net/packet/af_packet.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              PACKET - implements raw packet sockets.
   7  *
   8  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  11  *
  12  * Fixes:
  13  *              Alan Cox        :       verify_area() now used correctly
  14  *              Alan Cox        :       new skbuff lists, look ma no backlogs!
  15  *              Alan Cox        :       tidied skbuff lists.
  16  *              Alan Cox        :       Now uses generic datagram routines I
  17  *                                      added. Also fixed the peek/read crash
  18  *                                      from all old Linux datagram code.
  19  *              Alan Cox        :       Uses the improved datagram code.
  20  *              Alan Cox        :       Added NULL's for socket options.
  21  *              Alan Cox        :       Re-commented the code.
  22  *              Alan Cox        :       Use new kernel side addressing
  23  *              Rob Janssen     :       Correct MTU usage.
  24  *              Dave Platt      :       Counter leaks caused by incorrect
  25  *                                      interrupt locking and some slightly
  26  *                                      dubious gcc output. Can you read
  27  *                                      compiler: it said _VOLATILE_
  28  *      Richard Kooijman        :       Timestamp fixes.
  29  *              Alan Cox        :       New buffers. Use sk->mac.raw.
  30  *              Alan Cox        :       sendmsg/recvmsg support.
  31  *              Alan Cox        :       Protocol setting support
  32  *      Alexey Kuznetsov        :       Untied from IPv4 stack.
  33  *      Cyrus Durgin            :       Fixed kerneld for kmod.
  34  *      Michal Ostrowski        :       Module initialization cleanup.
  35  *         Ulises Alonso        :       Frame number limit removal and
  36  *                                      packet_set_ring memory leak.
  37  *              Eric Biederman  :       Allow for > 8 byte hardware addresses.
  38  *                                      The convention is that longer addresses
  39  *                                      will simply extend the hardware address
  40  *                                      byte arrays at the end of sockaddr_ll
  41  *                                      and packet_mreq.
  42  *              Johann Baudy    :       Added TX RING.
  43  *              Chetan Loke     :       Implemented TPACKET_V3 block abstraction
  44  *                                      layer.
  45  *                                      Copyright (C) 2011, <lokec@ccs.neu.edu>
  46  *
  47  *
  48  *              This program is free software; you can redistribute it and/or
  49  *              modify it under the terms of the GNU General Public License
  50  *              as published by the Free Software Foundation; either version
  51  *              2 of the License, or (at your option) any later version.
  52  *
  53  */
  54
  55 #include <linux/types.h>
  56 #include <linux/mm.h>
  57 #include <linux/capability.h>
  58 #include <linux/fcntl.h>
  59 #include <linux/socket.h>
  60 #include <linux/in.h>
  61 #include <linux/inet.h>
  62 #include <linux/netdevice.h>
  63 #include <linux/if_packet.h>
  64 #include <linux/wireless.h>
  65 #include <linux/kernel.h>
  66 #include <linux/kmod.h>
  67 #include <linux/slab.h>
  68 #include <linux/vmalloc.h>
  69 #include <net/net_namespace.h>
  70 #include <net/ip.h>
  71 #include <net/protocol.h>
  72 #include <linux/skbuff.h>
  73 #include <net/sock.h>
  74 #include <linux/errno.h>
  75 #include <linux/timer.h>
  76 #include <asm/uaccess.h>
  77 #include <asm/ioctls.h>
  78 #include <asm/page.h>
  79 #include <asm/cacheflush.h>
  80 #include <asm/io.h>
  81 #include <linux/proc_fs.h>
  82 #include <linux/seq_file.h>
  83 #include <linux/poll.h>
  84 #include <linux/module.h>
  85 #include <linux/init.h>
  86 #include <linux/mutex.h>
  87 #include <linux/if_vlan.h>
  88 #include <linux/virtio_net.h>
  89 #include <linux/errqueue.h>
  90 #include <linux/net_tstamp.h>
  91 #include <linux/reciprocal_div.h>
  92 #ifdef CONFIG_INET
  93 #include <net/inet_common.h>
  94 #endif
  95
  96 #include "internal.h"
  97
  98 /*
  99    Assumptions:
 100    - if device has no dev->hard_header routine, it adds and removes ll header
 101      inside itself. In this case ll header is invisible outside of device,
 102      but higher levels still should reserve dev->hard_header_len.
 103      Some devices are enough clever to reallocate skb, when header
 104      will not fit to reserved space (tunnel), another ones are silly
 105      (PPP).
 106    - packet socket receives packets with pulled ll header,
 107      so that SOCK_RAW should push it back.
 108
 109 On receive:
 110 -----------
 111
 112 Incoming, dev->hard_header!=NULL
 113    mac_header -> ll header
 114    data       -> data
 115
 116 Outgoing, dev->hard_header!=NULL
 117    mac_header -> ll header
 118    data       -> ll header
 119
 120 Incoming, dev->hard_header==NULL
 121    mac_header -> UNKNOWN position. It is very likely, that it points to ll
 122                  header.  PPP makes it, that is wrong, because introduce
 123                  assymetry between rx and tx paths.
 124    data       -> data
 125
 126 Outgoing, dev->hard_header==NULL
 127    mac_header -> data. ll header is still not built!
 128    data       -> data
 129
 130 Resume
 131   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 132
 133
 134 On transmit:
 135 ------------
 136
 137 dev->hard_header != NULL
 138    mac_header -> ll header
 139    data       -> ll header
 140
 141 dev->hard_header == NULL (ll header is added by device, we cannot control it)
 142    mac_header -> data
 143    data       -> data
 144
 145    We should set nh.raw on output to correct posistion,
 146    packet classifier depends on it.
 147  */
 148
 149 /* Private packet socket structures. */
 150
 151 /* identical to struct packet_mreq except it has
 152  * a longer address field.
 153  */
 154 struct packet_mreq_max {
 155         int             mr_ifindex;
 156         unsigned short  mr_type;
 157         unsigned short  mr_alen;
 158         unsigned char   mr_address[MAX_ADDR_LEN];
 159 };
 160
 161 union tpacket_uhdr {
 162         struct tpacket_hdr  *h1;
 163         struct tpacket2_hdr *h2;
 164         struct tpacket3_hdr *h3;
 165         void *raw;
 166 };
 167
 168 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 169                 int closing, int tx_ring);
 170
 171 #define V3_ALIGNMENT    (8)
 172
 173 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 174
 175 #define BLK_PLUS_PRIV(sz_of_priv) \
 176         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 177
 178 #define PGV_FROM_VMALLOC 1
 179
 180 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
 181 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
 182 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
 183 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
 184 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
 185 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
 186 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
 187
 188 struct packet_sock;
 189 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
 190 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 191                        struct packet_type *pt, struct net_device *orig_dev);
 192
 193 static void *packet_previous_frame(struct packet_sock *po,
 194                 struct packet_ring_buffer *rb,
 195                 int status);
 196 static void packet_increment_head(struct packet_ring_buffer *buff);
 197 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
 198                         struct tpacket_block_desc *);
 199 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 200                         struct packet_sock *);
 201 static void prb_retire_current_block(struct tpacket_kbdq_core *,
 202                 struct packet_sock *, unsigned int status);
 203 static int prb_queue_frozen(struct tpacket_kbdq_core *);
 204 static void prb_open_block(struct tpacket_kbdq_core *,
 205                 struct tpacket_block_desc *);
 206 static void prb_retire_rx_blk_timer_expired(unsigned long);
 207 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 208 static void prb_init_blk_timer(struct packet_sock *,
 209                 struct tpacket_kbdq_core *,
 210                 void (*func) (unsigned long));
 211 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 212 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 213                 struct tpacket3_hdr *);
 214 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 215                 struct tpacket3_hdr *);
 216 static void packet_flush_mclist(struct sock *sk);
 217
 218 struct packet_skb_cb {
 219         unsigned int origlen;
 220         union {
 221                 struct sockaddr_pkt pkt;
 222                 struct sockaddr_ll ll;
 223         } sa;
 224 };
 225
 226 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 227
 228 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 229 #define GET_PBLOCK_DESC(x, bid) \
 230         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 231 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 232         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 233 #define GET_NEXT_PRB_BLK_NUM(x) \
 234         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 235         ((x)->kactive_blk_num+1) : 0)
 236
 237 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 238 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 239
 240 /* 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
 248         if (!po->running) {
 249                 if (po->fanout) {
 250                         __fanout_link(sk, po);
 251                 } else {
 252                         dev_add_pack(&po->prot_hook);
 253                         rcu_assign_pointer(po->cached_dev, po->prot_hook.dev);
 254                 }
 255
 256                 sock_hold(sk);
 257                 po->running = 1;
 258         }
 259 }
 260
 261 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
 262  * held.   If the sync parameter is true, we will temporarily drop
 263  * the po->bind_lock and do a synchronize_net to make sure no
 264  * asynchronous packet processing paths still refer to the elements
 265  * of po->prot_hook.  If the sync parameter is false, it is the
 266  * callers responsibility to take care of this.
 267  */
 268 static void __unregister_prot_hook(struct sock *sk, bool sync)
 269 {
 270         struct packet_sock *po = pkt_sk(sk);
 271
 272         po->running = 0;
 273         if (po->fanout) {
 274                 __fanout_unlink(sk, po);
 275         } else {
 276                 __dev_remove_pack(&po->prot_hook);
 277                 RCU_INIT_POINTER(po->cached_dev, NULL);
 278         }
 279
 280         __sock_put(sk);
 281
 282         if (sync) {
 283                 spin_unlock(&po->bind_lock);
 284                 synchronize_net();
 285                 spin_lock(&po->bind_lock);
 286         }
 287 }
 288
 289 static void unregister_prot_hook(struct sock *sk, bool sync)
 290 {
 291         struct packet_sock *po = pkt_sk(sk);
 292
 293         if (po->running)
 294                 __unregister_prot_hook(sk, sync);
 295 }
 296
 297 static inline __pure struct page *pgv_to_page(void *addr)
 298 {
 299         if (is_vmalloc_addr(addr))
 300                 return vmalloc_to_page(addr);
 301         return virt_to_page(addr);
 302 }
 303
 304 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 305 {
 306         union tpacket_uhdr h;
 307
 308         h.raw = frame;
 309         switch (po->tp_version) {
 310         case TPACKET_V1:
 311                 h.h1->tp_status = status;
 312                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 313                 break;
 314         case TPACKET_V2:
 315                 h.h2->tp_status = status;
 316                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 317                 break;
 318         case TPACKET_V3:
 319         default:
 320                 WARN(1, "TPACKET version not supported.\n");
 321                 BUG();
 322         }
 323
 324         smp_wmb();
 325 }
 326
 327 static int __packet_get_status(struct packet_sock *po, void *frame)
 328 {
 329         union tpacket_uhdr h;
 330
 331         smp_rmb();
 332
 333         h.raw = frame;
 334         switch (po->tp_version) {
 335         case TPACKET_V1:
 336                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 337                 return h.h1->tp_status;
 338         case TPACKET_V2:
 339                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 340                 return h.h2->tp_status;
 341         case TPACKET_V3:
 342         default:
 343                 WARN(1, "TPACKET version not supported.\n");
 344                 BUG();
 345                 return 0;
 346         }
 347 }
 348
 349 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
 350                                    unsigned int flags)
 351 {
 352         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 353
 354         if (shhwtstamps) {
 355                 if ((flags & SOF_TIMESTAMPING_SYS_HARDWARE) &&
 356                     ktime_to_timespec_cond(shhwtstamps->syststamp, ts))
 357                         return TP_STATUS_TS_SYS_HARDWARE;
 358                 if ((flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
 359                     ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
 360                         return TP_STATUS_TS_RAW_HARDWARE;
 361         }
 362
 363         if (ktime_to_timespec_cond(skb->tstamp, ts))
 364                 return TP_STATUS_TS_SOFTWARE;
 365
 366         return 0;
 367 }
 368
 369 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
 370                                     struct sk_buff *skb)
 371 {
 372         union tpacket_uhdr h;
 373         struct timespec ts;
 374         __u32 ts_status;
 375
 376         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
 377                 return 0;
 378
 379         h.raw = frame;
 380         switch (po->tp_version) {
 381         case TPACKET_V1:
 382                 h.h1->tp_sec = ts.tv_sec;
 383                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
 384                 break;
 385         case TPACKET_V2:
 386                 h.h2->tp_sec = ts.tv_sec;
 387                 h.h2->tp_nsec = ts.tv_nsec;
 388                 break;
 389         case TPACKET_V3:
 390         default:
 391                 WARN(1, "TPACKET version not supported.\n");
 392                 BUG();
 393         }
 394
 395         /* one flush is safe, as both fields always lie on the same cacheline */
 396         flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
 397         smp_wmb();
 398
 399         return ts_status;
 400 }
 401
 402 static void *packet_lookup_frame(struct packet_sock *po,
 403                 struct packet_ring_buffer *rb,
 404                 unsigned int position,
 405                 int status)
 406 {
 407         unsigned int pg_vec_pos, frame_offset;
 408         union tpacket_uhdr h;
 409
 410         pg_vec_pos = position / rb->frames_per_block;
 411         frame_offset = position % rb->frames_per_block;
 412
 413         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 414                 (frame_offset * rb->frame_size);
 415
 416         if (status != __packet_get_status(po, h.raw))
 417                 return NULL;
 418
 419         return h.raw;
 420 }
 421
 422 static void *packet_current_frame(struct packet_sock *po,
 423                 struct packet_ring_buffer *rb,
 424                 int status)
 425 {
 426         return packet_lookup_frame(po, rb, rb->head, status);
 427 }
 428
 429 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 430 {
 431         del_timer_sync(&pkc->retire_blk_timer);
 432 }
 433
 434 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 435                 int tx_ring,
 436                 struct sk_buff_head *rb_queue)
 437 {
 438         struct tpacket_kbdq_core *pkc;
 439
 440         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 441
 442         spin_lock_bh(&rb_queue->lock);
 443         pkc->delete_blk_timer = 1;
 444         spin_unlock_bh(&rb_queue->lock);
 445
 446         prb_del_retire_blk_timer(pkc);
 447 }
 448
 449 static void prb_init_blk_timer(struct packet_sock *po,
 450                 struct tpacket_kbdq_core *pkc,
 451                 void (*func) (unsigned long))
 452 {
 453         init_timer(&pkc->retire_blk_timer);
 454         pkc->retire_blk_timer.data = (long)po;
 455         pkc->retire_blk_timer.function = func;
 456         pkc->retire_blk_timer.expires = jiffies;
 457 }
 458
 459 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
 460 {
 461         struct tpacket_kbdq_core *pkc;
 462
 463         if (tx_ring)
 464                 BUG();
 465
 466         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 467         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
 468 }
 469
 470 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 471                                 int blk_size_in_bytes)
 472 {
 473         struct net_device *dev;
 474         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 475         struct ethtool_cmd ecmd;
 476         int err;
 477         u32 speed;
 478
 479         rtnl_lock();
 480         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 481         if (unlikely(!dev)) {
 482                 rtnl_unlock();
 483                 return DEFAULT_PRB_RETIRE_TOV;
 484         }
 485         err = __ethtool_get_settings(dev, &ecmd);
 486         speed = ethtool_cmd_speed(&ecmd);
 487         rtnl_unlock();
 488         if (!err) {
 489                 /*
 490                  * If the link speed is so slow you don't really
 491                  * need to worry about perf anyways
 492                  */
 493                 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
 494                         return DEFAULT_PRB_RETIRE_TOV;
 495                 } else {
 496                         msec = 1;
 497                         div = speed / 1000;
 498                 }
 499         }
 500
 501         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 502
 503         if (div)
 504                 mbits /= div;
 505
 506         tmo = mbits * msec;
 507
 508         if (div)
 509                 return tmo+1;
 510         return tmo;
 511 }
 512
 513 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 514                         union tpacket_req_u *req_u)
 515 {
 516         p1->feature_req_word = req_u->req3.tp_feature_req_word;
 517 }
 518
 519 static void init_prb_bdqc(struct packet_sock *po,
 520                         struct packet_ring_buffer *rb,
 521                         struct pgv *pg_vec,
 522                         union tpacket_req_u *req_u, int tx_ring)
 523 {
 524         struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
 525         struct tpacket_block_desc *pbd;
 526
 527         memset(p1, 0x0, sizeof(*p1));
 528
 529         p1->knxt_seq_num = 1;
 530         p1->pkbdq = pg_vec;
 531         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 532         p1->pkblk_start = pg_vec[0].buffer;
 533         p1->kblk_size = req_u->req3.tp_block_size;
 534         p1->knum_blocks = req_u->req3.tp_block_nr;
 535         p1->hdrlen = po->tp_hdrlen;
 536         p1->version = po->tp_version;
 537         p1->last_kactive_blk_num = 0;
 538         po->stats.stats3.tp_freeze_q_cnt = 0;
 539         if (req_u->req3.tp_retire_blk_tov)
 540                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 541         else
 542                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 543                                                 req_u->req3.tp_block_size);
 544         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 545         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 546
 547         prb_init_ft_ops(p1, req_u);
 548         prb_setup_retire_blk_timer(po, tx_ring);
 549         prb_open_block(p1, pbd);
 550 }
 551
 552 /*  Do NOT update the last_blk_num first.
