net/packet/af_packet.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              PACKET - implements raw packet sockets.
   7  *
   8  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  11  *
  12  * Fixes:
  13  *              Alan Cox        :       verify_area() now used correctly
  14  *              Alan Cox        :       new skbuff lists, look ma no backlogs!
  15  *              Alan Cox        :       tidied skbuff lists.
  16  *              Alan Cox        :       Now uses generic datagram routines I
  17  *                                      added. Also fixed the peek/read crash
  18  *                                      from all old Linux datagram code.
  19  *              Alan Cox        :       Uses the improved datagram code.
  20  *              Alan Cox        :       Added NULL's for socket options.
  21  *              Alan Cox        :       Re-commented the code.
  22  *              Alan Cox        :       Use new kernel side addressing
  23  *              Rob Janssen     :       Correct MTU usage.
  24  *              Dave Platt      :       Counter leaks caused by incorrect
  25  *                                      interrupt locking and some slightly
  26  *                                      dubious gcc output. Can you read
  27  *                                      compiler: it said _VOLATILE_
  28  *      Richard Kooijman        :       Timestamp fixes.
  29  *              Alan Cox        :       New buffers. Use sk->mac.raw.
  30  *              Alan Cox        :       sendmsg/recvmsg support.
  31  *              Alan Cox        :       Protocol setting support
  32  *      Alexey Kuznetsov        :       Untied from IPv4 stack.
  33  *      Cyrus Durgin            :       Fixed kerneld for kmod.
  34  *      Michal Ostrowski        :       Module initialization cleanup.
  35  *         Ulises Alonso        :       Frame number limit removal and
  36  *                                      packet_set_ring memory leak.
  37  *              Eric Biederman  :       Allow for > 8 byte hardware addresses.
  38  *                                      The convention is that longer addresses
  39  *                                      will simply extend the hardware address
  40  *                                      byte arrays at the end of sockaddr_ll
  41  *                                      and packet_mreq.
  42  *              Johann Baudy    :       Added TX RING.
  43  *              Chetan Loke     :       Implemented TPACKET_V3 block abstraction
  44  *                                      layer.
  45  *                                      Copyright (C) 2011, <lokec@ccs.neu.edu>
  46  *
  47  *
  48  *              This program is free software; you can redistribute it and/or
  49  *              modify it under the terms of the GNU General Public License
  50  *              as published by the Free Software Foundation; either version
  51  *              2 of the License, or (at your option) any later version.
  52  *
  53  */
  54
  55 #include <linux/types.h>
  56 #include <linux/mm.h>
  57 #include <linux/capability.h>
  58 #include <linux/fcntl.h>
  59 #include <linux/socket.h>
  60 #include <linux/in.h>
  61 #include <linux/inet.h>
  62 #include <linux/netdevice.h>
  63 #include <linux/if_packet.h>
  64 #include <linux/wireless.h>
  65 #include <linux/kernel.h>
  66 #include <linux/kmod.h>
  67 #include <linux/slab.h>
  68 #include <linux/vmalloc.h>
  69 #include <net/net_namespace.h>
  70 #include <net/ip.h>
  71 #include <net/protocol.h>
  72 #include <linux/skbuff.h>
  73 #include <net/sock.h>
  74 #include <linux/errno.h>
  75 #include <linux/timer.h>
  76 #include <asm/uaccess.h>
  77 #include <asm/ioctls.h>
  78 #include <asm/page.h>
  79 #include <asm/cacheflush.h>
  80 #include <asm/io.h>
  81 #include <linux/proc_fs.h>
  82 #include <linux/seq_file.h>
  83 #include <linux/poll.h>
  84 #include <linux/module.h>
  85 #include <linux/init.h>
  86 #include <linux/mutex.h>
  87 #include <linux/if_vlan.h>
  88 #include <linux/virtio_net.h>
  89 #include <linux/errqueue.h>
  90 #include <linux/net_tstamp.h>
  91
  92 #ifdef CONFIG_INET
  93 #include <net/inet_common.h>
  94 #endif
  95
  96 #include "internal.h"
  97
  98 /*
  99    Assumptions:
 100    - if device has no dev->hard_header routine, it adds and removes ll header
 101      inside itself. In this case ll header is invisible outside of device,
 102      but higher levels still should reserve dev->hard_header_len.
 103      Some devices are enough clever to reallocate skb, when header
 104      will not fit to reserved space (tunnel), another ones are silly
 105      (PPP).
 106    - packet socket receives packets with pulled ll header,
 107      so that SOCK_RAW should push it back.
 108
 109 On receive:
 110 -----------
 111
 112 Incoming, dev->hard_header!=NULL
 113    mac_header -> ll header
 114    data       -> data
 115
 116 Outgoing, dev->hard_header!=NULL
 117    mac_header -> ll header
 118    data       -> ll header
 119
 120 Incoming, dev->hard_header==NULL
 121    mac_header -> UNKNOWN position. It is very likely, that it points to ll
 122                  header.  PPP makes it, that is wrong, because introduce
 123                  assymetry between rx and tx paths.
 124    data       -> data
 125
 126 Outgoing, dev->hard_header==NULL
 127    mac_header -> data. ll header is still not built!
 128    data       -> data
 129
 130 Resume
 131   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 132
 133
 134 On transmit:
 135 ------------
 136
 137 dev->hard_header != NULL
 138    mac_header -> ll header
 139    data       -> ll header
 140
 141 dev->hard_header == NULL (ll header is added by device, we cannot control it)
 142    mac_header -> data
 143    data       -> data
 144
 145    We should set nh.raw on output to correct posistion,
 146    packet classifier depends on it.
 147  */
 148
 149 /* Private packet socket structures. */
 150
 151 /* identical to struct packet_mreq except it has
 152  * a longer address field.
 153  */
 154 struct packet_mreq_max {
 155         int             mr_ifindex;
 156         unsigned short  mr_type;
 157         unsigned short  mr_alen;
 158         unsigned char   mr_address[MAX_ADDR_LEN];
 159 };
 160
 161 union tpacket_uhdr {
 162         struct tpacket_hdr  *h1;
 163         struct tpacket2_hdr *h2;
 164         struct tpacket3_hdr *h3;
 165         void *raw;
 166 };
 167
 168 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 169                 int closing, int tx_ring);
 170
 171 #define V3_ALIGNMENT    (8)
 172
 173 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 174
 175 #define BLK_PLUS_PRIV(sz_of_priv) \
 176         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 177
 178 #define PGV_FROM_VMALLOC 1
 179
 180 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
 181 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
 182 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
 183 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
 184 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
 185 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
 186 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
 187
 188 struct packet_sock;
 189 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
 190 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 191                        struct packet_type *pt, struct net_device *orig_dev);
 192
 193 static void *packet_previous_frame(struct packet_sock *po,
 194                 struct packet_ring_buffer *rb,
 195                 int status);
 196 static void packet_increment_head(struct packet_ring_buffer *buff);
 197 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
 198                         struct tpacket_block_desc *);
 199 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 200                         struct packet_sock *);
 201 static void prb_retire_current_block(struct tpacket_kbdq_core *,
 202                 struct packet_sock *, unsigned int status);
 203 static int prb_queue_frozen(struct tpacket_kbdq_core *);
 204 static void prb_open_block(struct tpacket_kbdq_core *,
 205                 struct tpacket_block_desc *);
 206 static void prb_retire_rx_blk_timer_expired(unsigned long);
 207 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 208 static void prb_init_blk_timer(struct packet_sock *,
 209                 struct tpacket_kbdq_core *,
 210                 void (*func) (unsigned long));
 211 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 212 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 213                 struct tpacket3_hdr *);
 214 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 215                 struct tpacket3_hdr *);
 216 static void packet_flush_mclist(struct sock *sk);
 217
 218 struct packet_skb_cb {
 219         unsigned int origlen;
 220         union {
 221                 struct sockaddr_pkt pkt;
 222                 struct sockaddr_ll ll;
 223         } sa;
 224 };
 225
 226 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 227
 228 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 229 #define GET_PBLOCK_DESC(x, bid) \
 230         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 231 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 232         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 233 #define GET_NEXT_PRB_BLK_NUM(x) \
 234         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 235         ((x)->kactive_blk_num+1) : 0)
 236
 237 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 238 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 239
 240 /* register_prot_hook must be invoked with the po->bind_lock held,
 241  * or from a context in which asynchronous accesses to the packet
 242  * socket is not possible (packet_create()).
 243  */
 244 static void register_prot_hook(struct sock *sk)
 245 {
 246         struct packet_sock *po = pkt_sk(sk);
 247         if (!po->running) {
 248                 if (po->fanout)
 249                         __fanout_link(sk, po);
 250                 else
 251                         dev_add_pack(&po->prot_hook);
 252                 sock_hold(sk);
 253                 po->running = 1;
 254         }
 255 }
 256
 257 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
 258  * held.   If the sync parameter is true, we will temporarily drop
 259  * the po->bind_lock and do a synchronize_net to make sure no
 260  * asynchronous packet processing paths still refer to the elements
 261  * of po->prot_hook.  If the sync parameter is false, it is the
 262  * callers responsibility to take care of this.
 263  */
 264 static void __unregister_prot_hook(struct sock *sk, bool sync)
 265 {
 266         struct packet_sock *po = pkt_sk(sk);
 267
 268         po->running = 0;
 269         if (po->fanout)
 270                 __fanout_unlink(sk, po);
 271         else
 272                 __dev_remove_pack(&po->prot_hook);
 273         __sock_put(sk);
 274
 275         if (sync) {
 276                 spin_unlock(&po->bind_lock);
 277                 synchronize_net();
 278                 spin_lock(&po->bind_lock);
 279         }
 280 }
 281
 282 static void unregister_prot_hook(struct sock *sk, bool sync)
 283 {
 284         struct packet_sock *po = pkt_sk(sk);
 285
 286         if (po->running)
 287                 __unregister_prot_hook(sk, sync);
 288 }
 289
 290 static inline __pure struct page *pgv_to_page(void *addr)
 291 {
 292         if (is_vmalloc_addr(addr))
 293                 return vmalloc_to_page(addr);
 294         return virt_to_page(addr);
 295 }
 296
 297 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 298 {
 299         union tpacket_uhdr h;
 300
 301         h.raw = frame;
 302         switch (po->tp_version) {
 303         case TPACKET_V1:
 304                 h.h1->tp_status = status;
 305                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 306                 break;
 307         case TPACKET_V2:
 308                 h.h2->tp_status = status;
 309                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 310                 break;
 311         case TPACKET_V3:
 312         default:
 313                 WARN(1, "TPACKET version not supported.\n");
 314                 BUG();
 315         }
 316
 317         smp_wmb();
 318 }
 319
 320 static int __packet_get_status(struct packet_sock *po, void *frame)
 321 {
 322         union tpacket_uhdr h;
 323
 324         smp_rmb();
 325
 326         h.raw = frame;
 327         switch (po->tp_version) {
 328         case TPACKET_V1:
 329                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 330                 return h.h1->tp_status;
 331         case TPACKET_V2:
 332                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 333                 return h.h2->tp_status;
 334         case TPACKET_V3:
 335         default:
 336                 WARN(1, "TPACKET version not supported.\n");
 337                 BUG();
 338                 return 0;
 339         }
 340 }
 341
 342 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
 343                                    unsigned int flags)
 344 {
 345         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 346
 347         if (shhwtstamps) {
 348                 if ((flags & SOF_TIMESTAMPING_SYS_HARDWARE) &&
 349                     ktime_to_timespec_cond(shhwtstamps->syststamp, ts))
 350                         return TP_STATUS_TS_SYS_HARDWARE;
 351                 if ((flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
 352                     ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
 353                         return TP_STATUS_TS_RAW_HARDWARE;
 354         }
 355
 356         if (ktime_to_timespec_cond(skb->tstamp, ts))
 357                 return TP_STATUS_TS_SOFTWARE;
 358
 359         return 0;
 360 }
 361
 362 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
 363                                     struct sk_buff *skb)
 364 {
 365         union tpacket_uhdr h;
 366         struct timespec ts;
 367         __u32 ts_status;
 368
 369         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
 370                 return 0;
 371
 372         h.raw = frame;
 373         switch (po->tp_version) {
 374         case TPACKET_V1:
 375                 h.h1->tp_sec = ts.tv_sec;
 376                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
 377                 break;
 378         case TPACKET_V2:
 379                 h.h2->tp_sec = ts.tv_sec;
 380                 h.h2->tp_nsec = ts.tv_nsec;
 381                 break;
 382         case TPACKET_V3:
 383         default:
 384                 WARN(1, "TPACKET version not supported.\n");
 385                 BUG();
 386         }
 387
 388         /* one flush is safe, as both fields always lie on the same cacheline */
 389         flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
 390         smp_wmb();
 391
 392         return ts_status;
 393 }
 394
 395 static void *packet_lookup_frame(struct packet_sock *po,
 396                 struct packet_ring_buffer *rb,
 397                 unsigned int position,
 398                 int status)
 399 {
 400         unsigned int pg_vec_pos, frame_offset;
 401         union tpacket_uhdr h;
 402
 403         pg_vec_pos = position / rb->frames_per_block;
 404         frame_offset = position % rb->frames_per_block;
 405
 406         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 407                 (frame_offset * rb->frame_size);
 408
 409         if (status != __packet_get_status(po, h.raw))
 410                 return NULL;
 411
 412         return h.raw;
 413 }
 414
 415 static void *packet_current_frame(struct packet_sock *po,
 416                 struct packet_ring_buffer *rb,
 417                 int status)
 418 {
 419         return packet_lookup_frame(po, rb, rb->head, status);
 420 }
 421
 422 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 423 {
 424         del_timer_sync(&pkc->retire_blk_timer);
 425 }
 426
 427 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 428                 int tx_ring,
 429                 struct sk_buff_head *rb_queue)
 430 {
 431         struct tpacket_kbdq_core *pkc;
 432
 433         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 434
 435         spin_lock(&rb_queue->lock);
 436         pkc->delete_blk_timer = 1;
 437         spin_unlock(&rb_queue->lock);
 438
 439         prb_del_retire_blk_timer(pkc);
 440 }
 441
 442 static void prb_init_blk_timer(struct packet_sock *po,
 443                 struct tpacket_kbdq_core *pkc,
 444                 void (*func) (unsigned long))
 445 {
 446         init_timer(&pkc->retire_blk_timer);
 447         pkc->retire_blk_timer.data = (long)po;
 448         pkc->retire_blk_timer.function = func;
 449         pkc->retire_blk_timer.expires = jiffies;
 450 }
 451
 452 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
 453 {
 454         struct tpacket_kbdq_core *pkc;
 455
 456         if (tx_ring)
 457                 BUG();
 458
 459         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 460         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
 461 }
 462
 463 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 464                                 int blk_size_in_bytes)
 465 {
 466         struct net_device *dev;
 467         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 468         struct ethtool_cmd ecmd;
 469         int err;
 470         u32 speed;
 471
 472         rtnl_lock();
 473         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 474         if (unlikely(!dev)) {
 475                 rtnl_unlock();
 476                 return DEFAULT_PRB_RETIRE_TOV;
 477         }
 478         err = __ethtool_get_settings(dev, &ecmd);
 479         speed = ethtool_cmd_speed(&ecmd);
 480         rtnl_unlock();
 481         if (!err) {
 482                 /*
 483                  * If the link speed is so slow you don't really
 484                  * need to worry about perf anyways
 485                  */
 486                 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
 487                         return DEFAULT_PRB_RETIRE_TOV;
 488                 } else {
 489                         msec = 1;
 490                         div = speed / 1000;
 491                 }
 492         }
 493
 494         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 495
 496         if (div)
 497                 mbits /= div;
 498
 499         tmo = mbits * msec;
 500
 501         if (div)
 502                 return tmo+1;
 503         return tmo;
 504 }
 505
 506 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 507                         union tpacket_req_u *req_u)
 508 {
 509         p1->feature_req_word = req_u->req3.tp_feature_req_word;
 510 }
 511
 512 static void init_prb_bdqc(struct packet_sock *po,
 513                         struct packet_ring_buffer *rb,
 514                         struct pgv *pg_vec,
 515                         union tpacket_req_u *req_u, int tx_ring)
 516 {
 517         struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
 518         struct tpacket_block_desc *pbd;
 519
 520         memset(p1, 0x0, sizeof(*p1));
 521
 522         p1->knxt_seq_num = 1;
 523         p1->pkbdq = pg_vec;
 524         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 525         p1->pkblk_start = pg_vec[0].buffer;
 526         p1->kblk_size = req_u->req3.tp_block_size;
 527         p1->knum_blocks = req_u->req3.tp_block_nr;
 528         p1->hdrlen = po->tp_hdrlen;
 529         p1->version = po->tp_version;
 530         p1->last_kactive_blk_num = 0;
 531         po->stats.stats3.tp_freeze_q_cnt = 0;
 532         if (req_u->req3.tp_retire_blk_tov)
 533                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 534         else
 535                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 536                                                 req_u->req3.tp_block_size);
 537         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 538         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 539
 540         prb_init_ft_ops(p1, req_u);
 541         prb_setup_retire_blk_timer(po, tx_ring);
 542         prb_open_block(p1, pbd);
 543 }
 544
 545 /*  Do NOT update the last_blk_num first.
