIZE);
3517
3518         if (buffer)
3519                 return buffer;
3520
3521         /*
3522          * vmalloc failed, lets dig into swap here
3523          */
3524         gfp_flags &= ~__GFP_NORETRY;
3525         buffer = (char *)__get_free_pages(gfp_flags, order);
3526         if (buffer)
3527                 return buffer;
3528
3529         /*
3530          * complete and utter failure
3531          */
3532         return NULL;
3533 }
3534
3535 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3536 {
3537         unsigned int block_nr = req->tp_block_nr;
3538         struct pgv *pg_vec;
3539         int i;
3540
3541         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3542         if (unlikely(!pg_vec))
3543                 goto out;
3544
3545         for (i = 0; i < block_nr; i++) {
3546                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3547                 if (unlikely(!pg_vec[i].buffer))
3548                         goto out_free_pgvec;
3549         }
3550
3551 out:
3552         return pg_vec;
3553
3554 out_free_pgvec:
3555         free_pg_vec(pg_vec, order, block_nr);
3556         pg_vec = NULL;
3557         goto out;
3558 }
3559
3560 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3561                 int closing, int tx_ring)
3562 {
3563         struct pgv *pg_vec = NULL;
3564         struct packet_sock *po = pkt_sk(sk);
3565         int was_running, order = 0;
3566         struct packet_ring_buffer *rb;
3567         struct sk_buff_head *rb_queue;
3568         __be16 num;
3569         int err = -EINVAL;
3570         /* Added to avoid minimal code churn */
3571         struct tpacket_req *req = &req_u->req;
3572
3573         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3574         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3575                 WARN(1, "Tx-ring is not supported.\n");
3576                 goto out;
3577         }
3578
3579         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3580         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3581
3582         err = -EBUSY;
3583         if (!closing) {
3584                 if (atomic_read(&po->mapped))
3585                         goto out;
3586                 if (atomic_read(&rb->pending))
3587                         goto out;
3588         }
3589
3590         if (req->tp_block_nr) {
3591                 /* Sanity tests and some calculations */
3592                 err = -EBUSY;
3593                 if (unlikely(rb->pg_vec))
3594                         goto out;
3595
3596                 switch (po->tp_version) {
3597                 case TPACKET_V1:
3598                         po->tp_hdrlen = TPACKET_HDRLEN;
3599                         break;
3600                 case TPACKET_V2:
3601                         po->tp_hdrlen = TPACKET2_HDRLEN;
3602                         break;
3603                 case TPACKET_V3:
3604                         po->tp_hdrlen = TPACKET3_HDRLEN;
3605                         break;
3606                 }
3607
3608                 err = -EINVAL;
3609                 if (unlikely((int)req->tp_block_size <= 0))
3610                         goto out;
3611                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3612                         goto out;
3613                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3614                                         po->tp_reserve))
3615                         goto out;
3616                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3617                         goto out;
3618
3619                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3620                 if (unlikely(rb->frames_per_block <= 0))
3621                         goto out;
3622                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3623                                         req->tp_frame_nr))
3624                         goto out;
3625
3626                 err = -ENOMEM;
3627                 order = get_order(req->tp_block_size);
3628                 pg_vec = alloc_pg_vec(req, order);
3629                 if (unlikely(!pg_vec))
3630                         goto out;
3631                 switch (po->tp_version) {
3632                 case TPACKET_V3:
3633                 /* Transmit path is not supported. We checked
3634                  * it above but just being paranoid
3635                  */
3636                         if (!tx_ring)
3637                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3638                                 break;
3639                 default:
3640                         break;
3641                 }
3642         }
3643         /* Done */
3644         else {
3645                 err = -EINVAL;
3646                 if (unlikely(req->tp_frame_nr))
3647                         goto out;
3648         }
3649
3650         lock_sock(sk);
3651
3652         /* Detach socket from network */
3653         spin_lock(&po->bind_lock);
3654         was_running = po->running;
