net/packet/af_packet.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              PACKET - implements raw packet sockets.
   7  *
   8  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  11  *
  12  * Fixes:
  13  *              Alan Cox        :       verify_area() now used correctly
  14  *              Alan Cox        :       new skbuff lists, look ma no backlogs!
  15  *              Alan Cox        :       tidied skbuff lists.
  16  *              Alan Cox        :       Now uses generic datagram routines I
  17  *                                      added. Also fixed the peek/read crash
  18  *                                      from all old Linux datagram code.
  19  *              Alan Cox        :       Uses the improved datagram code.
  20  *              Alan Cox        :       Added NULL's for socket options.
  21  *              Alan Cox        :       Re-commented the code.
  22  *              Alan Cox        :       Use new kernel side addressing
  23  *              Rob Janssen     :       Correct MTU usage.
  24  *              Dave Platt      :       Counter leaks caused by incorrect
  25  *                                      interrupt locking and some slightly
  26  *                                      dubious gcc output. Can you read
  27  *                                      compiler: it said _VOLATILE_
  28  *      Richard Kooijman        :       Timestamp fixes.
  29  *              Alan Cox        :       New buffers. Use sk->mac.raw.
  30  *              Alan Cox        :       sendmsg/recvmsg support.
  31  *              Alan Cox        :       Protocol setting support
  32  *      Alexey Kuznetsov        :       Untied from IPv4 stack.
  33  *      Cyrus Durgin            :       Fixed kerneld for kmod.
  34  *      Michal Ostrowski        :       Module initialization cleanup.
  35  *         Ulises Alonso        :       Frame number limit removal and
  36  *                                      packet_set_ring memory leak.
  37  *              Eric Biederman  :       Allow for > 8 byte hardware addresses.
  38  *                                      The convention is that longer addresses
  39  *                                      will simply extend the hardware address
  40  *                                      byte arrays at the end of sockaddr_ll
  41  *                                      and packet_mreq.
  42  *              Johann Baudy    :       Added TX RING.
  43  *              Chetan Loke     :       Implemented TPACKET_V3 block abstraction
  44  *                                      layer.
  45  *                                      Copyright (C) 2011, <lokec@ccs.neu.edu>
  46  *
  47  *
  48  *              This program is free software; you can redistribute it and/or
  49  *              modify it under the terms of the GNU General Public License
  50  *              as published by the Free Software Foundation; either version
  51  *              2 of the License, or (at your option) any later version.
  52  *
  53  */
  54
  55 #include <linux/types.h>
  56 #include <linux/mm.h>
  57 #include <linux/capability.h>
  58 #include <linux/fcntl.h>
  59 #include <linux/socket.h>
  60 #include <linux/in.h>
  61 #include <linux/inet.h>
  62 #include <linux/netdevice.h>
  63 #include <linux/if_packet.h>
  64 #include <linux/wireless.h>
  65 #include <linux/kernel.h>
  66 #include <linux/kmod.h>
  67 #include <linux/slab.h>
  68 #include <linux/vmalloc.h>
  69 #include <net/net_namespace.h>
  70 #include <net/ip.h>
  71 #include <net/protocol.h>
  72 #include <linux/skbuff.h>
  73 #include <net/sock.h>
  74 #include <linux/errno.h>
  75 #include <linux/timer.h>
  76 #include <asm/uaccess.h>
  77 #include <asm/ioctls.h>
  78 #include <asm/page.h>
  79 #include <asm/cacheflush.h>
  80 #include <asm/io.h>
  81 #include <linux/proc_fs.h>
  82 #include <linux/seq_file.h>
  83 #include <linux/poll.h>
  84 #include <linux/module.h>
  85 #include <linux/init.h>
  86 #include <linux/mutex.h>
  87 #include <linux/if_vlan.h>
  88 #include <linux/virtio_net.h>
  89 #include <linux/errqueue.h>
  90 #include <linux/net_tstamp.h>
  91 #include <linux/reciprocal_div.h>
  92 #ifdef CONFIG_INET
  93 #include <net/inet_common.h>
  94 #endif
  95
  96 #include "internal.h"
  97
  98 /*
  99    Assumptions:
 100    - if device has no dev->hard_header routine, it adds and removes ll header
 101      inside itself. In this case ll header is invisible outside of device,
 102      but higher levels still should reserve dev->hard_header_len.
 103      Some devices are enough clever to reallocate skb, when header
 104      will not fit to reserved space (tunnel), another ones are silly
 105      (PPP).
 106    - packet socket receives packets with pulled ll header,
 107      so that SOCK_RAW should push it back.
 108
 109 On receive:
 110 -----------
 111
 112 Incoming, dev->hard_header!=NULL
 113    mac_header -> ll header
 114    data       -> data
 115
 116 Outgoing, dev->hard_header!=NULL
 117    mac_header -> ll header
 118    data       -> ll header
 119
 120 Incoming, dev->hard_header==NULL
 121    mac_header -> UNKNOWN position. It is very likely, that it points to ll
 122                  header.  PPP makes it, that is wrong, because introduce
 123                  assymetry between rx and tx paths.
 124    data       -> data
 125
 126 Outgoing, dev->hard_header==NULL
 127    mac_header -> data. ll header is still not built!
 128    data       -> data
 129
 130 Resume
 131   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 132
 133
 134 On transmit:
 135 ------------
 136
 137 dev->hard_header != NULL
 138    mac_header -> ll header
 139    data       -> ll header
 140
 141 dev->hard_header == NULL (ll header is added by device, we cannot control it)
 142    mac_header -> data
 143    data       -> data
 144
 145    We should set nh.raw on output to correct posistion,
 146    packet classifier depends on it.
 147  */
 148
 149 /* Private packet socket structures. */
 150
 151 /* identical to struct packet_mreq except it has
 152  * a longer address field.
 153  */
 154 struct packet_mreq_max {
 155         int             mr_ifindex;
 156         unsigned short  mr_type;
 157         unsigned short  mr_alen;
 158         unsigned char   mr_address[MAX_ADDR_LEN];
 159 };
 160
 161 union tpacket_uhdr {
 162         struct tpacket_hdr  *h1;
 163         struct tpacket2_hdr *h2;
 164         struct tpacket3_hdr *h3;
 165         void *raw;
 166 };
 167
 168 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 169                 int closing, int tx_ring);
 170
 171 #define V3_ALIGNMENT    (8)
 172
 173 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 174
 175 #define BLK_PLUS_PRIV(sz_of_priv) \
 176         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 177
 178 #define PGV_FROM_VMALLOC 1
 179
 180 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
 181 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
 182 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
 183 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
 184 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
 185 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
 186 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
 187
 188 struct packet_sock;
 189 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
 190 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 191                        struct packet_type *pt, struct net_device *orig_dev);
 192
 193 static void *packet_previous_frame(struct packet_sock *po,
 194                 struct packet_ring_buffer *rb,
 195                 int status);
 196 static void packet_increment_head(struct packet_ring_buffer *buff);
 197 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
 198                         struct tpacket_block_desc *);
 199 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 200                         struct packet_sock *);
 201 static void prb_retire_current_block(struct tpacket_kbdq_core *,
 202                 struct packet_sock *, unsigned int status);
 203 static int prb_queue_frozen(struct tpacket_kbdq_core *);
 204 static void prb_open_block(struct tpacket_kbdq_core *,
 205                 struct tpacket_block_desc *);
 206 static void prb_retire_rx_blk_timer_expired(unsigned long);
 207 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 208 static void prb_init_blk_timer(struct packet_sock *,
 209                 struct tpacket_kbdq_core *,
 210                 void (*func) (unsigned long));
 211 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 212 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 213                 struct tpacket3_hdr *);
 214 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 215                 struct tpacket3_hdr *);
 216 static void packet_flush_mclist(struct sock *sk);
 217
 218 struct packet_skb_cb {
 219         unsigned int origlen;
 220         union {
 221                 struct sockaddr_pkt pkt;
 222                 struct sockaddr_ll ll;
 223         } sa;
 224 };
 225
 226 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 227
 228 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 229 #define GET_PBLOCK_DESC(x, bid) \
 230         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 231 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 232         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 233 #define GET_NEXT_PRB_BLK_NUM(x) \
 234         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 235         ((x)->kactive_blk_num+1) : 0)
 236
 237 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 238 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 239
 240 /* register_prot_hook must be invoked with the po->bind_lock held,
 241  * or from a context in which asynchronous accesses to the packet
 242  * socket is not possible (packet_create()).
 243  */
 244 static void register_prot_hook(struct sock *sk)
 245 {
 246         struct packet_sock *po = pkt_sk(sk);
 247         if (!po->running) {
 248                 if (po->fanout)
 249                         __fanout_link(sk, po);
 250                 else
 251                         dev_add_pack(&po->prot_hook);
 252                 sock_hold(sk);
 253                 po->running = 1;
 254         }
 255 }
 256
 257 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
 258  * held.   If the sync parameter is true, we will temporarily drop
 259  * the po->bind_lock and do a synchronize_net to make sure no
 260  * asynchronous packet processing paths still refer to the elements
 261  * of po->prot_hook.  If the sync parameter is false, it is the
 262  * callers responsibility to take care of this.
 263  */
 264 static void __unregister_prot_hook(struct sock *sk, bool sync)
 265 {
 266         struct packet_sock *po = pkt_sk(sk);
 267
 268         po->running = 0;
 269         if (po->fanout)
 270                 __fanout_unlink(sk, po);
 271         else
 272                 __dev_remove_pack(&po->prot_hook);
 273         __sock_put(sk);
 274
 275         if (sync) {
 276                 spin_unlock(&po->bind_lock);
 277                 synchronize_net();
 278                 spin_lock(&po->bind_lock);
 279         }
 280 }
 281
 282 static void unregister_prot_hook(struct sock *sk, bool sync)
 283 {
 284         struct packet_sock *po = pkt_sk(sk);
 285
 286         if (po->running)
 287                 __unregister_prot_hook(sk, sync);
 288 }
 289
 290 static inline __pure struct page *pgv_to_page(void *addr)
 291 {
 292         if (is_vmalloc_addr(addr))
 293                 return vmalloc_to_page(addr);
 294         return virt_to_page(addr);
 295 }
 296
 297 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 298 {
 299         union tpacket_uhdr h;
 300
 301         h.raw = frame;
 302         switch (po->tp_version) {
 303         case TPACKET_V1:
 304                 h.h1->tp_status = status;
 305                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 306                 break;
 307         case TPACKET_V2:
 308                 h.h2->tp_status = status;
 309                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 310                 break;
 311         case TPACKET_V3:
 312         default:
 313                 WARN(1, "TPACKET version not supported.\n");
 314                 BUG();
 315         }
 316
 317         smp_wmb();
 318 }
 319
 320 static int __packet_get_status(struct packet_sock *po, void *frame)
 321 {
 322         union tpacket_uhdr h;
 323
 324         smp_rmb();
 325
 326         h.raw = frame;
 327         switch (po->tp_version) {
 328         case TPACKET_V1:
 329                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 330                 return h.h1->tp_status;
 331         case TPACKET_V2:
 332                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 333                 return h.h2->tp_status;
 334         case TPACKET_V3:
 335         default:
 336                 WARN(1, "TPACKET version not supported.\n");
 337                 BUG();
 338                 return 0;
 339         }
 340 }
 341
 342 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
 343                                    unsigned int flags)
 344 {
 345         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 346
 347         if (shhwtstamps) {
 348                 if ((flags & SOF_TIMESTAMPING_SYS_HARDWARE) &&
 349                     ktime_to_timespec_cond(shhwtstamps->syststamp, ts))
 350                         return TP_STATUS_TS_SYS_HARDWARE;
 351                 if ((flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
 352                     ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
 353                         return TP_STATUS_TS_RAW_HARDWARE;
 354         }
 355
 356         if (ktime_to_timespec_cond(skb->tstamp, ts))
 357                 return TP_STATUS_TS_SOFTWARE;
 358
 359         return 0;
 360 }
 361
 362 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
 363                                     struct sk_buff *skb)
 364 {
 365         union tpacket_uhdr h;
 366         struct timespec ts;
 367         __u32 ts_status;
 368
 369         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
 370                 return 0;
 371
 372         h.raw = frame;
 373         switch (po->tp_version) {
 374         case TPACKET_V1:
 375                 h.h1->tp_sec = ts.tv_sec;
 376                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
 377                 break;
 378         case TPACKET_V2:
 379                 h.h2->tp_sec = ts.tv_sec;
 380                 h.h2->tp_nsec = ts.tv_nsec;
 381                 break;
 382         case TPACKET_V3:
 383         default:
 384                 WARN(1, "TPACKET version not supported.\n");
 385                 BUG();
 386         }
 387
 388         /* one flush is safe, as both fields always lie on the same cacheline */
 389         flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
 390         smp_wmb();
 391
 392         return ts_status;
 393 }
 394
 395 static void *packet_lookup_frame(struct packet_sock *po,
 396                 struct packet_ring_buffer *rb,
 397                 unsigned int position,
 398                 int status)
 399 {
 400         unsigned int pg_vec_pos, frame_offset;
 401         union tpacket_uhdr h;
 402
 403         pg_vec_pos = position / rb->frames_per_block;
 404         frame_offset = position % rb->frames_per_block;
 405
 406         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 407                 (frame_offset * rb->frame_size);
 408
 409         if (status != __packet_get_status(po, h.raw))
 410                 return NULL;
 411
 412         return h.raw;
 413 }
 414
 415 static void *packet_current_frame(struct packet_sock *po,
 416                 struct packet_ring_buffer *rb,
 417                 int status)
 418 {
 419         return packet_lookup_frame(po, rb, rb->head, status);
 420 }
 421
 422 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 423 {
 424         del_timer_sync(&pkc->retire_blk_timer);
 425 }
 426
 427 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 428                 int tx_ring,
 429                 struct sk_buff_head *rb_queue)
 430 {
 431         struct tpacket_kbdq_core *pkc;
 432
 433         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 434
 435         spin_lock(&rb_queue->lock);
 436         pkc->delete_blk_timer = 1;
 437         spin_unlock(&rb_queue->lock);
 438
 439         prb_del_retire_blk_timer(pkc);
 440 }
 441
 442 static void prb_init_blk_timer(struct packet_sock *po,
 443                 struct tpacket_kbdq_core *pkc,
 444                 void (*func) (unsigned long))
 445 {
 446         init_timer(&pkc->retire_blk_timer);
 447         pkc->retire_blk_timer.data = (long)po;
 448         pkc->retire_blk_timer.function = func;
 449         pkc->retire_blk_timer.expires = jiffies;
 450 }
 451
 452 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
 453 {
 454         struct tpacket_kbdq_core *pkc;
 455
 456         if (tx_ring)
 457                 BUG();
 458
 459         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 460         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
 461 }
 462
 463 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 464                                 int blk_size_in_bytes)
 465 {
 466         struct net_device *dev;
 467         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 468         struct ethtool_cmd ecmd;
 469         int err;
 470         u32 speed;
 471
 472         rtnl_lock();
 473         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 474         if (unlikely(!dev)) {
 475                 rtnl_unlock();
 476                 return DEFAULT_PRB_RETIRE_TOV;
 477         }
 478         err = __ethtool_get_settings(dev, &ecmd);
 479         speed = ethtool_cmd_speed(&ecmd);
 480         rtnl_unlock();
 481         if (!err) {
 482                 /*
 483                  * If the link speed is so slow you don't really
 484                  * need to worry about perf anyways
 485                  */
 486                 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
 487                         return DEFAULT_PRB_RETIRE_TOV;
 488                 } else {
 489                         msec = 1;
 490                         div = speed / 1000;
 491                 }
 492         }
 493
 494         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 495
 496         if (div)
 497                 mbits /= div;
 498
 499         tmo = mbits * msec;
 500
 501         if (div)
 502                 return tmo+1;
 503         return tmo;
 504 }
 505
 506 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 507                         union tpacket_req_u *req_u)
 508 {
 509         p1->feature_req_word = req_u->req3.tp_feature_req_word;
 510 }
 511
 512 static void init_prb_bdqc(struct packet_sock *po,
 513                         struct packet_ring_buffer *rb,
 514                         struct pgv *pg_vec,
 515                         union tpacket_req_u *req_u, int tx_ring)
 516 {
 517         struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
 518         struct tpacket_block_desc *pbd;
 519
 520         memset(p1, 0x0, sizeof(*p1));
 521
 522         p1->knxt_seq_num = 1;
 523         p1->pkbdq = pg_vec;
 524         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 525         p1->pkblk_start = pg_vec[0].buffer;
 526         p1->kblk_size = req_u->req3.tp_block_size;
 527         p1->knum_blocks = req_u->req3.tp_block_nr;
 528         p1->hdrlen = po->tp_hdrlen;
 529         p1->version = po->tp_version;
 530         p1->last_kactive_blk_num = 0;
 531         po->stats.stats3.tp_freeze_q_cnt = 0;
 532         if (req_u->req3.tp_retire_blk_tov)
 533                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 534         else
 535                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 536                                                 req_u->req3.tp_block_size);
 537         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 538         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 539
 540         prb_init_ft_ops(p1, req_u);
 541         prb_setup_retire_blk_timer(po, tx_ring);
 542         prb_open_block(p1, pbd);
 543 }
 544
 545 /*  Do NOT update the last_blk_num first.