 546  *  Assumes sk_buff_head lock is held.
 547  */
 548 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 549 {
 550         mod_timer(&pkc->retire_blk_timer,
 551                         jiffies + pkc->tov_in_jiffies);
 552         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 553 }
 554
 555 /*
 556  * Timer logic:
 557  * 1) We refresh the timer only when we open a block.
 558  *    By doing this we don't waste cycles refreshing the timer
 559  *        on packet-by-packet basis.
 560  *
 561  * With a 1MB block-size, on a 1Gbps line, it will take
 562  * i) ~8 ms to fill a block + ii) memcpy etc.
 563  * In this cut we are not accounting for the memcpy time.
 564  *
 565  * So, if the user sets the 'tmo' to 10ms then the timer
 566  * will never fire while the block is still getting filled
 567  * (which is what we want). However, the user could choose
 568  * to close a block early and that's fine.
 569  *
 570  * But when the timer does fire, we check whether or not to refresh it.
 571  * Since the tmo granularity is in msecs, it is not too expensive
 572  * to refresh the timer, lets say every '8' msecs.
 573  * Either the user can set the 'tmo' or we can derive it based on
 574  * a) line-speed and b) block-size.
 575  * prb_calc_retire_blk_tmo() calculates the tmo.
 576  *
 577  */
 578 static void prb_retire_rx_blk_timer_expired(unsigned long data)
 579 {
 580         struct packet_sock *po = (struct packet_sock *)data;
 581         struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
 582         unsigned int frozen;
 583         struct tpacket_block_desc *pbd;
 584
 585         spin_lock(&po->sk.sk_receive_queue.lock);
 586
 587         frozen = prb_queue_frozen(pkc);
 588         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 589
 590         if (unlikely(pkc->delete_blk_timer))
 591                 goto out;
 592
 593         /* We only need to plug the race when the block is partially filled.
 594          * tpacket_rcv:
 595          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 596          *              copy_bits() is in progress ...
 597          *              timer fires on other cpu:
 598          *              we can't retire the current block because copy_bits
 599          *              is in progress.
 600          *
 601          */
 602         if (BLOCK_NUM_PKTS(pbd)) {
 603                 while (atomic_read(&pkc->blk_fill_in_prog)) {
 604                         /* Waiting for skb_copy_bits to finish... */
 605                         cpu_relax();
 606                 }
 607         }
 608
 609         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 610                 if (!frozen) {
 611                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 612                         if (!prb_dispatch_next_block(pkc, po))
 613                                 goto refresh_timer;
 614                         else
 615                                 goto out;
 616                 } else {
 617                         /* Case 1. Queue was frozen because user-space was
 618                          *         lagging behind.
 619                          */
 620                         if (prb_curr_blk_in_use(pkc, pbd)) {
 621                                 /*
 622                                  * Ok, user-space is still behind.
 623                                  * So just refresh the timer.
 624                                  */
 625                                 goto refresh_timer;
 626                         } else {
 627                                /* Case 2. queue was frozen,user-space caught up,
 628                                 * now the link went idle && the timer fired.
 629                                 * We don't have a block to close.So we open this
 630                                 * block and restart the timer.
 631                                 * opening a block thaws the queue,restarts timer
 632                                 * Thawing/timer-refresh is a side effect.
 633                                 */
 634                                 prb_open_block(pkc, pbd);
 635                                 goto out;
 636                         }
 637                 }
 638         }
 639
 640 refresh_timer:
 641         _prb_refresh_rx_retire_blk_timer(pkc);
 642
 643 out:
 644         spin_unlock(&po->sk.sk_receive_queue.lock);
 645 }
 646
 647 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 648                 struct tpacket_block_desc *pbd1, __u32 status)
 649 {
 650         /* Flush everything minus the block header */
 651
 652 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 653         u8 *start, *end;
 654
 655         start = (u8 *)pbd1;
 656
 657         /* Skip the block header(we know header WILL fit in 4K) */
 658         start += PAGE_SIZE;
 659
 660         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 661         for (; start < end; start += PAGE_SIZE)
 662                 flush_dcache_page(pgv_to_page(start));
 663
 664         smp_wmb();
 665 #endif
 666
 667         /* Now update the block status. */
 668
 669         BLOCK_STATUS(pbd1) = status;
 670
 671         /* Flush the block header */
 672
 673 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 674         start = (u8 *)pbd1;
 675         flush_dcache_page(pgv_to_page(start));
 676
 677         smp_wmb();
 678 #endif
 679 }
 680
 681 /*
 682  * Side effect:
 683  *
 684  * 1) flush the block
 685  * 2) Increment active_blk_num
 686  *
 687  * Note:We DONT refresh the timer on purpose.
 688  *      Because almost always the next block will be opened.
 689  */
 690 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 691                 struct tpacket_block_desc *pbd1,
 692                 struct packet_sock *po, unsigned int stat)
 693 {
 694         __u32 status = TP_STATUS_USER | stat;
 695
 696         struct tpacket3_hdr *last_pkt;
 697         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 698
 699         if (po->stats.stats3.tp_drops)
 700                 status |= TP_STATUS_LOSING;
 701
 702         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 703         last_pkt->tp_next_offset = 0;
 704
 705         /* Get the ts of the last pkt */
 706         if (BLOCK_NUM_PKTS(pbd1)) {
 707                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 708                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 709         } else {
 710                 /* Ok, we tmo'd - so get the current time */
 711                 struct timespec ts;
 712                 getnstimeofday(&ts);
 713                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 714                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 715         }
 716
 717         smp_wmb();
 718
 719         /* Flush the block */
 720         prb_flush_block(pkc1, pbd1, status);
 721
 722         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 723 }
 724
 725 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 726 {
 727         pkc->reset_pending_on_curr_blk = 0;
 728 }
 729
 730 /*
 731  * Side effect of opening a block:
 732  *
 733  * 1) prb_queue is thawed.
 734  * 2) retire_blk_timer is refreshed.
 735  *
 736  */
 737 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 738         struct tpacket_block_desc *pbd1)
 739 {
 740         struct timespec ts;
 741         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 742
 743         smp_rmb();
 744
 745         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd1))) {
 746
 747                 /* We could have just memset this but we will lose the
 748                  * flexibility of making the priv area sticky
 749                  */
 750                 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 751                 BLOCK_NUM_PKTS(pbd1) = 0;
 752                 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 753                 getnstimeofday(&ts);
 754                 h1->ts_first_pkt.ts_sec = ts.tv_sec;
 755                 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 756                 pkc1->pkblk_start = (char *)pbd1;
 757                 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 758                 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 759                 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 760                 pbd1->version = pkc1->version;
 761                 pkc1->prev = pkc1->nxt_offset;
 762                 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 763                 prb_thaw_queue(pkc1);
 764                 _prb_refresh_rx_retire_blk_timer(pkc1);
 765
 766                 smp_wmb();
 767
 768                 return;
 769         }
 770
 771         WARN(1, "ERROR block:%p is NOT FREE status:%d kactive_blk_num:%d\n",
 772                 pbd1, BLOCK_STATUS(pbd1), pkc1->kactive_blk_num);
 773         dump_stack();
 774         BUG();
 775 }
 776
 777 /*
 778  * Queue freeze logic:
 779  * 1) Assume tp_block_nr = 8 blocks.
 780  * 2) At time 't0', user opens Rx ring.
 781  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 782  * 4) user-space is either sleeping or processing block '0'.
 783  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 784  *    it will close block-7,loop around and try to fill block '0'.
 785  *    call-flow:
 786  *    __packet_lookup_frame_in_block
 787  *      prb_retire_current_block()
 788  *      prb_dispatch_next_block()
 789  *        |->(BLOCK_STATUS == USER) evaluates to true
 790  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 791  * 6) Now there are two cases:
 792  *    6.1) Link goes idle right after the queue is frozen.
 793  *         But remember, the last open_block() refreshed the timer.
 794  *         When this timer expires,it will refresh itself so that we can
 795  *         re-open block-0 in near future.
 796  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 797  *         case and __packet_lookup_frame_in_block will check if block-0
 798  *         is free and can now be re-used.
 799  */
 800 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 801                                   struct packet_sock *po)
 802 {
 803         pkc->reset_pending_on_curr_blk = 1;
 804         po->stats.stats3.tp_freeze_q_cnt++;
 805 }
 806
 807 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 808
 809 /*
 810  * If the next block is free then we will dispatch it
 811  * and return a good offset.
 812  * Else, we will freeze the queue.
 813  * So, caller must check the return value.
 814  */
 815 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 816                 struct packet_sock *po)
 817 {
 818         struct tpacket_block_desc *pbd;
 819
 820         smp_rmb();
 821
 822         /* 1. Get current block num */
 823         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 824
 825         /* 2. If this block is currently in_use then freeze the queue */
 826         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 827                 prb_freeze_queue(pkc, po);
 828                 return NULL;
 829         }
 830
 831         /*
 832          * 3.
 833          * open this block and return the offset where the first packet
 834          * needs to get stored.
 835          */
 836         prb_open_block(pkc, pbd);
 837         return (void *)pkc->nxt_offset;
 838 }
 839
 840 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 841                 struct packet_sock *po, unsigned int status)
 842 {
 843         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 844
 845         /* retire/close the current block */
 846         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 847                 /*
 848                  * Plug the case where copy_bits() is in progress on
 849                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 850                  * have space to copy the pkt in the current block and
 851                  * called prb_retire_current_block()
 852                  *
 853                  * We don't need to worry about the TMO case because
 854                  * the timer-handler already handled this case.
 855                  */
 856                 if (!(status & TP_STATUS_BLK_TMO)) {
 857                         while (atomic_read(&pkc->blk_fill_in_prog)) {
 858                                 /* Waiting for skb_copy_bits to finish... */
 859                                 cpu_relax();
 860                         }
 861                 }
 862                 prb_close_block(pkc, pbd, po, status);
 863                 return;
 864         }
 865
 866         WARN(1, "ERROR-pbd[%d]:%p\n", pkc->kactive_blk_num, pbd);
 867         dump_stack();
 868         BUG();
 869 }
 870
 871 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
 872                                       struct tpacket_block_desc *pbd)
 873 {
 874         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 875 }
 876
 877 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 878 {
 879         return pkc->reset_pending_on_curr_blk;
 880 }
 881
 882 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 883 {
 884         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 885         atomic_dec(&pkc->blk_fill_in_prog);
 886 }
 887
 888 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 889                         struct tpacket3_hdr *ppd)
 890 {
 891         ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
 892 }
 893
 894 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 895                         struct tpacket3_hdr *ppd)
 896 {
 897         ppd->hv1.tp_rxhash = 0;
 898 }
 899
 900 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
 901                         struct tpacket3_hdr *ppd)
 902 {
 903         if (vlan_tx_tag_present(pkc->skb)) {
 904                 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
 905                 ppd->tp_status = TP_STATUS_VLAN_VALID;
 906         } else {
 907                 ppd->hv1.tp_vlan_tci = 0;
 908                 ppd->tp_status = TP_STATUS_AVAILABLE;
 909         }
 910 }
 911
 912 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
 913                         struct tpacket3_hdr *ppd)
 914 {
 915         prb_fill_vlan_info(pkc, ppd);
 916
 917         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
 918                 prb_fill_rxhash(pkc, ppd);
 919         else
 920                 prb_clear_rxhash(pkc, ppd);
 921 }
 922
 923 static void prb_fill_curr_block(char *curr,
 924                                 struct tpacket_kbdq_core *pkc,
 925                                 struct tpacket_block_desc *pbd,
 926                                 unsigned int len)
 927 {
 928         struct tpacket3_hdr *ppd;
 929
 930         ppd  = (struct tpacket3_hdr *)curr;
 931         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
 932         pkc->prev = curr;
 933         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
 934         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
 935         BLOCK_NUM_PKTS(pbd) += 1;
 936         atomic_inc(&pkc->blk_fill_in_prog);
 937         prb_run_all_ft_ops(pkc, ppd);
 938 }
 939
 940 /* Assumes caller has the sk->rx_queue.lock */
 941 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
 942                                             struct sk_buff *skb,
 943                                                 int status,
 944                                             unsigned int len
 945                                             )
 946 {
 947         struct tpacket_kbdq_core *pkc;
 948         struct tpacket_block_desc *pbd;
 949         char *curr, *end;
 950
 951         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 952         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 953
 954         /* Queue is frozen when user space is lagging behind */
 955         if (prb_queue_frozen(pkc)) {
 956                 /*
 957                  * Check if that last block which caused the queue to freeze,
 958                  * is still in_use by user-space.
 959                  */
 960                 if (prb_curr_blk_in_use(pkc, pbd)) {
 961                         /* Can't record this packet */
 962                         return NULL;
 963                 } else {
 964                         /*
 965                          * Ok, the block was released by user-space.
 966                          * Now let's open that block.
 967                          * opening a block also thaws the queue.
 968                          * Thawing is a side effect.