3655         num = po->num;
3656         if (was_running) {
3657                 po->num = 0;
3658                 __unregister_prot_hook(sk, false);
3659         }
3660         spin_unlock(&po->bind_lock);
3661
3662         synchronize_net();
3663
3664         err = -EBUSY;
3665         mutex_lock(&po->pg_vec_lock);
3666         if (closing || atomic_read(&po->mapped) == 0) {
3667                 err = 0;
3668                 spin_lock_bh(&rb_queue->lock);
3669                 swap(rb->pg_vec, pg_vec);
3670                 rb->frame_max = (req->tp_frame_nr - 1);
3671                 rb->head = 0;
3672                 rb->frame_size = req->tp_frame_size;
3673                 spin_unlock_bh(&rb_queue->lock);
3674
3675                 swap(rb->pg_vec_order, order);
3676                 swap(rb->pg_vec_len, req->tp_block_nr);
3677
3678                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3679                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3680                                                 tpacket_rcv : packet_rcv;
3681                 skb_queue_purge(rb_queue);
3682                 if (atomic_read(&po->mapped))
3683                         pr_err("packet_mmap: vma is busy: %d\n",
3684                                atomic_read(&po->mapped));
3685         }
3686         mutex_unlock(&po->pg_vec_lock);
3687
3688         spin_lock(&po->bind_lock);
3689         if (was_running) {
3690                 po->num = num;
3691                 register_prot_hook(sk);
3692         }
3693         spin_unlock(&po->bind_lock);
3694         if (closing && (po->tp_version > TPACKET_V2)) {
3695                 /* Because we don't support block-based V3 on tx-ring */
3696                 if (!tx_ring)
3697                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3698         }
3699         release_sock(sk);
3700
3701         if (pg_vec)
3702                 free_pg_vec(pg_vec, order, req->tp_block_nr);
3703 out:
3704         return err;
3705 }
3706
3707 static int packet_mmap(struct file *file, struct socket *sock,
3708                 struct vm_area_struct *vma)
3709 {
3710         struct sock *sk = sock->sk;
3711         struct packet_sock *po = pkt_sk(sk);
3712         unsigned long size, expected_size;
3713         struct packet_ring_buffer *rb;
3714         unsigned long start;
3715         int err = -EINVAL;
3716         int i;
3717
3718         if (vma->vm_pgoff)
3719                 return -EINVAL;
3720
3721         mutex_lock(&po->pg_vec_lock);
3722
3723         expected_size = 0;
3724         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3725                 if (rb->pg_vec) {
3726                         expected_size += rb->pg_vec_len
3727                                                 * rb->pg_vec_pages
3728                                                 * PAGE_SIZE;
3729                 }
3730         }
3731
3732         if (expected_size == 0)
3733                 goto out;
3734
3735         size = vma->vm_end - vma->vm_start;
3736         if (size != expected_size)
3737                 goto out;
3738
3739         start = vma->vm_start;
3740         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3741                 if (rb->pg_vec == NULL)
3742                         continue;
3743
3744                 for (i = 0; i < rb->pg_vec_len; i++) {
3745                         struct page *page;
3746                         void *kaddr = rb->pg_vec[i].buffer;
3747                         int pg_num;
3748
3749                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3750                                 page = pgv_to_page(kaddr);
3751                                 err = vm_insert_page(vma, start, page);
3752                                 if (unlikely(err))
3753                                         goto out;
3754                                 start += PAGE_SIZE;
3755                                 kaddr += PAGE_SIZE;
3756                         }
3757                 }
3758         }
3759
3760         atomic_inc(&po->mapped);
3761         vma->vm_ops = &packet_mmap_ops;
3762         err = 0;
3763
3764 out:
3765         mutex_unlock(&po->pg_vec_lock);
3766         return err;
3767 }
3768
3769 static const struct proto_ops packet_ops_spkt = {
3770         .family =       PF_PACKET,
3771         .owner =        THIS_MODULE,
3772         .release =      packet_release,
3773         .bind =         packet_bind_spkt,
3774         .connect =      sock_no_connect,
3775         .socketpair =   sock_no_socketpair,
3776         .accept =       sock_no_accept,
3777         .getname =      packet_getname_spkt,
3778         .poll =         datagram_poll,
3779         .ioctl =        packet_ioctl,
3780         .listen =       sock_no_listen,
3781         .shutdown =     sock_no_shutdown,
3782         .setsockopt =   sock_no_setsockopt,
3783         .getsockopt =   sock_no_getsockopt,
3784         .sendmsg =      packet_sendmsg_spkt,
3785         .recvmsg =      packet_recvmsg,