 546  *  Assumes sk_buff_head lock is held.
 547  */
 548 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 549 {
 550         mod_timer(&pkc->retire_blk_timer,
 551                         jiffies + pkc->tov_in_jiffies);
 552         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 553 }
 554
 555 /*
 556  * Timer logic:
 557  * 1) We refresh the timer only when we open a block.
 558  *    By doing this we don't waste cycles refreshing the timer
 559  *        on packet-by-packet basis.
 560  *
 561  * With a 1MB block-size, on a 1Gbps line, it will take
 562  * i) ~8 ms to fill a block + ii) memcpy etc.
 563  * In this cut we are not accounting for the memcpy time.
 564  *
 565  * So, if the user sets the 'tmo' to 10ms then the timer
 566  * will never fire while the block is still getting filled
 567  * (which is what we want). However, the user could choose
 568  * to close a block early and that's fine.
 569  *
 570  * But when the timer does fire, we check whether or not to refresh it.
 571  * Since the tmo granularity is in msecs, it is not too expensive
 572  * to refresh the timer, lets say every '8' msecs.
 573  * Either the user can set the 'tmo' or we can derive it based on
 574  * a) line-speed and b) block-size.
 575  * prb_calc_retire_blk_tmo() calculates the tmo.
 576  *
 577  */
 578 static void prb_retire_rx_blk_timer_expired(unsigned long data)
 579 {
 580         struct packet_sock *po = (struct packet_sock *)data;
 581         struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
 582         unsigned int frozen;
 583         struct tpacket_block_desc *pbd;
 584
 585         spin_lock(&po->sk.sk_receive_queue.lock);
 586
 587         frozen = prb_queue_frozen(pkc);
 588         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 589
 590         if (unlikely(pkc->delete_blk_timer))
 591                 goto out;
 592
 593         /* We only need to plug the race when the block is partially filled.
 594          * tpacket_rcv:
 595          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 596          *              copy_bits() is in progress ...
 597          *              timer fires on other cpu:
 598          *              we can't retire the current block because copy_bits
 599          *              is in progress.
 600          *
 601          */
 602         if (BLOCK_NUM_PKTS(pbd)) {
 603                 while (atomic_read(&pkc->blk_fill_in_prog)) {
 604                         /* Waiting for skb_copy_bits to finish... */
 605                         cpu_relax();
 606                 }
 607         }
 608
 609         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 610                 if (!frozen) {
 611                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 612                         if (!prb_dispatch_next_block(pkc, po))
 613                                 goto refresh_timer;
 614                         else
 615                                 goto out;
 616                 } else {
 617                         /* Case 1. Queue was frozen because user-space was
 618                          *         lagging behind.
 619                          */
 620                         if (prb_curr_blk_in_use(pkc, pbd)) {
 621                                 /*
 622                                  * Ok, user-space is still behind.
 623                                  * So just refresh the timer.
 624                                  */
 625                                 goto refresh_timer;
 626                         } else {
 627                                /* Case 2. queue was frozen,user-space caught up,
 628                                 * now the link went idle && the timer fired.
 629                                 * We don't have a block to close.So we open this
 630                                 * block and restart the timer.
 631                                 * opening a block thaws the queue,restarts timer
 632                                 * Thawing/timer-refresh is a side effect.
 633                                 */
 634                                 prb_open_block(pkc, pbd);
 635                                 goto out;
 636                         }
 637                 }
 638         }
 639
 640 refresh_timer:
 641         _prb_refresh_rx_retire_blk_timer(pkc);
 642
 643 out:
 644         spin_unlock(&po->sk.sk_receive_queue.lock);
 645 }
 646
 647 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 648                 struct tpacket_block_desc *pbd1, __u32 status)
 649 {
 650         /* Flush everything minus the block header */
 651
 652 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 653         u8 *start, *end;
 654
 655         start = (u8 *)pbd1;
 656
 657         /* Skip the block header(we know header WILL fit in 4K) */
 658         start += PAGE_SIZE;
 659
 660         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 661         for (; start < end; start += PAGE_SIZE)
 662                 flush_dcache_page(pgv_to_page(start));
 663
 664         smp_wmb();
 665 #endif
 666
 667         /* Now update the block status. */
 668
 669         BLOCK_STATUS(pbd1) = status;
 670
 671         /* Flush the block header */
 672
 673 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 674         start = (u8 *)pbd1;
 675         flush_dcache_page(pgv_to_page(start));
 676
 677         smp_wmb();
 678 #endif
 679 }
 680
 681 /*
 682  * Side effect:
 683  *
 684  * 1) flush the block
 685  * 2) Increment active_blk_num
 686  *
 687  * Note:We DONT refresh the timer on purpose.
 688  *      Because almost always the next block will be opened.
 689  */
 690 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 691                 struct tpacket_block_desc *pbd1,
 692                 struct packet_sock *po, unsigned int stat)
 693 {
 694         __u32 status = TP_STATUS_USER | stat;
 695
 696         struct tpacket3_hdr *last_pkt;
 697         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 698
 699         if (po->stats.stats3.tp_drops)
 700                 status |= TP_STATUS_LOSING;
 701
 702         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 703         last_pkt->tp_next_offset = 0;
 704
 705         /* Get the ts of the last pkt */
 706         if (BLOCK_NUM_PKTS(pbd1)) {
 707                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 708                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 709         } else {
 710                 /* Ok, we tmo'd - so get the current time */
 711                 struct timespec ts;
 712                 getnstimeofday(&ts);
 713                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 714                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 715         }
 716
 717         smp_wmb();
 718
 719         /* Flush the block */
 720         prb_flush_block(pkc1, pbd1, status);
 721
 722         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 723 }
 724
 725 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 726 {
 727         pkc->reset_pending_on_curr_blk = 0;
 728 }
 729
 730 /*
 731  * Side effect of opening a block:
 732  *
 733  * 1) prb_queue is thawed.
 734  * 2) retire_blk_timer is refreshed.
 735  *
 736  */
 737 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 738         struct tpacket_block_desc *pbd1)
 739 {
 740         struct timespec ts;
 741         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 742
 743         smp_rmb();
 744
 745         /* We could have just memset this but we will lose the
 746          * flexibility of making the priv area sticky
 747          */
 748
 749         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 750         BLOCK_NUM_PKTS(pbd1) = 0;
 751         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 752
 753         getnstimeofday(&ts);
 754
 755         h1->ts_first_pkt.ts_sec = ts.tv_sec;
 756         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 757
 758         pkc1->pkblk_start = (char *)pbd1;
 759         pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 760
 761         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 762         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 763
 764         pbd1->version = pkc1->version;
 765         pkc1->prev = pkc1->nxt_offset;
 766         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 767
 768         prb_thaw_queue(pkc1);
 769         _prb_refresh_rx_retire_blk_timer(pkc1);
 770
 771         smp_wmb();
 772 }
 773
 774 /*
 775  * Queue freeze logic:
 776  * 1) Assume tp_block_nr = 8 blocks.
 777  * 2) At time 't0', user opens Rx ring.
 778  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 779  * 4) user-space is either sleeping or processing block '0'.
 780  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 781  *    it will close block-7,loop around and try to fill block '0'.
 782  *    call-flow:
 783  *    __packet_lookup_frame_in_block
 784  *      prb_retire_current_block()
 785  *      prb_dispatch_next_block()
 786  *        |->(BLOCK_STATUS == USER) evaluates to true
 787  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 788  * 6) Now there are two cases:
 789  *    6.1) Link goes idle right after the queue is frozen.
 790  *         But remember, the last open_block() refreshed the timer.
 791  *         When this timer expires,it will refresh itself so that we can
 792  *         re-open block-0 in near future.
 793  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 794  *         case and __packet_lookup_frame_in_block will check if block-0
 795  *         is free and can now be re-used.
 796  */
 797 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 798                                   struct packet_sock *po)
 799 {
 800         pkc->reset_pending_on_curr_blk = 1;
 801         po->stats.stats3.tp_freeze_q_cnt++;
 802 }
 803
 804 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 805
 806 /*
 807  * If the next block is free then we will dispatch it
 808  * and return a good offset.
 809  * Else, we will freeze the queue.
 810  * So, caller must check the return value.
 811  */
 812 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 813                 struct packet_sock *po)
 814 {
 815         struct tpacket_block_desc *pbd;
 816
 817         smp_rmb();
 818
 819         /* 1. Get current block num */
 820         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 821
 822         /* 2. If this block is currently in_use then freeze the queue */
 823         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 824                 prb_freeze_queue(pkc, po);
 825                 return NULL;
 826         }
 827
 828         /*
 829          * 3.
 830          * open this block and return the offset where the first packet
 831          * needs to get stored.
 832          */
 833         prb_open_block(pkc, pbd);
 834         return (void *)pkc->nxt_offset;
 835 }
 836
 837 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 838                 struct packet_sock *po, unsigned int status)
 839 {
 840         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 841
 842         /* retire/close the current block */
 843         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 844                 /*
 845                  * Plug the case where copy_bits() is in progress on
 846                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 847                  * have space to copy the pkt in the current block and
 848                  * called prb_retire_current_block()
 849                  *
 850                  * We don't need to worry about the TMO case because
 851                  * the timer-handler already handled this case.
