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