 969                          */
 970                         prb_open_block(pkc, pbd);
 971                 }
 972         }
 973
 974         smp_mb();
 975         curr = pkc->nxt_offset;
 976         pkc->skb = skb;
 977         end = (char *)pbd + pkc->kblk_size;
 978
 979         /* first try the current block */
 980         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
 981                 prb_fill_curr_block(curr, pkc, pbd, len);
 982                 return (void *)curr;
 983         }
 984
 985         /* Ok, close the current block */
 986         prb_retire_current_block(pkc, po, 0);
 987
 988         /* Now, try to dispatch the next block */
 989         curr = (char *)prb_dispatch_next_block(pkc, po);
 990         if (curr) {
 991                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 992                 prb_fill_curr_block(curr, pkc, pbd, len);
 993                 return (void *)curr;
 994         }
 995
 996         /*
 997          * No free blocks are available.user_space hasn't caught up yet.
 998          * Queue was just frozen and now this packet will get dropped.
 999          */
1000         return NULL;
1001 }
1002
1003 static void *packet_current_rx_frame(struct packet_sock *po,
1004                                             struct sk_buff *skb,
1005                                             int status, unsigned int len)
1006 {
1007         char *curr = NULL;
1008         switch (po->tp_version) {
1009         case TPACKET_V1:
1010         case TPACKET_V2:
1011                 curr = packet_lookup_frame(po, &po->rx_ring,
1012                                         po->rx_ring.head, status);
1013                 return curr;
1014         case TPACKET_V3:
1015                 return __packet_lookup_frame_in_block(po, skb, status, len);
1016         default:
1017                 WARN(1, "TPACKET version not supported\n");
1018                 BUG();
1019                 return NULL;
1020         }
1021 }
1022
1023 static void *prb_lookup_block(struct packet_sock *po,
1024                                      struct packet_ring_buffer *rb,
1025                                      unsigned int idx,
1026                                      int status)
1027 {
1028         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1029         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1030
1031         if (status != BLOCK_STATUS(pbd))
1032                 return NULL;
1033         return pbd;
1034 }
1035
1036 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1037 {
1038         unsigned int prev;
1039         if (rb->prb_bdqc.kactive_blk_num)
1040                 prev = rb->prb_bdqc.kactive_blk_num-1;
1041         else
1042                 prev = rb->prb_bdqc.knum_blocks-1;
1043         return prev;
1044 }
1045
1046 /* Assumes caller has held the rx_queue.lock */
1047 static void *__prb_previous_block(struct packet_sock *po,
1048                                          struct packet_ring_buffer *rb,
1049                                          int status)
1050 {
1051         unsigned int previous = prb_previous_blk_num(rb);
1052         return prb_lookup_block(po, rb, previous, status);
1053 }
1054
1055 static void *packet_previous_rx_frame(struct packet_sock *po,
1056                                              struct packet_ring_buffer *rb,
1057                                              int status)
1058 {
1059         if (po->tp_version <= TPACKET_V2)
1060                 return packet_previous_frame(po, rb, status);
1061
1062         return __prb_previous_block(po, rb, status);
1063 }
1064
1065 static void packet_increment_rx_head(struct packet_sock *po,
1066                                             struct packet_ring_buffer *rb)
1067 {
1068         switch (po->tp_version) {
1069         case TPACKET_V1:
1070         case TPACKET_V2:
1071                 return packet_increment_head(rb);
1072         case TPACKET_V3:
1073         default:
1074                 WARN(1, "TPACKET version not supported.\n");
1075                 BUG();
1076                 return;
1077         }
1078 }
1079
1080 static void *packet_previous_frame(struct packet_sock *po,
1081                 struct packet_ring_buffer *rb,
1082                 int status)
1083 {
1084         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1085         return packet_lookup_frame(po, rb, previous, status);
1086 }
1087
1088 static void packet_increment_head(struct packet_ring_buffer *buff)
1089 {
1090         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1091 }
1092
1093 static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1094 {
1095         struct sock *sk = &po->sk;
1096         bool has_room;
1097
1098         if (po->prot_hook.func != tpacket_rcv)
1099                 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1100                         <= sk->sk_rcvbuf;
1101
1102         spin_lock(&sk->sk_receive_queue.lock);
1103         if (po->tp_version == TPACKET_V3)
1104                 has_room = prb_lookup_block(po, &po->rx_ring,
1105                                             po->rx_ring.prb_bdqc.kactive_blk_num,
1106                                             TP_STATUS_KERNEL);
1107         else
1108                 has_room = packet_lookup_frame(po, &po->rx_ring,
1109                                                po->rx_ring.head,
1110                                                TP_STATUS_KERNEL);
1111         spin_unlock(&sk->sk_receive_queue.lock);
1112
1113         return has_room;
1114 }
1115
1116 static void packet_sock_destruct(struct sock *sk)
1117 {
1118         skb_queue_purge(&sk->sk_error_queue);
1119
1120         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1121         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1122
1123         if (!sock_flag(sk, SOCK_DEAD)) {
1124                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1125                 return;
1126         }
1127
1128         sk_refcnt_debug_dec(sk);
1129 }
1130
1131 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1132 {
1133         int x = atomic_read(&f->rr_cur) + 1;
1134
1135         if (x >= num)
1136                 x = 0;
1137
1138         return x;
1139 }
1140
1141 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1142                                       struct sk_buff *skb,
1143                                       unsigned int num)
1144 {
1145         return (((u64)skb->rxhash) * num) >> 32;
1146 }
1147
1148 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1149                                     struct sk_buff *skb,
1150                                     unsigned int num)
1151 {
1152         int cur, old;
1153
1154         cur = atomic_read(&f->rr_cur);
1155         while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1156                                      fanout_rr_next(f, num))) != cur)
1157                 cur = old;
1158         return cur;
1159 }
1160
1161 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1162                                      struct sk_buff *skb,
1163                                      unsigned int num)
1164 {
1165         return smp_processor_id() % num;
1166 }
1167
1168 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1169                                           struct sk_buff *skb,
1170                                           unsigned int idx, unsigned int skip,
1171                                           unsigned int num)
1172 {
1173         unsigned int i, j;
1174
1175         i = j = min_t(int, f->next[idx], num - 1);
1176         do {
1177                 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1178                         if (i != j)
1179                                 f->next[idx] = i;
1180                         return i;
1181                 }
1182                 if (++i == num)
1183                         i = 0;
1184         } while (i != j);
1185
1186         return idx;
1187 }
1188
1189 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1190 {
1191         return f->flags & (flag >> 8);
1192 }
1193
1194 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1195                              struct packet_type *pt, struct net_device *orig_dev)
1196 {
1197         struct packet_fanout *f = pt->af_packet_priv;
1198         unsigned int num = f->num_members;
1199         struct packet_sock *po;
1200         unsigned int idx;
1201
1202         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1203             !num) {
1204                 kfree_skb(skb);
1205                 return 0;
1206         }
1207
1208         switch (f->type) {
1209         case PACKET_FANOUT_HASH:
1210         default:
1211                 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1212                         skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1213                         if (!skb)
1214                                 return 0;
1215                 }
1216                 skb_get_rxhash(skb);
1217                 idx = fanout_demux_hash(f, skb, num);
1218                 break;
1219         case PACKET_FANOUT_LB:
1220                 idx = fanout_demux_lb(f, skb, num);
1221                 break;
1222         case PACKET_FANOUT_CPU:
1223                 idx = fanout_demux_cpu(f, skb, num);
1224                 break;
1225         case PACKET_FANOUT_ROLLOVER:
1226                 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1227                 break;
1228         }
1229
1230         po = pkt_sk(f->arr[idx]);
1231         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1232             unlikely(!packet_rcv_has_room(po, skb))) {
1233                 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1234                 po = pkt_sk(f->arr[idx]);
1235         }
1236
1237         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1238 }
1239
1240 DEFINE_MUTEX(fanout_mutex);
1241 EXPORT_SYMBOL_GPL(fanout_mutex);
1242 static LIST_HEAD(fanout_list);
1243
1244 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1245 {
1246         struct packet_fanout *f = po->fanout;
1247
1248         spin_lock(&f->lock);
1249         f->arr[f->num_members] = sk;
1250         smp_wmb();
1251         f->num_members++;
1252         spin_unlock(&f->lock);
1253 }
1254
1255 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1256 {
1257         struct packet_fanout *f = po->fanout;
1258         int i;
1259
1260         spin_lock(&f->lock);
1261         for (i = 0; i < f->num_members; i++) {
1262                 if (f->arr[i] == sk)
1263                         break;
1264         }
1265         BUG_ON(i >= f->num_members);
1266         f->arr[i] = f->arr[f->num_members - 1];
1267         f->num_members--;
1268         spin_unlock(&f->lock);
1269 }
1270
1271 static bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1272 {
1273         if (ptype->af_packet_priv == (void*)((struct packet_sock *)sk)->fanout)
1274                 return true;
1275
1276         return false;
1277 }
1278
1279 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1280 {
1281         struct packet_sock *po = pkt_sk(sk);
1282         struct packet_fanout *f, *match;
1283         u8 type = type_flags & 0xff;
1284         u8 flags = type_flags >> 8;
1285         int err;
1286
1287         switch (type) {
1288         case PACKET_FANOUT_ROLLOVER:
1289                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1290                         return -EINVAL;
1291         case PACKET_FANOUT_HASH:
1292         case PACKET_FANOUT_LB:
1293         case PACKET_FANOUT_CPU:
1294                 break;
1295         default:
1296                 return -EINVAL;
1297         }
1298
1299         if (!po->running)
1300                 return -EINVAL;
1301
1302         if (po->fanout)
1303                 return -EALREADY;
1304
1305         mutex_lock(&fanout_mutex);
1306         match = NULL;
1307         list_for_each_entry(f, &fanout_list, list) {
1308                 if (f->id == id &&
1309                     read_pnet(&f->net) == sock_net(sk)) {
1310                         match = f;
1311                         break;
1312                 }
1313         }
1314         err = -EINVAL;
1315         if (match && match->flags != flags)
1316                 goto out;
1317         if (!match) {
1318                 err = -ENOMEM;
1319                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1320                 if (!match)
1321                         goto out;
1322                 write_pnet(&match->net, sock_net(sk));
1323                 match->id = id;
1324                 match->type = type;
1325                 match->flags = flags;
1326                 atomic_set(&match->rr_cur, 0);
1327                 INIT_LIST_HEAD(&match->list);
1328                 spin_lock_init(&match->lock);
1329                 atomic_set(&match->sk_ref, 0);
1330                 match->prot_hook.type = po->prot_hook.type;
1331                 match->prot_hook.dev = po->prot_hook.dev;
1332                 match->prot_hook.func = packet_rcv_fanout;
1333                 match->prot_hook.af_packet_priv = match;
1334                 match->prot_hook.id_match = match_fanout_group;
1335                 dev_add_pack(&match->prot_hook);
1336                 list_add(&match->list, &fanout_list);
1337         }
1338         err = -EINVAL;
1339         if (match->type == type &&
1340             match->prot_hook.type == po->prot_hook.type &&
1341             match->prot_hook.dev == po->prot_hook.dev) {
1342                 err = -ENOSPC;
1343                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1344                         __dev_remove_pack(&po->prot_hook);
1345                         po->fanout = match;
1346                         atomic_inc(&match->sk_ref);
1347                         __fanout_link(sk, po);
1348                         err = 0;
1349                 }
1350         }
1351 out:
1352         mutex_unlock(&fanout_mutex);
1353         return err;
1354 }
1355
1356 static void fanout_release(struct sock *sk)
1357 {
1358         struct packet_sock *po = pkt_sk(sk);
1359         struct packet_fanout *f;
1360
1361         f = po->fanout;
1362         if (!f)
1363                 return;
1364
1365         mutex_lock(&fanout_mutex);
1366         po->fanout = NULL;
1367
1368         if (atomic_dec_and_test(&f->sk_ref)) {
1369                 list_del(&f->list);
1370                 dev_remove_pack(&f->prot_hook);
1371                 kfree(f);
1372         }
1373         mutex_unlock(&fanout_mutex);
1374 }
1375
1376 static const struct proto_ops packet_ops;
1377
1378 static const struct proto_ops packet_ops_spkt;
1379
1380 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1381                            struct packet_type *pt, struct net_device *orig_dev)
1382 {
1383         struct sock *sk;
1384         struct sockaddr_pkt *spkt;
1385
1386         /*
1387          *      When we registered the protocol we saved the socket in the data
1388          *      field for just this event.
1389          */
1390
1391         sk = pt->af_packet_priv;
1392
1393         /*
1394          *      Yank back the headers [hope the device set this
1395          *      right or kerboom...]
1396          *
1397          *      Incoming packets have ll header pulled,
1398          *      push it back.
1399          *
1400          *      For outgoing ones skb->data == skb_mac_header(skb)
1401          *      so that this procedure is noop.
1402          */
1403
1404         if (skb->pkt_type == PACKET_LOOPBACK)
1405                 goto out;
1406
1407         if (!net_eq(dev_net(dev), sock_net(sk)))
1408                 goto out;
1409
1410         skb = skb_share_check(skb, GFP_ATOMIC);
1411         if (skb == NULL)
1412                 goto oom;
1413
1414         /* drop any routing info */
1415         skb_dst_drop(skb);
1416
1417         /* drop conntrack reference */
1418         nf_reset(skb);
1419
1420         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1421
1422         skb_push(skb, skb->data - skb_mac_header(skb));
1423
1424         /*
1425          *      The SOCK_PACKET socket receives _all_ frames.
1426          */
1427
1428         spkt->spkt_family = dev->type;
1429         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1430         spkt->spkt_protocol = skb->protocol;
1431
1432         /*
1433          *      Charge the memory to the socket. This is done specifically
1434          *      to prevent sockets using all the memory up.
1435          */
1436
1437         if (sock_queue_rcv_skb(sk, skb) == 0)
1438                 return 0;
1439
1440 out:
1441         kfree_skb(skb);
1442 oom:
1443         return 0;
1444 }
1445
1446
1447 /*
1448  *      Output a raw packet to a device layer. This bypasses all the other
1449  *      protocol layers and you must therefore supply it with a complete frame
1450  */
1451
1452 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1453                                struct msghdr *msg, size_t len)
1454 {
1455         struct sock *sk = sock->sk;
1456         struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1457         struct sk_buff *skb = NULL;
1458         struct net_device *dev;
1459         __be16 proto = 0;
1460         int err;
1461         int extra_len = 0;
1462
1463         /*
1464          *      Get and verify the address.
1465          */
1466
1467         if (saddr) {
1468                 if (msg->msg_namelen < sizeof(struct sockaddr))
1469                         return -EINVAL;
1470                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1471                         proto = saddr->spkt_protocol;
1472         } else
1473                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1474
1475         /*
1476          *      Find the device first to size check it
1477          */
1478
1479         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1480 retry:
1481         rcu_read_lock();
1482         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1483         err = -ENODEV;
1484         if (dev == NULL)
1485                 goto out_unlock;
1486
1487         err = -ENETDOWN;
1488         if (!(dev->flags & IFF_UP))
1489                 goto out_unlock;
1490
1491         /*
1492          * You may not queue a frame bigger than the mtu. This is the lowest level
1493          * raw protocol and you must do your own fragmentation at this level.