 852                  */
 853                 if (!(status & TP_STATUS_BLK_TMO)) {
 854                         while (atomic_read(&pkc->blk_fill_in_prog)) {
 855                                 /* Waiting for skb_copy_bits to finish... */
 856                                 cpu_relax();
 857                         }
 858                 }
 859                 prb_close_block(pkc, pbd, po, status);
 860                 return;
 861         }
 862 }
 863
 864 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
 865                                       struct tpacket_block_desc *pbd)
 866 {
 867         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 868 }
 869
 870 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 871 {
 872         return pkc->reset_pending_on_curr_blk;
 873 }
 874
 875 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 876 {
 877         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 878         atomic_dec(&pkc->blk_fill_in_prog);
 879 }
 880
 881 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 882                         struct tpacket3_hdr *ppd)
 883 {
 884         ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
 885 }
 886
 887 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 888                         struct tpacket3_hdr *ppd)
 889 {
 890         ppd->hv1.tp_rxhash = 0;
 891 }
 892
 893 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
 894                         struct tpacket3_hdr *ppd)
 895 {
 896         if (vlan_tx_tag_present(pkc->skb)) {
 897                 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
 898                 ppd->tp_status = TP_STATUS_VLAN_VALID;
 899         } else {
 900                 ppd->hv1.tp_vlan_tci = 0;
 901                 ppd->tp_status = TP_STATUS_AVAILABLE;
 902         }
 903 }
 904
 905 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
 906                         struct tpacket3_hdr *ppd)
 907 {
 908         prb_fill_vlan_info(pkc, ppd);
 909
 910         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
 911                 prb_fill_rxhash(pkc, ppd);
 912         else
 913                 prb_clear_rxhash(pkc, ppd);
 914 }
 915
 916 static void prb_fill_curr_block(char *curr,
 917                                 struct tpacket_kbdq_core *pkc,
 918                                 struct tpacket_block_desc *pbd,
 919                                 unsigned int len)
 920 {
 921         struct tpacket3_hdr *ppd;
 922
 923         ppd  = (struct tpacket3_hdr *)curr;
 924         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
 925         pkc->prev = curr;
 926         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
 927         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
 928         BLOCK_NUM_PKTS(pbd) += 1;
 929         atomic_inc(&pkc->blk_fill_in_prog);
 930         prb_run_all_ft_ops(pkc, ppd);
 931 }
 932
 933 /* Assumes caller has the sk->rx_queue.lock */
 934 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
 935                                             struct sk_buff *skb,
 936                                                 int status,
 937                                             unsigned int len
 938                                             )
 939 {
 940         struct tpacket_kbdq_core *pkc;
 941         struct tpacket_block_desc *pbd;
 942         char *curr, *end;
 943
 944         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 945         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 946
 947         /* Queue is frozen when user space is lagging behind */
 948         if (prb_queue_frozen(pkc)) {
 949                 /*
 950                  * Check if that last block which caused the queue to freeze,
 951                  * is still in_use by user-space.
 952                  */
 953                 if (prb_curr_blk_in_use(pkc, pbd)) {
 954                         /* Can't record this packet */
 955                         return NULL;
 956                 } else {
 957                         /*
 958                          * Ok, the block was released by user-space.
 959                          * Now let's open that block.
 960                          * opening a block also thaws the queue.
 961                          * Thawing is a side effect.
 962                          */
 963                         prb_open_block(pkc, pbd);
 964                 }
 965         }
 966
 967         smp_mb();
 968         curr = pkc->nxt_offset;
 969         pkc->skb = skb;
 970         end = (char *)pbd + pkc->kblk_size;
 971
 972         /* first try the current block */
 973         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
 974                 prb_fill_curr_block(curr, pkc, pbd, len);
 975                 return (void *)curr;
 976         }
 977
 978         /* Ok, close the current block */
 979         prb_retire_current_block(pkc, po, 0);
 980
 981         /* Now, try to dispatch the next block */
 982         curr = (char *)prb_dispatch_next_block(pkc, po);
 983         if (curr) {
 984                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 985                 prb_fill_curr_block(curr, pkc, pbd, len);
 986                 return (void *)curr;
 987         }
 988
 989         /*
 990          * No free blocks are available.user_space hasn't caught up yet.
 991          * Queue was just frozen and now this packet will get dropped.
 992          */
 993         return NULL;
 994 }
 995
 996 static void *packet_current_rx_frame(struct packet_sock *po,
 997                                             struct sk_buff *skb,
 998                                             int status, unsigned int len)
 999 {
1000         char *curr = NULL;
1001         switch (po->tp_version) {
1002         case TPACKET_V1:
1003         case TPACKET_V2:
1004                 curr = packet_lookup_frame(po, &po->rx_ring,
1005                                         po->rx_ring.head, status);
1006                 return curr;
1007         case TPACKET_V3:
1008                 return __packet_lookup_frame_in_block(po, skb, status, len);
1009         default:
1010                 WARN(1, "TPACKET version not supported\n");
1011                 BUG();
1012                 return NULL;
1013         }
1014 }
1015
1016 static void *prb_lookup_block(struct packet_sock *po,
1017                                      struct packet_ring_buffer *rb,
1018                                      unsigned int idx,
1019                                      int status)
1020 {
1021         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1022         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1023
1024         if (status != BLOCK_STATUS(pbd))
1025                 return NULL;
1026         return pbd;
1027 }
1028
1029 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1030 {
1031         unsigned int prev;
1032         if (rb->prb_bdqc.kactive_blk_num)
1033                 prev = rb->prb_bdqc.kactive_blk_num-1;
1034         else
1035                 prev = rb->prb_bdqc.knum_blocks-1;
1036         return prev;
1037 }
1038
1039 /* Assumes caller has held the rx_queue.lock */
1040 static void *__prb_previous_block(struct packet_sock *po,
1041                                          struct packet_ring_buffer *rb,
1042                                          int status)
1043 {
1044         unsigned int previous = prb_previous_blk_num(rb);
1045         return prb_lookup_block(po, rb, previous, status);
1046 }
1047
1048 static void *packet_previous_rx_frame(struct packet_sock *po,
1049                                              struct packet_ring_buffer *rb,
1050                                              int status)
1051 {
1052         if (po->tp_version <= TPACKET_V2)
1053                 return packet_previous_frame(po, rb, status);
1054
1055         return __prb_previous_block(po, rb, status);
1056 }
1057
1058 static void packet_increment_rx_head(struct packet_sock *po,
1059                                             struct packet_ring_buffer *rb)
1060 {
1061         switch (po->tp_version) {
1062         case TPACKET_V1:
1063         case TPACKET_V2:
1064                 return packet_increment_head(rb);
1065         case TPACKET_V3:
1066         default:
1067                 WARN(1, "TPACKET version not supported.\n");
1068                 BUG();
1069                 return;
1070         }
1071 }
1072
1073 static void *packet_previous_frame(struct packet_sock *po,
1074                 struct packet_ring_buffer *rb,
1075                 int status)
1076 {
1077         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1078         return packet_lookup_frame(po, rb, previous, status);
1079 }
1080
1081 static void packet_increment_head(struct packet_ring_buffer *buff)
1082 {
1083         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1084 }
1085
1086 static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1087 {
1088         struct sock *sk = &po->sk;
1089         bool has_room;
1090
1091         if (po->prot_hook.func != tpacket_rcv)
1092                 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1093                         <= sk->sk_rcvbuf;
1094
1095         spin_lock(&sk->sk_receive_queue.lock);
1096         if (po->tp_version == TPACKET_V3)
1097                 has_room = prb_lookup_block(po, &po->rx_ring,
1098                                             po->rx_ring.prb_bdqc.kactive_blk_num,
1099                                             TP_STATUS_KERNEL);
1100         else
1101                 has_room = packet_lookup_frame(po, &po->rx_ring,
1102                                                po->rx_ring.head,
1103                                                TP_STATUS_KERNEL);
1104         spin_unlock(&sk->sk_receive_queue.lock);
1105
1106         return has_room;
1107 }
1108
1109 static void packet_sock_destruct(struct sock *sk)
1110 {
1111         skb_queue_purge(&sk->sk_error_queue);
1112
1113         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1114         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1115
1116         if (!sock_flag(sk, SOCK_DEAD)) {
1117                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1118                 return;
1119         }
1120
1121         sk_refcnt_debug_dec(sk);
1122 }
1123
1124 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1125 {
1126         int x = atomic_read(&f->rr_cur) + 1;
1127
1128         if (x >= num)
1129                 x = 0;
1130
1131         return x;
1132 }
1133
1134 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1135                                       struct sk_buff *skb,
1136                                       unsigned int num)
1137 {
1138         return reciprocal_divide(skb->rxhash, num);
1139 }
1140
1141 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1142                                     struct sk_buff *skb,
1143                                     unsigned int num)
1144 {
1145         int cur, old;
1146
1147         cur = atomic_read(&f->rr_cur);
1148         while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1149                                      fanout_rr_next(f, num))) != cur)
1150                 cur = old;
1151         return cur;
1152 }
1153
1154 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1155                                      struct sk_buff *skb,
1156                                      unsigned int num)
1157 {
1158         return smp_processor_id() % num;
1159 }
1160
1161 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1162                                      struct sk_buff *skb,
1163                                      unsigned int num)
1164 {
1165         return reciprocal_divide(prandom_u32(), num);
1166 }
1167
1168 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1169                                           struct sk_buff *skb,
1170                                           unsigned int idx, unsigned int skip,
1171                                           unsigned int num)
1172 {
1173         unsigned int i, j;
1174
1175         i = j = min_t(int, f->next[idx], num - 1);
1176         do {
1177                 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1178                         if (i != j)
1179                                 f->next[idx] = i;
1180                         return i;
1181                 }
1182                 if (++i == num)
1183                         i = 0;
1184         } while (i != j);
1185
1186         return idx;
1187 }
1188
1189 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1190 {
1191         return f->flags & (flag >> 8);
1192 }
1193
1194 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1195                              struct packet_type *pt, struct net_device *orig_dev)
1196 {
1197         struct packet_fanout *f = pt->af_packet_priv;
1198         unsigned int num = f->num_members;
1199         struct packet_sock *po;
1200         unsigned int idx;
1201
1202         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1203             !num) {
1204                 kfree_skb(skb);
1205                 return 0;
1206         }
1207
1208         switch (f->type) {
1209         case PACKET_FANOUT_HASH:
1210         default:
1211                 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1212                         skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1213                         if (!skb)
1214                                 return 0;
1215                 }
1216                 skb_get_rxhash(skb);
1217                 idx = fanout_demux_hash(f, skb, num);
1218                 break;
1219         case PACKET_FANOUT_LB:
1220                 idx = fanout_demux_lb(f, skb, num);
1221                 break;
1222         case PACKET_FANOUT_CPU:
1223                 idx = fanout_demux_cpu(f, skb, num);
1224                 break;
1225         case PACKET_FANOUT_RND:
1226                 idx = fanout_demux_rnd(f, skb, num);
1227                 break;
1228         case PACKET_FANOUT_ROLLOVER:
1229                 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1230                 break;
1231         }
1232
1233         po = pkt_sk(f->arr[idx]);
1234         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1235             unlikely(!packet_rcv_has_room(po, skb))) {
1236                 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1237                 po = pkt_sk(f->arr[idx]);
1238         }
1239
1240         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1241 }
1242
1243 DEFINE_MUTEX(fanout_mutex);
1244 EXPORT_SYMBOL_GPL(fanout_mutex);
1245 static LIST_HEAD(fanout_list);
1246
1247 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1248 {
1249         struct packet_fanout *f = po->fanout;
1250
1251         spin_lock(&f->lock);
1252         f->arr[f->num_members] = sk;
1253         smp_wmb();
1254         f->num_members++;
1255         spin_unlock(&f->lock);
1256 }
1257
1258 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1259 {
1260         struct packet_fanout *f = po->fanout;
1261         int i;
1262
1263         spin_lock(&f->lock);
1264         for (i = 0; i < f->num_members; i++) {
1265                 if (f->arr[i] == sk)
1266                         break;
1267         }
1268         BUG_ON(i >= f->num_members);
1269         f->arr[i] = f->arr[f->num_members - 1];
1270         f->num_members--;
1271         spin_unlock(&f->lock);
1272 }
1273
1274 static bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1275 {
1276         if (ptype->af_packet_priv == (void*)((struct packet_sock *)sk)->fanout)
1277                 return true;
1278
1279         return false;
1280 }
1281
1282 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1283 {
1284         struct packet_sock *po = pkt_sk(sk);
1285         struct packet_fanout *f, *match;
1286         u8 type = type_flags & 0xff;
1287         u8 flags = type_flags >> 8;
1288         int err;
1289
1290         switch (type) {
1291         case PACKET_FANOUT_ROLLOVER:
1292                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1293                         return -EINVAL;
1294         case PACKET_FANOUT_HASH:
1295         case PACKET_FANOUT_LB:
1296         case PACKET_FANOUT_CPU:
1297         case PACKET_FANOUT_RND:
1298                 break;
1299         default:
1300                 return -EINVAL;
1301         }
1302
1303         if (!po->running)
1304                 return -EINVAL;
1305
1306         if (po->fanout)
1307                 return -EALREADY;
1308
1309         mutex_lock(&fanout_mutex);
1310         match = NULL;
1311         list_for_each_entry(f, &fanout_list, list) {
1312                 if (f->id == id &&
1313                     read_pnet(&f->net) == sock_net(sk)) {
1314                         match = f;
1315                         break;
1316                 }
1317         }
1318         err = -EINVAL;
1319         if (match && match->flags != flags)
1320                 goto out;
1321         if (!match) {
1322                 err = -ENOMEM;
1323                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1324                 if (!match)
1325                         goto out;
1326                 write_pnet(&match->net, sock_net(sk));
1327                 match->id = id;
1328                 match->type = type;
1329                 match->flags = flags;
1330                 atomic_set(&match->rr_cur, 0);
1331                 INIT_LIST_HEAD(&match->list);
1332                 spin_lock_init(&match->lock);
1333                 atomic_set(&match->sk_ref, 0);
1334                 match->prot_hook.type = po->prot_hook.type;
1335                 match->prot_hook.dev = po->prot_hook.dev;
1336                 match->prot_hook.func = packet_rcv_fanout;
1337                 match->prot_hook.af_packet_priv = match;
1338                 match->prot_hook.id_match = match_fanout_group;
1339                 dev_add_pack(&match->prot_hook);
1340                 list_add(&match->list, &fanout_list);
1341         }
1342         err = -EINVAL;
1343         if (match->type == type &&
1344             match->prot_hook.type == po->prot_hook.type &&
1345             match->prot_hook.dev == po->prot_hook.dev) {
1346                 err = -ENOSPC;
1347                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1348                         __dev_remove_pack(&po->prot_hook);
1349                         po->fanout = match;
1350                         atomic_inc(&match->sk_ref);
1351                         __fanout_link(sk, po);
1352                         err = 0;
1353                 }
1354         }
1355 out:
1356         mutex_unlock(&fanout_mutex);
1357         return err;
1358 }
1359
1360 static void fanout_release(struct sock *sk)
1361 {
1362         struct packet_sock *po = pkt_sk(sk);
1363         struct packet_fanout *f;
1364
1365         f = po->fanout;
1366         if (!f)
1367                 return;
1368
1369         mutex_lock(&fanout_mutex);
1370         po->fanout = NULL;
1371
1372         if (atomic_dec_and_test(&f->sk_ref)) {
1373                 list_del(&f->list);
1374                 dev_remove_pack(&f->prot_hook);
1375                 kfree(f);
1376         }
1377         mutex_unlock(&fanout_mutex);
1378 }
1379
1380 static const struct proto_ops packet_ops;
1381
1382 static const struct proto_ops packet_ops_spkt;
1383
1384 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1385                            struct packet_type *pt, struct net_device *orig_dev)
1386 {
1387         struct sock *sk;
1388         struct sockaddr_pkt *spkt;
1389
1390         /*
1391          *      When we registered the protocol we saved the socket in the data
1392          *      field for just this event.