 553  *  Assumes sk_buff_head lock is held.
 554  */
 555 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 556 {
 557         mod_timer(&pkc->retire_blk_timer,
 558                         jiffies + pkc->tov_in_jiffies);
 559         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 560 }
 561
 562 /*
 563  * Timer logic:
 564  * 1) We refresh the timer only when we open a block.
 565  *    By doing this we don't waste cycles refreshing the timer
 566  *        on packet-by-packet basis.
 567  *
 568  * With a 1MB block-size, on a 1Gbps line, it will take
 569  * i) ~8 ms to fill a block + ii) memcpy etc.
 570  * In this cut we are not accounting for the memcpy time.
 571  *
 572  * So, if the user sets the 'tmo' to 10ms then the timer
 573  * will never fire while the block is still getting filled
 574  * (which is what we want). However, the user could choose
 575  * to close a block early and that's fine.
 576  *
 577  * But when the timer does fire, we check whether or not to refresh it.
 578  * Since the tmo granularity is in msecs, it is not too expensive
 579  * to refresh the timer, lets say every '8' msecs.
 580  * Either the user can set the 'tmo' or we can derive it based on
 581  * a) line-speed and b) block-size.
 582  * prb_calc_retire_blk_tmo() calculates the tmo.
 583  *
 584  */
 585 static void prb_retire_rx_blk_timer_expired(unsigned long data)
 586 {
 587         struct packet_sock *po = (struct packet_sock *)data;
 588         struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
 589         unsigned int frozen;
 590         struct tpacket_block_desc *pbd;
 591
 592         spin_lock(&po->sk.sk_receive_queue.lock);
 593
 594         frozen = prb_queue_frozen(pkc);
 595         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 596
 597         if (unlikely(pkc->delete_blk_timer))
 598                 goto out;
 599
 600         /* We only need to plug the race when the block is partially filled.
 601          * tpacket_rcv:
 602          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 603          *              copy_bits() is in progress ...
 604          *              timer fires on other cpu:
 605          *              we can't retire the current block because copy_bits
 606          *              is in progress.
 607          *
 608          */
 609         if (BLOCK_NUM_PKTS(pbd)) {
 610                 while (atomic_read(&pkc->blk_fill_in_prog)) {
 611                         /* Waiting for skb_copy_bits to finish... */
 612                         cpu_relax();
 613                 }
 614         }
 615
 616         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 617                 if (!frozen) {
 618                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 619                         if (!prb_dispatch_next_block(pkc, po))
 620                                 goto refresh_timer;
 621                         else
 622                                 goto out;
 623                 } else {
 624                         /* Case 1. Queue was frozen because user-space was
 625                          *         lagging behind.
 626                          */
 627                         if (prb_curr_blk_in_use(pkc, pbd)) {
 628                                 /*
 629                                  * Ok, user-space is still behind.
 630                                  * So just refresh the timer.
 631                                  */
 632                                 goto refresh_timer;
 633                         } else {
 634                                /* Case 2. queue was frozen,user-space caught up,
 635                                 * now the link went idle && the timer fired.
 636                                 * We don't have a block to close.So we open this
 637                                 * block and restart the timer.
 638                                 * opening a block thaws the queue,restarts timer
 639                                 * Thawing/timer-refresh is a side effect.
 640                                 */
 641                                 prb_open_block(pkc, pbd);
 642                                 goto out;
 643                         }
 644                 }
 645         }
 646
 647 refresh_timer:
 648         _prb_refresh_rx_retire_blk_timer(pkc);
 649
 650 out:
 651         spin_unlock(&po->sk.sk_receive_queue.lock);
 652 }
 653
 654 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 655                 struct tpacket_block_desc *pbd1, __u32 status)
 656 {
 657         /* Flush everything minus the block header */
 658
 659 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 660         u8 *start, *end;
 661
 662         start = (u8 *)pbd1;
 663
 664         /* Skip the block header(we know header WILL fit in 4K) */
 665         start += PAGE_SIZE;
 666
 667         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 668         for (; start < end; start += PAGE_SIZE)
 669                 flush_dcache_page(pgv_to_page(start));
 670
 671         smp_wmb();
 672 #endif
 673
 674         /* Now update the block status. */
 675
 676         BLOCK_STATUS(pbd1) = status;
 677
 678         /* Flush the block header */
 679
 680 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 681         start = (u8 *)pbd1;
 682         flush_dcache_page(pgv_to_page(start));
 683
 684         smp_wmb();
 685 #endif
 686 }
 687
 688 /*
 689  * Side effect:
 690  *
 691  * 1) flush the block
 692  * 2) Increment active_blk_num
 693  *
 694  * Note:We DONT refresh the timer on purpose.
 695  *      Because almost always the next block will be opened.
 696  */
 697 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 698                 struct tpacket_block_desc *pbd1,
 699                 struct packet_sock *po, unsigned int stat)
 700 {
 701         __u32 status = TP_STATUS_USER | stat;
 702
 703         struct tpacket3_hdr *last_pkt;
 704         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 705
 706         if (po->stats.stats3.tp_drops)
 707                 status |= TP_STATUS_LOSING;
 708
 709         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 710         last_pkt->tp_next_offset = 0;
 711
 712         /* Get the ts of the last pkt */
 713         if (BLOCK_NUM_PKTS(pbd1)) {
 714                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 715                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 716         } else {
 717                 /* Ok, we tmo'd - so get the current time */
 718                 struct timespec ts;
 719                 getnstimeofday(&ts);
 720                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 721                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 722         }
 723
 724         smp_wmb();
 725
 726         /* Flush the block */
 727         prb_flush_block(pkc1, pbd1, status);
 728
 729         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 730 }
 731
 732 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 733 {
 734         pkc->reset_pending_on_curr_blk = 0;
 735 }
 736
 737 /*
 738  * Side effect of opening a block:
 739  *
 740  * 1) prb_queue is thawed.
 741  * 2) retire_blk_timer is refreshed.
 742  *
 743  */
 744 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 745         struct tpacket_block_desc *pbd1)
 746 {
 747         struct timespec ts;
 748         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 749
 750         smp_rmb();
 751
 752         /* We could have just memset this but we will lose the
 753          * flexibility of making the priv area sticky
 754          */
 755
 756         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 757         BLOCK_NUM_PKTS(pbd1) = 0;
 758         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 759
 760         getnstimeofday(&ts);
 761
 762         h1->ts_first_pkt.ts_sec = ts.tv_sec;
 763         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 764
 765         pkc1->pkblk_start = (char *)pbd1;
 766         pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 767
 768         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 769         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 770
 771         pbd1->version = pkc1->version;
 772         pkc1->prev = pkc1->nxt_offset;
 773         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 774
 775         prb_thaw_queue(pkc1);
 776         _prb_refresh_rx_retire_blk_timer(pkc1);
 777
 778         smp_wmb();
 779 }
 780
 781 /*
 782  * Queue freeze logic:
 783  * 1) Assume tp_block_nr = 8 blocks.
 784  * 2) At time 't0', user opens Rx ring.
 785  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 786  * 4) user-space is either sleeping or processing block '0'.
 787  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 788  *    it will close block-7,loop around and try to fill block '0'.
 789  *    call-flow:
 790  *    __packet_lookup_frame_in_block
 791  *      prb_retire_current_block()
 792  *      prb_dispatch_next_block()
 793  *        |->(BLOCK_STATUS == USER) evaluates to true
 794  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 795  * 6) Now there are two cases:
 796  *    6.1) Link goes idle right after the queue is frozen.
 797  *         But remember, the last open_block() refreshed the timer.
 798  *         When this timer expires,it will refresh itself so that we can
 799  *         re-open block-0 in near future.
 800  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 801  *         case and __packet_lookup_frame_in_block will check if block-0
 802  *         is free and can now be re-used.
 803  */
 804 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 805                                   struct packet_sock *po)
 806 {
 807         pkc->reset_pending_on_curr_blk = 1;
 808         po->stats.stats3.tp_freeze_q_cnt++;
 809 }
 810
 811 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 812
 813 /*
 814  * If the next block is free then we will dispatch it
 815  * and return a good offset.
 816  * Else, we will freeze the queue.
 817  * So, caller must check the return value.
 818  */
 819 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 820                 struct packet_sock *po)
 821 {
 822         struct tpacket_block_desc *pbd;
 823
 824         smp_rmb();
 825
 826         /* 1. Get current block num */
 827         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 828
 829         /* 2. If this block is currently in_use then freeze the queue */
 830         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 831                 prb_freeze_queue(pkc, po);
 832                 return NULL;
 833         }
 834
 835         /*
 836          * 3.
 837          * open this block and return the offset where the first packet
 838          * needs to get stored.
 839          */
 840         prb_open_block(pkc, pbd);
 841         return (void *)pkc->nxt_offset;
 842 }
 843
 844 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 845                 struct packet_sock *po, unsigned int status)
 846 {
 847         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 848
 849         /* retire/close the current block */
 850         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 851                 /*
 852                  * Plug the case where copy_bits() is in progress on
 853                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 854                  * have space to copy the pkt in the current block and
 855                  * called prb_retire_current_block()
 856                  *
 857                  * We don't need to worry about the TMO case because
 858                  * the timer-handler already handled this case.