1494          */
1495
1496         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1497                 if (!netif_supports_nofcs(dev)) {
1498                         err = -EPROTONOSUPPORT;
1499                         goto out_unlock;
1500                 }
1501                 extra_len = 4; /* We're doing our own CRC */
1502         }
1503
1504         err = -EMSGSIZE;
1505         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1506                 goto out_unlock;
1507
1508         if (!skb) {
1509                 size_t reserved = LL_RESERVED_SPACE(dev);
1510                 int tlen = dev->needed_tailroom;
1511                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1512
1513                 rcu_read_unlock();
1514                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1515                 if (skb == NULL)
1516                         return -ENOBUFS;
1517                 /* FIXME: Save some space for broken drivers that write a hard
1518                  * header at transmission time by themselves. PPP is the notable
1519                  * one here. This should really be fixed at the driver level.
1520                  */
1521                 skb_reserve(skb, reserved);
1522                 skb_reset_network_header(skb);
1523
1524                 /* Try to align data part correctly */
1525                 if (hhlen) {
1526                         skb->data -= hhlen;
1527                         skb->tail -= hhlen;
1528                         if (len < hhlen)
1529                                 skb_reset_network_header(skb);
1530                 }
1531                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1532                 if (err)
1533                         goto out_free;
1534                 goto retry;
1535         }
1536
1537         if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1538                 /* Earlier code assumed this would be a VLAN pkt,
1539                  * double-check this now that we have the actual
1540                  * packet in hand.
1541                  */
1542                 struct ethhdr *ehdr;
1543                 skb_reset_mac_header(skb);
1544                 ehdr = eth_hdr(skb);
1545                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1546                         err = -EMSGSIZE;
1547                         goto out_unlock;
1548                 }
1549         }
1550
1551         skb->protocol = proto;
1552         skb->dev = dev;
1553         skb->priority = sk->sk_priority;
1554         skb->mark = sk->sk_mark;
1555
1556         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1557
1558         if (unlikely(extra_len == 4))
1559                 skb->no_fcs = 1;
1560
1561         skb_probe_transport_header(skb, 0);
1562
1563         dev_queue_xmit(skb);
1564         rcu_read_unlock();
1565         return len;
1566
1567 out_unlock:
1568         rcu_read_unlock();
1569 out_free:
1570         kfree_skb(skb);
1571         return err;
1572 }
1573
1574 static unsigned int run_filter(const struct sk_buff *skb,
1575                                       const struct sock *sk,
1576                                       unsigned int res)
1577 {
1578         struct sk_filter *filter;
1579
1580         rcu_read_lock();
1581         filter = rcu_dereference(sk->sk_filter);
1582         if (filter != NULL)
1583                 res = SK_RUN_FILTER(filter, skb);
1584         rcu_read_unlock();
1585
1586         return res;
1587 }
1588
1589 /*
1590  * This function makes lazy skb cloning in hope that most of packets
1591  * are discarded by BPF.
1592  *
1593  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1594  * and skb->cb are mangled. It works because (and until) packets
1595  * falling here are owned by current CPU. Output packets are cloned
1596  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1597  * sequencially, so that if we return skb to original state on exit,
1598  * we will not harm anyone.
1599  */
1600
1601 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1602                       struct packet_type *pt, struct net_device *orig_dev)
1603 {
1604         struct sock *sk;
1605         struct sockaddr_ll *sll;
1606         struct packet_sock *po;
1607         u8 *skb_head = skb->data;
1608         int skb_len = skb->len;
1609         unsigned int snaplen, res;
1610
1611         if (skb->pkt_type == PACKET_LOOPBACK)
1612                 goto drop;
1613
1614         sk = pt->af_packet_priv;
1615         po = pkt_sk(sk);
1616
1617         if (!net_eq(dev_net(dev), sock_net(sk)))
1618                 goto drop;
1619
1620         skb->dev = dev;
1621
1622         if (dev->header_ops) {
1623                 /* The device has an explicit notion of ll header,
1624                  * exported to higher levels.
1625                  *
1626                  * Otherwise, the device hides details of its frame
1627                  * structure, so that corresponding packet head is
1628                  * never delivered to user.
1629                  */
1630                 if (sk->sk_type != SOCK_DGRAM)
1631                         skb_push(skb, skb->data - skb_mac_header(skb));
1632                 else if (skb->pkt_type == PACKET_OUTGOING) {
1633                         /* Special case: outgoing packets have ll header at head */
1634                         skb_pull(skb, skb_network_offset(skb));
1635                 }
1636         }
1637
1638         snaplen = skb->len;
1639
1640         res = run_filter(skb, sk, snaplen);
1641         if (!res)
1642                 goto drop_n_restore;
1643         if (snaplen > res)
1644                 snaplen = res;
1645
1646         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1647                 goto drop_n_acct;
1648
1649         if (skb_shared(skb)) {
1650                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1651                 if (nskb == NULL)
1652                         goto drop_n_acct;
1653
1654                 if (skb_head != skb->data) {
1655                         skb->data = skb_head;
1656                         skb->len = skb_len;
1657                 }
1658                 consume_skb(skb);
1659                 skb = nskb;
1660         }
1661
1662         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1663                      sizeof(skb->cb));
1664
1665         sll = &PACKET_SKB_CB(skb)->sa.ll;
1666         sll->sll_family = AF_PACKET;
1667         sll->sll_hatype = dev->type;
1668         sll->sll_protocol = skb->protocol;
1669         sll->sll_pkttype = skb->pkt_type;
1670         if (unlikely(po->origdev))
1671                 sll->sll_ifindex = orig_dev->ifindex;
1672         else
1673                 sll->sll_ifindex = dev->ifindex;
1674
1675         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1676
1677         PACKET_SKB_CB(skb)->origlen = skb->len;
1678
1679         if (pskb_trim(skb, snaplen))
1680                 goto drop_n_acct;
1681
1682         skb_set_owner_r(skb, sk);
1683         skb->dev = NULL;
1684         skb_dst_drop(skb);
1685
1686         /* drop conntrack reference */
1687         nf_reset(skb);
1688
1689         spin_lock(&sk->sk_receive_queue.lock);
1690         po->stats.stats1.tp_packets++;
1691         skb->dropcount = atomic_read(&sk->sk_drops);
1692         __skb_queue_tail(&sk->sk_receive_queue, skb);
1693         spin_unlock(&sk->sk_receive_queue.lock);
1694         sk->sk_data_ready(sk, skb->len);
1695         return 0;
1696
1697 drop_n_acct:
1698         spin_lock(&sk->sk_receive_queue.lock);
1699         po->stats.stats1.tp_drops++;
1700         atomic_inc(&sk->sk_drops);
1701         spin_unlock(&sk->sk_receive_queue.lock);
1702
1703 drop_n_restore:
1704         if (skb_head != skb->data && skb_shared(skb)) {
1705                 skb->data = skb_head;
1706                 skb->len = skb_len;
1707         }
1708 drop:
1709         consume_skb(skb);
1710         return 0;
1711 }
1712
1713 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1714                        struct packet_type *pt, struct net_device *orig_dev)
1715 {
1716         struct sock *sk;
1717         struct packet_sock *po;
1718         struct sockaddr_ll *sll;
1719         union tpacket_uhdr h;
1720         u8 *skb_head = skb->data;
1721         int skb_len = skb->len;
1722         unsigned int snaplen, res;
1723         unsigned long status = TP_STATUS_USER;
1724         unsigned short macoff, netoff, hdrlen;
1725         struct sk_buff *copy_skb = NULL;
1726         struct timespec ts;
1727         __u32 ts_status;
1728
1729         if (skb->pkt_type == PACKET_LOOPBACK)
1730                 goto drop;
1731
1732         sk = pt->af_packet_priv;
1733         po = pkt_sk(sk);
1734
1735         if (!net_eq(dev_net(dev), sock_net(sk)))
1736                 goto drop;
1737
1738         if (dev->header_ops) {
1739                 if (sk->sk_type != SOCK_DGRAM)
1740                         skb_push(skb, skb->data - skb_mac_header(skb));
1741                 else if (skb->pkt_type == PACKET_OUTGOING) {
1742                         /* Special case: outgoing packets have ll header at head */
1743                         skb_pull(skb, skb_network_offset(skb));
1744                 }
1745         }
1746
1747         if (skb->ip_summed == CHECKSUM_PARTIAL)
1748                 status |= TP_STATUS_CSUMNOTREADY;
1749
1750         snaplen = skb->len;
1751
1752         res = run_filter(skb, sk, snaplen);
1753         if (!res)
1754                 goto drop_n_restore;
1755         if (snaplen > res)
1756                 snaplen = res;
1757
1758         if (sk->sk_type == SOCK_DGRAM) {
1759                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1760                                   po->tp_reserve;
1761         } else {
1762                 unsigned int maclen = skb_network_offset(skb);
1763                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1764                                        (maclen < 16 ? 16 : maclen)) +
1765                         po->tp_reserve;
1766                 macoff = netoff - maclen;
1767         }
1768         if (po->tp_version <= TPACKET_V2) {
1769                 if (macoff + snaplen > po->rx_ring.frame_size) {
1770                         if (po->copy_thresh &&
1771                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1772                                 if (skb_shared(skb)) {
1773                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1774                                 } else {
1775                                         copy_skb = skb_get(skb);
1776                                         skb_head = skb->data;
1777                                 }
1778                                 if (copy_skb)
1779                                         skb_set_owner_r(copy_skb, sk);
1780                         }
1781                         snaplen = po->rx_ring.frame_size - macoff;
1782                         if ((int)snaplen < 0)
1783                                 snaplen = 0;
1784                 }
1785         }
1786         spin_lock(&sk->sk_receive_queue.lock);
1787         h.raw = packet_current_rx_frame(po, skb,
1788                                         TP_STATUS_KERNEL, (macoff+snaplen));
1789         if (!h.raw)
1790                 goto ring_is_full;
1791         if (po->tp_version <= TPACKET_V2) {
1792                 packet_increment_rx_head(po, &po->rx_ring);
1793         /*
1794          * LOSING will be reported till you read the stats,
1795          * because it's COR - Clear On Read.
1796          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1797          * at packet level.
1798          */
1799                 if (po->stats.stats1.tp_drops)
1800                         status |= TP_STATUS_LOSING;
1801         }
1802         po->stats.stats1.tp_packets++;
1803         if (copy_skb) {
1804                 status |= TP_STATUS_COPY;
1805                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1806         }
1807         spin_unlock(&sk->sk_receive_queue.lock);
1808
1809         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1810
1811         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
1812                 getnstimeofday(&ts);
1813
1814         status |= ts_status;
1815
1816         switch (po->tp_version) {
1817         case TPACKET_V1:
1818                 h.h1->tp_len = skb->len;
1819                 h.h1->tp_snaplen = snaplen;
1820                 h.h1->tp_mac = macoff;
1821                 h.h1->tp_net = netoff;
1822                 h.h1->tp_sec = ts.tv_sec;
1823                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
1824                 hdrlen = sizeof(*h.h1);
1825                 break;
1826         case TPACKET_V2:
1827                 h.h2->tp_len = skb->len;
1828                 h.h2->tp_snaplen = snaplen;
1829                 h.h2->tp_mac = macoff;
1830                 h.h2->tp_net = netoff;
1831                 h.h2->tp_sec = ts.tv_sec;
1832                 h.h2->tp_nsec = ts.tv_nsec;
1833                 if (vlan_tx_tag_present(skb)) {
1834                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1835                         status |= TP_STATUS_VLAN_VALID;
1836                 } else {
1837                         h.h2->tp_vlan_tci = 0;
1838                 }
1839                 h.h2->tp_padding = 0;
1840                 hdrlen = sizeof(*h.h2);
1841                 break;
1842         case TPACKET_V3:
1843                 /* tp_nxt_offset,vlan are already populated above.