3786         .mmap =         sock_no_mmap,
3787         .sendpage =     sock_no_sendpage,
3788 };
3789
3790 static const struct proto_ops packet_ops = {
3791         .family =       PF_PACKET,
3792         .owner =        THIS_MODULE,
3793         .release =      packet_release,
3794         .bind =         packet_bind,
3795         .connect =      sock_no_connect,
3796         .socketpair =   sock_no_socketpair,
3797         .accept =       sock_no_accept,
3798         .getname =      packet_getname,
3799         .poll =         packet_poll,
3800         .ioctl =        packet_ioctl,
3801         .listen =       sock_no_listen,
3802         .shutdown =     sock_no_shutdown,
3803         .setsockopt =   packet_setsockopt,
3804         .getsockopt =   packet_getsockopt,
3805         .sendmsg =      packet_sendmsg,
3806         .recvmsg =      packet_recvmsg,
3807         .mmap =         packet_mmap,
3808         .sendpage =     sock_no_sendpage,
3809 };
3810
3811 static const struct net_proto_family packet_family_ops = {
3812         .family =       PF_PACKET,
3813         .create =       packet_create,
3814         .owner  =       THIS_MODULE,
3815 };
3816
3817 static struct notifier_block packet_netdev_notifier = {
3818         .notifier_call =        packet_notifier,
3819 };
3820
3821 #ifdef CONFIG_PROC_FS
3822
3823 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3824         __acquires(RCU)
3825 {
3826         struct net *net = seq_file_net(seq);
3827
3828         rcu_read_lock();
3829         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3830 }
3831
3832 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3833 {
3834         struct net *net = seq_file_net(seq);
3835         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3836 }
3837
3838 static void packet_seq_stop(struct seq_file *seq, void *v)
3839         __releases(RCU)
3840 {
3841         rcu_read_unlock();
3842 }
3843
3844 static int packet_seq_show(struct seq_file *seq, void *v)
3845 {
3846         if (v == SEQ_START_TOKEN)
3847                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
3848         else {
3849                 struct sock *s = sk_entry(v);
3850                 const struct packet_sock *po = pkt_sk(s);
3851
3852                 seq_printf(seq,
3853                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
3854                            s,
3855                            atomic_read(&s->sk_refcnt),
3856                            s->sk_type,
3857                            ntohs(po->num),
3858                            po->ifindex,
3859                            po->running,
3860                            atomic_read(&s->sk_rmem_alloc),
3861                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
3862                            sock_i_ino(s));
3863         }
3864
3865         return 0;
3866 }
3867
3868 static const struct seq_operations packet_seq_ops = {
3869         .start  = packet_seq_start,
3870         .next   = packet_seq_next,
3871         .stop   = packet_seq_stop,
3872         .show   = packet_seq_show,
3873 };
3874
3875 static int packet_seq_open(struct inode *inode, struct file *file)
3876 {
3877         return seq_open_net(inode, file, &packet_seq_ops,
3878                             sizeof(struct seq_net_private));
3879 }
3880
3881 static const struct file_operations packet_seq_fops = {
3882         .owner          = THIS_MODULE,
3883         .open           = packet_seq_open,
3884         .read           = seq_read,
3885         .llseek         = seq_lseek,
3886         .release        = seq_release_net,
3887 };
3888
3889 #endif
3890
3891 static int __net_init packet_net_init(struct net *net)
3892 {
3893         mutex_init(&net->packet.sklist_lock);
3894         INIT_HLIST_HEAD(&net->packet.sklist);
3895
3896         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
3897                 return -ENOMEM;
3898
3899         return 0;
3900 }
3901
3902 static void __net_exit packet_net_exit(struct net *net)
3903 {
3904         remove_proc_entry("packet", net->proc_net);
3905 }
3906
3907 static struct pernet_operations packet_net_ops = {
3908         .init = packet_net_init,
3909         .exit = packet_net_exit,
3910 };
3911
3912
3913 static void __exit packet_exit(void)
3914 {
3915         unregister_netdevice_notifier(&packet_netdev_notifier);
3916         unregister_pernet_subsys(&packet_net_ops);
3917         sock_unregister(PF_PACKET);
3918         proto_unregister(&packet_proto);
3919 }
3920
3921 static int __init packet_init(void)
3922 {
3923         int rc = proto_register(&packet_proto, 0);
3924
3925         if (rc != 0)
3926                 goto out;
3927
3928         sock_register(&packet_family_ops);
3929         register_pernet_subsys(&packet_net_ops);
3930         register_netdevice_notifier(&packet_netdev_notifier);
3931 out:
3932         return rc;
3933 }
3934
3935 module_init(packet_init);
3936 module_exit(packet_exit);
3937 MODULE_LICENSE("GPL");
3938 MODULE_ALIAS_NETPROTO(PF_PACKET);