 859                  */
 860                 if (!(status & TP_STATUS_BLK_TMO)) {
 861                         while (atomic_read(&pkc->blk_fill_in_prog)) {
 862                                 /* Waiting for skb_copy_bits to finish... */
 863                                 cpu_relax();
 864                         }
 865                 }
 866                 prb_close_block(pkc, pbd, po, status);
 867                 return;
 868         }
 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 reciprocal_divide(skb->rxhash, num);
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_rnd(struct packet_fanout *f,
1169                                      struct sk_buff *skb,
1170                                      unsigned int num)
1171 {
1172         return reciprocal_divide(prandom_u32(), num);
1173 }
1174
1175 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1176                                           struct sk_buff *skb,
1177                                           unsigned int idx, unsigned int skip,
1178                                           unsigned int num)
1179 {
1180         unsigned int i, j;
1181
1182         i = j = min_t(int, f->next[idx], num - 1);
1183         do {
1184                 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1185                         if (i != j)
1186                                 f->next[idx] = i;
1187                         return i;
1188                 }
1189                 if (++i == num)
1190                         i = 0;
1191         } while (i != j);
1192
1193         return idx;
1194 }
1195
1196 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1197 {
1198         return f->flags & (flag >> 8);
1199 }
1200
1201 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1202                              struct packet_type *pt, struct net_device *orig_dev)
1203 {
1204         struct packet_fanout *f = pt->af_packet_priv;
1205         unsigned int num = f->num_members;
1206         struct packet_sock *po;
1207         unsigned int idx;
1208
1209         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1210             !num) {
1211                 kfree_skb(skb);
1212                 return 0;
1213         }
1214
1215         switch (f->type) {
1216         case PACKET_FANOUT_HASH:
1217         default:
1218                 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1219                         skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1220                         if (!skb)
1221                                 return 0;
1222                 }
1223                 skb_get_rxhash(skb);
1224                 idx = fanout_demux_hash(f, skb, num);
1225                 break;
1226         case PACKET_FANOUT_LB:
1227                 idx = fanout_demux_lb(f, skb, num);
1228                 break;
1229         case PACKET_FANOUT_CPU:
1230                 idx = fanout_demux_cpu(f, skb, num);
1231                 break;
1232         case PACKET_FANOUT_RND:
1233                 idx = fanout_demux_rnd(f, skb, num);
1234                 break;
1235         case PACKET_FANOUT_ROLLOVER:
1236                 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1237                 break;
1238         }
1239
1240         po = pkt_sk(f->arr[idx]);
1241         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1242             unlikely(!packet_rcv_has_room(po, skb))) {
1243                 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1244                 po = pkt_sk(f->arr[idx]);
1245         }
1246
1247         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1248 }
1249
1250 DEFINE_MUTEX(fanout_mutex);
1251 EXPORT_SYMBOL_GPL(fanout_mutex);
1252 static LIST_HEAD(fanout_list);
1253
1254 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1255 {
1256         struct packet_fanout *f = po->fanout;
1257
1258         spin_lock(&f->lock);
1259         f->arr[f->num_members] = sk;
1260         smp_wmb();
1261         f->num_members++;
1262         spin_unlock(&f->lock);
1263 }
1264
1265 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1266 {
1267         struct packet_fanout *f = po->fanout;
1268         int i;
1269
1270         spin_lock(&f->lock);
1271         for (i = 0; i < f->num_members; i++) {
1272                 if (f->arr[i] == sk)
1273                         break;
1274         }
1275         BUG_ON(i >= f->num_members);
1276         f->arr[i] = f->arr[f->num_members - 1];
1277         f->num_members--;
1278         spin_unlock(&f->lock);
1279 }
1280
1281 static bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1282 {
1283         if (ptype->af_packet_priv == (void*)((struct packet_sock *)sk)->fanout)
1284                 return true;
1285
1286         return false;
1287 }
1288
1289 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1290 {
1291         struct packet_sock *po = pkt_sk(sk);
1292         struct packet_fanout *f, *match;
1293         u8 type = type_flags & 0xff;
1294         u8 flags = type_flags >> 8;
1295         int err;
1296
1297         switch (type) {
1298         case PACKET_FANOUT_ROLLOVER:
1299                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1300                         return -EINVAL;
1301         case PACKET_FANOUT_HASH:
1302         case PACKET_FANOUT_LB:
1303         case PACKET_FANOUT_CPU:
1304         case PACKET_FANOUT_RND:
1305                 break;
1306         default:
1307                 return -EINVAL;
1308         }
1309
1310         if (!po->running)
1311                 return -EINVAL;
1312
1313         if (po->fanout)
1314                 return -EALREADY;
1315
1316         mutex_lock(&fanout_mutex);
1317         match = NULL;
1318         list_for_each_entry(f, &fanout_list, list) {
1319                 if (f->id == id &&
1320                     read_pnet(&f->net) == sock_net(sk)) {
1321                         match = f;
1322                         break;
1323                 }
1324         }
1325         err = -EINVAL;
1326         if (match && match->flags != flags)
1327                 goto out;
1328         if (!match) {
1329                 err = -ENOMEM;
1330                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1331                 if (!match)
1332                         goto out;
1333                 write_pnet(&match->net, sock_net(sk));
1334                 match->id = id;
1335                 match->type = type;
1336                 match->flags = flags;
1337                 atomic_set(&match->rr_cur, 0);
1338                 INIT_LIST_HEAD(&match->list);
1339                 spin_lock_init(&match->lock);
1340                 atomic_set(&match->sk_ref, 0);
1341                 match->prot_hook.type = po->prot_hook.type;
1342                 match->prot_hook.dev = po->prot_hook.dev;
1343                 match->prot_hook.func = packet_rcv_fanout;
1344                 match->prot_hook.af_packet_priv = match;
1345                 match->prot_hook.id_match = match_fanout_group;
1346                 dev_add_pack(&match->prot_hook);
1347                 list_add(&match->list, &fanout_list);
1348         }
1349         err = -EINVAL;
1350         if (match->type == type &&
1351             match->prot_hook.type == po->prot_hook.type &&
1352             match->prot_hook.dev == po->prot_hook.dev) {
1353                 err = -ENOSPC;
1354                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1355                         __dev_remove_pack(&po->prot_hook);
1356                         po->fanout = match;
1357                         atomic_inc(&match->sk_ref);
1358                         __fanout_link(sk, po);
1359                         err = 0;
1360                 }
1361         }
1362 out:
1363         mutex_unlock(&fanout_mutex);
1364         return err;
1365 }
1366
1367 static void fanout_release(struct sock *sk)
1368 {
1369         struct packet_sock *po = pkt_sk(sk);
1370         struct packet_fanout *f;
1371
1372         f = po->fanout;
1373         if (!f)
1374                 return;
1375
1376         mutex_lock(&fanout_mutex);
1377         po->fanout = NULL;
1378
1379         if (atomic_dec_and_test(&f->sk_ref)) {
1380                 list_del(&f->list);
1381                 dev_remove_pack(&f->prot_hook);
1382                 kfree(f);
1383         }
1384         mutex_unlock(&fanout_mutex);
1385 }
1386
1387 static const struct proto_ops packet_ops;
1388
1389 static const struct proto_ops packet_ops_spkt;
1390
1391 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1392                            struct packet_type *pt, struct net_device *orig_dev)
1393 {
1394         struct sock *sk;
1395         struct sockaddr_pkt *spkt;
1396
1397         /*
1398          *      When we registered the protocol we saved the socket in the data
1399          *      field for just this event.
1400          */
1401
1402         sk = pt->af_packet_priv;
1403
1404         /*
1405          *      Yank back the headers [hope the device set this
1406          *      right or kerboom...]
1407          *
1408          *      Incoming packets have ll header pulled,
1409          *      push it back.
1410          *
1411          *      For outgoing ones skb->data == skb_mac_header(skb)
1412          *      so that this procedure is noop.
1413          */
1414
1415         if (skb->pkt_type == PACKET_LOOPBACK)
1416                 goto out;
1417
1418         if (!net_eq(dev_net(dev), sock_net(sk)))
1419                 goto out;
1420
1421         skb = skb_share_check(skb, GFP_ATOMIC);
1422         if (skb == NULL)
1423                 goto oom;
1424
1425         /* drop any routing info */
1426         skb_dst_drop(skb);
1427
1428         /* drop conntrack reference */
1429         nf_reset(skb);
1430
1431         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1432
1433         skb_push(skb, skb->data - skb_mac_header(skb));
1434
1435         /*
1436          *      The SOCK_PACKET socket receives _all_ frames.
1437          */
1438
1439         spkt->spkt_family = dev->type;
1440         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1441         spkt->spkt_protocol = skb->protocol;
1442
1443         /*
1444          *      Charge the memory to the socket. This is done specifically
1445          *      to prevent sockets using all the memory up.
1446          */
1447
1448         if (sock_queue_rcv_skb(sk, skb) == 0)
1449                 return 0;
1450
1451 out:
1452         kfree_skb(skb);
1453 oom:
1454         return 0;
1455 }
1456
1457
1458 /*
1459  *      Output a raw packet to a device layer. This bypasses all the other
1460  *      protocol layers and you must therefore supply it with a complete frame
1461  */
1462
1463 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1464                                struct msghdr *msg, size_t len)
1465 {
1466         struct sock *sk = sock->sk;
1467         struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1468         struct sk_buff *skb = NULL;
1469         struct net_device *dev;
1470         __be16 proto = 0;
1471         int err;
1472         int extra_len = 0;
1473
1474         /*
1475          *      Get and verify the address.
1476          */
1477
1478         if (saddr) {
1479                 if (msg->msg_namelen < sizeof(struct sockaddr))
1480                         return -EINVAL;
1481                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1482                         proto = saddr->spkt_protocol;
1483         } else
1484                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1485
1486         /*
1487          *      Find the device first to size check it
1488          */
1489
1490         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1491 retry:
1492         rcu_read_lock();
1493         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1494         err = -ENODEV;
1495         if (dev == NULL)
1496                 goto out_unlock;
1497
1498         err = -ENETDOWN;
1499         if (!(dev->flags & IFF_UP))
1500                 goto out_unlock;
1501
1502         /*
1503          * You may not queue a frame bigger than the mtu. This is the lowest level
1504          * raw protocol and you must do your own fragmentation at this level.
1505          */
1506
1507         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1508                 if (!netif_supports_nofcs(dev)) {
1509                         err = -EPROTONOSUPPORT;
1510                         goto out_unlock;
1511                 }
1512                 extra_len = 4; /* We're doing our own CRC */
1513         }
1514
1515         err = -EMSGSIZE;
1516         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1517                 goto out_unlock;
1518
1519         if (!skb) {
1520                 size_t reserved = LL_RESERVED_SPACE(dev);
1521                 int tlen = dev->needed_tailroom;
1522                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1523
1524                 rcu_read_unlock();
1525                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1526                 if (skb == NULL)
1527                         return -ENOBUFS;
1528                 /* FIXME: Save some space for broken drivers that write a hard
1529                  * header at transmission time by themselves. PPP is the notable
1530                  * one here. This should really be fixed at the driver level.
1531                  */
1532                 skb_reserve(skb, reserved);
1533                 skb_reset_network_header(skb);
1534
1535                 /* Try to align data part correctly */
1536                 if (hhlen) {
1537                         skb->data -= hhlen;
1538                         skb->tail -= hhlen;
1539                         if (len < hhlen)
1540                                 skb_reset_network_header(skb);
1541                 }
1542                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1543                 if (err)
1544                         goto out_free;
1545                 goto retry;
1546         }
1547
1548         if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1549                 /* Earlier code assumed this would be a VLAN pkt,
1550                  * double-check this now that we have the actual
1551                  * packet in hand.
1552                  */
1553                 struct ethhdr *ehdr;
1554                 skb_reset_mac_header(skb);
1555                 ehdr = eth_hdr(skb);
1556                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1557                         err = -EMSGSIZE;
1558                         goto out_unlock;
1559                 }
1560         }
1561
1562         skb->protocol = proto;
1563         skb->dev = dev;
1564         skb->priority = sk->sk_priority;
1565         skb->mark = sk->sk_mark;
1566
1567         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1568
1569         if (unlikely(extra_len == 4))
1570                 skb->no_fcs = 1;
1571
1572         skb_probe_transport_header(skb, 0);
1573
1574         dev_queue_xmit(skb);
1575         rcu_read_unlock();
1576         return len;
1577
1578 out_unlock:
1579         rcu_read_unlock();
1580 out_free:
1581         kfree_skb(skb);
1582         return err;
1583 }
1584
1585 static unsigned int run_filter(const struct sk_buff *skb,
1586                                       const struct sock *sk,
1587                                       unsigned int res)
1588 {
1589         struct sk_filter *filter;
1590
1591         rcu_read_lock();
1592         filter = rcu_dereference(sk->sk_filter);
1593         if (filter != NULL)
1594                 res = SK_RUN_FILTER(filter, skb);
1595         rcu_read_unlock();
1596
1597         return res;
1598 }
1599
1600 /*
1601  * This function makes lazy skb cloning in hope that most of packets
1602  * are discarded by BPF.
1603  *
1604  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1605  * and skb->cb are mangled. It works because (and until) packets
1606  * falling here are owned by current CPU. Output packets are cloned
1607  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1608  * sequencially, so that if we return skb to original state on exit,
1609  * we will not harm anyone.
1610  */
1611
1612 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1613                       struct packet_type *pt, struct net_device *orig_dev)
1614 {
1615         struct sock *sk;
1616         struct sockaddr_ll *sll;
1617         struct packet_sock *po;
1618         u8 *skb_head = skb->data;
1619         int skb_len = skb->len;
1620         unsigned int snaplen, res;
1621
1622         if (skb->pkt_type == PACKET_LOOPBACK)
1623                 goto drop;
1624
1625         sk = pt->af_packet_priv;
1626         po = pkt_sk(sk);
1627
1628         if (!net_eq(dev_net(dev), sock_net(sk)))
1629                 goto drop;
1630
1631         skb->dev = dev;
1632
1633         if (dev->header_ops) {
1634                 /* The device has an explicit notion of ll header,
1635                  * exported to higher levels.
1636                  *
1637                  * Otherwise, the device hides details of its frame
1638                  * structure, so that corresponding packet head is
1639                  * never delivered to user.