1844                  * So DONT clear those fields here
1845                  */
1846                 h.h3->tp_status |= status;
1847                 h.h3->tp_len = skb->len;
1848                 h.h3->tp_snaplen = snaplen;
1849                 h.h3->tp_mac = macoff;
1850                 h.h3->tp_net = netoff;
1851                 h.h3->tp_sec  = ts.tv_sec;
1852                 h.h3->tp_nsec = ts.tv_nsec;
1853                 hdrlen = sizeof(*h.h3);
1854                 break;
1855         default:
1856                 BUG();
1857         }
1858
1859         sll = h.raw + TPACKET_ALIGN(hdrlen);
1860         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1861         sll->sll_family = AF_PACKET;
1862         sll->sll_hatype = dev->type;
1863         sll->sll_protocol = skb->protocol;
1864         sll->sll_pkttype = skb->pkt_type;
1865         if (unlikely(po->origdev))
1866                 sll->sll_ifindex = orig_dev->ifindex;
1867         else
1868                 sll->sll_ifindex = dev->ifindex;
1869
1870         smp_mb();
1871 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1872         {
1873                 u8 *start, *end;
1874
1875                 if (po->tp_version <= TPACKET_V2) {
1876                         end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1877                                 + macoff + snaplen);
1878                         for (start = h.raw; start < end; start += PAGE_SIZE)
1879                                 flush_dcache_page(pgv_to_page(start));
1880                 }
1881                 smp_wmb();
1882         }
1883 #endif
1884         if (po->tp_version <= TPACKET_V2)
1885                 __packet_set_status(po, h.raw, status);
1886         else
1887                 prb_clear_blk_fill_status(&po->rx_ring);
1888
1889         sk->sk_data_ready(sk, 0);
1890
1891 drop_n_restore:
1892         if (skb_head != skb->data && skb_shared(skb)) {
1893                 skb->data = skb_head;
1894                 skb->len = skb_len;
1895         }
1896 drop:
1897         kfree_skb(skb);
1898         return 0;
1899
1900 ring_is_full:
1901         po->stats.stats1.tp_drops++;
1902         spin_unlock(&sk->sk_receive_queue.lock);
1903
1904         sk->sk_data_ready(sk, 0);
1905         kfree_skb(copy_skb);
1906         goto drop_n_restore;
1907 }
1908
1909 static void tpacket_destruct_skb(struct sk_buff *skb)
1910 {
1911         struct packet_sock *po = pkt_sk(skb->sk);
1912         void *ph;
1913
1914         if (likely(po->tx_ring.pg_vec)) {
1915                 __u32 ts;
1916
1917                 ph = skb_shinfo(skb)->destructor_arg;
1918                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1919                 atomic_dec(&po->tx_ring.pending);
1920
1921                 ts = __packet_set_timestamp(po, ph, skb);
1922                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
1923         }
1924
1925         sock_wfree(skb);
1926 }
1927
1928 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1929                 void *frame, struct net_device *dev, int size_max,
1930                 __be16 proto, unsigned char *addr, int hlen)
1931 {
1932         union tpacket_uhdr ph;
1933         int to_write, offset, len, tp_len, nr_frags, len_max;
1934         struct socket *sock = po->sk.sk_socket;
1935         struct page *page;
1936         void *data;
1937         int err;
1938
1939         ph.raw = frame;
1940
1941         skb->protocol = proto;
1942         skb->dev = dev;
1943         skb->priority = po->sk.sk_priority;
1944         skb->mark = po->sk.sk_mark;
1945         sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
1946         skb_shinfo(skb)->destructor_arg = ph.raw;
1947
1948         switch (po->tp_version) {
1949         case TPACKET_V2:
1950                 tp_len = ph.h2->tp_len;
1951                 break;
1952         default:
1953                 tp_len = ph.h1->tp_len;
1954                 break;
1955         }
1956         if (unlikely(tp_len > size_max)) {
1957                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1958                 return -EMSGSIZE;
1959         }
1960
1961         skb_reserve(skb, hlen);
1962         skb_reset_network_header(skb);
1963         skb_probe_transport_header(skb, 0);
1964
1965         if (po->tp_tx_has_off) {
1966                 int off_min, off_max, off;
1967                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
1968                 off_max = po->tx_ring.frame_size - tp_len;
1969                 if (sock->type == SOCK_DGRAM) {
1970                         switch (po->tp_version) {
1971                         case TPACKET_V2:
1972                                 off = ph.h2->tp_net;
1973                                 break;
1974                         default:
1975                                 off = ph.h1->tp_net;
1976                                 break;
1977                         }
1978                 } else {
1979                         switch (po->tp_version) {
1980                         case TPACKET_V2:
1981                                 off = ph.h2->tp_mac;
1982                                 break;
1983                         default:
1984                                 off = ph.h1->tp_mac;
1985                                 break;
1986                         }
1987                 }
1988                 if (unlikely((off < off_min) || (off_max < off)))
1989                         return -EINVAL;
1990                 data = ph.raw + off;
1991         } else {
1992                 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1993         }
1994         to_write = tp_len;
1995
1996         if (sock->type == SOCK_DGRAM) {
1997                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
1998                                 NULL, tp_len);
1999                 if (unlikely(err < 0))
2000                         return -EINVAL;
2001         } else if (dev->hard_header_len) {
2002                 /* net device doesn't like empty head */
2003                 if (unlikely(tp_len <= dev->hard_header_len)) {
2004                         pr_err("packet size is too short (%d < %d)\n",
2005                                tp_len, dev->hard_header_len);
2006                         return -EINVAL;
2007                 }
2008
2009                 skb_push(skb, dev->hard_header_len);
2010                 err = skb_store_bits(skb, 0, data,
2011                                 dev->hard_header_len);
2012                 if (unlikely(err))
2013                         return err;
2014
2015                 data += dev->hard_header_len;
2016                 to_write -= dev->hard_header_len;
2017         }
2018
2019         offset = offset_in_page(data);
2020         len_max = PAGE_SIZE - offset;
2021         len = ((to_write > len_max) ? len_max : to_write);
2022
2023         skb->data_len = to_write;
2024         skb->len += to_write;
2025         skb->truesize += to_write;
2026         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2027
2028         while (likely(to_write)) {
2029                 nr_frags = skb_shinfo(skb)->nr_frags;
2030
2031                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2032                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2033                                MAX_SKB_FRAGS);
2034                         return -EFAULT;
2035                 }
2036
2037                 page = pgv_to_page(data);
2038                 data += len;
2039                 flush_dcache_page(page);
2040                 get_page(page);
2041                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2042                 to_write -= len;
2043                 offset = 0;
2044                 len_max = PAGE_SIZE;
2045                 len = ((to_write > len_max) ? len_max : to_write);
2046         }
2047
2048         return tp_len;
2049 }
2050
2051 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2052 {
2053         struct sk_buff *skb;
2054         struct net_device *dev;
2055         __be16 proto;
2056         bool need_rls_dev = false;
2057         int err, reserve = 0;
2058         void *ph;
2059         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2060         int tp_len, size_max;
2061         unsigned char *addr;
2062         int len_sum = 0;
2063         int status = TP_STATUS_AVAILABLE;
2064         int hlen, tlen;
2065
2066         mutex_lock(&po->pg_vec_lock);
2067
2068         if (saddr == NULL) {
2069                 dev = po->prot_hook.dev;
2070                 proto   = po->num;
2071                 addr    = NULL;
2072         } else {
2073                 err = -EINVAL;
2074                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2075                         goto out;
2076                 if (msg->msg_namelen < (saddr->sll_halen
2077                                         + offsetof(struct sockaddr_ll,
2078                                                 sll_addr)))
2079                         goto out;
2080                 proto   = saddr->sll_protocol;
2081                 addr    = saddr->sll_addr;
2082                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2083                 need_rls_dev = true;
2084         }
2085
2086         err = -ENXIO;
2087         if (unlikely(dev == NULL))
2088                 goto out;
2089
2090         reserve = dev->hard_header_len;
2091
2092         err = -ENETDOWN;
2093         if (unlikely(!(dev->flags & IFF_UP)))
2094                 goto out_put;
2095
2096         size_max = po->tx_ring.frame_size
2097                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2098
2099         if (size_max > dev->mtu + reserve)
2100                 size_max = dev->mtu + reserve;
2101
2102         do {
2103                 ph = packet_current_frame(po, &po->tx_ring,
2104                                 TP_STATUS_SEND_REQUEST);
2105
2106                 if (unlikely(ph == NULL)) {
2107                         schedule();
2108                         continue;
2109                 }
2110
2111                 status = TP_STATUS_SEND_REQUEST;
2112                 hlen = LL_RESERVED_SPACE(dev);
2113                 tlen = dev->needed_tailroom;
2114                 skb = sock_alloc_send_skb(&po->sk,
2115                                 hlen + tlen + sizeof(struct sockaddr_ll),
2116                                 0, &err);
2117
2118                 if (unlikely(skb == NULL))
2119                         goto out_status;
2120
2121                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2122                                 addr, hlen);
2123
2124                 if (unlikely(tp_len < 0)) {
2125                         if (po->tp_loss) {
2126                                 __packet_set_status(po, ph,
2127                                                 TP_STATUS_AVAILABLE);
2128                                 packet_increment_head(&po->tx_ring);
2129                                 kfree_skb(skb);
2130                                 continue;
2131                         } else {
2132                                 status = TP_STATUS_WRONG_FORMAT;
2133                                 err = tp_len;
2134                                 goto out_status;
2135                         }
2136                 }
2137
2138                 skb->destructor = tpacket_destruct_skb;
2139                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2140                 atomic_inc(&po->tx_ring.pending);
2141
2142                 status = TP_STATUS_SEND_REQUEST;
2143                 err = dev_queue_xmit(skb);
2144                 if (unlikely(err > 0)) {
2145                         err = net_xmit_errno(err);
2146                         if (err && __packet_get_status(po, ph) ==
2147                                    TP_STATUS_AVAILABLE) {
2148                                 /* skb was destructed already */
2149                                 skb = NULL;
2150                                 goto out_status;
2151                         }
2152                         /*
2153                          * skb was dropped but not destructed yet;
2154                          * let's treat it like congestion or err < 0
2155                          */
2156                         err = 0;
2157                 }
2158                 packet_increment_head(&po->tx_ring);
2159                 len_sum += tp_len;
2160         } while (likely((ph != NULL) ||
2161                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2162                          (atomic_read(&po->tx_ring.pending))))
2163                 );
2164
2165         err = len_sum;
2166         goto out_put;
2167
2168 out_status:
2169         __packet_set_status(po, ph, status);
2170         kfree_skb(skb);
2171 out_put:
2172         if (need_rls_dev)
2173                 dev_put(dev);
2174 out:
2175         mutex_unlock(&po->pg_vec_lock);
2176         return err;
2177 }
2178
2179 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2180                                         size_t reserve, size_t len,
2181                                         size_t linear, int noblock,
2182                                         int *err)
2183 {
2184         struct sk_buff *skb;
2185
2186         /* Under a page?  Don't bother with paged skb. */
2187         if (prepad + len < PAGE_SIZE || !linear)
2188                 linear = len;
2189
2190         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2191                                    err);
2192         if (!skb)
2193                 return NULL;
2194
2195         skb_reserve(skb, reserve);
2196         skb_put(skb, linear);
2197         skb->data_len = len - linear;
2198         skb->len += len - linear;
2199
2200         return skb;
2201 }
2202
2203 static int packet_snd(struct socket *sock,
2204                           struct msghdr *msg, size_t len)
2205 {
2206         struct sock *sk = sock->sk;
2207         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2208         struct sk_buff *skb;
2209         struct net_device *dev;
2210         __be16 proto;
2211         bool need_rls_dev = false;
2212         unsigned char *addr;
2213         int err, reserve = 0;
2214         struct virtio_net_hdr vnet_hdr = { 0 };
2215         int offset = 0;
2216         int vnet_hdr_len;
2217         struct packet_sock *po = pkt_sk(sk);
2218         unsigned short gso_type = 0;
2219         int hlen, tlen;
2220         int extra_len = 0;
2221
2222         /*
2223          *      Get and verify the address.
2224          */
2225
2226         if (saddr == NULL) {
2227                 dev = po->prot_hook.dev;
2228                 proto   = po->num;
2229                 addr    = NULL;
2230         } else {
2231                 err = -EINVAL;
2232                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2233                         goto out;
2234                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2235                         goto out;
2236                 proto   = saddr->sll_protocol;
2237                 addr    = saddr->sll_addr;
2238                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2239                 need_rls_dev = true;
2240         }
2241
2242         err = -ENXIO;
2243         if (dev == NULL)
2244                 goto out_unlock;
2245         if (sock->type == SOCK_RAW)
2246                 reserve = dev->hard_header_len;
2247
2248         err = -ENETDOWN;
2249         if (!(dev->flags & IFF_UP))
2250                 goto out_unlock;
2251
2252         if (po->has_vnet_hdr) {
2253                 vnet_hdr_len = sizeof(vnet_hdr);
2254
2255                 err = -EINVAL;
2256                 if (len < vnet_hdr_len)
2257                         goto out_unlock;
2258
2259                 len -= vnet_hdr_len;
2260
2261                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2262                                        vnet_hdr_len);
2263                 if (err < 0)
2264                         goto out_unlock;
2265
2266                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2267                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2268                       vnet_hdr.hdr_len))
2269                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
2270                                                  vnet_hdr.csum_offset + 2;
2271
2272                 err = -EINVAL;
2273                 if (vnet_hdr.hdr_len > len)
2274                         goto out_unlock;
2275
2276                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2277                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2278                         case VIRTIO_NET_HDR_GSO_TCPV4:
2279                                 gso_type = SKB_GSO_TCPV4;
2280                                 break;
2281                         case VIRTIO_NET_HDR_GSO_TCPV6:
2282                                 gso_type = SKB_GSO_TCPV6;
2283                                 break;
2284                         case VIRTIO_NET_HDR_GSO_UDP:
2285                                 gso_type = SKB_GSO_UDP;
2286                                 break;
2287                         default:
2288                                 goto out_unlock;
2289                         }
2290
2291                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2292                                 gso_type |= SKB_GSO_TCP_ECN;
2293
2294                         if (vnet_hdr.gso_size == 0)
2295                                 goto out_unlock;
2296
2297                 }
2298         }
2299
2300         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2301                 if (!netif_supports_nofcs(dev)) {
2302                         err = -EPROTONOSUPPORT;
2303                         goto out_unlock;
2304                 }
2305                 extra_len = 4; /* We're doing our own CRC */
2306         }
2307
2308         err = -EMSGSIZE;
2309         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2310                 goto out_unlock;
2311
2312         err = -ENOBUFS;
2313         hlen = LL_RESERVED_SPACE(dev);
2314         tlen = dev->needed_tailroom;
2315         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2316                                msg->msg_flags & MSG_DONTWAIT, &err);
2317         if (skb == NULL)
2318                 goto out_unlock;
2319
2320         skb_set_network_header(skb, reserve);
2321
2322         err = -EINVAL;
2323         if (sock->type == SOCK_DGRAM &&
2324             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2325                 goto out_free;
2326
2327         /* Returns -EFAULT on error */
2328         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2329         if (err)
2330                 goto out_free;
2331
2332         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2333
2334         if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2335                 /* Earlier code assumed this would be a VLAN pkt,
2336                  * double-check this now that we have the actual
2337                  * packet in hand.
2338                  */
2339                 struct ethhdr *ehdr;
2340                 skb_reset_mac_header(skb);
2341                 ehdr = eth_hdr(skb);
2342                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2343                         err = -EMSGSIZE;
2344                         goto out_free;
2345                 }
2346         }
2347
2348         skb->protocol = proto;
2349         skb->dev = dev;
2350         skb->priority = sk->sk_priority;
2351         skb->mark = sk->sk_mark;
2352
2353         if (po->has_vnet_hdr) {
2354                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2355                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2356                                                   vnet_hdr.csum_offset)) {
2357                                 err = -EINVAL;
2358                                 goto out_free;
2359                         }
2360                 }
2361
2362                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2363                 skb_shinfo(skb)->gso_type = gso_type;
2364
2365                 /* Header must be checked, and gso_segs computed. */
2366                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2367                 skb_shinfo(skb)->gso_segs = 0;
2368
2369                 len += vnet_hdr_len;
2370         }
2371
2372         skb_probe_transport_header(skb, reserve);
2373
2374         if (unlikely(extra_len == 4))
2375                 skb->no_fcs = 1;
2376
2377         /*
2378          *      Now send it
2379          */
2380
2381         err = dev_queue_xmit(skb);
2382         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2383                 goto out_unlock;
2384
2385         if (need_rls_dev)
2386                 dev_put(dev);
2387
2388         return len;
2389
2390 out_free:
2391         kfree_skb(skb);
2392 out_unlock:
2393         if (dev && need_rls_dev)
2394                 dev_put(dev);
2395 out:
2396         return err;
2397 }
2398
2399 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2400                 struct msghdr *msg, size_t len)
2401 {
2402         struct sock *sk = sock->sk;
2403         struct packet_sock *po = pkt_sk(sk);
2404         if (po->tx_ring.pg_vec)
2405                 return tpacket_snd(po, msg);
2406         else
2407                 return packet_snd(sock, msg, len);
2408 }
2409
2410 /*
2411  *      Close a PACKET socket. This is fairly simple. We immediately go
2412  *      to 'closed' state and remove our protocol entry in the device list.
2413  */
2414
2415 static int packet_release(struct socket *sock)
2416 {
2417         struct sock *sk = sock->sk;
2418         struct packet_sock *po;
2419         struct net *net;
2420         union tpacket_req_u req_u;
2421
2422         if (!sk)
2423                 return 0;
2424
2425         net = sock_net(sk);
2426         po = pkt_sk(sk);
2427
2428         mutex_lock(&net->packet.sklist_lock);
2429         sk_del_node_init_rcu(sk);
2430         mutex_unlock(&net->packet.sklist_lock);
2431
2432         preempt_disable();
2433         sock_prot_inuse_add(net, sk->sk_prot, -1);
2434         preempt_enable();
2435
2436         spin_lock(&po->bind_lock);
2437         unregister_prot_hook(sk, false);
2438         if (po->prot_hook.dev) {
2439                 dev_put(po->prot_hook.dev);
2440                 po->prot_hook.dev = NULL;
2441         }
2442         spin_unlock(&po->bind_lock);
2443
2444         packet_flush_mclist(sk);
2445
2446         if (po->rx_ring.pg_vec) {
2447                 memset(&req_u, 0, sizeof(req_u));
2448                 packet_set_ring(sk, &req_u, 1, 0);
2449         }
2450
2451         if (po->tx_ring.pg_vec) {
2452                 memset(&req_u, 0, sizeof(req_u));
2453                 packet_set_ring(sk, &req_u, 1, 1);
2454         }
2455
2456         fanout_release(sk);
2457
2458         synchronize_net();
2459         /*
2460          *      Now the socket is dead. No more input will appear.