1393          */
1394
1395         sk = pt->af_packet_priv;
1396
1397         /*
1398          *      Yank back the headers [hope the device set this
1399          *      right or kerboom...]
1400          *
1401          *      Incoming packets have ll header pulled,
1402          *      push it back.
1403          *
1404          *      For outgoing ones skb->data == skb_mac_header(skb)
1405          *      so that this procedure is noop.
1406          */
1407
1408         if (skb->pkt_type == PACKET_LOOPBACK)
1409                 goto out;
1410
1411         if (!net_eq(dev_net(dev), sock_net(sk)))
1412                 goto out;
1413
1414         skb = skb_share_check(skb, GFP_ATOMIC);
1415         if (skb == NULL)
1416                 goto oom;
1417
1418         /* drop any routing info */
1419         skb_dst_drop(skb);
1420
1421         /* drop conntrack reference */
1422         nf_reset(skb);
1423
1424         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1425
1426         skb_push(skb, skb->data - skb_mac_header(skb));
1427
1428         /*
1429          *      The SOCK_PACKET socket receives _all_ frames.
1430          */
1431
1432         spkt->spkt_family = dev->type;
1433         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1434         spkt->spkt_protocol = skb->protocol;
1435
1436         /*
1437          *      Charge the memory to the socket. This is done specifically
1438          *      to prevent sockets using all the memory up.
1439          */
1440
1441         if (sock_queue_rcv_skb(sk, skb) == 0)
1442                 return 0;
1443
1444 out:
1445         kfree_skb(skb);
1446 oom:
1447         return 0;
1448 }
1449
1450
1451 /*
1452  *      Output a raw packet to a device layer. This bypasses all the other
1453  *      protocol layers and you must therefore supply it with a complete frame
1454  */
1455
1456 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1457                                struct msghdr *msg, size_t len)
1458 {
1459         struct sock *sk = sock->sk;
1460         struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1461         struct sk_buff *skb = NULL;
1462         struct net_device *dev;
1463         __be16 proto = 0;
1464         int err;
1465         int extra_len = 0;
1466
1467         /*
1468          *      Get and verify the address.
1469          */
1470
1471         if (saddr) {
1472                 if (msg->msg_namelen < sizeof(struct sockaddr))
1473                         return -EINVAL;
1474                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1475                         proto = saddr->spkt_protocol;
1476         } else
1477                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1478
1479         /*
1480          *      Find the device first to size check it
1481          */
1482
1483         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1484 retry:
1485         rcu_read_lock();
1486         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1487         err = -ENODEV;
1488         if (dev == NULL)
1489                 goto out_unlock;
1490
1491         err = -ENETDOWN;
1492         if (!(dev->flags & IFF_UP))
1493                 goto out_unlock;
1494
1495         /*
1496          * You may not queue a frame bigger than the mtu. This is the lowest level
1497          * raw protocol and you must do your own fragmentation at this level.
1498          */
1499
1500         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1501                 if (!netif_supports_nofcs(dev)) {
1502                         err = -EPROTONOSUPPORT;
1503                         goto out_unlock;
1504                 }
1505                 extra_len = 4; /* We're doing our own CRC */
1506         }
1507
1508         err = -EMSGSIZE;
1509         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1510                 goto out_unlock;
1511
1512         if (!skb) {
1513                 size_t reserved = LL_RESERVED_SPACE(dev);
1514                 int tlen = dev->needed_tailroom;
1515                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1516
1517                 rcu_read_unlock();
1518                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1519                 if (skb == NULL)
1520                         return -ENOBUFS;
1521                 /* FIXME: Save some space for broken drivers that write a hard
1522                  * header at transmission time by themselves. PPP is the notable
1523                  * one here. This should really be fixed at the driver level.
1524                  */
1525                 skb_reserve(skb, reserved);
1526                 skb_reset_network_header(skb);
1527
1528                 /* Try to align data part correctly */
1529                 if (hhlen) {
1530                         skb->data -= hhlen;
1531                         skb->tail -= hhlen;
1532                         if (len < hhlen)
1533                                 skb_reset_network_header(skb);
1534                 }
1535                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1536                 if (err)
1537                         goto out_free;
1538                 goto retry;
1539         }
1540
1541         if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1542                 /* Earlier code assumed this would be a VLAN pkt,
1543                  * double-check this now that we have the actual
1544                  * packet in hand.
1545                  */
1546                 struct ethhdr *ehdr;
1547                 skb_reset_mac_header(skb);
1548                 ehdr = eth_hdr(skb);
1549                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1550                         err = -EMSGSIZE;
1551                         goto out_unlock;
1552                 }
1553         }
1554
1555         skb->protocol = proto;
1556         skb->dev = dev;
1557         skb->priority = sk->sk_priority;
1558         skb->mark = sk->sk_mark;
1559
1560         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1561
1562         if (unlikely(extra_len == 4))
1563                 skb->no_fcs = 1;
1564
1565         skb_probe_transport_header(skb, 0);
1566
1567         dev_queue_xmit(skb);
1568         rcu_read_unlock();
1569         return len;
1570
1571 out_unlock:
1572         rcu_read_unlock();
1573 out_free:
1574         kfree_skb(skb);
1575         return err;
1576 }
1577
1578 static unsigned int run_filter(const struct sk_buff *skb,
1579                                       const struct sock *sk,
1580                                       unsigned int res)
1581 {
1582         struct sk_filter *filter;
1583
1584         rcu_read_lock();
1585         filter = rcu_dereference(sk->sk_filter);
1586         if (filter != NULL)
1587                 res = SK_RUN_FILTER(filter, skb);
1588         rcu_read_unlock();
1589
1590         return res;
1591 }
1592
1593 /*
1594  * This function makes lazy skb cloning in hope that most of packets
1595  * are discarded by BPF.
1596  *
1597  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1598  * and skb->cb are mangled. It works because (and until) packets
1599  * falling here are owned by current CPU. Output packets are cloned
1600  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1601  * sequencially, so that if we return skb to original state on exit,
1602  * we will not harm anyone.
1603  */
1604
1605 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1606                       struct packet_type *pt, struct net_device *orig_dev)
1607 {
1608         struct sock *sk;
1609         struct sockaddr_ll *sll;
1610         struct packet_sock *po;
1611         u8 *skb_head = skb->data;
1612         int skb_len = skb->len;
1613         unsigned int snaplen, res;
1614
1615         if (skb->pkt_type == PACKET_LOOPBACK)
1616                 goto drop;
1617
1618         sk = pt->af_packet_priv;
1619         po = pkt_sk(sk);
1620
1621         if (!net_eq(dev_net(dev), sock_net(sk)))
1622                 goto drop;
1623
1624         skb->dev = dev;
1625
1626         if (dev->header_ops) {
1627                 /* The device has an explicit notion of ll header,
1628                  * exported to higher levels.
1629                  *
1630                  * Otherwise, the device hides details of its frame
1631                  * structure, so that corresponding packet head is
1632                  * never delivered to user.
1633                  */
1634                 if (sk->sk_type != SOCK_DGRAM)
1635                         skb_push(skb, skb->data - skb_mac_header(skb));
1636                 else if (skb->pkt_type == PACKET_OUTGOING) {
1637                         /* Special case: outgoing packets have ll header at head */
1638                         skb_pull(skb, skb_network_offset(skb));
1639                 }
1640         }
1641
1642         snaplen = skb->len;
1643
1644         res = run_filter(skb, sk, snaplen);
1645         if (!res)
1646                 goto drop_n_restore;
1647         if (snaplen > res)
1648                 snaplen = res;
1649
1650         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1651                 goto drop_n_acct;
1652
1653         if (skb_shared(skb)) {
1654                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1655                 if (nskb == NULL)
1656                         goto drop_n_acct;
1657
1658                 if (skb_head != skb->data) {
1659                         skb->data = skb_head;
1660                         skb->len = skb_len;
1661                 }
1662                 consume_skb(skb);
1663                 skb = nskb;
1664         }
1665
1666         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1667                      sizeof(skb->cb));
1668
1669         sll = &PACKET_SKB_CB(skb)->sa.ll;
1670         sll->sll_family = AF_PACKET;
1671         sll->sll_hatype = dev->type;
1672         sll->sll_protocol = skb->protocol;
1673         sll->sll_pkttype = skb->pkt_type;
1674         if (unlikely(po->origdev))
1675                 sll->sll_ifindex = orig_dev->ifindex;
1676         else
1677                 sll->sll_ifindex = dev->ifindex;
1678
1679         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1680
1681         PACKET_SKB_CB(skb)->origlen = skb->len;
1682
1683         if (pskb_trim(skb, snaplen))
1684                 goto drop_n_acct;
1685
1686         skb_set_owner_r(skb, sk);
1687         skb->dev = NULL;
1688         skb_dst_drop(skb);
1689
1690         /* drop conntrack reference */
1691         nf_reset(skb);
1692
1693         spin_lock(&sk->sk_receive_queue.lock);
1694         po->stats.stats1.tp_packets++;
1695         skb->dropcount = atomic_read(&sk->sk_drops);
1696         __skb_queue_tail(&sk->sk_receive_queue, skb);
1697         spin_unlock(&sk->sk_receive_queue.lock);
1698         sk->sk_data_ready(sk, skb->len);
1699         return 0;
1700
1701 drop_n_acct:
1702         spin_lock(&sk->sk_receive_queue.lock);
1703         po->stats.stats1.tp_drops++;
1704         atomic_inc(&sk->sk_drops);
1705         spin_unlock(&sk->sk_receive_queue.lock);
1706
1707 drop_n_restore:
1708         if (skb_head != skb->data && skb_shared(skb)) {
1709                 skb->data = skb_head;
1710                 skb->len = skb_len;
1711         }
1712 drop:
1713         consume_skb(skb);
1714         return 0;
1715 }
1716
1717 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1718                        struct packet_type *pt, struct net_device *orig_dev)
1719 {
1720         struct sock *sk;
1721         struct packet_sock *po;
1722         struct sockaddr_ll *sll;
1723         union tpacket_uhdr h;
1724         u8 *skb_head = skb->data;
1725         int skb_len = skb->len;
1726         unsigned int snaplen, res;
1727         unsigned long status = TP_STATUS_USER;
1728         unsigned short macoff, netoff, hdrlen;
1729         struct sk_buff *copy_skb = NULL;
1730         struct timespec ts;
1731         __u32 ts_status;
1732
1733         if (skb->pkt_type == PACKET_LOOPBACK)
1734                 goto drop;
1735
1736         sk = pt->af_packet_priv;
1737         po = pkt_sk(sk);
1738
1739         if (!net_eq(dev_net(dev), sock_net(sk)))
1740                 goto drop;
1741
1742         if (dev->header_ops) {
1743                 if (sk->sk_type != SOCK_DGRAM)
1744                         skb_push(skb, skb->data - skb_mac_header(skb));
1745                 else if (skb->pkt_type == PACKET_OUTGOING) {
1746                         /* Special case: outgoing packets have ll header at head */
1747                         skb_pull(skb, skb_network_offset(skb));
1748                 }
1749         }
1750
1751         if (skb->ip_summed == CHECKSUM_PARTIAL)
1752                 status |= TP_STATUS_CSUMNOTREADY;
1753
1754         snaplen = skb->len;
1755
1756         res = run_filter(skb, sk, snaplen);
1757         if (!res)
1758                 goto drop_n_restore;
1759         if (snaplen > res)
1760                 snaplen = res;
1761
1762         if (sk->sk_type == SOCK_DGRAM) {
1763                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1764                                   po->tp_reserve;
1765         } else {
1766                 unsigned int maclen = skb_network_offset(skb);
1767                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1768                                        (maclen < 16 ? 16 : maclen)) +
1769                         po->tp_reserve;
1770                 macoff = netoff - maclen;
1771         }
1772         if (po->tp_version <= TPACKET_V2) {
1773                 if (macoff + snaplen > po->rx_ring.frame_size) {
1774                         if (po->copy_thresh &&
1775                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1776                                 if (skb_shared(skb)) {
1777                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1778                                 } else {
1779                                         copy_skb = skb_get(skb);
1780                                         skb_head = skb->data;
1781                                 }
1782                                 if (copy_skb)
1783                                         skb_set_owner_r(copy_skb, sk);
1784                         }
1785                         snaplen = po->rx_ring.frame_size - macoff;
1786                         if ((int)snaplen < 0)
1787                                 snaplen = 0;
1788                 }
1789         }
1790         spin_lock(&sk->sk_receive_queue.lock);
1791         h.raw = packet_current_rx_frame(po, skb,
1792                                         TP_STATUS_KERNEL, (macoff+snaplen));
1793         if (!h.raw)
1794                 goto ring_is_full;
1795         if (po->tp_version <= TPACKET_V2) {
1796                 packet_increment_rx_head(po, &po->rx_ring);
1797         /*
1798          * LOSING will be reported till you read the stats,
1799          * because it's COR - Clear On Read.
1800          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1801          * at packet level.