1640                  */
1641                 if (sk->sk_type != SOCK_DGRAM)
1642                         skb_push(skb, skb->data - skb_mac_header(skb));
1643                 else if (skb->pkt_type == PACKET_OUTGOING) {
1644                         /* Special case: outgoing packets have ll header at head */
1645                         skb_pull(skb, skb_network_offset(skb));
1646                 }
1647         }
1648
1649         snaplen = skb->len;
1650
1651         res = run_filter(skb, sk, snaplen);
1652         if (!res)
1653                 goto drop_n_restore;
1654         if (snaplen > res)
1655                 snaplen = res;
1656
1657         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1658                 goto drop_n_acct;
1659
1660         if (skb_shared(skb)) {
1661                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1662                 if (nskb == NULL)
1663                         goto drop_n_acct;
1664
1665                 if (skb_head != skb->data) {
1666                         skb->data = skb_head;
1667                         skb->len = skb_len;
1668                 }
1669                 consume_skb(skb);
1670                 skb = nskb;
1671         }
1672
1673         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1674                      sizeof(skb->cb));
1675
1676         sll = &PACKET_SKB_CB(skb)->sa.ll;
1677         sll->sll_family = AF_PACKET;
1678         sll->sll_hatype = dev->type;
1679         sll->sll_protocol = skb->protocol;
1680         sll->sll_pkttype = skb->pkt_type;
1681         if (unlikely(po->origdev))
1682                 sll->sll_ifindex = orig_dev->ifindex;
1683         else
1684                 sll->sll_ifindex = dev->ifindex;
1685
1686         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1687
1688         PACKET_SKB_CB(skb)->origlen = skb->len;
1689
1690         if (pskb_trim(skb, snaplen))
1691                 goto drop_n_acct;
1692
1693         skb_set_owner_r(skb, sk);
1694         skb->dev = NULL;
1695         skb_dst_drop(skb);
1696
1697         /* drop conntrack reference */
1698         nf_reset(skb);
1699
1700         spin_lock(&sk->sk_receive_queue.lock);
1701         po->stats.stats1.tp_packets++;
1702         skb->dropcount = atomic_read(&sk->sk_drops);
1703         __skb_queue_tail(&sk->sk_receive_queue, skb);
1704         spin_unlock(&sk->sk_receive_queue.lock);
1705         sk->sk_data_ready(sk, skb->len);
1706         return 0;
1707
1708 drop_n_acct:
1709         spin_lock(&sk->sk_receive_queue.lock);
1710         po->stats.stats1.tp_drops++;
1711         atomic_inc(&sk->sk_drops);
1712         spin_unlock(&sk->sk_receive_queue.lock);
1713
1714 drop_n_restore:
1715         if (skb_head != skb->data && skb_shared(skb)) {
1716                 skb->data = skb_head;
1717                 skb->len = skb_len;
1718         }
1719 drop:
1720         consume_skb(skb);
1721         return 0;
1722 }
1723
1724 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1725                        struct packet_type *pt, struct net_device *orig_dev)
1726 {
1727         struct sock *sk;
1728         struct packet_sock *po;
1729         struct sockaddr_ll *sll;
1730         union tpacket_uhdr h;
1731         u8 *skb_head = skb->data;
1732         int skb_len = skb->len;
1733         unsigned int snaplen, res;
1734         unsigned long status = TP_STATUS_USER;
1735         unsigned short macoff, netoff, hdrlen;
1736         struct sk_buff *copy_skb = NULL;
1737         struct timespec ts;
1738         __u32 ts_status;
1739
1740         if (skb->pkt_type == PACKET_LOOPBACK)
1741                 goto drop;
1742
1743         sk = pt->af_packet_priv;
1744         po = pkt_sk(sk);
1745
1746         if (!net_eq(dev_net(dev), sock_net(sk)))
1747                 goto drop;
1748
1749         if (dev->header_ops) {
1750                 if (sk->sk_type != SOCK_DGRAM)
1751                         skb_push(skb, skb->data - skb_mac_header(skb));
1752                 else if (skb->pkt_type == PACKET_OUTGOING) {
1753                         /* Special case: outgoing packets have ll header at head */
1754                         skb_pull(skb, skb_network_offset(skb));
1755                 }
1756         }
1757
1758         if (skb->ip_summed == CHECKSUM_PARTIAL)
1759                 status |= TP_STATUS_CSUMNOTREADY;
1760
1761         snaplen = skb->len;
1762
1763         res = run_filter(skb, sk, snaplen);
1764         if (!res)
1765                 goto drop_n_restore;
1766         if (snaplen > res)
1767                 snaplen = res;
1768
1769         if (sk->sk_type == SOCK_DGRAM) {
1770                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1771                                   po->tp_reserve;
1772         } else {
1773                 unsigned int maclen = skb_network_offset(skb);
1774                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1775                                        (maclen < 16 ? 16 : maclen)) +
1776                         po->tp_reserve;
1777                 macoff = netoff - maclen;
1778         }
1779         if (po->tp_version <= TPACKET_V2) {
1780                 if (macoff + snaplen > po->rx_ring.frame_size) {
1781                         if (po->copy_thresh &&
1782                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1783                                 if (skb_shared(skb)) {
1784                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1785                                 } else {
1786                                         copy_skb = skb_get(skb);
1787                                         skb_head = skb->data;
1788                                 }
1789                                 if (copy_skb)
1790                                         skb_set_owner_r(copy_skb, sk);
1791                         }
1792                         snaplen = po->rx_ring.frame_size - macoff;
1793                         if ((int)snaplen < 0)
1794                                 snaplen = 0;
1795                 }
1796         }
1797         spin_lock(&sk->sk_receive_queue.lock);
1798         h.raw = packet_current_rx_frame(po, skb,
1799                                         TP_STATUS_KERNEL, (macoff+snaplen));
1800         if (!h.raw)
1801                 goto ring_is_full;
1802         if (po->tp_version <= TPACKET_V2) {
1803                 packet_increment_rx_head(po, &po->rx_ring);
1804         /*
1805          * LOSING will be reported till you read the stats,
1806          * because it's COR - Clear On Read.
1807          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1808          * at packet level.
1809          */
1810                 if (po->stats.stats1.tp_drops)
1811                         status |= TP_STATUS_LOSING;
1812         }
1813         po->stats.stats1.tp_packets++;
1814         if (copy_skb) {
1815                 status |= TP_STATUS_COPY;
1816                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1817         }
1818         spin_unlock(&sk->sk_receive_queue.lock);
1819
1820         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1821
1822         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
1823                 getnstimeofday(&ts);
1824
1825         status |= ts_status;
1826
1827         switch (po->tp_version) {
1828         case TPACKET_V1:
1829                 h.h1->tp_len = skb->len;
1830                 h.h1->tp_snaplen = snaplen;
1831                 h.h1->tp_mac = macoff;
1832                 h.h1->tp_net = netoff;
1833                 h.h1->tp_sec = ts.tv_sec;
1834                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
1835                 hdrlen = sizeof(*h.h1);
1836                 break;
1837         case TPACKET_V2:
1838                 h.h2->tp_len = skb->len;
1839                 h.h2->tp_snaplen = snaplen;
1840                 h.h2->tp_mac = macoff;
1841                 h.h2->tp_net = netoff;
1842                 h.h2->tp_sec = ts.tv_sec;
1843                 h.h2->tp_nsec = ts.tv_nsec;
1844                 if (vlan_tx_tag_present(skb)) {
1845                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1846                         status |= TP_STATUS_VLAN_VALID;
1847                 } else {
1848                         h.h2->tp_vlan_tci = 0;
1849                 }
1850                 h.h2->tp_padding = 0;
1851                 hdrlen = sizeof(*h.h2);
1852                 break;
1853         case TPACKET_V3:
1854                 /* tp_nxt_offset,vlan are already populated above.
1855                  * So DONT clear those fields here
1856                  */
1857                 h.h3->tp_status |= status;
1858                 h.h3->tp_len = skb->len;
1859                 h.h3->tp_snaplen = snaplen;
1860                 h.h3->tp_mac = macoff;
1861                 h.h3->tp_net = netoff;
1862                 h.h3->tp_sec  = ts.tv_sec;
1863                 h.h3->tp_nsec = ts.tv_nsec;
1864                 hdrlen = sizeof(*h.h3);
1865                 break;
1866         default:
1867                 BUG();
1868         }
1869
1870         sll = h.raw + TPACKET_ALIGN(hdrlen);
1871         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1872         sll->sll_family = AF_PACKET;
1873         sll->sll_hatype = dev->type;
1874         sll->sll_protocol = skb->protocol;
1875         sll->sll_pkttype = skb->pkt_type;
1876         if (unlikely(po->origdev))
1877                 sll->sll_ifindex = orig_dev->ifindex;
1878         else
1879                 sll->sll_ifindex = dev->ifindex;
1880
1881         smp_mb();
1882 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1883         {
1884                 u8 *start, *end;
1885
1886                 if (po->tp_version <= TPACKET_V2) {
1887                         end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1888                                 + macoff + snaplen);
1889                         for (start = h.raw; start < end; start += PAGE_SIZE)
1890                                 flush_dcache_page(pgv_to_page(start));
1891                 }
1892                 smp_wmb();
1893         }
1894 #endif
1895         if (po->tp_version <= TPACKET_V2)
1896                 __packet_set_status(po, h.raw, status);
1897         else
1898                 prb_clear_blk_fill_status(&po->rx_ring);
1899
1900         sk->sk_data_ready(sk, 0);
1901
1902 drop_n_restore:
1903         if (skb_head != skb->data && skb_shared(skb)) {
1904                 skb->data = skb_head;
1905                 skb->len = skb_len;
1906         }
1907 drop:
1908         kfree_skb(skb);
1909         return 0;
1910
1911 ring_is_full:
1912         po->stats.stats1.tp_drops++;
1913         spin_unlock(&sk->sk_receive_queue.lock);
1914
1915         sk->sk_data_ready(sk, 0);
1916         kfree_skb(copy_skb);
1917         goto drop_n_restore;
1918 }
1919
1920 static void tpacket_destruct_skb(struct sk_buff *skb)
1921 {
1922         struct packet_sock *po = pkt_sk(skb->sk);
1923         void *ph;
1924
1925         if (likely(po->tx_ring.pg_vec)) {
1926                 __u32 ts;
1927
1928                 ph = skb_shinfo(skb)->destructor_arg;
1929                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1930                 atomic_dec(&po->tx_ring.pending);
1931
1932                 ts = __packet_set_timestamp(po, ph, skb);
1933                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
1934         }
1935
1936         sock_wfree(skb);
1937 }
1938
1939 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1940                 void *frame, struct net_device *dev, int size_max,
1941                 __be16 proto, unsigned char *addr, int hlen)
1942 {
1943         union tpacket_uhdr ph;
1944         int to_write, offset, len, tp_len, nr_frags, len_max;
1945         struct socket *sock = po->sk.sk_socket;
1946         struct page *page;
1947         void *data;
1948         int err;
1949
1950         ph.raw = frame;
1951
1952         skb->protocol = proto;
1953         skb->dev = dev;
1954         skb->priority = po->sk.sk_priority;
1955         skb->mark = po->sk.sk_mark;
1956         sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
1957         skb_shinfo(skb)->destructor_arg = ph.raw;
1958
1959         switch (po->tp_version) {
1960         case TPACKET_V2:
1961                 tp_len = ph.h2->tp_len;
1962                 break;
1963         default:
1964                 tp_len = ph.h1->tp_len;
1965                 break;
1966         }
1967         if (unlikely(tp_len > size_max)) {
1968                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1969                 return -EMSGSIZE;
1970         }
1971
1972         skb_reserve(skb, hlen);
1973         skb_reset_network_header(skb);
1974         skb_probe_transport_header(skb, 0);
1975
1976         if (po->tp_tx_has_off) {
1977                 int off_min, off_max, off;
1978                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
1979                 off_max = po->tx_ring.frame_size - tp_len;
1980                 if (sock->type == SOCK_DGRAM) {
1981                         switch (po->tp_version) {
1982                         case TPACKET_V2:
1983                                 off = ph.h2->tp_net;
1984                                 break;
1985                         default:
1986                                 off = ph.h1->tp_net;
1987                                 break;
1988                         }
1989                 } else {
1990                         switch (po->tp_version) {
1991                         case TPACKET_V2:
1992                                 off = ph.h2->tp_mac;
1993                                 break;
1994                         default:
1995                                 off = ph.h1->tp_mac;
1996                                 break;
1997                         }
1998                 }
1999                 if (unlikely((off < off_min) || (off_max < off)))
2000                         return -EINVAL;
2001                 data = ph.raw + off;
2002         } else {
2003                 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2004         }
2005         to_write = tp_len;
2006
2007         if (sock->type == SOCK_DGRAM) {
2008                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2009                                 NULL, tp_len);
2010                 if (unlikely(err < 0))
2011                         return -EINVAL;
2012         } else if (dev->hard_header_len) {
2013                 /* net device doesn't like empty head */
2014                 if (unlikely(tp_len <= dev->hard_header_len)) {
2015                         pr_err("packet size is too short (%d < %d)\n",
2016                                tp_len, dev->hard_header_len);
2017                         return -EINVAL;
2018                 }
2019
2020                 skb_push(skb, dev->hard_header_len);
2021                 err = skb_store_bits(skb, 0, data,
2022                                 dev->hard_header_len);
2023                 if (unlikely(err))
2024                         return err;
2025
2026                 data += dev->hard_header_len;
2027                 to_write -= dev->hard_header_len;
2028         }
2029
2030         offset = offset_in_page(data);
2031         len_max = PAGE_SIZE - offset;
2032         len = ((to_write > len_max) ? len_max : to_write);
2033
2034         skb->data_len = to_write;
2035         skb->len += to_write;
2036         skb->truesize += to_write;
2037         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2038
2039         while (likely(to_write)) {
2040                 nr_frags = skb_shinfo(skb)->nr_frags;
2041
2042                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2043                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2044                                MAX_SKB_FRAGS);
2045                         return -EFAULT;
2046                 }
2047
2048                 page = pgv_to_page(data);
2049                 data += len;
2050                 flush_dcache_page(page);
2051                 get_page(page);
2052                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2053                 to_write -= len;
2054                 offset = 0;
2055                 len_max = PAGE_SIZE;
2056                 len = ((to_write > len_max) ? len_max : to_write);
2057         }
2058
2059         return tp_len;
2060 }
2061
2062 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
2063 {
2064         struct net_device *dev;
2065
2066         rcu_read_lock();
2067         dev = rcu_dereference(po->cached_dev);
2068         if (dev)
2069                 dev_hold(dev);
2070         rcu_read_unlock();
2071
2072         return dev;
2073 }
2074
2075 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2076 {
2077         struct sk_buff *skb;
2078         struct net_device *dev;
2079         __be16 proto;
2080         int err, reserve = 0;
2081         void *ph;
2082         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2083         int tp_len, size_max;
2084         unsigned char *addr;
2085         int len_sum = 0;
2086         int status = TP_STATUS_AVAILABLE;
2087         int hlen, tlen;
2088
2089         mutex_lock(&po->pg_vec_lock);
2090
2091         if (saddr == NULL) {
2092                 dev     = packet_cached_dev_get(po);
2093                 proto   = po->num;
2094                 addr    = NULL;
2095         } else {
2096                 err = -EINVAL;