2461          */
2462         sock_orphan(sk);
2463         sock->sk = NULL;
2464
2465         /* Purge queues */
2466
2467         skb_queue_purge(&sk->sk_receive_queue);
2468         sk_refcnt_debug_release(sk);
2469
2470         sock_put(sk);
2471         return 0;
2472 }
2473
2474 /*
2475  *      Attach a packet hook.
2476  */
2477
2478 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2479 {
2480         struct packet_sock *po = pkt_sk(sk);
2481
2482         if (po->fanout) {
2483                 if (dev)
2484                         dev_put(dev);
2485
2486                 return -EINVAL;
2487         }
2488
2489         lock_sock(sk);
2490
2491         spin_lock(&po->bind_lock);
2492         unregister_prot_hook(sk, true);
2493         po->num = protocol;
2494         po->prot_hook.type = protocol;
2495         if (po->prot_hook.dev)
2496                 dev_put(po->prot_hook.dev);
2497         po->prot_hook.dev = dev;
2498
2499         po->ifindex = dev ? dev->ifindex : 0;
2500
2501         if (protocol == 0)
2502                 goto out_unlock;
2503
2504         if (!dev || (dev->flags & IFF_UP)) {
2505                 register_prot_hook(sk);
2506         } else {
2507                 sk->sk_err = ENETDOWN;
2508                 if (!sock_flag(sk, SOCK_DEAD))
2509                         sk->sk_error_report(sk);
2510         }
2511
2512 out_unlock:
2513         spin_unlock(&po->bind_lock);
2514         release_sock(sk);
2515         return 0;
2516 }
2517
2518 /*
2519  *      Bind a packet socket to a device
2520  */
2521
2522 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2523                             int addr_len)
2524 {
2525         struct sock *sk = sock->sk;
2526         char name[15];
2527         struct net_device *dev;
2528         int err = -ENODEV;
2529
2530         /*
2531          *      Check legality
2532          */
2533
2534         if (addr_len != sizeof(struct sockaddr))
2535                 return -EINVAL;
2536         strlcpy(name, uaddr->sa_data, sizeof(name));
2537
2538         dev = dev_get_by_name(sock_net(sk), name);
2539         if (dev)
2540                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2541         return err;
2542 }
2543
2544 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2545 {
2546         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2547         struct sock *sk = sock->sk;
2548         struct net_device *dev = NULL;
2549         int err;
2550
2551
2552         /*
2553          *      Check legality
2554          */
2555
2556         if (addr_len < sizeof(struct sockaddr_ll))
2557                 return -EINVAL;
2558         if (sll->sll_family != AF_PACKET)
2559                 return -EINVAL;
2560
2561         if (sll->sll_ifindex) {
2562                 err = -ENODEV;
2563                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2564                 if (dev == NULL)
2565                         goto out;
2566         }
2567         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2568
2569 out:
2570         return err;
2571 }
2572
2573 static struct proto packet_proto = {
2574         .name     = "PACKET",
2575         .owner    = THIS_MODULE,
2576         .obj_size = sizeof(struct packet_sock),
2577 };
2578
2579 /*
2580  *      Create a packet of type SOCK_PACKET.
2581  */
2582
2583 static int packet_create(struct net *net, struct socket *sock, int protocol,
2584                          int kern)
2585 {
2586         struct sock *sk;
2587         struct packet_sock *po;
2588         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2589         int err;
2590
2591         if (!ns_capable(net->user_ns, CAP_NET_RAW))
2592                 return -EPERM;
2593         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2594             sock->type != SOCK_PACKET)
2595                 return -ESOCKTNOSUPPORT;
2596
2597         sock->state = SS_UNCONNECTED;
2598
2599         err = -ENOBUFS;
2600         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2601         if (sk == NULL)
2602                 goto out;
2603
2604         sock->ops = &packet_ops;
2605         if (sock->type == SOCK_PACKET)
2606                 sock->ops = &packet_ops_spkt;
2607
2608         sock_init_data(sock, sk);
2609
2610         po = pkt_sk(sk);
2611         sk->sk_family = PF_PACKET;
2612         po->num = proto;
2613
2614         sk->sk_destruct = packet_sock_destruct;
2615         sk_refcnt_debug_inc(sk);
2616
2617         /*
2618          *      Attach a protocol block
2619          */
2620
2621         spin_lock_init(&po->bind_lock);
2622         mutex_init(&po->pg_vec_lock);
2623         po->prot_hook.func = packet_rcv;
2624
2625         if (sock->type == SOCK_PACKET)
2626                 po->prot_hook.func = packet_rcv_spkt;
2627
2628         po->prot_hook.af_packet_priv = sk;
2629
2630         if (proto) {
2631                 po->prot_hook.type = proto;
2632                 register_prot_hook(sk);
2633         }
2634
2635         mutex_lock(&net->packet.sklist_lock);
2636         sk_add_node_rcu(sk, &net->packet.sklist);
2637         mutex_unlock(&net->packet.sklist_lock);
2638
2639         preempt_disable();
2640         sock_prot_inuse_add(net, &packet_proto, 1);
2641         preempt_enable();
2642
2643         return 0;
2644 out:
2645         return err;
2646 }
2647
2648 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2649 {
2650         struct sock_exterr_skb *serr;
2651         struct sk_buff *skb, *skb2;
2652         int copied, err;
2653
2654         err = -EAGAIN;
2655         skb = skb_dequeue(&sk->sk_error_queue);
2656         if (skb == NULL)
2657                 goto out;
2658
2659         copied = skb->len;
2660         if (copied > len) {
2661                 msg->msg_flags |= MSG_TRUNC;
2662                 copied = len;
2663         }
2664         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2665         if (err)
2666                 goto out_free_skb;
2667
2668         sock_recv_timestamp(msg, sk, skb);
2669
2670         serr = SKB_EXT_ERR(skb);
2671         put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2672                  sizeof(serr->ee), &serr->ee);
2673
2674         msg->msg_flags |= MSG_ERRQUEUE;
2675         err = copied;
2676
2677         /* Reset and regenerate socket error */
2678         spin_lock_bh(&sk->sk_error_queue.lock);
2679         sk->sk_err = 0;
2680         if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2681                 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2682                 spin_unlock_bh(&sk->sk_error_queue.lock);
2683                 sk->sk_error_report(sk);
2684         } else
2685                 spin_unlock_bh(&sk->sk_error_queue.lock);
2686
2687 out_free_skb:
2688         kfree_skb(skb);
2689 out:
2690         return err;
2691 }
2692
2693 /*
2694  *      Pull a packet from our receive queue and hand it to the user.
2695  *      If necessary we block.
2696  */
2697
2698 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2699                           struct msghdr *msg, size_t len, int flags)
2700 {
2701         struct sock *sk = sock->sk;
2702         struct sk_buff *skb;
2703         int copied, err;
2704         struct sockaddr_ll *sll;
2705         int vnet_hdr_len = 0;
2706
2707         err = -EINVAL;
2708         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2709                 goto out;
2710
2711 #if 0
2712         /* What error should we return now? EUNATTACH? */
2713         if (pkt_sk(sk)->ifindex < 0)
2714                 return -ENODEV;
2715 #endif
2716
2717         if (flags & MSG_ERRQUEUE) {
2718                 err = packet_recv_error(sk, msg, len);
2719                 goto out;
2720         }
2721
2722         /*
2723          *      Call the generic datagram receiver. This handles all sorts
2724          *      of horrible races and re-entrancy so we can forget about it
2725          *      in the protocol layers.
2726          *
2727          *      Now it will return ENETDOWN, if device have just gone down,
2728          *      but then it will block.
2729          */
2730
2731         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2732
2733         /*
2734          *      An error occurred so return it. Because skb_recv_datagram()
2735          *      handles the blocking we don't see and worry about blocking
2736          *      retries.
2737          */
2738
2739         if (skb == NULL)
2740                 goto out;
2741
2742         if (pkt_sk(sk)->has_vnet_hdr) {
2743                 struct virtio_net_hdr vnet_hdr = { 0 };
2744
2745                 err = -EINVAL;
2746                 vnet_hdr_len = sizeof(vnet_hdr);
2747                 if (len < vnet_hdr_len)
2748                         goto out_free;
2749
2750                 len -= vnet_hdr_len;
2751
2752                 if (skb_is_gso(skb)) {
2753                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2754
2755                         /* This is a hint as to how much should be linear. */
2756                         vnet_hdr.hdr_len = skb_headlen(skb);
2757                         vnet_hdr.gso_size = sinfo->gso_size;
2758                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2759                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2760                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2761                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2762                         else if (sinfo->gso_type & SKB_GSO_UDP)
2763                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2764                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2765                                 goto out_free;
2766                         else
2767                                 BUG();
2768                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2769                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2770                 } else
2771                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2772
2773                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2774                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2775                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2776                         vnet_hdr.csum_offset = skb->csum_offset;
2777                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2778                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2779                 } /* else everything is zero */
2780
2781                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2782                                      vnet_hdr_len);
2783                 if (err < 0)
2784                         goto out_free;
2785         }
2786
2787         /*
2788          *      If the address length field is there to be filled in, we fill
2789          *      it in now.
2790          */
2791
2792         sll = &PACKET_SKB_CB(skb)->sa.ll;
2793         if (sock->type == SOCK_PACKET)
2794                 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2795         else
2796                 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2797
2798         /*
2799          *      You lose any data beyond the buffer you gave. If it worries a
2800          *      user program they can ask the device for its MTU anyway.
2801          */
2802
2803         copied = skb->len;
2804         if (copied > len) {
2805                 copied = len;
2806                 msg->msg_flags |= MSG_TRUNC;
2807         }
2808
2809         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2810         if (err)
2811                 goto out_free;
2812
2813         sock_recv_ts_and_drops(msg, sk, skb);
2814
2815         if (msg->msg_name)
2816                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2817                        msg->msg_namelen);
2818
2819         if (pkt_sk(sk)->auxdata) {
2820                 struct tpacket_auxdata aux;
2821
2822                 aux.tp_status = TP_STATUS_USER;
2823                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2824                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2825                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2826                 aux.tp_snaplen = skb->len;
2827                 aux.tp_mac = 0;
2828                 aux.tp_net = skb_network_offset(skb);
2829                 if (vlan_tx_tag_present(skb)) {
2830                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2831                         aux.tp_status |= TP_STATUS_VLAN_VALID;
2832                 } else {
2833                         aux.tp_vlan_tci = 0;
2834                 }
2835                 aux.tp_padding = 0;
2836                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2837         }
2838
2839         /*
2840          *      Free or return the buffer as appropriate. Again this
2841          *      hides all the races and re-entrancy issues from us.