1802          */
1803                 if (po->stats.stats1.tp_drops)
1804                         status |= TP_STATUS_LOSING;
1805         }
1806         po->stats.stats1.tp_packets++;
1807         if (copy_skb) {
1808                 status |= TP_STATUS_COPY;
1809                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1810         }
1811         spin_unlock(&sk->sk_receive_queue.lock);
1812
1813         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1814
1815         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
1816                 getnstimeofday(&ts);
1817
1818         status |= ts_status;
1819
1820         switch (po->tp_version) {
1821         case TPACKET_V1:
1822                 h.h1->tp_len = skb->len;
1823                 h.h1->tp_snaplen = snaplen;
1824                 h.h1->tp_mac = macoff;
1825                 h.h1->tp_net = netoff;
1826                 h.h1->tp_sec = ts.tv_sec;
1827                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
1828                 hdrlen = sizeof(*h.h1);
1829                 break;
1830         case TPACKET_V2:
1831                 h.h2->tp_len = skb->len;
1832                 h.h2->tp_snaplen = snaplen;
1833                 h.h2->tp_mac = macoff;
1834                 h.h2->tp_net = netoff;
1835                 h.h2->tp_sec = ts.tv_sec;
1836                 h.h2->tp_nsec = ts.tv_nsec;
1837                 if (vlan_tx_tag_present(skb)) {
1838                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1839                         status |= TP_STATUS_VLAN_VALID;
1840                 } else {
1841                         h.h2->tp_vlan_tci = 0;
1842                 }
1843                 h.h2->tp_padding = 0;
1844                 hdrlen = sizeof(*h.h2);
1845                 break;
1846         case TPACKET_V3:
1847                 /* tp_nxt_offset,vlan are already populated above.
1848                  * So DONT clear those fields here
1849                  */
1850                 h.h3->tp_status |= status;
1851                 h.h3->tp_len = skb->len;
1852                 h.h3->tp_snaplen = snaplen;
1853                 h.h3->tp_mac = macoff;
1854                 h.h3->tp_net = netoff;
1855                 h.h3->tp_sec  = ts.tv_sec;
1856                 h.h3->tp_nsec = ts.tv_nsec;
1857                 hdrlen = sizeof(*h.h3);
1858                 break;
1859         default:
1860                 BUG();
1861         }
1862
1863         sll = h.raw + TPACKET_ALIGN(hdrlen);
1864         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1865         sll->sll_family = AF_PACKET;
1866         sll->sll_hatype = dev->type;
1867         sll->sll_protocol = skb->protocol;
1868         sll->sll_pkttype = skb->pkt_type;
1869         if (unlikely(po->origdev))
1870                 sll->sll_ifindex = orig_dev->ifindex;
1871         else
1872                 sll->sll_ifindex = dev->ifindex;
1873
1874         smp_mb();
1875 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1876         {
1877                 u8 *start, *end;
1878
1879                 if (po->tp_version <= TPACKET_V2) {
1880                         end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1881                                 + macoff + snaplen);
1882                         for (start = h.raw; start < end; start += PAGE_SIZE)
1883                                 flush_dcache_page(pgv_to_page(start));
1884                 }
1885                 smp_wmb();
1886         }
1887 #endif
1888         if (po->tp_version <= TPACKET_V2)
1889                 __packet_set_status(po, h.raw, status);
1890         else
1891                 prb_clear_blk_fill_status(&po->rx_ring);
1892
1893         sk->sk_data_ready(sk, 0);
1894
1895 drop_n_restore:
1896         if (skb_head != skb->data && skb_shared(skb)) {
1897                 skb->data = skb_head;
1898                 skb->len = skb_len;
1899         }
1900 drop:
1901         kfree_skb(skb);
1902         return 0;
1903
1904 ring_is_full:
1905         po->stats.stats1.tp_drops++;
1906         spin_unlock(&sk->sk_receive_queue.lock);
1907
1908         sk->sk_data_ready(sk, 0);
1909         kfree_skb(copy_skb);
1910         goto drop_n_restore;
1911 }
1912
1913 static void tpacket_destruct_skb(struct sk_buff *skb)
1914 {
1915         struct packet_sock *po = pkt_sk(skb->sk);
1916         void *ph;
1917
1918         if (likely(po->tx_ring.pg_vec)) {
1919                 __u32 ts;
1920
1921                 ph = skb_shinfo(skb)->destructor_arg;
1922                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1923                 atomic_dec(&po->tx_ring.pending);
1924
1925                 ts = __packet_set_timestamp(po, ph, skb);
1926                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
1927         }
1928
1929         sock_wfree(skb);
1930 }
1931
1932 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1933                 void *frame, struct net_device *dev, int size_max,
1934                 __be16 proto, unsigned char *addr, int hlen)
1935 {
1936         union tpacket_uhdr ph;
1937         int to_write, offset, len, tp_len, nr_frags, len_max;
1938         struct socket *sock = po->sk.sk_socket;
1939         struct page *page;
1940         void *data;
1941         int err;
1942
1943         ph.raw = frame;
1944
1945         skb->protocol = proto;
1946         skb->dev = dev;
1947         skb->priority = po->sk.sk_priority;
1948         skb->mark = po->sk.sk_mark;
1949         sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
1950         skb_shinfo(skb)->destructor_arg = ph.raw;
1951
1952         switch (po->tp_version) {
1953         case TPACKET_V2:
1954                 tp_len = ph.h2->tp_len;
1955                 break;
1956         default:
1957                 tp_len = ph.h1->tp_len;
1958                 break;
1959         }
1960         if (unlikely(tp_len > size_max)) {
1961                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1962                 return -EMSGSIZE;
1963         }
1964
1965         skb_reserve(skb, hlen);
1966         skb_reset_network_header(skb);
1967         skb_probe_transport_header(skb, 0);
1968
1969         if (po->tp_tx_has_off) {
1970                 int off_min, off_max, off;
1971                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
1972                 off_max = po->tx_ring.frame_size - tp_len;
1973                 if (sock->type == SOCK_DGRAM) {
1974                         switch (po->tp_version) {
1975                         case TPACKET_V2:
1976                                 off = ph.h2->tp_net;
1977                                 break;
1978                         default:
1979                                 off = ph.h1->tp_net;
1980                                 break;
1981                         }
1982                 } else {
1983                         switch (po->tp_version) {
1984                         case TPACKET_V2:
1985                                 off = ph.h2->tp_mac;
1986                                 break;
1987                         default:
1988                                 off = ph.h1->tp_mac;
1989                                 break;
1990                         }
1991                 }
1992                 if (unlikely((off < off_min) || (off_max < off)))
1993                         return -EINVAL;
1994                 data = ph.raw + off;
1995         } else {
1996                 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1997         }
1998         to_write = tp_len;
1999
2000         if (sock->type == SOCK_DGRAM) {
2001                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2002                                 NULL, tp_len);
2003                 if (unlikely(err < 0))
2004                         return -EINVAL;
2005         } else if (dev->hard_header_len) {
2006                 /* net device doesn't like empty head */
2007                 if (unlikely(tp_len <= dev->hard_header_len)) {
2008                         pr_err("packet size is too short (%d < %d)\n",
2009                                tp_len, dev->hard_header_len);
2010                         return -EINVAL;
2011                 }
2012
2013                 skb_push(skb, dev->hard_header_len);
2014                 err = skb_store_bits(skb, 0, data,
2015                                 dev->hard_header_len);
2016                 if (unlikely(err))
2017                         return err;
2018
2019                 data += dev->hard_header_len;
2020                 to_write -= dev->hard_header_len;
2021         }
2022
2023         offset = offset_in_page(data);
2024         len_max = PAGE_SIZE - offset;
2025         len = ((to_write > len_max) ? len_max : to_write);
2026
2027         skb->data_len = to_write;
2028         skb->len += to_write;
2029         skb->truesize += to_write;
2030         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2031
2032         while (likely(to_write)) {
2033                 nr_frags = skb_shinfo(skb)->nr_frags;
2034
2035                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2036                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2037                                MAX_SKB_FRAGS);
2038                         return -EFAULT;
2039                 }
2040
2041                 page = pgv_to_page(data);
2042                 data += len;
2043                 flush_dcache_page(page);
2044                 get_page(page);
2045                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2046                 to_write -= len;
2047                 offset = 0;
2048                 len_max = PAGE_SIZE;
2049                 len = ((to_write > len_max) ? len_max : to_write);
2050         }
2051
2052         return tp_len;
2053 }
2054
2055 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2056 {
2057         struct sk_buff *skb;
2058         struct net_device *dev;
2059         __be16 proto;
2060         bool need_rls_dev = false;
2061         int err, reserve = 0;
2062         void *ph;
2063         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2064         int tp_len, size_max;
2065         unsigned char *addr;
2066         int len_sum = 0;
2067         int status = TP_STATUS_AVAILABLE;
2068         int hlen, tlen;
2069
2070         mutex_lock(&po->pg_vec_lock);
2071
2072         if (saddr == NULL) {
2073                 dev = po->prot_hook.dev;
2074                 proto   = po->num;
2075                 addr    = NULL;
2076         } else {
2077                 err = -EINVAL;
2078                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2079                         goto out;
2080                 if (msg->msg_namelen < (saddr->sll_halen
2081                                         + offsetof(struct sockaddr_ll,
2082                                                 sll_addr)))
2083                         goto out;
2084                 proto   = saddr->sll_protocol;
2085                 addr    = saddr->sll_addr;
2086                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2087                 need_rls_dev = true;
2088         }
2089
2090         err = -ENXIO;
2091         if (unlikely(dev == NULL))
2092                 goto out;
2093
2094         reserve = dev->hard_header_len;
2095
2096         err = -ENETDOWN;
2097         if (unlikely(!(dev->flags & IFF_UP)))
2098                 goto out_put;
2099
2100         size_max = po->tx_ring.frame_size
2101                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2102
2103         if (size_max > dev->mtu + reserve)
2104                 size_max = dev->mtu + reserve;
2105
2106         do {
2107                 ph = packet_current_frame(po, &po->tx_ring,
2108                                 TP_STATUS_SEND_REQUEST);
2109
2110                 if (unlikely(ph == NULL)) {
2111                         schedule();
2112                         continue;
2113                 }
2114
2115                 status = TP_STATUS_SEND_REQUEST;
2116                 hlen = LL_RESERVED_SPACE(dev);
2117                 tlen = dev->needed_tailroom;
2118                 skb = sock_alloc_send_skb(&po->sk,
2119                                 hlen + tlen + sizeof(struct sockaddr_ll),
2120                                 0, &err);
2121
2122                 if (unlikely(skb == NULL))
2123                         goto out_status;
2124
2125                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2126                                 addr, hlen);
2127
2128                 if (unlikely(tp_len < 0)) {
2129                         if (po->tp_loss) {
2130                                 __packet_set_status(po, ph,
2131                                                 TP_STATUS_AVAILABLE);
2132                                 packet_increment_head(&po->tx_ring);
2133                                 kfree_skb(skb);
2134                                 continue;
2135                         } else {
2136                                 status = TP_STATUS_WRONG_FORMAT;
2137                                 err = tp_len;
2138                                 goto out_status;
2139                         }
2140                 }
2141
2142                 skb->destructor = tpacket_destruct_skb;
2143                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2144                 atomic_inc(&po->tx_ring.pending);
2145
2146                 status = TP_STATUS_SEND_REQUEST;
2147                 err = dev_queue_xmit(skb);
2148                 if (unlikely(err > 0)) {
2149                         err = net_xmit_errno(err);
2150                         if (err && __packet_get_status(po, ph) ==
2151                                    TP_STATUS_AVAILABLE) {
2152                                 /* skb was destructed already */
2153                                 skb = NULL;
2154                                 goto out_status;
2155                         }
2156                         /*
2157                          * skb was dropped but not destructed yet;
2158                          * let's treat it like congestion or err < 0
2159                          */
2160                         err = 0;
2161                 }
2162                 packet_increment_head(&po->tx_ring);
2163                 len_sum += tp_len;
2164         } while (likely((ph != NULL) ||
2165                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2166                          (atomic_read(&po->tx_ring.pending))))
2167                 );
2168
2169         err = len_sum;
2170         goto out_put;
2171
2172 out_status:
2173         __packet_set_status(po, ph, status);
2174         kfree_skb(skb);
2175 out_put:
2176         if (need_rls_dev)
2177                 dev_put(dev);
2178 out:
2179         mutex_unlock(&po->pg_vec_lock);
2180         return err;
2181 }
2182
2183 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2184                                         size_t reserve, size_t len,
2185                                         size_t linear, int noblock,
2186                                         int *err)
2187 {
2188         struct sk_buff *skb;
2189
2190         /* Under a page?  Don't bother with paged skb. */
2191         if (prepad + len < PAGE_SIZE || !linear)
2192                 linear = len;
2193
2194         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2195                                    err, 0);
2196         if (!skb)
2197                 return NULL;
2198
2199         skb_reserve(skb, reserve);
2200         skb_put(skb, linear);
2201         skb->data_len = len - linear;
2202         skb->len += len - linear;
2203
2204         return skb;
2205 }
2206
2207 static int packet_snd(struct socket *sock,
2208                           struct msghdr *msg, size_t len)
2209 {
2210         struct sock *sk = sock->sk;
2211         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2212         struct sk_buff *skb;
2213         struct net_device *dev;
2214         __be16 proto;
2215         bool need_rls_dev = false;
2216         unsigned char *addr;
2217         int err, reserve = 0;
2218         struct virtio_net_hdr vnet_hdr = { 0 };
2219         int offset = 0;
2220         int vnet_hdr_len;
2221         struct packet_sock *po = pkt_sk(sk);
2222         unsigned short gso_type = 0;
2223         int hlen, tlen;
2224         int extra_len = 0;
2225
2226         /*
2227          *      Get and verify the address.