2097                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2098                         goto out;
2099                 if (msg->msg_namelen < (saddr->sll_halen
2100                                         + offsetof(struct sockaddr_ll,
2101                                                 sll_addr)))
2102                         goto out;
2103                 proto   = saddr->sll_protocol;
2104                 addr    = saddr->sll_addr;
2105                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2106         }
2107
2108         err = -ENXIO;
2109         if (unlikely(dev == NULL))
2110                 goto out;
2111         err = -ENETDOWN;
2112         if (unlikely(!(dev->flags & IFF_UP)))
2113                 goto out_put;
2114
2115         reserve = dev->hard_header_len;
2116
2117         size_max = po->tx_ring.frame_size
2118                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2119
2120         if (size_max > dev->mtu + reserve)
2121                 size_max = dev->mtu + reserve;
2122
2123         do {
2124                 ph = packet_current_frame(po, &po->tx_ring,
2125                                 TP_STATUS_SEND_REQUEST);
2126
2127                 if (unlikely(ph == NULL)) {
2128                         schedule();
2129                         continue;
2130                 }
2131
2132                 status = TP_STATUS_SEND_REQUEST;
2133                 hlen = LL_RESERVED_SPACE(dev);
2134                 tlen = dev->needed_tailroom;
2135                 skb = sock_alloc_send_skb(&po->sk,
2136                                 hlen + tlen + sizeof(struct sockaddr_ll),
2137                                 0, &err);
2138
2139                 if (unlikely(skb == NULL))
2140                         goto out_status;
2141
2142                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2143                                 addr, hlen);
2144
2145                 if (unlikely(tp_len < 0)) {
2146                         if (po->tp_loss) {
2147                                 __packet_set_status(po, ph,
2148                                                 TP_STATUS_AVAILABLE);
2149                                 packet_increment_head(&po->tx_ring);
2150                                 kfree_skb(skb);
2151                                 continue;
2152                         } else {
2153                                 status = TP_STATUS_WRONG_FORMAT;
2154                                 err = tp_len;
2155                                 goto out_status;
2156                         }
2157                 }
2158
2159                 skb->destructor = tpacket_destruct_skb;
2160                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2161                 atomic_inc(&po->tx_ring.pending);
2162
2163                 status = TP_STATUS_SEND_REQUEST;
2164                 err = dev_queue_xmit(skb);
2165                 if (unlikely(err > 0)) {
2166                         err = net_xmit_errno(err);
2167                         if (err && __packet_get_status(po, ph) ==
2168                                    TP_STATUS_AVAILABLE) {
2169                                 /* skb was destructed already */
2170                                 skb = NULL;
2171                                 goto out_status;
2172                         }
2173                         /*
2174                          * skb was dropped but not destructed yet;
2175                          * let's treat it like congestion or err < 0
2176                          */
2177                         err = 0;
2178                 }
2179                 packet_increment_head(&po->tx_ring);
2180                 len_sum += tp_len;
2181         } while (likely((ph != NULL) ||
2182                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2183                          (atomic_read(&po->tx_ring.pending))))
2184                 );
2185
2186         err = len_sum;
2187         goto out_put;
2188
2189 out_status:
2190         __packet_set_status(po, ph, status);
2191         kfree_skb(skb);
2192 out_put:
2193         dev_put(dev);
2194 out:
2195         mutex_unlock(&po->pg_vec_lock);
2196         return err;
2197 }
2198
2199 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2200                                         size_t reserve, size_t len,
2201                                         size_t linear, int noblock,
2202                                         int *err)
2203 {
2204         struct sk_buff *skb;
2205
2206         /* Under a page?  Don't bother with paged skb. */
2207         if (prepad + len < PAGE_SIZE || !linear)
2208                 linear = len;
2209
2210         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2211                                    err, 0);
2212         if (!skb)
2213                 return NULL;
2214
2215         skb_reserve(skb, reserve);
2216         skb_put(skb, linear);
2217         skb->data_len = len - linear;
2218         skb->len += len - linear;
2219
2220         return skb;
2221 }
2222
2223 static int packet_snd(struct socket *sock,
2224                           struct msghdr *msg, size_t len)
2225 {
2226         struct sock *sk = sock->sk;
2227         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2228         struct sk_buff *skb;
2229         struct net_device *dev;
2230         __be16 proto;
2231         unsigned char *addr;
2232         int err, reserve = 0;
2233         struct virtio_net_hdr vnet_hdr = { 0 };
2234         int offset = 0;
2235         int vnet_hdr_len;
2236         struct packet_sock *po = pkt_sk(sk);
2237         unsigned short gso_type = 0;
2238         int hlen, tlen;
2239         int extra_len = 0;
2240
2241         /*
2242          *      Get and verify the address.
2243          */
2244
2245         if (saddr == NULL) {
2246                 dev     = packet_cached_dev_get(po);
2247                 proto   = po->num;
2248                 addr    = NULL;
2249         } else {
2250                 err = -EINVAL;
2251                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2252                         goto out;
2253                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2254                         goto out;
2255                 proto   = saddr->sll_protocol;
2256                 addr    = saddr->sll_addr;
2257                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2258         }
2259
2260         err = -ENXIO;
2261         if (unlikely(dev == NULL))
2262                 goto out_unlock;
2263         err = -ENETDOWN;
2264         if (unlikely(!(dev->flags & IFF_UP)))
2265                 goto out_unlock;
2266
2267         if (sock->type == SOCK_RAW)
2268                 reserve = dev->hard_header_len;
2269         if (po->has_vnet_hdr) {
2270                 vnet_hdr_len = sizeof(vnet_hdr);
2271
2272                 err = -EINVAL;
2273                 if (len < vnet_hdr_len)
2274                         goto out_unlock;
2275
2276                 len -= vnet_hdr_len;
2277
2278                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2279                                        vnet_hdr_len);
2280                 if (err < 0)
2281                         goto out_unlock;
2282
2283                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2284                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2285                       vnet_hdr.hdr_len))
2286                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
2287                                                  vnet_hdr.csum_offset + 2;
2288
2289                 err = -EINVAL;
2290                 if (vnet_hdr.hdr_len > len)
2291                         goto out_unlock;
2292
2293                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2294                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2295                         case VIRTIO_NET_HDR_GSO_TCPV4:
2296                                 gso_type = SKB_GSO_TCPV4;
2297                                 break;
2298                         case VIRTIO_NET_HDR_GSO_TCPV6:
2299                                 gso_type = SKB_GSO_TCPV6;
2300                                 break;
2301                         case VIRTIO_NET_HDR_GSO_UDP:
2302                                 gso_type = SKB_GSO_UDP;
2303                                 break;
2304                         default:
2305                                 goto out_unlock;
2306                         }
2307
2308                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2309                                 gso_type |= SKB_GSO_TCP_ECN;
2310
2311                         if (vnet_hdr.gso_size == 0)
2312                                 goto out_unlock;
2313
2314                 }
2315         }
2316
2317         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2318                 if (!netif_supports_nofcs(dev)) {
2319                         err = -EPROTONOSUPPORT;
2320                         goto out_unlock;
2321                 }
2322                 extra_len = 4; /* We're doing our own CRC */
2323         }
2324
2325         err = -EMSGSIZE;
2326         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2327                 goto out_unlock;
2328
2329         err = -ENOBUFS;
2330         hlen = LL_RESERVED_SPACE(dev);
2331         tlen = dev->needed_tailroom;
2332         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2333                                msg->msg_flags & MSG_DONTWAIT, &err);
2334         if (skb == NULL)
2335                 goto out_unlock;
2336
2337         skb_set_network_header(skb, reserve);
2338
2339         err = -EINVAL;
2340         if (sock->type == SOCK_DGRAM &&
2341             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2342                 goto out_free;
2343
2344         /* Returns -EFAULT on error */
2345         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2346         if (err)
2347                 goto out_free;
2348
2349         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2350
2351         if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2352                 /* Earlier code assumed this would be a VLAN pkt,
2353                  * double-check this now that we have the actual
2354                  * packet in hand.
2355                  */
2356                 struct ethhdr *ehdr;
2357                 skb_reset_mac_header(skb);
2358                 ehdr = eth_hdr(skb);
2359                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2360                         err = -EMSGSIZE;
2361                         goto out_free;
2362                 }
2363         }
2364
2365         skb->protocol = proto;
2366         skb->dev = dev;
2367         skb->priority = sk->sk_priority;
2368         skb->mark = sk->sk_mark;
2369
2370         if (po->has_vnet_hdr) {
2371                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2372                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2373                                                   vnet_hdr.csum_offset)) {
2374                                 err = -EINVAL;
2375                                 goto out_free;
2376                         }
2377                 }
2378
2379                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2380                 skb_shinfo(skb)->gso_type = gso_type;
2381
2382                 /* Header must be checked, and gso_segs computed. */
2383                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2384                 skb_shinfo(skb)->gso_segs = 0;
2385
2386                 len += vnet_hdr_len;
2387         }
2388
2389         skb_probe_transport_header(skb, reserve);
2390
2391         if (unlikely(extra_len == 4))
2392                 skb->no_fcs = 1;
2393
2394         /*
2395          *      Now send it
2396          */
2397
2398         err = dev_queue_xmit(skb);
2399         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2400                 goto out_unlock;
2401
2402         dev_put(dev);
2403
2404         return len;
2405
2406 out_free:
2407         kfree_skb(skb);
2408 out_unlock:
2409         if (dev)
2410                 dev_put(dev);
2411 out:
2412         return err;
2413 }
2414
2415 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2416                 struct msghdr *msg, size_t len)
2417 {
2418         struct sock *sk = sock->sk;
2419         struct packet_sock *po = pkt_sk(sk);
2420         if (po->tx_ring.pg_vec)
2421                 return tpacket_snd(po, msg);
2422         else
2423                 return packet_snd(sock, msg, len);
2424 }
2425
2426 /*
2427  *      Close a PACKET socket. This is fairly simple. We immediately go
2428  *      to 'closed' state and remove our protocol entry in the device list.
2429  */
2430
2431 static int packet_release(struct socket *sock)
2432 {
2433         struct sock *sk = sock->sk;
2434         struct packet_sock *po;
2435         struct net *net;
2436         union tpacket_req_u req_u;
2437
2438         if (!sk)
2439                 return 0;
2440
2441         net = sock_net(sk);
2442         po = pkt_sk(sk);
2443
2444         mutex_lock(&net->packet.sklist_lock);
2445         sk_del_node_init_rcu(sk);
2446         mutex_unlock(&net->packet.sklist_lock);
2447
2448         preempt_disable();
2449         sock_prot_inuse_add(net, sk->sk_prot, -1);
2450         preempt_enable();
2451
2452         spin_lock(&po->bind_lock);
2453         unregister_prot_hook(sk, false);
2454         if (po->prot_hook.dev) {
2455                 dev_put(po->prot_hook.dev);
2456                 po->prot_hook.dev = NULL;
2457         }
2458         spin_unlock(&po->bind_lock);
2459
2460         packet_flush_mclist(sk);
2461
2462         if (po->rx_ring.pg_vec) {
2463                 memset(&req_u, 0, sizeof(req_u));
2464                 packet_set_ring(sk, &req_u, 1, 0);
2465         }
2466
2467         if (po->tx_ring.pg_vec) {
2468                 memset(&req_u, 0, sizeof(req_u));
2469                 packet_set_ring(sk, &req_u, 1, 1);
2470         }
2471
2472         fanout_release(sk);
2473
2474         synchronize_net();
2475         /*
2476          *      Now the socket is dead. No more input will appear.
2477          */
2478         sock_orphan(sk);
2479         sock->sk = NULL;
2480
2481         /* Purge queues */
2482
2483         skb_queue_purge(&sk->sk_receive_queue);
2484         sk_refcnt_debug_release(sk);
2485
2486         sock_put(sk);
2487         return 0;
2488 }
2489
2490 /*
2491  *      Attach a packet hook.
2492  */
2493
2494 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2495 {
2496         struct packet_sock *po = pkt_sk(sk);
2497
2498         if (po->fanout) {
2499                 if (dev)
2500                         dev_put(dev);
2501
2502                 return -EINVAL;
2503         }
2504
2505         lock_sock(sk);
2506
2507         spin_lock(&po->bind_lock);
2508         unregister_prot_hook(sk, true);
2509         po->num = protocol;
2510         po->prot_hook.type = protocol;
2511         if (po->prot_hook.dev)
2512                 dev_put(po->prot_hook.dev);
2513         po->prot_hook.dev = dev;
2514
2515         po->ifindex = dev ? dev->ifindex : 0;
2516
2517         if (protocol == 0)
2518                 goto out_unlock;
2519
2520         if (!dev || (dev->flags & IFF_UP)) {
2521                 register_prot_hook(sk);
2522         } else {
2523                 sk->sk_err = ENETDOWN;
2524                 if (!sock_flag(sk, SOCK_DEAD))
2525                         sk->sk_error_report(sk);
2526         }
2527
2528 out_unlock:
2529         spin_unlock(&po->bind_lock);
2530         release_sock(sk);
2531         return 0;
2532 }
2533
2534 /*
2535  *      Bind a packet socket to a device
2536  */
2537
2538 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2539                             int addr_len)
2540 {
2541         struct sock *sk = sock->sk;
2542         char name[15];
2543         struct net_device *dev;
2544         int err = -ENODEV;
2545
2546         /*
2547          *      Check legality
2548          */
2549
2550         if (addr_len != sizeof(struct sockaddr))
2551                 return -EINVAL;
2552         strlcpy(name, uaddr->sa_data, sizeof(name));
2553
2554         dev = dev_get_by_name(sock_net(sk), name);
2555         if (dev)
2556                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2557         return err;
2558 }
2559
2560 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2561 {
2562         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2563         struct sock *sk = sock->sk;
2564         struct net_device *dev = NULL;
2565         int err;
2566
2567
2568         /*
2569          *      Check legality
2570          */
2571
2572         if (addr_len < sizeof(struct sockaddr_ll))
2573                 return -EINVAL;
2574         if (sll->sll_family != AF_PACKET)
2575                 return -EINVAL;
2576
2577         if (sll->sll_ifindex) {
2578                 err = -ENODEV;
2579                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2580                 if (dev == NULL)
2581                         goto out;
2582         }
2583         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2584
2585 out:
2586         return err;
2587 }
2588
2589 static struct proto packet_proto = {
2590         .name     = "PACKET",
2591         .owner    = THIS_MODULE,
2592         .obj_size = sizeof(struct packet_sock),
2593 };
2594
2595 /*
2596  *      Create a packet of type SOCK_PACKET.