2842          */
2843         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2844
2845 out_free:
2846         skb_free_datagram(sk, skb);
2847 out:
2848         return err;
2849 }
2850
2851 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2852                                int *uaddr_len, int peer)
2853 {
2854         struct net_device *dev;
2855         struct sock *sk = sock->sk;
2856
2857         if (peer)
2858                 return -EOPNOTSUPP;
2859
2860         uaddr->sa_family = AF_PACKET;
2861         rcu_read_lock();
2862         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2863         if (dev)
2864                 strncpy(uaddr->sa_data, dev->name, 14);
2865         else
2866                 memset(uaddr->sa_data, 0, 14);
2867         rcu_read_unlock();
2868         *uaddr_len = sizeof(*uaddr);
2869
2870         return 0;
2871 }
2872
2873 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2874                           int *uaddr_len, int peer)
2875 {
2876         struct net_device *dev;
2877         struct sock *sk = sock->sk;
2878         struct packet_sock *po = pkt_sk(sk);
2879         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2880
2881         if (peer)
2882                 return -EOPNOTSUPP;
2883
2884         sll->sll_family = AF_PACKET;
2885         sll->sll_ifindex = po->ifindex;
2886         sll->sll_protocol = po->num;
2887         sll->sll_pkttype = 0;
2888         rcu_read_lock();
2889         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2890         if (dev) {
2891                 sll->sll_hatype = dev->type;
2892                 sll->sll_halen = dev->addr_len;
2893                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2894         } else {
2895                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
2896                 sll->sll_halen = 0;
2897         }
2898         rcu_read_unlock();
2899         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2900
2901         return 0;
2902 }
2903
2904 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2905                          int what)
2906 {
2907         switch (i->type) {
2908         case PACKET_MR_MULTICAST:
2909                 if (i->alen != dev->addr_len)
2910                         return -EINVAL;
2911                 if (what > 0)
2912                         return dev_mc_add(dev, i->addr);
2913                 else
2914                         return dev_mc_del(dev, i->addr);
2915                 break;
2916         case PACKET_MR_PROMISC:
2917                 return dev_set_promiscuity(dev, what);
2918                 break;
2919         case PACKET_MR_ALLMULTI:
2920                 return dev_set_allmulti(dev, what);
2921                 break;
2922         case PACKET_MR_UNICAST:
2923                 if (i->alen != dev->addr_len)
2924                         return -EINVAL;
2925                 if (what > 0)
2926                         return dev_uc_add(dev, i->addr);
2927                 else
2928                         return dev_uc_del(dev, i->addr);
2929                 break;
2930         default:
2931                 break;
2932         }
2933         return 0;
2934 }
2935
2936 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2937 {
2938         for ( ; i; i = i->next) {
2939                 if (i->ifindex == dev->ifindex)
2940                         packet_dev_mc(dev, i, what);
2941         }
2942 }
2943
2944 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2945 {
2946         struct packet_sock *po = pkt_sk(sk);
2947         struct packet_mclist *ml, *i;
2948         struct net_device *dev;
2949         int err;
2950
2951         rtnl_lock();
2952
2953         err = -ENODEV;
2954         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2955         if (!dev)
2956                 goto done;
2957
2958         err = -EINVAL;
2959         if (mreq->mr_alen > dev->addr_len)
2960                 goto done;
2961
2962         err = -ENOBUFS;
2963         i = kmalloc(sizeof(*i), GFP_KERNEL);
2964         if (i == NULL)
2965                 goto done;
2966
2967         err = 0;
2968         for (ml = po->mclist; ml; ml = ml->next) {
2969                 if (ml->ifindex == mreq->mr_ifindex &&
2970                     ml->type == mreq->mr_type &&
2971                     ml->alen == mreq->mr_alen &&
2972                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2973                         ml->count++;
2974                         /* Free the new element ... */
2975                         kfree(i);
2976                         goto done;
2977                 }
2978         }
2979
2980         i->type = mreq->mr_type;
2981         i->ifindex = mreq->mr_ifindex;
2982         i->alen = mreq->mr_alen;
2983         memcpy(i->addr, mreq->mr_address, i->alen);
2984         i->count = 1;
2985         i->next = po->mclist;
2986         po->mclist = i;
2987         err = packet_dev_mc(dev, i, 1);
2988         if (err) {
2989                 po->mclist = i->next;
2990                 kfree(i);
2991         }
2992
2993 done:
2994         rtnl_unlock();
2995         return err;
2996 }
2997
2998 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2999 {
3000         struct packet_mclist *ml, **mlp;
3001
3002         rtnl_lock();
3003
3004         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3005                 if (ml->ifindex == mreq->mr_ifindex &&
3006                     ml->type == mreq->mr_type &&
3007                     ml->alen == mreq->mr_alen &&
3008                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3009                         if (--ml->count == 0) {
3010                                 struct net_device *dev;
3011                                 *mlp = ml->next;
3012                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3013                                 if (dev)
3014                                         packet_dev_mc(dev, ml, -1);
3015                                 kfree(ml);
3016                         }
3017                         rtnl_unlock();
3018                         return 0;
3019                 }
3020         }
3021         rtnl_unlock();
3022         return -EADDRNOTAVAIL;
3023 }
3024
3025 static void packet_flush_mclist(struct sock *sk)
3026 {
3027         struct packet_sock *po = pkt_sk(sk);
3028         struct packet_mclist *ml;
3029
3030         if (!po->mclist)
3031                 return;
3032
3033         rtnl_lock();
3034         while ((ml = po->mclist) != NULL) {
3035                 struct net_device *dev;
3036
3037                 po->mclist = ml->next;
3038                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3039                 if (dev != NULL)
3040                         packet_dev_mc(dev, ml, -1);
3041                 kfree(ml);
3042         }
3043         rtnl_unlock();
3044 }
3045
3046 static int
3047 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3048 {
3049         struct sock *sk = sock->sk;
3050         struct packet_sock *po = pkt_sk(sk);
3051         int ret;
3052
3053         if (level != SOL_PACKET)
3054                 return -ENOPROTOOPT;
3055
3056         switch (optname) {
3057         case PACKET_ADD_MEMBERSHIP:
3058         case PACKET_DROP_MEMBERSHIP:
3059         {
3060                 struct packet_mreq_max mreq;
3061                 int len = optlen;
3062                 memset(&mreq, 0, sizeof(mreq));
3063                 if (len < sizeof(struct packet_mreq))
3064                         return -EINVAL;
3065                 if (len > sizeof(mreq))
3066                         len = sizeof(mreq);
3067                 if (copy_from_user(&mreq, optval, len))
3068                         return -EFAULT;
3069                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3070                         return -EINVAL;
3071                 if (optname == PACKET_ADD_MEMBERSHIP)
3072                         ret = packet_mc_add(sk, &mreq);
3073                 else
3074                         ret = packet_mc_drop(sk, &mreq);
3075                 return ret;
3076         }
3077
3078         case PACKET_RX_RING:
3079         case PACKET_TX_RING:
3080         {
3081                 union tpacket_req_u req_u;
3082                 int len;
3083
3084                 switch (po->tp_version) {
3085                 case TPACKET_V1:
3086                 case TPACKET_V2:
3087                         len = sizeof(req_u.req);
3088                         break;
3089                 case TPACKET_V3:
3090                 default:
3091                         len = sizeof(req_u.req3);
3092                         break;
3093                 }
3094                 if (optlen < len)
3095                         return -EINVAL;
3096                 if (pkt_sk(sk)->has_vnet_hdr)
3097                         return -EINVAL;
3098                 if (copy_from_user(&req_u.req, optval, len))
3099                         return -EFAULT;
3100                 return packet_set_ring(sk, &req_u, 0,
3101                         optname == PACKET_TX_RING);
3102         }
3103         case PACKET_COPY_THRESH:
3104         {
3105                 int val;
3106
3107                 if (optlen != sizeof(val))
3108                         return -EINVAL;
3109                 if (copy_from_user(&val, optval, sizeof(val)))
3110                         return -EFAULT;
3111
3112                 pkt_sk(sk)->copy_thresh = val;
3113                 return 0;
3114         }
3115         case PACKET_VERSION:
3116         {
3117                 int val;
3118
3119                 if (optlen != sizeof(val))
3120                         return -EINVAL;
3121                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3122                         return -EBUSY;
3123                 if (copy_from_user(&val, optval, sizeof(val)))
3124                         return -EFAULT;
3125                 switch (val) {
3126                 case TPACKET_V1:
3127                 case TPACKET_V2:
3128                 case TPACKET_V3:
3129                         po->tp_version = val;
3130                         return 0;
3131                 default:
3132                         return -EINVAL;
3133                 }
3134         }
3135         case PACKET_RESERVE:
3136         {
3137                 unsigned int val;
3138
3139                 if (optlen != sizeof(val))
3140                         return -EINVAL;
3141                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3142                         return -EBUSY;
3143                 if (copy_from_user(&val, optval, sizeof(val)))
3144                         return -EFAULT;
3145                 po->tp_reserve = val;
3146                 return 0;
3147         }
3148         case PACKET_LOSS:
3149         {
3150                 unsigned int val;
3151
3152                 if (optlen != sizeof(val))
3153                         return -EINVAL;
3154                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3155                         return -EBUSY;
3156                 if (copy_from_user(&val, optval, sizeof(val)))
3157                         return -EFAULT;
3158                 po->tp_loss = !!val;
3159                 return 0;
3160         }
3161         case PACKET_AUXDATA:
3162         {
3163                 int val;
3164
3165                 if (optlen < sizeof(val))
3166                         return -EINVAL;
3167                 if (copy_from_user(&val, optval, sizeof(val)))
3168                         return -EFAULT;
3169
3170                 po->auxdata = !!val;
3171                 return 0;
3172         }
3173         case PACKET_ORIGDEV:
3174         {
3175                 int val;
3176
3177                 if (optlen < sizeof(val))
3178                         return -EINVAL;
3179                 if (copy_from_user(&val, optval, sizeof(val)))
3180                         return -EFAULT;
3181
3182                 po->origdev = !!val;
3183                 return 0;
3184         }
3185         case PACKET_VNET_HDR:
3186         {
3187                 int val;
3188
3189                 if (sock->type != SOCK_RAW)
3190                         return -EINVAL;
3191                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3192                         return -EBUSY;
3193                 if (optlen < sizeof(val))
3194                         return -EINVAL;
3195                 if (copy_from_user(&val, optval, sizeof(val)))
3196                         return -EFAULT;
3197
3198                 po->has_vnet_hdr = !!val;
3199                 return 0;
3200         }
3201         case PACKET_TIMESTAMP:
3202         {
3203                 int val;
3204
3205                 if (optlen != sizeof(val))
3206                         return -EINVAL;
3207                 if (copy_from_user(&val, optval, sizeof(val)))
3208                         return -EFAULT;
3209
3210                 po->tp_tstamp = val;
3211                 return 0;
3212         }
3213         case PACKET_FANOUT:
3214         {
3215                 int val;
3216
3217                 if (optlen != sizeof(val))
3218                         return -EINVAL;
3219                 if (copy_from_user(&val, optval, sizeof(val)))
3220                         return -EFAULT;
3221
3222                 return fanout_add(sk, val & 0xffff, val >> 16);
3223         }
3224         case PACKET_TX_HAS_OFF:
3225         {
3226                 unsigned int val;
3227
3228                 if (optlen != sizeof(val))
3229                         return -EINVAL;
3230                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3231                         return -EBUSY;
3232                 if (copy_from_user(&val, optval, sizeof(val)))
3233                         return -EFAULT;
3234                 po->tp_tx_has_off = !!val;
3235                 return 0;
3236         }
3237         default:
3238                 return -ENOPROTOOPT;
3239         }
3240 }
3241
3242 static int packet_getsockopt(struct socket *sock, int level, int optname,
3243                              char __user *optval, int __user *optlen)
3244 {
3245         int len;
3246         int val, lv = sizeof(val);
3247         struct sock *sk = sock->sk;
3248         struct packet_sock *po = pkt_sk(sk);
3249         void *data = &val;
3250         union tpacket_stats_u st;
3251
3252         if (level != SOL_PACKET)
3253                 return -ENOPROTOOPT;
3254
3255         if (get_user(len, optlen))
3256                 return -EFAULT;
3257
3258         if (len < 0)
3259                 return -EINVAL;
3260
3261         switch (optname) {
3262         case PACKET_STATISTICS:
3263                 spin_lock_bh(&sk->sk_receive_queue.lock);
3264                 memcpy(&st, &po->stats, sizeof(st));
3265                 memset(&po->stats, 0, sizeof(po->stats));
3266                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3267
3268                 if (po->tp_version == TPACKET_V3) {
3269                         lv = sizeof(struct tpacket_stats_v3);
3270                         data = &st.stats3;
3271                 } else {
3272                         lv = sizeof(struct tpacket_stats);
3273                         data = &st.stats1;
3274                 }
3275
3276                 break;
3277         case PACKET_AUXDATA:
3278                 val = po->auxdata;
3279                 break;
3280         case PACKET_ORIGDEV:
3281                 val = po->origdev;
3282                 break;
3283         case PACKET_VNET_HDR:
3284                 val = po->has_vnet_hdr;
3285                 break;
3286         case PACKET_VERSION:
3287                 val = po->tp_version;
3288                 break;
3289         case PACKET_HDRLEN:
3290                 if (len > sizeof(int))
3291                         len = sizeof(int);
3292                 if (copy_from_user(&val, optval, len))
3293                         return -EFAULT;
3294                 switch (val) {
3295                 case TPACKET_V1:
3296                         val = sizeof(struct tpacket_hdr);
3297                         break;
3298                 case TPACKET_V2:
3299                         val = sizeof(struct tpacket2_hdr);
3300                         break;
3301                 case TPACKET_V3:
3302                         val = sizeof(struct tpacket3_hdr);
3303                         break;
3304                 default:
3305                         return -EINVAL;
3306                 }
3307                 break;
3308         case PACKET_RESERVE:
3309                 val = po->tp_reserve;
3310                 break;
3311         case PACKET_LOSS:
3312                 val = po->tp_loss;
3313                 break;
3314         case PACKET_TIMESTAMP:
3315                 val = po->tp_tstamp;
3316                 break;
3317         case PACKET_FANOUT:
3318                 val = (po->fanout ?
3319                        ((u32)po->fanout->id |
3320                         ((u32)po->fanout->type << 16) |
3321                         ((u32)po->fanout->flags << 24)) :
3322                        0);
3323                 break;
3324         case PACKET_TX_HAS_OFF:
3325                 val = po->tp_tx_has_off;
3326                 break;
3327         default:
3328                 return -ENOPROTOOPT;
3329         }
3330
3331         if (len > lv)
3332                 len = lv;
3333         if (put_user(len, optlen))
3334                 return -EFAULT;
3335         if (copy_to_user(optval, data, len))
3336                 return -EFAULT;
3337         return 0;
3338 }
3339
3340
3341 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3342 {
3343         struct sock *sk;
3344         struct net_device *dev = data;
3345         struct net *net = dev_net(dev);
3346
3347         rcu_read_lock();
3348         sk_for_each_rcu(sk, &net->packet.sklist) {
3349                 struct packet_sock *po = pkt_sk(sk);
3350
3351                 switch (msg) {
3352                 case NETDEV_UNREGISTER:
3353                         if (po->mclist)
3354                                 packet_dev_mclist(dev, po->mclist, -1);
3355                         /* fallthrough */
3356
3357                 case NETDEV_DOWN:
3358                         if (dev->ifindex == po->ifindex) {
3359                                 spin_lock(&po->bind_lock);
3360                                 if (po->running) {
3361                                         __unregister_prot_hook(sk, false);
3362                                         sk->sk_err = ENETDOWN;
3363                                         if (!sock_flag(sk, SOCK_DEAD))
3364                                                 sk->sk_error_report(sk);
3365                                 }
3366                                 if (msg == NETDEV_UNREGISTER) {
3367                                         po->ifindex = -1;
3368                                         if (po->prot_hook.dev)
3369                                                 dev_put(po->prot_hook.dev);
3370                                         po->prot_hook.dev = NULL;
3371                                 }
3372                                 spin_unlock(&po->bind_lock);
3373                         }
3374                         break;
3375                 case NETDEV_UP:
3376                         if (dev->ifindex == po->ifindex) {
3377                                 spin_lock(&po->bind_lock);
3378                                 if (po->num)
3379                                         register_prot_hook(sk);
3380                                 spin_unlock(&po->bind_lock);
3381                         }
3382                         break;
3383                 }
3384         }
3385         rcu_read_unlock();
3386         return NOTIFY_DONE;
3387 }
3388
3389
3390 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3391                         unsigned long arg)
3392 {
3393         struct sock *sk = sock->sk;
3394
3395         switch (cmd) {
3396         case SIOCOUTQ:
3397         {
3398                 int amount = sk_wmem_alloc_get(sk);
3399
3400                 return put_user(amount, (int __user *)arg);
3401         }
3402         case SIOCINQ:
3403         {
3404                 struct sk_buff *skb;
3405                 int amount = 0;
3406
3407                 spin_lock_bh(&sk->sk_receive_queue.lock);
3408                 skb = skb_peek(&sk->sk_receive_queue);
3409                 if (skb)
3410                         amount = skb->len;
3411                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3412                 return put_user(amount, (int __user *)arg);
3413         }
3414         case SIOCGSTAMP:
3415                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3416         case SIOCGSTAMPNS:
3417                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3418
3419 #ifdef CONFIG_INET
3420         case SIOCADDRT:
3421         case SIOCDELRT:
3422         case SIOCDARP:
3423         case SIOCGARP:
3424         case SIOCSARP:
3425         case SIOCGIFADDR:
3426         case SIOCSIFADDR:
3427         case SIOCGIFBRDADDR:
3428         case SIOCSIFBRDADDR:
3429         case SIOCGIFNETMASK:
3430         case SIOCSIFNETMASK:
3431         case SIOCGIFDSTADDR:
3432         case SIOCSIFDSTADDR:
3433         case SIOCSIFFLAGS:
3434                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3435 #endif
3436
3437         default:
3438                 return -ENOIOCTLCMD;
3439         }
3440         return 0;
3441 }
3442
3443 static unsigned int packet_poll(struct file *file, struct socket *sock,
3444                                 poll_table *wait)
3445 {
3446         struct sock *sk = sock->sk;
3447         struct packet_sock *po = pkt_sk(sk);
3448         unsigned int mask = datagram_poll(file, sock, wait);
3449
3450         spin_lock_bh(&sk->sk_receive_queue.lock);
3451         if (po->rx_ring.pg_vec) {
3452                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3453                         TP_STATUS_KERNEL))
3454                         mask |= POLLIN | POLLRDNORM;
3455         }
3456         spin_unlock_bh(&sk->sk_receive_queue.lock);
3457         spin_lock_bh(&sk->sk_write_queue.lock);
3458         if (po->tx_ring.pg_vec) {
3459                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3460                         mask |= POLLOUT | POLLWRNORM;
3461         }
3462         spin_unlock_bh(&sk->sk_write_queue.lock);
3463         return mask;
3464 }
3465
3466
3467 /* Dirty? Well, I still did not learn better way to account
3468  * for user mmaps.