2228          */
2229
2230         if (saddr == NULL) {
2231                 dev = po->prot_hook.dev;
2232                 proto   = po->num;
2233                 addr    = NULL;
2234         } else {
2235                 err = -EINVAL;
2236                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2237                         goto out;
2238                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2239                         goto out;
2240                 proto   = saddr->sll_protocol;
2241                 addr    = saddr->sll_addr;
2242                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2243                 need_rls_dev = true;
2244         }
2245
2246         err = -ENXIO;
2247         if (dev == NULL)
2248                 goto out_unlock;
2249         if (sock->type == SOCK_RAW)
2250                 reserve = dev->hard_header_len;
2251
2252         err = -ENETDOWN;
2253         if (!(dev->flags & IFF_UP))
2254                 goto out_unlock;
2255
2256         if (po->has_vnet_hdr) {
2257                 vnet_hdr_len = sizeof(vnet_hdr);
2258
2259                 err = -EINVAL;
2260                 if (len < vnet_hdr_len)
2261                         goto out_unlock;
2262
2263                 len -= vnet_hdr_len;
2264
2265                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2266                                        vnet_hdr_len);
2267                 if (err < 0)
2268                         goto out_unlock;
2269
2270                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2271                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2272                       vnet_hdr.hdr_len))
2273                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
2274                                                  vnet_hdr.csum_offset + 2;
2275
2276                 err = -EINVAL;
2277                 if (vnet_hdr.hdr_len > len)
2278                         goto out_unlock;
2279
2280                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2281                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2282                         case VIRTIO_NET_HDR_GSO_TCPV4:
2283                                 gso_type = SKB_GSO_TCPV4;
2284                                 break;
2285                         case VIRTIO_NET_HDR_GSO_TCPV6:
2286                                 gso_type = SKB_GSO_TCPV6;
2287                                 break;
2288                         case VIRTIO_NET_HDR_GSO_UDP:
2289                                 gso_type = SKB_GSO_UDP;
2290                                 break;
2291                         default:
2292                                 goto out_unlock;
2293                         }
2294
2295                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2296                                 gso_type |= SKB_GSO_TCP_ECN;
2297
2298                         if (vnet_hdr.gso_size == 0)
2299                                 goto out_unlock;
2300
2301                 }
2302         }
2303
2304         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2305                 if (!netif_supports_nofcs(dev)) {
2306                         err = -EPROTONOSUPPORT;
2307                         goto out_unlock;
2308                 }
2309                 extra_len = 4; /* We're doing our own CRC */
2310         }
2311
2312         err = -EMSGSIZE;
2313         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2314                 goto out_unlock;
2315
2316         err = -ENOBUFS;
2317         hlen = LL_RESERVED_SPACE(dev);
2318         tlen = dev->needed_tailroom;
2319         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2320                                msg->msg_flags & MSG_DONTWAIT, &err);
2321         if (skb == NULL)
2322                 goto out_unlock;
2323
2324         skb_set_network_header(skb, reserve);
2325
2326         err = -EINVAL;
2327         if (sock->type == SOCK_DGRAM &&
2328             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2329                 goto out_free;
2330
2331         /* Returns -EFAULT on error */
2332         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2333         if (err)
2334                 goto out_free;
2335
2336         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2337
2338         if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2339                 /* Earlier code assumed this would be a VLAN pkt,
2340                  * double-check this now that we have the actual
2341                  * packet in hand.
2342                  */
2343                 struct ethhdr *ehdr;
2344                 skb_reset_mac_header(skb);
2345                 ehdr = eth_hdr(skb);
2346                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2347                         err = -EMSGSIZE;
2348                         goto out_free;
2349                 }
2350         }
2351
2352         skb->protocol = proto;
2353         skb->dev = dev;
2354         skb->priority = sk->sk_priority;
2355         skb->mark = sk->sk_mark;
2356
2357         if (po->has_vnet_hdr) {
2358                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2359                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2360                                                   vnet_hdr.csum_offset)) {
2361                                 err = -EINVAL;
2362                                 goto out_free;
2363                         }
2364                 }
2365
2366                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2367                 skb_shinfo(skb)->gso_type = gso_type;
2368
2369                 /* Header must be checked, and gso_segs computed. */
2370                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2371                 skb_shinfo(skb)->gso_segs = 0;
2372
2373                 len += vnet_hdr_len;
2374         }
2375
2376         skb_probe_transport_header(skb, reserve);
2377
2378         if (unlikely(extra_len == 4))
2379                 skb->no_fcs = 1;
2380
2381         /*
2382          *      Now send it
2383          */
2384
2385         err = dev_queue_xmit(skb);
2386         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2387                 goto out_unlock;
2388
2389         if (need_rls_dev)
2390                 dev_put(dev);
2391
2392         return len;
2393
2394 out_free:
2395         kfree_skb(skb);
2396 out_unlock:
2397         if (dev && need_rls_dev)
2398                 dev_put(dev);
2399 out:
2400         return err;
2401 }
2402
2403 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2404                 struct msghdr *msg, size_t len)
2405 {
2406         struct sock *sk = sock->sk;
2407         struct packet_sock *po = pkt_sk(sk);
2408         if (po->tx_ring.pg_vec)
2409                 return tpacket_snd(po, msg);
2410         else
2411                 return packet_snd(sock, msg, len);
2412 }
2413
2414 /*
2415  *      Close a PACKET socket. This is fairly simple. We immediately go
2416  *      to 'closed' state and remove our protocol entry in the device list.
2417  */
2418
2419 static int packet_release(struct socket *sock)
2420 {
2421         struct sock *sk = sock->sk;
2422         struct packet_sock *po;
2423         struct net *net;
2424         union tpacket_req_u req_u;
2425
2426         if (!sk)
2427                 return 0;
2428
2429         net = sock_net(sk);
2430         po = pkt_sk(sk);
2431
2432         mutex_lock(&net->packet.sklist_lock);
2433         sk_del_node_init_rcu(sk);
2434         mutex_unlock(&net->packet.sklist_lock);
2435
2436         preempt_disable();
2437         sock_prot_inuse_add(net, sk->sk_prot, -1);
2438         preempt_enable();
2439
2440         spin_lock(&po->bind_lock);
2441         unregister_prot_hook(sk, false);
2442         if (po->prot_hook.dev) {
2443                 dev_put(po->prot_hook.dev);
2444                 po->prot_hook.dev = NULL;
2445         }
2446         spin_unlock(&po->bind_lock);
2447
2448         packet_flush_mclist(sk);
2449
2450         if (po->rx_ring.pg_vec) {
2451                 memset(&req_u, 0, sizeof(req_u));
2452                 packet_set_ring(sk, &req_u, 1, 0);
2453         }
2454
2455         if (po->tx_ring.pg_vec) {
2456                 memset(&req_u, 0, sizeof(req_u));
2457                 packet_set_ring(sk, &req_u, 1, 1);
2458         }
2459
2460         fanout_release(sk);
2461
2462         synchronize_net();
2463         /*
2464          *      Now the socket is dead. No more input will appear.
2465          */
2466         sock_orphan(sk);
2467         sock->sk = NULL;
2468
2469         /* Purge queues */
2470
2471         skb_queue_purge(&sk->sk_receive_queue);
2472         sk_refcnt_debug_release(sk);
2473
2474         sock_put(sk);
2475         return 0;
2476 }
2477
2478 /*
2479  *      Attach a packet hook.
2480  */
2481
2482 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2483 {
2484         struct packet_sock *po = pkt_sk(sk);
2485
2486         if (po->fanout) {
2487                 if (dev)
2488                         dev_put(dev);
2489
2490                 return -EINVAL;
2491         }
2492
2493         lock_sock(sk);
2494
2495         spin_lock(&po->bind_lock);
2496         unregister_prot_hook(sk, true);
2497         po->num = protocol;
2498         po->prot_hook.type = protocol;
2499         if (po->prot_hook.dev)
2500                 dev_put(po->prot_hook.dev);
2501         po->prot_hook.dev = dev;
2502
2503         po->ifindex = dev ? dev->ifindex : 0;
2504
2505         if (protocol == 0)
2506                 goto out_unlock;
2507
2508         if (!dev || (dev->flags & IFF_UP)) {
2509                 register_prot_hook(sk);
2510         } else {
2511                 sk->sk_err = ENETDOWN;
2512                 if (!sock_flag(sk, SOCK_DEAD))
2513                         sk->sk_error_report(sk);
2514         }
2515
2516 out_unlock:
2517         spin_unlock(&po->bind_lock);
2518         release_sock(sk);
2519         return 0;
2520 }
2521
2522 /*
2523  *      Bind a packet socket to a device
2524  */
2525
2526 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2527                             int addr_len)
2528 {
2529         struct sock *sk = sock->sk;
2530         char name[15];
2531         struct net_device *dev;
2532         int err = -ENODEV;
2533
2534         /*
2535          *      Check legality
2536          */
2537
2538         if (addr_len != sizeof(struct sockaddr))
2539                 return -EINVAL;
2540         strlcpy(name, uaddr->sa_data, sizeof(name));
2541
2542         dev = dev_get_by_name(sock_net(sk), name);
2543         if (dev)
2544                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2545         return err;
2546 }
2547
2548 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2549 {
2550         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2551         struct sock *sk = sock->sk;
2552         struct net_device *dev = NULL;
2553         int err;
2554
2555
2556         /*
2557          *      Check legality
2558          */
2559
2560         if (addr_len < sizeof(struct sockaddr_ll))
2561                 return -EINVAL;
2562         if (sll->sll_family != AF_PACKET)
2563                 return -EINVAL;
2564
2565         if (sll->sll_ifindex) {
2566                 err = -ENODEV;
2567                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2568                 if (dev == NULL)
2569                         goto out;
2570         }
2571         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2572
2573 out:
2574         return err;
2575 }
2576
2577 static struct proto packet_proto = {
2578         .name     = "PACKET",
2579         .owner    = THIS_MODULE,
2580         .obj_size = sizeof(struct packet_sock),
2581 };
2582
2583 /*
2584  *      Create a packet of type SOCK_PACKET.
2585  */
2586
2587 static int packet_create(struct net *net, struct socket *sock, int protocol,
2588                          int kern)
2589 {
2590         struct sock *sk;
2591         struct packet_sock *po;
2592         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2593         int err;
2594
2595         if (!ns_capable(net->user_ns, CAP_NET_RAW))
2596                 return -EPERM;
2597         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2598             sock->type != SOCK_PACKET)
2599                 return -ESOCKTNOSUPPORT;
2600
2601         sock->state = SS_UNCONNECTED;
2602
2603         err = -ENOBUFS;
2604         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2605         if (sk == NULL)
2606                 goto out;
2607
2608         sock->ops = &packet_ops;
2609         if (sock->type == SOCK_PACKET)
2610                 sock->ops = &packet_ops_spkt;
2611
2612         sock_init_data(sock, sk);
2613
2614         po = pkt_sk(sk);
2615         sk->sk_family = PF_PACKET;
2616         po->num = proto;
2617
2618         sk->sk_destruct = packet_sock_destruct;
2619         sk_refcnt_debug_inc(sk);
2620
2621         /*
2622          *      Attach a protocol block
2623          */
2624
2625         spin_lock_init(&po->bind_lock);
2626         mutex_init(&po->pg_vec_lock);
2627         po->prot_hook.func = packet_rcv;
2628
2629         if (sock->type == SOCK_PACKET)
2630                 po->prot_hook.func = packet_rcv_spkt;
2631
2632         po->prot_hook.af_packet_priv = sk;
2633
2634         if (proto) {
2635                 po->prot_hook.type = proto;
2636                 register_prot_hook(sk);
2637         }
2638
2639         mutex_lock(&net->packet.sklist_lock);
2640         sk_add_node_rcu(sk, &net->packet.sklist);
2641         mutex_unlock(&net->packet.sklist_lock);
2642
2643         preempt_disable();
2644         sock_prot_inuse_add(net, &packet_proto, 1);
2645         preempt_enable();
2646
2647         return 0;
2648 out:
2649         return err;
2650 }
2651
2652 /*
2653  *      Pull a packet from our receive queue and hand it to the user.
2654  *      If necessary we block.
2655  */
2656
2657 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2658                           struct msghdr *msg, size_t len, int flags)
2659 {
2660         struct sock *sk = sock->sk;
2661         struct sk_buff *skb;
2662         int copied, err;
2663         struct sockaddr_ll *sll;
2664         int vnet_hdr_len = 0;
2665
2666         err = -EINVAL;
2667         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2668                 goto out;
2669
2670 #if 0
2671         /* What error should we return now? EUNATTACH? */
2672         if (pkt_sk(sk)->ifindex < 0)
2673                 return -ENODEV;
2674 #endif
2675
2676         if (flags & MSG_ERRQUEUE) {
2677                 err = sock_recv_errqueue(sk, msg, len,
2678                                          SOL_PACKET, PACKET_TX_TIMESTAMP);
2679                 goto out;
2680         }
2681
2682         /*
2683          *      Call the generic datagram receiver. This handles all sorts
2684          *      of horrible races and re-entrancy so we can forget about it
2685          *      in the protocol layers.
2686          *
2687          *      Now it will return ENETDOWN, if device have just gone down,
2688          *      but then it will block.
2689          */
2690
2691         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2692
2693         /*
2694          *      An error occurred so return it. Because skb_recv_datagram()
2695          *      handles the blocking we don't see and worry about blocking
2696          *      retries.
2697          */
2698
2699         if (skb == NULL)
2700                 goto out;
2701
2702         if (pkt_sk(sk)->has_vnet_hdr) {
2703                 struct virtio_net_hdr vnet_hdr = { 0 };
2704
2705                 err = -EINVAL;
2706                 vnet_hdr_len = sizeof(vnet_hdr);
2707                 if (len < vnet_hdr_len)
2708                         goto out_free;
2709
2710                 len -= vnet_hdr_len;
2711
2712                 if (skb_is_gso(skb)) {
2713                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2714
2715                         /* This is a hint as to how much should be linear. */
2716                         vnet_hdr.hdr_len = skb_headlen(skb);
2717                         vnet_hdr.gso_size = sinfo->gso_size;
2718                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2719                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2720                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2721                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2722                         else if (sinfo->gso_type & SKB_GSO_UDP)
2723                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2724                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2725                                 goto out_free;
2726                         else
2727                                 BUG();
2728                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2729                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2730                 } else
2731                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2732
2733                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2734                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2735                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2736                         vnet_hdr.csum_offset = skb->csum_offset;
2737                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2738                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2739                 } /* else everything is zero */
2740
2741                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2742                                      vnet_hdr_len);
2743                 if (err < 0)
2744                         goto out_free;
2745         }
2746
2747         /*
2748          *      If the address length field is there to be filled in, we fill
2749          *      it in now.