2597  */
2598
2599 static int packet_create(struct net *net, struct socket *sock, int protocol,
2600                          int kern)
2601 {
2602         struct sock *sk;
2603         struct packet_sock *po;
2604         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2605         int err;
2606
2607         if (!ns_capable(net->user_ns, CAP_NET_RAW))
2608                 return -EPERM;
2609         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2610             sock->type != SOCK_PACKET)
2611                 return -ESOCKTNOSUPPORT;
2612
2613         sock->state = SS_UNCONNECTED;
2614
2615         err = -ENOBUFS;
2616         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2617         if (sk == NULL)
2618                 goto out;
2619
2620         sock->ops = &packet_ops;
2621         if (sock->type == SOCK_PACKET)
2622                 sock->ops = &packet_ops_spkt;
2623
2624         sock_init_data(sock, sk);
2625
2626         po = pkt_sk(sk);
2627         sk->sk_family = PF_PACKET;
2628         po->num = proto;
2629         RCU_INIT_POINTER(po->cached_dev, NULL);
2630
2631         sk->sk_destruct = packet_sock_destruct;
2632         sk_refcnt_debug_inc(sk);
2633
2634         /*
2635          *      Attach a protocol block
2636          */
2637
2638         spin_lock_init(&po->bind_lock);
2639         mutex_init(&po->pg_vec_lock);
2640         po->prot_hook.func = packet_rcv;
2641
2642         if (sock->type == SOCK_PACKET)
2643                 po->prot_hook.func = packet_rcv_spkt;
2644
2645         po->prot_hook.af_packet_priv = sk;
2646
2647         if (proto) {
2648                 po->prot_hook.type = proto;
2649                 register_prot_hook(sk);
2650         }
2651
2652         mutex_lock(&net->packet.sklist_lock);
2653         sk_add_node_rcu(sk, &net->packet.sklist);
2654         mutex_unlock(&net->packet.sklist_lock);
2655
2656         preempt_disable();
2657         sock_prot_inuse_add(net, &packet_proto, 1);
2658         preempt_enable();
2659
2660         return 0;
2661 out:
2662         return err;
2663 }
2664
2665 /*
2666  *      Pull a packet from our receive queue and hand it to the user.
2667  *      If necessary we block.
2668  */
2669
2670 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2671                           struct msghdr *msg, size_t len, int flags)
2672 {
2673         struct sock *sk = sock->sk;
2674         struct sk_buff *skb;
2675         int copied, err;
2676         int vnet_hdr_len = 0;
2677
2678         err = -EINVAL;
2679         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2680                 goto out;
2681
2682 #if 0
2683         /* What error should we return now? EUNATTACH? */
2684         if (pkt_sk(sk)->ifindex < 0)
2685                 return -ENODEV;
2686 #endif
2687
2688         if (flags & MSG_ERRQUEUE) {
2689                 err = sock_recv_errqueue(sk, msg, len,
2690                                          SOL_PACKET, PACKET_TX_TIMESTAMP);
2691                 goto out;
2692         }
2693
2694         /*
2695          *      Call the generic datagram receiver. This handles all sorts
2696          *      of horrible races and re-entrancy so we can forget about it
2697          *      in the protocol layers.
2698          *
2699          *      Now it will return ENETDOWN, if device have just gone down,
2700          *      but then it will block.
2701          */
2702
2703         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2704
2705         /*
2706          *      An error occurred so return it. Because skb_recv_datagram()
2707          *      handles the blocking we don't see and worry about blocking
2708          *      retries.
2709          */
2710
2711         if (skb == NULL)
2712                 goto out;
2713
2714         if (pkt_sk(sk)->has_vnet_hdr) {
2715                 struct virtio_net_hdr vnet_hdr = { 0 };
2716
2717                 err = -EINVAL;
2718                 vnet_hdr_len = sizeof(vnet_hdr);
2719                 if (len < vnet_hdr_len)
2720                         goto out_free;
2721
2722                 len -= vnet_hdr_len;
2723
2724                 if (skb_is_gso(skb)) {
2725                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2726
2727                         /* This is a hint as to how much should be linear. */
2728                         vnet_hdr.hdr_len = skb_headlen(skb);
2729                         vnet_hdr.gso_size = sinfo->gso_size;
2730                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2731                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2732                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2733                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2734                         else if (sinfo->gso_type & SKB_GSO_UDP)
2735                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2736                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2737                                 goto out_free;
2738                         else
2739                                 BUG();
2740                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2741                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2742                 } else
2743                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2744
2745                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2746                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2747                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2748                         vnet_hdr.csum_offset = skb->csum_offset;
2749                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2750                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2751                 } /* else everything is zero */
2752
2753                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2754                                      vnet_hdr_len);
2755                 if (err < 0)
2756                         goto out_free;
2757         }
2758
2759         /* You lose any data beyond the buffer you gave. If it worries
2760          * a user program they can ask the device for its MTU
2761          * anyway.
2762          */
2763         copied = skb->len;
2764         if (copied > len) {
2765                 copied = len;
2766                 msg->msg_flags |= MSG_TRUNC;
2767         }
2768
2769         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2770         if (err)
2771                 goto out_free;
2772
2773         sock_recv_ts_and_drops(msg, sk, skb);
2774
2775         if (msg->msg_name) {
2776                 /* If the address length field is there to be filled
2777                  * in, we fill it in now.
2778                  */
2779                 if (sock->type == SOCK_PACKET) {
2780                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
2781                 } else {
2782                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2783                         msg->msg_namelen = sll->sll_halen +
2784                                 offsetof(struct sockaddr_ll, sll_addr);
2785                 }
2786                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2787                        msg->msg_namelen);
2788         }
2789
2790         if (pkt_sk(sk)->auxdata) {
2791                 struct tpacket_auxdata aux;
2792
2793                 aux.tp_status = TP_STATUS_USER;
2794                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2795                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2796                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2797                 aux.tp_snaplen = skb->len;
2798                 aux.tp_mac = 0;
2799                 aux.tp_net = skb_network_offset(skb);
2800                 if (vlan_tx_tag_present(skb)) {
2801                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2802                         aux.tp_status |= TP_STATUS_VLAN_VALID;
2803                 } else {
2804                         aux.tp_vlan_tci = 0;
2805                 }
2806                 aux.tp_padding = 0;
2807                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2808         }
2809
2810         /*
2811          *      Free or return the buffer as appropriate. Again this
2812          *      hides all the races and re-entrancy issues from us.
2813          */
2814         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2815
2816 out_free:
2817         skb_free_datagram(sk, skb);
2818 out:
2819         return err;
2820 }
2821
2822 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2823                                int *uaddr_len, int peer)
2824 {
2825         struct net_device *dev;
2826         struct sock *sk = sock->sk;
2827
2828         if (peer)
2829                 return -EOPNOTSUPP;
2830
2831         uaddr->sa_family = AF_PACKET;
2832         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
2833         rcu_read_lock();
2834         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2835         if (dev)
2836                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
2837         rcu_read_unlock();
2838         *uaddr_len = sizeof(*uaddr);
2839
2840         return 0;
2841 }
2842
2843 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2844                           int *uaddr_len, int peer)
2845 {
2846         struct net_device *dev;
2847         struct sock *sk = sock->sk;
2848         struct packet_sock *po = pkt_sk(sk);
2849         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2850
2851         if (peer)
2852                 return -EOPNOTSUPP;
2853
2854         sll->sll_family = AF_PACKET;
2855         sll->sll_ifindex = po->ifindex;
2856         sll->sll_protocol = po->num;
2857         sll->sll_pkttype = 0;
2858         rcu_read_lock();
2859         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2860         if (dev) {
2861                 sll->sll_hatype = dev->type;
2862                 sll->sll_halen = dev->addr_len;
2863                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2864         } else {
2865                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
2866                 sll->sll_halen = 0;
2867         }
2868         rcu_read_unlock();
2869         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2870
2871         return 0;
2872 }
2873
2874 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2875                          int what)
2876 {
2877         switch (i->type) {
2878         case PACKET_MR_MULTICAST:
2879                 if (i->alen != dev->addr_len)
2880                         return -EINVAL;
2881                 if (what > 0)
2882                         return dev_mc_add(dev, i->addr);
2883                 else
2884                         return dev_mc_del(dev, i->addr);
2885                 break;
2886         case PACKET_MR_PROMISC:
2887                 return dev_set_promiscuity(dev, what);
2888                 break;
2889         case PACKET_MR_ALLMULTI:
2890                 return dev_set_allmulti(dev, what);
2891                 break;
2892         case PACKET_MR_UNICAST:
2893                 if (i->alen != dev->addr_len)
2894                         return -EINVAL;
2895                 if (what > 0)
2896                         return dev_uc_add(dev, i->addr);
2897                 else
2898                         return dev_uc_del(dev, i->addr);
2899                 break;
2900         default:
2901                 break;
2902         }
2903         return 0;
2904 }
2905
2906 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2907 {
2908         for ( ; i; i = i->next) {
2909                 if (i->ifindex == dev->ifindex)
2910                         packet_dev_mc(dev, i, what);
2911         }
2912 }
2913
2914 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2915 {
2916         struct packet_sock *po = pkt_sk(sk);
2917         struct packet_mclist *ml, *i;
2918         struct net_device *dev;
2919         int err;
2920
2921         rtnl_lock();
2922
2923         err = -ENODEV;
2924         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2925         if (!dev)
2926                 goto done;
2927
2928         err = -EINVAL;
2929         if (mreq->mr_alen > dev->addr_len)
2930                 goto done;
2931
2932         err = -ENOBUFS;
2933         i = kmalloc(sizeof(*i), GFP_KERNEL);
2934         if (i == NULL)
2935                 goto done;
2936
2937         err = 0;
2938         for (ml = po->mclist; ml; ml = ml->next) {
2939                 if (ml->ifindex == mreq->mr_ifindex &&
2940                     ml->type == mreq->mr_type &&
2941                     ml->alen == mreq->mr_alen &&
2942                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2943                         ml->count++;
2944                         /* Free the new element ... */
2945                         kfree(i);
2946                         goto done;
2947                 }
2948         }
2949
2950         i->type = mreq->mr_type;
2951         i->ifindex = mreq->mr_ifindex;
2952         i->alen = mreq->mr_alen;
2953         memcpy(i->addr, mreq->mr_address, i->alen);
2954         i->count = 1;
2955         i->next = po->mclist;
2956         po->mclist = i;
2957         err = packet_dev_mc(dev, i, 1);
2958         if (err) {
2959                 po->mclist = i->next;
2960                 kfree(i);
2961         }
2962
2963 done:
2964         rtnl_unlock();
2965         return err;
2966 }
2967
2968 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2969 {
2970         struct packet_mclist *ml, **mlp;
2971
2972         rtnl_lock();
2973
2974         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
2975                 if (ml->ifindex == mreq->mr_ifindex &&
2976                     ml->type == mreq->mr_type &&
2977                     ml->alen == mreq->mr_alen &&
2978                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2979                         if (--ml->count == 0) {
2980                                 struct net_device *dev;
2981                                 *mlp = ml->next;
2982                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2983                                 if (dev)
2984                                         packet_dev_mc(dev, ml, -1);
2985                                 kfree(ml);
2986                         }
2987                         rtnl_unlock();
2988                         return 0;
2989                 }
2990         }
2991         rtnl_unlock();
2992         return -EADDRNOTAVAIL;
2993 }
2994
2995 static void packet_flush_mclist(struct sock *sk)
2996 {
2997         struct packet_sock *po = pkt_sk(sk);
2998         struct packet_mclist *ml;
2999
3000         if (!po->mclist)
3001                 return;
3002
3003         rtnl_lock();
3004         while ((ml = po->mclist) != NULL) {
3005                 struct net_device *dev;
3006
3007                 po->mclist = ml->next;
3008                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3009                 if (dev != NULL)
3010                         packet_dev_mc(dev, ml, -1);
3011                 kfree(ml);
3012         }
3013         rtnl_unlock();
3014 }
3015
3016 static int
3017 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3018 {
3019         struct sock *sk = sock->sk;
3020         struct packet_sock *po = pkt_sk(sk);
3021         int ret;
3022
3023         if (level != SOL_PACKET)
3024                 return -ENOPROTOOPT;
3025
3026         switch (optname) {