3469  */
3470
3471 static void packet_mm_open(struct vm_area_struct *vma)
3472 {
3473         struct file *file = vma->vm_file;
3474         struct socket *sock = file->private_data;
3475         struct sock *sk = sock->sk;
3476
3477         if (sk)
3478                 atomic_inc(&pkt_sk(sk)->mapped);
3479 }
3480
3481 static void packet_mm_close(struct vm_area_struct *vma)
3482 {
3483         struct file *file = vma->vm_file;
3484         struct socket *sock = file->private_data;
3485         struct sock *sk = sock->sk;
3486
3487         if (sk)
3488                 atomic_dec(&pkt_sk(sk)->mapped);
3489 }
3490
3491 static const struct vm_operations_struct packet_mmap_ops = {
3492         .open   =       packet_mm_open,
3493         .close  =       packet_mm_close,
3494 };
3495
3496 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3497                         unsigned int len)
3498 {
3499         int i;
3500
3501         for (i = 0; i < len; i++) {
3502                 if (likely(pg_vec[i].buffer)) {
3503                         if (is_vmalloc_addr(pg_vec[i].buffer))
3504                                 vfree(pg_vec[i].buffer);
3505                         else
3506                                 free_pages((unsigned long)pg_vec[i].buffer,
3507                                            order);
3508                         pg_vec[i].buffer = NULL;
3509                 }
3510         }
3511         kfree(pg_vec);
3512 }
3513
3514 static char *alloc_one_pg_vec_page(unsigned long order)
3515 {
3516         char *buffer = NULL;
3517         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3518                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3519
3520         buffer = (char *) __get_free_pages(gfp_flags, order);
3521
3522         if (buffer)
3523                 return buffer;
3524
3525         /*
3526          * __get_free_pages failed, fall back to vmalloc
3527          */
3528         buffer = vzalloc((1 << order) * PAGE_SIZE);
3529
3530         if (buffer)
3531                 return buffer;
3532
3533         /*
3534          * vmalloc failed, lets dig into swap here
3535          */
3536         gfp_flags &= ~__GFP_NORETRY;
3537         buffer = (char *)__get_free_pages(gfp_flags, order);
3538         if (buffer)
3539                 return buffer;
3540
3541         /*
3542          * complete and utter failure
3543          */
3544         return NULL;
3545 }
3546
3547 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3548 {
3549         unsigned int block_nr = req->tp_block_nr;
3550         struct pgv *pg_vec;
3551         int i;
3552
3553         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3554         if (unlikely(!pg_vec))
3555                 goto out;
3556
3557         for (i = 0; i < block_nr; i++) {
3558                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3559                 if (unlikely(!pg_vec[i].buffer))
3560                         goto out_free_pgvec;
3561         }
3562
3563 out:
3564         return pg_vec;
3565
3566 out_free_pgvec:
3567         free_pg_vec(pg_vec, order, block_nr);
3568         pg_vec = NULL;
3569         goto out;
3570 }
3571
3572 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3573                 int closing, int tx_ring)
3574 {
3575         struct pgv *pg_vec = NULL;
3576         struct packet_sock *po = pkt_sk(sk);
3577         int was_running, order = 0;
3578         struct packet_ring_buffer *rb;
3579         struct sk_buff_head *rb_queue;
3580         __be16 num;
3581         int err = -EINVAL;
3582         /* Added to avoid minimal code churn */
3583         struct tpacket_req *req = &req_u->req;
3584
3585         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3586         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3587                 WARN(1, "Tx-ring is not supported.\n");
3588                 goto out;
3589         }
3590
3591         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3592         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3593
3594         err = -EBUSY;
3595         if (!closing) {
3596                 if (atomic_read(&po->mapped))
3597                         goto out;
3598                 if (atomic_read(&rb->pending))
3599                         goto out;
3600         }
3601
3602         if (req->tp_block_nr) {
3603                 /* Sanity tests and some calculations */
3604                 err = -EBUSY;
3605                 if (unlikely(rb->pg_vec))
3606                         goto out;
3607
3608                 switch (po->tp_version) {
3609                 case TPACKET_V1:
3610                         po->tp_hdrlen = TPACKET_HDRLEN;
3611                         break;
3612                 case TPACKET_V2:
3613                         po->tp_hdrlen = TPACKET2_HDRLEN;
3614                         break;
3615                 case TPACKET_V3:
3616                         po->tp_hdrlen = TPACKET3_HDRLEN;
3617                         break;
3618                 }
3619
3620                 err = -EINVAL;
3621                 if (unlikely((int)req->tp_block_size <= 0))
3622                         goto out;
3623                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3624                         goto out;
3625                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3626                                         po->tp_reserve))
3627                         goto out;
3628                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3629                         goto out;
3630
3631                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3632                 if (unlikely(rb->frames_per_block <= 0))
3633                         goto out;
3634                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3635                                         req->tp_frame_nr))
3636                         goto out;
3637
3638                 err = -ENOMEM;
3639                 order = get_order(req->tp_block_size);
3640                 pg_vec = alloc_pg_vec(req, order);
3641                 if (unlikely(!pg_vec))
3642                         goto out;
3643                 switch (po->tp_version) {
3644                 case TPACKET_V3:
3645                 /* Transmit path is not supported. We checked
3646                  * it above but just being paranoid
3647                  */
3648                         if (!tx_ring)
3649                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3650                                 break;
3651                 default:
3652                         break;
3653                 }
3654         }
3655         /* Done */
3656         else {
3657                 err = -EINVAL;
3658                 if (unlikely(req->tp_frame_nr))
3659                         goto out;
3660         }
3661
3662         lock_sock(sk);
3663
3664         /* Detach socket from network */
3665         spin_lock(&po->bind_lock);
3666         was_running = po->running;
3667         num = po->num;
3668         if (was_running) {
3669                 po->num = 0;
3670                 __unregister_prot_hook(sk, false);
3671         }
3672         spin_unlock(&po->bind_lock);
3673
3674         synchronize_net();
3675
3676         err = -EBUSY;
3677         mutex_lock(&po->pg_vec_lock);
3678         if (closing || atomic_read(&po->mapped) == 0) {
3679                 err = 0;
3680                 spin_lock_bh(&rb_queue->lock);
3681                 swap(rb->pg_vec, pg_vec);
3682                 rb->frame_max = (req->tp_frame_nr - 1);
3683                 rb->head = 0;
3684                 rb->frame_size = req->tp_frame_size;
3685                 spin_unlock_bh(&rb_queue->lock);
3686
3687                 swap(rb->pg_vec_order, order);
3688                 swap(rb->pg_vec_len, req->tp_block_nr);
3689
3690                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3691                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3692                                                 tpacket_rcv : packet_rcv;
3693                 skb_queue_purge(rb_queue);
3694                 if (atomic_read(&po->mapped))
3695                         pr_err("packet_mmap: vma is busy: %d\n",
3696                                atomic_read(&po->mapped));
3697         }
3698         mutex_unlock(&po->pg_vec_lock);
3699
3700         spin_lock(&po->bind_lock);
3701         if (was_running) {
3702                 po->num = num;
3703                 register_prot_hook(sk);
3704         }
3705         spin_unlock(&po->bind_lock);
3706         if (closing && (po->tp_version > TPACKET_V2)) {
3707                 /* Because we don't support block-based V3 on tx-ring */
3708                 if (!tx_ring)
3709                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3710         }
3711         release_sock(sk);
3712
3713         if (pg_vec)
3714                 free_pg_vec(pg_vec, order, req->tp_block_nr);
3715 out:
3716         return err;
3717 }
3718
3719 static int packet_mmap(struct file *file, struct socket *sock,
3720                 struct vm_area_struct *vma)
3721 {
3722         struct sock *sk = sock->sk;
3723         struct packet_sock *po = pkt_sk(sk);
3724         unsigned long size, expected_size;
3725         struct packet_ring_buffer *rb;
3726         unsigned long start;
3727         int err = -EINVAL;
3728         int i;
3729
3730         if (vma->vm_pgoff)
3731                 return -EINVAL;
3732
3733         mutex_lock(&po->pg_vec_lock);
3734
3735         expected_size = 0;
3736         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3737                 if (rb->pg_vec) {
3738                         expected_size += rb->pg_vec_len
3739                                                 * rb->pg_vec_pages
3740                                                 * PAGE_SIZE;
3741                 }
3742         }
3743
3744         if (expected_size == 0)
3745                 goto out;
3746
3747         size = vma->vm_end - vma->vm_start;
3748         if (size != expected_size)
3749                 goto out;
3750
3751         start = vma->vm_start;
3752         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3753                 if (rb->pg_vec == NULL)
3754                         continue;
3755
3756                 for (i = 0; i < rb->pg_vec_len; i++) {
3757                         struct page *page;
3758                         void *kaddr = rb->pg_vec[i].buffer;
3759                         int pg_num;
3760
3761                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3762                                 page = pgv_to_page(kaddr);
3763                                 err = vm_insert_page(vma, start, page);
3764                                 if (unlikely(err))
3765                                         goto out;
3766                                 start += PAGE_SIZE;
3767                                 kaddr += PAGE_SIZE;
3768                         }
3769                 }
3770         }
3771
3772         atomic_inc(&po->mapped);
3773         vma->vm_ops = &packet_mmap_ops;
3774         err = 0;
3775
3776 out:
3777         mutex_unlock(&po->pg_vec_lock);
3778         return err;
3779 }
3780
3781 static const struct proto_ops packet_ops_spkt = {
3782         .family =       PF_PACKET,
3783         .owner =        THIS_MODULE,
3784         .release =      packet_release,
3785         .bind =         packet_bind_spkt,
3786         .connect =      sock_no_connect,
3787         .socketpair =   sock_no_socketpair,
3788         .accept =       sock_no_accept,
3789         .getname =      packet_getname_spkt,
3790         .poll =         datagram_poll,
3791         .ioctl =        packet_ioctl,
3792         .listen =       sock_no_listen,
3793         .shutdown =     sock_no_shutdown,
3794         .setsockopt =   sock_no_setsockopt,
3795         .getsockopt =   sock_no_getsockopt,
3796         .sendmsg =      packet_sendmsg_spkt,
3797         .recvmsg =      packet_recvmsg,
3798         .mmap =         sock_no_mmap,
3799         .sendpage =     sock_no_sendpage,
3800 };
3801
3802 static const struct proto_ops packet_ops = {
3803         .family =       PF_PACKET,
3804         .owner =        THIS_MODULE,
3805         .release =      packet_release,
3806         .bind =         packet_bind,
3807         .connect =      sock_no_connect,
3808         .socketpair =   sock_no_socketpair,
3809         .accept =       sock_no_accept,
3810         .getname =      packet_getname,
3811         .poll =         packet_poll,
3812         .ioctl =        packet_ioctl,
3813         .listen =       sock_no_listen,
3814         .shutdown =     sock_no_shutdown,
3815         .setsockopt =   packet_setsockopt,
3816         .getsockopt =   packet_getsockopt,
3817         .sendmsg =      packet_sendmsg,
3818         .recvmsg =      packet_recvmsg,
3819         .mmap =         packet_mmap,
3820         .sendpage =     sock_no_sendpage,
3821 };
3822
3823 static const struct net_proto_family packet_family_ops = {
3824         .family =       PF_PACKET,
3825         .create =       packet_create,
3826         .owner  =       THIS_MODULE,
3827 };
3828
3829 static struct notifier_block packet_netdev_notifier = {
3830         .notifier_call =        packet_notifier,
3831 };
3832
3833 #ifdef CONFIG_PROC_FS
3834
3835 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3836         __acquires(RCU)
3837 {
3838         struct net *net = seq_file_net(seq);
3839
3840         rcu_read_lock();
3841         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3842 }
3843
3844 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3845 {
3846         struct net *net = seq_file_net(seq);
3847         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3848 }
3849
3850 static void packet_seq_stop(struct seq_file *seq, void *v)
3851         __releases(RCU)
3852 {
3853         rcu_read_unlock();
3854 }
3855
3856 static int packet_seq_show(struct seq_file *seq, void *v)
3857 {
3858         if (v == SEQ_START_TOKEN)
3859                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
3860         else {
3861                 struct sock *s = sk_entry(v);
3862                 const struct packet_sock *po = pkt_sk(s);
3863
3864                 seq_printf(seq,
3865                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
3866                            s,
3867                            atomic_read(&s->sk_refcnt),
3868                            s->sk_type,
3869                            ntohs(po->num),
3870                            po->ifindex,
3871                            po->running,
3872                            atomic_read(&s->sk_rmem_alloc),
3873                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
3874                            sock_i_ino(s));
3875         }
3876
3877         return 0;
3878 }
3879
3880 static const struct seq_operations packet_seq_ops = {
3881         .start  = packet_seq_start,
3882         .next   = packet_seq_next,
3883         .stop   = packet_seq_stop,
3884         .show   = packet_seq_show,
3885 };
3886
3887 static int packet_seq_open(struct inode *inode, struct file *file)
3888 {
3889         return seq_open_net(inode, file, &packet_seq_ops,
3890                             sizeof(struct seq_net_private));
3891 }
3892
3893 static const struct file_operations packet_seq_fops = {
3894         .owner          = THIS_MODULE,
3895         .open           = packet_seq_open,
3896         .read           = seq_read,
3897         .llseek         = seq_lseek,
3898         .release        = seq_release_net,
3899 };
3900
3901 #endif
3902
3903 static int __net_init packet_net_init(struct net *net)
3904 {
3905         mutex_init(&net->packet.sklist_lock);
3906         INIT_HLIST_HEAD(&net->packet.sklist);
3907
3908         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
3909                 return -ENOMEM;
3910
3911         return 0;
3912 }
3913
3914 static void __net_exit packet_net_exit(struct net *net)
3915 {
3916         remove_proc_entry("packet", net->proc_net);
3917 }
3918
3919 static struct pernet_operations packet_net_ops = {
3920         .init = packet_net_init,
3921         .exit = packet_net_exit,
3922 };
3923
3924
3925 static void __exit packet_exit(void)
3926 {
3927         unregister_netdevice_notifier(&packet_netdev_notifier);
3928         unregister_pernet_subsys(&packet_net_ops);
3929         sock_unregister(PF_PACKET);
3930         proto_unregister(&packet_proto);
3931 }
3932
3933 static int __init packet_init(void)
3934 {
3935         int rc = proto_register(&packet_proto, 0);
3936
3937         if (rc != 0)
3938                 goto out;
3939
3940         sock_register(&packet_family_ops);
3941         register_pernet_subsys(&packet_net_ops);
3942         register_netdevice_notifier(&packet_netdev_notifier);
3943 out:
3944         return rc;
3945 }
3946
3947 module_init(packet_init);
3948 module_exit(packet_exit);
3949 MODULE_LICENSE("GPL");
3950 MODULE_ALIAS_NETPROTO(PF_PACKET);