2750          */
2751
2752         sll = &PACKET_SKB_CB(skb)->sa.ll;
2753         if (sock->type == SOCK_PACKET)
2754                 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2755         else
2756                 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2757
2758         /*
2759          *      You lose any data beyond the buffer you gave. If it worries a
2760          *      user program they can ask the device for its MTU anyway.
2761          */
2762
2763         copied = skb->len;
2764         if (copied > len) {
2765                 copied = len;
2766                 msg->msg_flags |= MSG_TRUNC;
2767         }
2768
2769         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2770         if (err)
2771                 goto out_free;
2772
2773         sock_recv_ts_and_drops(msg, sk, skb);
2774
2775         if (msg->msg_name)
2776                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2777                        msg->msg_namelen);
2778
2779         if (pkt_sk(sk)->auxdata) {
2780                 struct tpacket_auxdata aux;
2781
2782                 aux.tp_status = TP_STATUS_USER;
2783                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2784                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2785                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2786                 aux.tp_snaplen = skb->len;
2787                 aux.tp_mac = 0;
2788                 aux.tp_net = skb_network_offset(skb);
2789                 if (vlan_tx_tag_present(skb)) {
2790                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2791                         aux.tp_status |= TP_STATUS_VLAN_VALID;
2792                 } else {
2793                         aux.tp_vlan_tci = 0;
2794                 }
2795                 aux.tp_padding = 0;
2796                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2797         }
2798
2799         /*
2800          *      Free or return the buffer as appropriate. Again this
2801          *      hides all the races and re-entrancy issues from us.
2802          */
2803         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2804
2805 out_free:
2806         skb_free_datagram(sk, skb);
2807 out:
2808         return err;
2809 }
2810
2811 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2812                                int *uaddr_len, int peer)
2813 {
2814         struct net_device *dev;
2815         struct sock *sk = sock->sk;
2816
2817         if (peer)
2818                 return -EOPNOTSUPP;
2819
2820         uaddr->sa_family = AF_PACKET;
2821         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
2822         rcu_read_lock();
2823         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2824         if (dev)
2825                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
2826         rcu_read_unlock();
2827         *uaddr_len = sizeof(*uaddr);
2828
2829         return 0;
2830 }
2831
2832 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2833                           int *uaddr_len, int peer)
2834 {
2835         struct net_device *dev;
2836         struct sock *sk = sock->sk;
2837         struct packet_sock *po = pkt_sk(sk);
2838         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2839
2840         if (peer)
2841                 return -EOPNOTSUPP;
2842
2843         sll->sll_family = AF_PACKET;
2844         sll->sll_ifindex = po->ifindex;
2845         sll->sll_protocol = po->num;
2846         sll->sll_pkttype = 0;
2847         rcu_read_lock();
2848         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2849         if (dev) {
2850                 sll->sll_hatype = dev->type;
2851                 sll->sll_halen = dev->addr_len;
2852                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2853         } else {
2854                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
2855                 sll->sll_halen = 0;
2856         }
2857         rcu_read_unlock();
2858         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2859
2860         return 0;
2861 }
2862
2863 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2864                          int what)
2865 {
2866         switch (i->type) {
2867         case PACKET_MR_MULTICAST:
2868                 if (i->alen != dev->addr_len)
2869                         return -EINVAL;
2870                 if (what > 0)
2871                         return dev_mc_add(dev, i->addr);
2872                 else
2873                         return dev_mc_del(dev, i->addr);
2874                 break;
2875         case PACKET_MR_PROMISC:
2876                 return dev_set_promiscuity(dev, what);
2877                 break;
2878         case PACKET_MR_ALLMULTI:
2879                 return dev_set_allmulti(dev, what);
2880                 break;
2881         case PACKET_MR_UNICAST:
2882                 if (i->alen != dev->addr_len)
2883                         return -EINVAL;
2884                 if (what > 0)
2885                         return dev_uc_add(dev, i->addr);
2886                 else
2887                         return dev_uc_del(dev, i->addr);
2888                 break;
2889         default:
2890                 break;
2891         }
2892         return 0;
2893 }
2894
2895 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2896 {
2897         for ( ; i; i = i->next) {
2898                 if (i->ifindex == dev->ifindex)
2899                         packet_dev_mc(dev, i, what);
2900         }
2901 }
2902
2903 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2904 {
2905         struct packet_sock *po = pkt_sk(sk);
2906         struct packet_mclist *ml, *i;
2907         struct net_device *dev;
2908         int err;
2909
2910         rtnl_lock();
2911
2912         err = -ENODEV;
2913         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2914         if (!dev)
2915                 goto done;
2916
2917         err = -EINVAL;
2918         if (mreq->mr_alen > dev->addr_len)
2919                 goto done;
2920
2921         err = -ENOBUFS;
2922         i = kmalloc(sizeof(*i), GFP_KERNEL);
2923         if (i == NULL)
2924                 goto done;
2925
2926         err = 0;
2927         for (ml = po->mclist; ml; ml = ml->next) {
2928                 if (ml->ifindex == mreq->mr_ifindex &&
2929                     ml->type == mreq->mr_type &&
2930                     ml->alen == mreq->mr_alen &&
2931                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2932                         ml->count++;
2933                         /* Free the new element ... */
2934                         kfree(i);
2935                         goto done;
2936                 }
2937         }
2938
2939         i->type = mreq->mr_type;
2940         i->ifindex = mreq->mr_ifindex;
2941         i->alen = mreq->mr_alen;
2942         memcpy(i->addr, mreq->mr_address, i->alen);
2943         i->count = 1;
2944         i->next = po->mclist;
2945         po->mclist = i;
2946         err = packet_dev_mc(dev, i, 1);
2947         if (err) {
2948                 po->mclist = i->next;
2949                 kfree(i);
2950         }
2951
2952 done:
2953         rtnl_unlock();
2954         return err;
2955 }
2956
2957 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2958 {
2959         struct packet_mclist *ml, **mlp;
2960
2961         rtnl_lock();
2962
2963         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
2964                 if (ml->ifindex == mreq->mr_ifindex &&
2965                     ml->type == mreq->mr_type &&
2966                     ml->alen == mreq->mr_alen &&
2967                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2968                         if (--ml->count == 0) {
2969                                 struct net_device *dev;
2970                                 *mlp = ml->next;
2971                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2972                                 if (dev)
2973                                         packet_dev_mc(dev, ml, -1);
2974                                 kfree(ml);
2975                         }
2976                         rtnl_unlock();
2977                         return 0;
2978                 }
2979         }
2980         rtnl_unlock();
2981         return -EADDRNOTAVAIL;
2982 }
2983
2984 static void packet_flush_mclist(struct sock *sk)
2985 {
2986         struct packet_sock *po = pkt_sk(sk);
2987         struct packet_mclist *ml;
2988
2989         if (!po->mclist)
2990                 return;
2991
2992         rtnl_lock();
2993         while ((ml = po->mclist) != NULL) {
2994                 struct net_device *dev;
2995
2996                 po->mclist = ml->next;
2997                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2998                 if (dev != NULL)
2999                         packet_dev_mc(dev, ml, -1);
3000                 kfree(ml);
3001         }
3002         rtnl_unlock();
3003 }
3004
3005 static int
3006 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3007 {
3008         struct sock *sk = sock->sk;
3009         struct packet_sock *po = pkt_sk(sk);
3010         int ret;
3011
3012         if (level != SOL_PACKET)
3013                 return -ENOPROTOOPT;
3014
3015         switch (optname) {
3016         case PACKET_ADD_MEMBERSHIP:
3017         case PACKET_DROP_MEMBERSHIP:
3018         {
3019                 struct packet_mreq_max mreq;
3020                 int len = optlen;
3021                 memset(&mreq, 0, sizeof(mreq));
3022                 if (len < sizeof(struct packet_mreq))
3023                         return -EINVAL;
3024                 if (len > sizeof(mreq))
3025                         len = sizeof(mreq);
3026                 if (copy_from_user(&mreq, optval, len))
3027                         return -EFAULT;
3028                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3029                         return -EINVAL;
3030                 if (optname == PACKET_ADD_MEMBERSHIP)
3031                         ret = packet_mc_add(sk, &mreq);
3032                 else
3033                         ret = packet_mc_drop(sk, &mreq);
3034                 return ret;
3035         }
3036
3037         case PACKET_RX_RING:
3038         case PACKET_TX_RING:
3039         {
3040                 union tpacket_req_u req_u;
3041                 int len;
3042
3043                 switch (po->tp_version) {
3044                 case TPACKET_V1:
3045                 case TPACKET_V2:
3046                         len = sizeof(req_u.req);
3047                         break;
3048                 case TPACKET_V3:
3049                 default:
3050                         len = sizeof(req_u.req3);
3051                         break;
3052                 }
3053                 if (optlen < len)
3054                         return -EINVAL;
3055                 if (pkt_sk(sk)->has_vnet_hdr)
3056                         return -EINVAL;
3057                 if (copy_from_user(&req_u.req, optval, len))
3058                         return -EFAULT;
3059                 return packet_set_ring(sk, &req_u, 0,
3060                         optname == PACKET_TX_RING);
3061         }
3062         case PACKET_COPY_THRESH:
3063         {
3064                 int val;
3065
3066                 if (optlen != sizeof(val))
3067                         return -EINVAL;
3068                 if (copy_from_user(&val, optval, sizeof(val)))
3069                         return -EFAULT;
3070
3071                 pkt_sk(sk)->copy_thresh = val;
3072                 return 0;
3073         }
3074         case PACKET_VERSION:
3075         {
3076                 int val;
3077
3078                 if (optlen != sizeof(val))
3079                         return -EINVAL;
3080                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3081                         return -EBUSY;
3082                 if (copy_from_user(&val, optval, sizeof(val)))
3083                         return -EFAULT;
3084                 switch (val) {
3085                 case TPACKET_V1:
3086                 case TPACKET_V2:
3087                 case TPACKET_V3:
3088                         po->tp_version = val;
3089                         return 0;
3090                 default:
3091                         return -EINVAL;
3092                 }
3093         }
3094         case PACKET_RESERVE:
3095         {
3096                 unsigned int val;
3097
3098                 if (optlen != sizeof(val))
3099                         return -EINVAL;
3100                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3101                         return -EBUSY;
3102                 if (copy_from_user(&val, optval, sizeof(val)))
3103                         return -EFAULT;
3104                 po->tp_reserve = val;
3105                 return 0;
3106         }
3107         case PACKET_LOSS:
3108         {
3109                 unsigned int val;
3110
3111                 if (optlen != sizeof(val))
3112                         return -EINVAL;
3113                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3114                         return -EBUSY;
3115                 if (copy_from_user(&val, optval, sizeof(val)))
3116                         return -EFAULT;
3117                 po->tp_loss = !!val;
3118                 return 0;
3119         }
3120         case PACKET_AUXDATA:
3121         {
3122                 int val;
3123
3124                 if (optlen < sizeof(val))
3125                         return -EINVAL;
3126                 if (copy_from_user(&val, optval, sizeof(val)))
3127                         return -EFAULT;
3128
3129                 po->auxdata = !!val;
3130                 return 0;
3131         }
3132         case PACKET_ORIGDEV:
3133         {
3134                 int val;
3135
3136                 if (optlen < sizeof(val))
3137                         return -EINVAL;
3138                 if (copy_from_user(&val, optval, sizeof(val)))
3139                         return -EFAULT;
3140
3141                 po->origdev = !!val;
3142                 return 0;
3143         }
3144         case PACKET_VNET_HDR:
3145         {
3146                 int val;
3147
3148                 if (sock->type != SOCK_RAW)
3149                         return -EINVAL;
3150                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3151                         return -EBUSY;
3152                 if (optlen < sizeof(val))
3153                         return -EINVAL;
3154                 if (copy_from_user(&val, optval, sizeof(val)))
3155                         return -EFAULT;
3156
3157                 po->has_vnet_hdr = !!val;
3158                 return 0;
3159         }
3160         case PACKET_TIMESTAMP:
3161         {
3162                 int val;
3163
3164                 if (optlen != sizeof(val))
3165                         return -EINVAL;
3166                 if (copy_from_user(&val, optval, sizeof(val)))
3167                         return -EFAULT;
3168
3169                 po->tp_tstamp = val;
3170                 return 0;
3171         }
3172         case PACKET_FANOUT:
3173         {
3174                 int val;
3175
3176                 if (optlen != sizeof(val))
3177                         return -EINVAL;
3178                 if (copy_from_user(&val, optval, sizeof(val)))
3179                         return -EFAULT;
3180
3181                 return fanout_add(sk, val & 0xffff, val >> 16);
3182         }
3183         case PACKET_TX_HAS_OFF:
3184         {
3185                 unsigned int val;
3186
3187                 if (optlen != sizeof(val))
3188                         return -EINVAL;
3189                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3190                         return -EBUSY;
3191                 if (copy_from_user(&val, optval, sizeof(val)))
3192                         return -EFAULT;
3193                 po->tp_tx_has_off = !!val;
3194                 return 0;
3195         }
3196         default:
3197                 return -ENOPROTOOPT;
3198         }
3199 }
3200
3201 static int packet_getsockopt(struct socket *sock, int level, int optname,
3202                              char __user *optval, int __user *optlen)
3203 {
3204         int len;
3205         int val, lv = sizeof(val);
3206         struct sock *sk = sock->sk;
3207         struct packet_sock *po = pkt_sk(sk);
3208         void *data = &val;
3209         union tpacket_stats_u st;
3210
3211         if (level != SOL_PACKET)
3212                 return -ENOPROTOOPT;
3213
3214         if (get_user(len, optlen))
3215                 return -EFAULT;
3216
3217         if (len < 0)
3218                 return -EINVAL;
3219
3220         switch (optname) {
3221         case PACKET_STATISTICS:
3222                 spin_lock_bh(&sk->sk_receive_queue.lock);
3223                 memcpy(&st, &po->stats, sizeof(st));
3224                 memset(&po->stats, 0, sizeof(po->stats));
3225                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3226
3227                 if (po->tp_version == TPACKET_V3) {
3228                         lv = sizeof(struct tpacket_stats_v3);
3229                         st.stats3.tp_packets += st.stats3.tp_drops;
3230                         data = &st.stats3;
3231                 } else {
3232                         lv = sizeof(struct tpacket_stats);
3233                         st.stats1.tp_packets += st.stats1.tp_drops;
3234                         data = &st.stats1;
3235                 }
3236
3237                 break;
3238         case PACKET_AUXDATA:
3239                 val = po->auxdata;
3240                 break;
3241         case PACKET_ORIGDEV:
3242                 val = po->origdev;
3243                 break;
3244         case PACKET_VNET_HDR:
3245                 val = po->has_vnet_hdr;
3246                 break;
3247         case PACKET_VERSION:
3248                 val = po->tp_version;
3249                 break;
3250         case PACKET_HDRLEN:
3251                 if (len > sizeof(int))
3252                         len = sizeof(int);
3253                 if (copy_from_user(&val, optval, len))
3254                         return -EFAULT;
3255                 switch (val) {
3256                 case TPACKET_V1:
3257                         val = sizeof(struct tpacket_hdr);
3258                         break;
3259                 case TPACKET_V2:
3260                         val = sizeof(struct tpacket2_hdr);
3261                         break;
3262                 case TPACKET_V3:
3263                         val = sizeof(struct tpacket3_hdr);
3264                         break;
3265                 default:
3266                         return -EINVAL;
3267                 }
3268                 break;
3269         case PACKET_RESERVE:
3270                 val = po->tp_reserve;
3271                 break;
3272         case PACKET_LOSS:
3273                 val = po->tp_loss;
3274                 break;
3275         case PACKET_TIMESTAMP:
3276                 val = po->tp_tstamp;
3277                 break;
3278         case PACKET_FANOUT:
3279                 val = (po->fanout ?