3027         case PACKET_ADD_MEMBERSHIP:
3028         case PACKET_DROP_MEMBERSHIP:
3029         {
3030                 struct packet_mreq_max mreq;
3031                 int len = optlen;
3032                 memset(&mreq, 0, sizeof(mreq));
3033                 if (len < sizeof(struct packet_mreq))
3034                         return -EINVAL;
3035                 if (len > sizeof(mreq))
3036                         len = sizeof(mreq);
3037                 if (copy_from_user(&mreq, optval, len))
3038                         return -EFAULT;
3039                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3040                         return -EINVAL;
3041                 if (optname == PACKET_ADD_MEMBERSHIP)
3042                         ret = packet_mc_add(sk, &mreq);
3043                 else
3044                         ret = packet_mc_drop(sk, &mreq);
3045                 return ret;
3046         }
3047
3048         case PACKET_RX_RING:
3049         case PACKET_TX_RING:
3050         {
3051                 union tpacket_req_u req_u;
3052                 int len;
3053
3054                 switch (po->tp_version) {
3055                 case TPACKET_V1:
3056                 case TPACKET_V2:
3057                         len = sizeof(req_u.req);
3058                         break;
3059                 case TPACKET_V3:
3060                 default:
3061                         len = sizeof(req_u.req3);
3062                         break;
3063                 }
3064                 if (optlen < len)
3065                         return -EINVAL;
3066                 if (pkt_sk(sk)->has_vnet_hdr)
3067                         return -EINVAL;
3068                 if (copy_from_user(&req_u.req, optval, len))
3069                         return -EFAULT;
3070                 return packet_set_ring(sk, &req_u, 0,
3071                         optname == PACKET_TX_RING);
3072         }
3073         case PACKET_COPY_THRESH:
3074         {
3075                 int val;
3076
3077                 if (optlen != sizeof(val))
3078                         return -EINVAL;
3079                 if (copy_from_user(&val, optval, sizeof(val)))
3080                         return -EFAULT;
3081
3082                 pkt_sk(sk)->copy_thresh = val;
3083                 return 0;
3084         }
3085         case PACKET_VERSION:
3086         {
3087                 int val;
3088
3089                 if (optlen != sizeof(val))
3090                         return -EINVAL;
3091                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3092                         return -EBUSY;
3093                 if (copy_from_user(&val, optval, sizeof(val)))
3094                         return -EFAULT;
3095                 switch (val) {
3096                 case TPACKET_V1:
3097                 case TPACKET_V2:
3098                 case TPACKET_V3:
3099                         po->tp_version = val;
3100                         return 0;
3101                 default:
3102                         return -EINVAL;
3103                 }
3104         }
3105         case PACKET_RESERVE:
3106         {
3107                 unsigned int val;
3108
3109                 if (optlen != sizeof(val))
3110                         return -EINVAL;
3111                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3112                         return -EBUSY;
3113                 if (copy_from_user(&val, optval, sizeof(val)))
3114                         return -EFAULT;
3115                 po->tp_reserve = val;
3116                 return 0;
3117         }
3118         case PACKET_LOSS:
3119         {
3120                 unsigned int val;
3121
3122                 if (optlen != sizeof(val))
3123                         return -EINVAL;
3124                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3125                         return -EBUSY;
3126                 if (copy_from_user(&val, optval, sizeof(val)))
3127                         return -EFAULT;
3128                 po->tp_loss = !!val;
3129                 return 0;
3130         }
3131         case PACKET_AUXDATA:
3132         {
3133                 int val;
3134
3135                 if (optlen < sizeof(val))
3136                         return -EINVAL;
3137                 if (copy_from_user(&val, optval, sizeof(val)))
3138                         return -EFAULT;
3139
3140                 po->auxdata = !!val;
3141                 return 0;
3142         }
3143         case PACKET_ORIGDEV:
3144         {
3145                 int val;
3146
3147                 if (optlen < sizeof(val))
3148                         return -EINVAL;
3149                 if (copy_from_user(&val, optval, sizeof(val)))
3150                         return -EFAULT;
3151
3152                 po->origdev = !!val;
3153                 return 0;
3154         }
3155         case PACKET_VNET_HDR:
3156         {
3157                 int val;
3158
3159                 if (sock->type != SOCK_RAW)
3160                         return -EINVAL;
3161                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3162                         return -EBUSY;
3163                 if (optlen < sizeof(val))
3164                         return -EINVAL;
3165                 if (copy_from_user(&val, optval, sizeof(val)))
3166                         return -EFAULT;
3167
3168                 po->has_vnet_hdr = !!val;
3169                 return 0;
3170         }
3171         case PACKET_TIMESTAMP:
3172         {
3173                 int val;
3174
3175                 if (optlen != sizeof(val))
3176                         return -EINVAL;
3177                 if (copy_from_user(&val, optval, sizeof(val)))
3178                         return -EFAULT;
3179
3180                 po->tp_tstamp = val;
3181                 return 0;
3182         }
3183         case PACKET_FANOUT:
3184         {
3185                 int val;
3186
3187                 if (optlen != sizeof(val))
3188                         return -EINVAL;
3189                 if (copy_from_user(&val, optval, sizeof(val)))
3190                         return -EFAULT;
3191
3192                 return fanout_add(sk, val & 0xffff, val >> 16);
3193         }
3194         case PACKET_TX_HAS_OFF:
3195         {
3196                 unsigned int val;
3197
3198                 if (optlen != sizeof(val))
3199                         return -EINVAL;
3200                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3201                         return -EBUSY;
3202                 if (copy_from_user(&val, optval, sizeof(val)))
3203                         return -EFAULT;
3204                 po->tp_tx_has_off = !!val;
3205                 return 0;
3206         }
3207         default:
3208                 return -ENOPROTOOPT;
3209         }
3210 }
3211
3212 static int packet_getsockopt(struct socket *sock, int level, int optname,
3213                              char __user *optval, int __user *optlen)
3214 {
3215         int len;
3216         int val, lv = sizeof(val);
3217         struct sock *sk = sock->sk;
3218         struct packet_sock *po = pkt_sk(sk);
3219         void *data = &val;
3220         union tpacket_stats_u st;
3221
3222         if (level != SOL_PACKET)
3223                 return -ENOPROTOOPT;
3224
3225         if (get_user(len, optlen))
3226                 return -EFAULT;
3227
3228         if (len < 0)
3229                 return -EINVAL;
3230
3231         switch (optname) {
3232         case PACKET_STATISTICS:
3233                 spin_lock_bh(&sk->sk_receive_queue.lock);
3234                 memcpy(&st, &po->stats, sizeof(st));
3235                 memset(&po->stats, 0, sizeof(po->stats));
3236                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3237
3238                 if (po->tp_version == TPACKET_V3) {
3239                         lv = sizeof(struct tpacket_stats_v3);
3240                         st.stats3.tp_packets += st.stats3.tp_drops;
3241                         data = &st.stats3;
3242                 } else {
3243                         lv = sizeof(struct tpacket_stats);
3244                         st.stats1.tp_packets += st.stats1.tp_drops;
3245                         data = &st.stats1;
3246                 }
3247
3248                 break;
3249         case PACKET_AUXDATA:
3250                 val = po->auxdata;
3251                 break;
3252         case PACKET_ORIGDEV:
3253                 val = po->origdev;
3254                 break;
3255         case PACKET_VNET_HDR:
3256                 val = po->has_vnet_hdr;
3257                 break;
3258         case PACKET_VERSION:
3259                 val = po->tp_version;
3260                 break;
3261         case PACKET_HDRLEN:
3262                 if (len > sizeof(int))
3263                         len = sizeof(int);
3264                 if (copy_from_user(&val, optval, len))
3265                         return -EFAULT;
3266                 switch (val) {
3267                 case TPACKET_V1:
3268                         val = sizeof(struct tpacket_hdr);
3269                         break;
3270                 case TPACKET_V2:
3271                         val = sizeof(struct tpacket2_hdr);
3272                         break;
3273                 case TPACKET_V3:
3274                         val = sizeof(struct tpacket3_hdr);
3275                         break;
3276                 default:
3277                         return -EINVAL;
3278                 }
3279                 break;
3280         case PACKET_RESERVE:
3281                 val = po->tp_reserve;
3282                 break;
3283         case PACKET_LOSS:
3284                 val = po->tp_loss;
3285                 break;
3286         case PACKET_TIMESTAMP:
3287                 val = po->tp_tstamp;
3288                 break;
3289         case PACKET_FANOUT:
3290                 val = (po->fanout ?
3291                        ((u32)po->fanout->id |
3292                         ((u32)po->fanout->type << 16) |
3293                         ((u32)po->fanout->flags << 24)) :
3294                        0);
3295                 break;
3296         case PACKET_TX_HAS_OFF:
3297                 val = po->tp_tx_has_off;
3298                 break;
3299         default:
3300                 return -ENOPROTOOPT;
3301         }
3302
3303         if (len > lv)
3304                 len = lv;
3305         if (put_user(len, optlen))
3306                 return -EFAULT;
3307         if (copy_to_user(optval, data, len))
3308                 return -EFAULT;
3309         return 0;
3310 }
3311
3312
3313 static int packet_notifier(struct notifier_block *this,
3314                            unsigned long msg, void *ptr)
3315 {
3316         struct sock *sk;
3317         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3318         struct net *net = dev_net(dev);
3319
3320         rcu_read_lock();
3321         sk_for_each_rcu(sk, &net->packet.sklist) {
3322                 struct packet_sock *po = pkt_sk(sk);
3323
3324                 switch (msg) {
3325                 case NETDEV_UNREGISTER:
3326                         if (po->mclist)
3327                                 packet_dev_mclist(dev, po->mclist, -1);
3328                         /* fallthrough */
3329
3330                 case NETDEV_DOWN:
3331                         if (dev->ifindex == po->ifindex) {
3332                                 spin_lock(&po->bind_lock);
3333                                 if (po->running) {
3334                                         __unregister_prot_hook(sk, false);
3335                                         sk->sk_err = ENETDOWN;
3336                                         if (!sock_flag(sk, SOCK_DEAD))
3337                                                 sk->sk_error_report(sk);
3338                                 }
3339                                 if (msg == NETDEV_UNREGISTER) {
3340                                         po->ifindex = -1;
3341                                         if (po->prot_hook.dev)
3342                                                 dev_put(po->prot_hook.dev);
3343                                         po->prot_hook.dev = NULL;
3344                                 }
3345                                 spin_unlock(&po->bind_lock);
3346                         }
3347                         break;
3348                 case NETDEV_UP:
3349                         if (dev->ifindex == po->ifindex) {
3350                                 spin_lock(&po->bind_lock);
3351                                 if (po->num)
3352                                         register_prot_hook(sk);
3353                                 spin_unlock(&po->bind_lock);
3354                         }
3355                         break;
3356                 }
3357         }
3358         rcu_read_unlock();
3359         return NOTIFY_DONE;
3360 }
3361
3362
3363 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3364                         unsigned long arg)
3365 {
3366         struct sock *sk = sock->sk;
3367
3368         switch (cmd) {
3369         case SIOCOUTQ:
3370         {
3371                 int amount = sk_wmem_alloc_get(sk);
3372
3373                 return put_user(amount, (int __user *)arg);
3374         }
3375         case SIOCINQ:
3376         {
3377                 struct sk_buff *skb;
3378                 int amount = 0;
3379
3380                 spin_lock_bh(&sk->sk_receive_queue.lock);
3381                 skb = skb_peek(&sk->sk_receive_queue);
3382                 if (skb)
3383                         amount = skb->len;
3384                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3385                 return put_user(amount, (int __user *)arg);
3386         }
3387         case SIOCGSTAMP:
3388                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3389         case SIOCGSTAMPNS:
3390                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3391
3392 #ifdef CONFIG_INET
3393         case SIOCADDRT:
3394         case SIOCDELRT:
3395         case SIOCDARP:
3396         case SIOCGARP:
3397         case SIOCSARP:
3398         case SIOCGIFADDR:
3399         case SIOCSIFADDR:
3400         case SIOCGIFBRDADDR:
3401         case SIOCSIFBRDADDR:
3402         case SIOCGIFNETMASK:
3403         case SIOCSIFNETMASK:
3404         case SIOCGIFDSTADDR:
3405         case SIOCSIFDSTADDR:
3406         case SIOCSIFFLAGS:
3407                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3408 #endif
3409
3410         default:
3411                 return -ENOIOCTLCMD;
3412         }
3413         return 0;
3414 }
3415
3416 static unsigned int packet_poll(struct file *file, struct socket *sock,
3417                                 poll_table *wait)
3418 {
3419         struct sock *sk = sock->sk;
3420         struct packet_sock *po = pkt_sk(sk);
3421         unsigned int mask = datagram_poll(file, sock, wait);
3422
3423         spin_lock_bh(&sk->sk_receive_queue.lock);
3424         if (po->rx_ring.pg_vec) {
3425                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3426                         TP_STATUS_KERNEL))
3427                         mask |= POLLIN | POLLRDNORM;
3428         }
3429         spin_unlock_bh(&sk->sk_receive_queue.lock);
3430         spin_lock_bh(&sk->sk_write_queue.lock);
3431         if (po->tx_ring.pg_vec) {
3432                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3433                         mask |= POLLOUT | POLLWRNORM;
3434         }
3435         spin_unlock_bh(&sk->sk_write_queue.lock);
3436         return mask;
3437 }
3438
3439
3440 /* Dirty? Well, I still did not learn better way to account
3441  * for user mmaps.
3442  */
3443
3444 static void packet_mm_open(struct vm_area_struct *vma)
3445 {
3446         struct file *file = vma->vm_file;
3447         struct socket *sock = file->private_data;
3448         struct sock *sk = sock->sk;
3449
3450         if (sk)
3451                 atomic_inc(&pkt_sk(sk)->mapped);
3452 }
3453
3454 static void packet_mm_close(struct vm_area_struct *vma)
3455 {
3456         struct file *file = vma->vm_file;
3457         struct socket *sock = file->private_data;
3458         struct sock *sk = sock->sk;
3459
3460         if (sk)
3461                 atomic_dec(&pkt_sk(sk)->mapped);
3462 }
3463
3464 static const struct vm_operations_struct packet_mmap_ops = {
3465         .open   =       packet_mm_open,
3466         .close  =       packet_mm_close,
3467 };
3468
3469 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3470                         unsigned int len)
3471 {
3472         int i;
3473
3474         for (i = 0; i < len; i++) {
3475                 if (likely(pg_vec[i].buffer)) {
3476                         if (is_vmalloc_addr(pg_vec[i].buffer))
3477                                 vfree(pg_vec[i].buffer);
3478                         else
3479                                 free_pages((unsigned long)pg_vec[i].buffer,
3480                                            order);
3481                         pg_vec[i].buffer = NULL;
3482                 }
3483         }
3484         kfree(pg_vec);
3485 }
3486
3487 static char *alloc_one_pg_vec_page(unsigned long order)
3488 {
3489         char *buffer = NULL;
3490         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3491                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3492
3493         buffer = (char *) __get_free_pages(gfp_flags, order);
3494
3495         if (buffer)
3496                 return buffer;
3497
3498         /*
3499          * __get_free_pages failed, fall back to vmalloc
3500          */
3501         buffer = vzalloc((1 << order) * PAGE_SIZE);
3502
3503         if (buffer)
3504                 return buffer;
3505
3506         /*
3507          * vmalloc failed, lets dig into swap here
3508          */
3509         gfp_flags &= ~__GFP_NORETRY;
3510         buffer = (char *)__get_free_pages(gfp_flags, order);
3511         if (buffer)
3512                 return buffer;
3513
3514         /*
3515          * complete and utter failure
3516          */
3517         return NULL;
3518 }
3519
3520 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3521 {
3522         unsigned int block_nr = req->tp_block_nr;
3523         struct pgv *pg_vec;
3524         int i;
3525
3526         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3527         if (unlikely(!pg_vec))
3528                 goto out;
3529
3530         for (i = 0; i < block_nr; i++) {
3531                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3532                 if (unlikely(!pg_vec[i].buffer))
3533                         goto out_free_pgvec;
3534         }
3535
3536 out:
3537         return pg_vec;
3538
3539 out_free_pgvec:
3540         free_pg_vec(pg_vec, order, block_nr);
3541         pg_vec = NULL;
3542         goto out;
3543 }
3544
3545 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3546                 int closing, int tx_ring)
3547 {
3548         struct pgv *pg_vec = NULL;
3549         struct packet_sock *po = pkt_sk(sk);
3550         int was_running, order = 0;
3551         struct packet_ring_buffer *rb;
3552         struct sk_buff_head *rb_queue;
3553         __be16 num;
3554         int err = -EINVAL;
3555         /* Added to avoid minimal code churn */
3556         struct tpacket_req *req = &req_u->req;
3557
3558         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3559         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3560                 WARN(1, "Tx-ring is not supported.\n");
3561                 goto out;
3562         }
3563
3564         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3565         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3566
3567         err = -EBUSY;
3568         if (!closing) {
3569                 if (atomic_read(&po->mapped))
3570                         goto out;
3571                 if (atomic_read(&rb->pending))
3572                         goto out;
3573         }
3574
3575         if (req->tp_block_nr) {
3576                 /* Sanity tests and some calculations */
3577                 err = -EBUSY;
3578                 if (unlikely(rb->pg_vec))
3579                         goto out;
3580
3581                 switch (po->tp_version) {
3582                 case TPACKET_V1:
3583                         po->tp_hdrlen = TPACKET_HDRLEN;
3584                         break;
3585                 case TPACKET_V2:
3586                         po->tp_hdrlen = TPACKET2_HDRLEN;
3587                         break;
3588                 case TPACKET_V3:
3589                         po->tp_hdrlen = TPACKET3_HDRLEN;
3590                         break;
3591                 }
3592
3593                 err = -EINVAL;
3594                 if (unlikely((int)req->tp_block_size <= 0))
3595                         goto out;
3596                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3597                         goto out;
3598                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3599                                         po->tp_reserve))
3600                         goto out;
3601                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3602                         goto out;
3603
3604                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3605                 if (unlikely(rb->frames_per_block <= 0))
3606                         goto out;
3607                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3608                                         req->tp_frame_nr))
3609                         goto out;
3610
3611                 err = -ENOMEM;
3612                 order = get_order(req->tp_block_size);
3613                 pg_vec = alloc_pg_vec(req, order);
3614                 if (unlikely(!pg_vec))
3615                         goto out;
3616                 switch (po->tp_version) {
3617                 case TPACKET_V3:
3618                 /* Transmit path is not supported. We checked
3619                  * it above but just being paranoid
3620                  */
3621                         if (!tx_ring)
3622                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3623                                 break;
3624                 default:
3625                         break;
3626                 }
3627         }
3628         /* Done */
3629         else {
3630                 err = -EINVAL;
3631                 if (unlikely(req->tp_frame_nr))
3632                         goto out;
3633         }
3634
3635         lock_sock(sk);
3636
3637         /* Detach socket from network */
3638         spin_lock(&po->bind_lock);
3639         was_running = po->running;
3640         num = po->num;
3641         if (was_running) {
3642                 po->num = 0;
3643                 __unregister_prot_hook(sk, false);
3644         }
3645         spin_unlock(&po->bind_lock);
3646
3647         synchronize_net();
3648
3649         err = -EBUSY;
3650         mutex_lock(&po->pg_vec_lock);
3651         if (closing || atomic_read(&po->mapped) == 0) {
3652                 err = 0;
3653                 spin_lock_bh(&rb_queue->lock);
3654                 swap(rb->pg_vec, pg_vec);
3655                 rb->frame_max = (req->tp_frame_nr - 1);
3656                 rb->head = 0;
3657                 rb->frame_size = req->tp_frame_size;
3658                 spin_unlock_bh(&rb_queue->lock);
3659
3660                 swap(rb->pg_vec_order, order);
3661                 swap(rb->pg_vec_len, req->tp_block_nr);
3662
3663                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3664                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3665                                                 tpacket_rcv : packet_rcv;
3666                 skb_queue_purge(rb_queue);
3667                 if (atomic_read(&po->mapped))
3668                         pr_err("packet_mmap: vma is busy: %d\n",
3669                                atomic_read(&po->mapped));
3670         }
3671         mutex_unlock(&po->pg_vec_lock);
3672
3673         spin_lock(&po->bind_lock);
3674         if (was_running) {
3675                 po->num = num;
3676                 register_prot_hook(sk);
3677         }
3678         spin_unlock(&po->bind_lock);
3679         if (closing && (po->tp_version > TPACKET_V2)) {
3680                 /* Because we don't support block-based V3 on tx-ring */
3681                 if (!tx_ring)
3682                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3683         }
3684         release_sock(sk);
3685
3686         if (pg_vec)
3687                 free_pg_vec(pg_vec, order, req->tp_block_nr);
3688 out:
3689         return err;
3690 }
3691
3692 static int packet_mmap(struct file *file, struct socket *sock,
3693                 struct vm_area_struct *vma)
3694 {
3695         struct sock *sk = sock->sk;
3696         struct packet_sock *po = pkt_sk(sk);
3697         unsigned long size, expected_size;
3698         struct packet_ring_buffer *rb;
3699         unsigned long start;
3700         int err = -EINVAL;
3701         int i;
3702
3703         if (vma->vm_pgoff)
3704                 return -EINVAL;
3705
3706         mutex_lock(&po->pg_vec_lock);
3707
3708         expected_size = 0;
3709         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3710                 if (rb->pg_vec) {
3711                         expected_size += rb->pg_vec_len
3712                                                 * rb->pg_vec_pages
3713                                                 * PAGE_SIZE;
3714                 }
3715         }
3716
3717         if (expected_size == 0)
3718                 goto out;
3719
3720         size = vma->vm_end - vma->vm_start;
3721         if (size != expected_size)
3722                 goto out;
3723
3724         start = vma->vm_start;
3725         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3726                 if (rb->pg_vec == NULL)
3727                         continue;
3728
3729                 for (i = 0; i < rb->pg_vec_len; i++) {
3730                         struct page *page;
3731                         void *kaddr = rb->pg_vec[i].buffer;
3732                         int pg_num;
3733
3734                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3735                                 page = pgv_to_page(kaddr);
3736                                 err = vm_insert_page(vma, start, page);
3737                                 if (unlikely(err))
3738                                         goto out;
3739                                 start += PAGE_SIZE;
3740                                 kaddr += PAGE_SIZE;
3741                         }
3742                 }
3743         }
3744
3745         atomic_inc(&po->mapped);
3746         vma->vm_ops = &packet_mmap_ops;
3747         err = 0;
3748
3749 out:
3750         mutex_unlock(&po->pg_vec_lock);
3751         return err;
3752 }
3753
3754 static const struct proto_ops packet_ops_spkt = {
3755         .family =       PF_PACKET,
3756         .owner =        THIS_MODULE,
3757         .release =      packet_release,
3758         .bind =         packet_bind_spkt,
3759         .connect =      sock_no_connect,
3760         .socketpair =   sock_no_socketpair,
3761         .accept =       sock_no_accept,
3762         .getname =      packet_getname_spkt,
3763         .poll =         datagram_poll,
3764         .ioctl =        packet_ioctl,
3765         .listen =       sock_no_listen,
3766         .shutdown =     sock_no_shutdown,
3767         .setsockopt =   sock_no_setsockopt,
3768         .getsockopt =   sock_no_getsockopt,
3769         .sendmsg =      packet_sendmsg_spkt,
3770         .recvmsg =      packet_recvmsg,
3771         .mmap =         sock_no_mmap,
3772         .sendpage =     sock_no_sendpage,
3773 };
3774
3775 static const struct proto_ops packet_ops = {
3776         .family =       PF_PACKET,
3777         .owner =        THIS_MODULE,
3778         .release =      packet_release,
3779         .bind =         packet_bind,
3780         .connect =      sock_no_connect,
3781         .socketpair =   sock_no_socketpair,
3782         .accept =       sock_no_accept,
3783         .getname =      packet_getname,
3784         .poll =         packet_poll,
3785         .ioctl =        packet_ioctl,
3786         .listen =       sock_no_listen,
3787         .shutdown =     sock_no_shutdown,
3788         .setsockopt =   packet_setsockopt,
3789         .getsockopt =   packet_getsockopt,
3790         .sendmsg =      packet_sendmsg,
3791         .recvmsg =      packet_recvmsg,
3792         .mmap =         packet_mmap,
3793         .sendpage =     sock_no_sendpage,
3794 };
3795
3796 static const struct net_proto_family packet_family_ops = {
3797         .family =       PF_PACKET,
3798         .create =       packet_create,
3799         .owner  =       THIS_MODULE,
3800 };
3801
3802 static struct notifier_block packet_netdev_notifier = {
3803         .notifier_call =        packet_notifier,
3804 };
3805
3806 #ifdef CONFIG_PROC_FS
3807
3808 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3809         __acquires(RCU)
3810 {
3811         struct net *net = seq_file_net(seq);
3812
3813         rcu_read_lock();
3814         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3815 }
3816
3817 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3818 {
3819         struct net *net = seq_file_net(seq);
3820         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3821 }
3822
3823 static void packet_seq_stop(struct seq_file *seq, void *v)
3824         __releases(RCU)
3825 {
3826         rcu_read_unlock();
3827 }
3828
3829 static int packet_seq_show(struct seq_file *seq, void *v)
3830 {
3831         if (v == SEQ_START_TOKEN)
3832                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
3833         else {
3834                 struct sock *s = sk_entry(v);
3835                 const struct packet_sock *po = pkt_sk(s);
3836
3837                 seq_printf(seq,
3838                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
3839                            s,
3840                            atomic_read(&s->sk_refcnt),
3841                            s->sk_type,
3842                            ntohs(po->num),
3843                            po->ifindex,
3844                            po->running,
3845                            atomic_read(&s->sk_rmem_alloc),
3846                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
3847                            sock_i_ino(s));
3848         }
3849
3850         return 0;
3851 }
3852
3853 static const struct seq_operations packet_seq_ops = {
3854         .start  = packet_seq_start,
3855         .next   = packet_seq_next,
3856         .stop   = packet_seq_stop,
3857         .show   = packet_seq_show,
3858 };
3859
3860 static int packet_seq_open(struct inode *inode, struct file *file)
3861 {
3862         return seq_open_net(inode, file, &packet_seq_ops,
3863                             sizeof(struct seq_net_private));
3864 }
3865
3866 static const struct file_operations packet_seq_fops = {
3867         .owner          = THIS_MODULE,
3868         .open           = packet_seq_open,
3869         .read           = seq_read,
3870         .llseek         = seq_lseek,
3871         .release        = seq_release_net,
3872 };
3873
3874 #endif
3875
3876 static int __net_init packet_net_init(struct net *net)
3877 {
3878         mutex_init(&net->packet.sklist_lock);
3879         INIT_HLIST_HEAD(&net->packet.sklist);
3880
3881         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
3882                 return -ENOMEM;
3883
3884         return 0;
3885 }
3886
3887 static void __net_exit packet_net_exit(struct net *net)
3888 {
3889         remove_proc_entry("packet", net->proc_net);
3890 }
3891
3892 static struct pernet_operations packet_net_ops = {
3893         .init = packet_net_init,
3894         .exit = packet_net_exit,
3895 };
3896
3897
3898 static void __exit packet_exit(void)
3899 {
3900         unregister_netdevice_notifier(&packet_netdev_notifier);
3901         unregister_pernet_subsys(&packet_net_ops);
3902         sock_unregister(PF_PACKET);
3903         proto_unregister(&packet_proto);
3904 }
3905
3906 static int __init packet_init(void)
3907 {
3908         int rc = proto_register(&packet_proto, 0);
3909
3910         if (rc != 0)
3911                 goto out;
3912
3913         sock_register(&packet_family_ops);
3914         register_pernet_subsys(&packet_net_ops);
3915         register_netdevice_notifier(&packet_netdev_notifier);
3916 out:
3917         return rc;
3918 }
3919
3920 module_init(packet_init);
3921 module_exit(packet_exit);
3922 MODULE_LICENSE("GPL");
3923 MODULE_ALIAS_NETPROTO(PF_PACKET);