3280                        ((u32)po->fanout->id |
3281                         ((u32)po->fanout->type << 16) |
3282                         ((u32)po->fanout->flags << 24)) :
3283                        0);
3284                 break;
3285         case PACKET_TX_HAS_OFF:
3286                 val = po->tp_tx_has_off;
3287                 break;
3288         default:
3289                 return -ENOPROTOOPT;
3290         }
3291
3292         if (len > lv)
3293                 len = lv;
3294         if (put_user(len, optlen))
3295                 return -EFAULT;
3296         if (copy_to_user(optval, data, len))
3297                 return -EFAULT;
3298         return 0;
3299 }
3300
3301
3302 static int packet_notifier(struct notifier_block *this,
3303                            unsigned long msg, void *ptr)
3304 {
3305         struct sock *sk;
3306         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3307         struct net *net = dev_net(dev);
3308
3309         rcu_read_lock();
3310         sk_for_each_rcu(sk, &net->packet.sklist) {
3311                 struct packet_sock *po = pkt_sk(sk);
3312
3313                 switch (msg) {
3314                 case NETDEV_UNREGISTER:
3315                         if (po->mclist)
3316                                 packet_dev_mclist(dev, po->mclist, -1);
3317                         /* fallthrough */
3318
3319                 case NETDEV_DOWN:
3320                         if (dev->ifindex == po->ifindex) {
3321                                 spin_lock(&po->bind_lock);
3322                                 if (po->running) {
3323                                         __unregister_prot_hook(sk, false);
3324                                         sk->sk_err = ENETDOWN;
3325                                         if (!sock_flag(sk, SOCK_DEAD))
3326                                                 sk->sk_error_report(sk);
3327                                 }
3328                                 if (msg == NETDEV_UNREGISTER) {
3329                                         po->ifindex = -1;
3330                                         if (po->prot_hook.dev)
3331                                                 dev_put(po->prot_hook.dev);
3332                                         po->prot_hook.dev = NULL;
3333                                 }
3334                                 spin_unlock(&po->bind_lock);
3335                         }
3336                         break;
3337                 case NETDEV_UP:
3338                         if (dev->ifindex == po->ifindex) {
3339                                 spin_lock(&po->bind_lock);
3340                                 if (po->num)
3341                                         register_prot_hook(sk);
3342                                 spin_unlock(&po->bind_lock);
3343                         }
3344                         break;
3345                 }
3346         }
3347         rcu_read_unlock();
3348         return NOTIFY_DONE;
3349 }
3350
3351
3352 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3353                         unsigned long arg)
3354 {
3355         struct sock *sk = sock->sk;
3356
3357         switch (cmd) {
3358         case SIOCOUTQ:
3359         {
3360                 int amount = sk_wmem_alloc_get(sk);
3361
3362                 return put_user(amount, (int __user *)arg);
3363         }
3364         case SIOCINQ:
3365         {
3366                 struct sk_buff *skb;
3367                 int amount = 0;
3368
3369                 spin_lock_bh(&sk->sk_receive_queue.lock);
3370                 skb = skb_peek(&sk->sk_receive_queue);
3371                 if (skb)
3372                         amount = skb->len;
3373                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3374                 return put_user(amount, (int __user *)arg);
3375         }
3376         case SIOCGSTAMP:
3377                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3378         case SIOCGSTAMPNS:
3379                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3380
3381 #ifdef CONFIG_INET
3382         case SIOCADDRT:
3383         case SIOCDELRT:
3384         case SIOCDARP:
3385         case SIOCGARP:
3386         case SIOCSARP:
3387         case SIOCGIFADDR:
3388         case SIOCSIFADDR:
3389         case SIOCGIFBRDADDR:
3390         case SIOCSIFBRDADDR:
3391         case SIOCGIFNETMASK:
3392         case SIOCSIFNETMASK:
3393         case SIOCGIFDSTADDR:
3394         case SIOCSIFDSTADDR:
3395         case SIOCSIFFLAGS:
3396                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3397 #endif
3398
3399         default:
3400                 return -ENOIOCTLCMD;
3401         }
3402         return 0;
3403 }
3404
3405 static unsigned int packet_poll(struct file *file, struct socket *sock,
3406                                 poll_table *wait)
3407 {
3408         struct sock *sk = sock->sk;
3409         struct packet_sock *po = pkt_sk(sk);
3410         unsigned int mask = datagram_poll(file, sock, wait);
3411
3412         spin_lock_bh(&sk->sk_receive_queue.lock);
3413         if (po->rx_ring.pg_vec) {
3414                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3415                         TP_STATUS_KERNEL))
3416                         mask |= POLLIN | POLLRDNORM;
3417         }
3418         spin_unlock_bh(&sk->sk_receive_queue.lock);
3419         spin_lock_bh(&sk->sk_write_queue.lock);
3420         if (po->tx_ring.pg_vec) {
3421                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3422                         mask |= POLLOUT | POLLWRNORM;
3423         }
3424         spin_unlock_bh(&sk->sk_write_queue.lock);
3425         return mask;
3426 }
3427
3428
3429 /* Dirty? Well, I still did not learn better way to account
3430  * for user mmaps.
3431  */
3432
3433 static void packet_mm_open(struct vm_area_struct *vma)
3434 {
3435         struct file *file = vma->vm_file;
3436         struct socket *sock = file->private_data;
3437         struct sock *sk = sock->sk;
3438
3439         if (sk)
3440                 atomic_inc(&pkt_sk(sk)->mapped);
3441 }
3442
3443 static void packet_mm_close(struct vm_area_struct *vma)
3444 {
3445         struct file *file = vma->vm_file;
3446         struct socket *sock = file->private_data;
3447         struct sock *sk = sock->sk;
3448
3449         if (sk)
3450                 atomic_dec(&pkt_sk(sk)->mapped);
3451 }
3452
3453 static const struct vm_operations_struct packet_mmap_ops = {
3454         .open   =       packet_mm_open,
3455         .close  =       packet_mm_close,
3456 };
3457
3458 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3459                         unsigned int len)
3460 {
3461         int i;
3462
3463         for (i = 0; i < len; i++) {
3464                 if (likely(pg_vec[i].buffer)) {
3465                         if (is_vmalloc_addr(pg_vec[i].buffer))
3466                                 vfree(pg_vec[i].buffer);
3467                         else
3468                                 free_pages((unsigned long)pg_vec[i].buffer,
3469                                            order);
3470                         pg_vec[i].buffer = NULL;
3471                 }
3472         }
3473         kfree(pg_vec);
3474 }
3475
3476 static char *alloc_one_pg_vec_page(unsigned long order)
3477 {
3478         char *buffer = NULL;
3479         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3480                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3481
3482         buffer = (char *) __get_free_pages(gfp_flags, order);
3483
3484         if (buffer)
3485                 return buffer;
3486
3487         /*
3488          * __get_free_pages failed, fall back to vmalloc
3489          */
3490         buffer = vzalloc((1 << order) * PAGE_SIZE);
3491
3492         if (buffer)
3493                 return buffer;
3494
3495         /*
3496          * vmalloc failed, lets dig into swap here
3497          */
3498         gfp_flags &= ~__GFP_NORETRY;
3499         buffer = (char *)__get_free_pages(gfp_flags, order);
3500         if (buffer)
3501                 return buffer;
3502
3503         /*
3504          * complete and utter failure
3505          */
3506         return NULL;
3507 }
3508
3509 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3510 {
3511         unsigned int block_nr = req->tp_block_nr;
3512         struct pgv *pg_vec;
3513         int i;
3514
3515         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3516         if (unlikely(!pg_vec))
3517                 goto out;
3518
3519         for (i = 0; i < block_nr; i++) {
3520                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3521                 if (unlikely(!pg_vec[i].buffer))
3522                         goto out_free_pgvec;
3523         }
3524
3525 out:
3526         return pg_vec;
3527
3528 out_free_pgvec:
3529         free_pg_vec(pg_vec, order, block_nr);
3530         pg_vec = NULL;
3531         goto out;
3532 }
3533
3534 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3535                 int closing, int tx_ring)
3536 {
3537         struct pgv *pg_vec = NULL;
3538         struct packet_sock *po = pkt_sk(sk);
3539         int was_running, order = 0;
3540         struct packet_ring_buffer *rb;
3541         struct sk_buff_head *rb_queue;
3542         __be16 num;
3543         int err = -EINVAL;
3544         /* Added to avoid minimal code churn */
3545         struct tpacket_req *req = &req_u->req;
3546
3547         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3548         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3549                 WARN(1, "Tx-ring is not supported.\n");
3550                 goto out;
3551         }
3552
3553         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3554         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3555
3556         err = -EBUSY;
3557         if (!closing) {
3558                 if (atomic_read(&po->mapped))
3559                         goto out;
3560                 if (atomic_read(&rb->pending))
3561                         goto out;
3562         }
3563
3564         if (req->tp_block_nr) {
3565                 /* Sanity tests and some calculations */
3566                 err = -EBUSY;
3567                 if (unlikely(rb->pg_vec))
3568                         goto out;
3569
3570                 switch (po->tp_version) {
3571                 case TPACKET_V1:
3572                         po->tp_hdrlen = TPACKET_HDRLEN;
3573                         break;
3574                 case TPACKET_V2:
3575                         po->tp_hdrlen = TPACKET2_HDRLEN;
3576                         break;
3577                 case TPACKET_V3:
3578                         po->tp_hdrlen = TPACKET3_HDRLEN;
3579                         break;
3580                 }
3581
3582                 err = -EINVAL;
3583                 if (unlikely((int)req->tp_block_size <= 0))
3584                         goto out;
3585                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3586                         goto out;
3587                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3588                                         po->tp_reserve))
3589                         goto out;
3590                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3591                         goto out;
3592
3593                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3594                 if (unlikely(rb->frames_per_block <= 0))
3595                         goto out;
3596                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3597                                         req->tp_frame_nr))
3598                         goto out;
3599
3600                 err = -ENOMEM;
3601                 order = get_order(req->tp_block_size);
3602                 pg_vec = alloc_pg_vec(req, order);
3603                 if (unlikely(!pg_vec))
3604                         goto out;
3605                 switch (po->tp_version) {
3606                 case TPACKET_V3:
3607                 /* Transmit path is not supported. We checked
3608                  * it above but just being paranoid
3609                  */
3610                         if (!tx_ring)
3611                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3612                                 break;
3613                 default:
3614                         break;
3615                 }
3616         }
3617         /* Done */
3618         else {
3619                 err = -EINVAL;
3620                 if (unlikely(req->tp_frame_nr))
3621                         goto out;
3622         }
3623
3624         lock_sock(sk);
3625
3626         /* Detach socket from network */
3627         spin_lock(&po->bind_lock);
3628         was_running = po->running;
3629         num = po->num;
3630         if (was_running) {
3631                 po->num = 0;
3632                 __unregister_prot_hook(sk, false);
3633         }
3634         spin_unlock(&po->bind_lock);
3635
3636         synchronize_net();
3637
3638         err = -EBUSY;
3639         mutex_lock(&po->pg_vec_lock);
3640         if (closing || atomic_read(&po->mapped) == 0) {
3641                 err = 0;
3642                 spin_lock_bh(&rb_queue->lock);
3643                 swap(rb->pg_vec, pg_vec);
3644                 rb->frame_max = (req->tp_frame_nr - 1);
3645                 rb->head = 0;
3646                 rb->frame_size = req->tp_frame_size;
3647                 spin_unlock_bh(&rb_queue->lock);
3648
3649                 swap(rb->pg_vec_order, order);
3650                 swap(rb->pg_vec_len, req->tp_block_nr);
3651
3652                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3653                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3654                                                 tpacket_rcv : packet_rcv;
3655                 skb_queue_purge(rb_queue);
3656                 if (atomic_read(&po->mapped))
3657                         pr_err("packet_mmap: vma is busy: %d\n",
3658                                atomic_read(&po->mapped));
3659         }
3660         mutex_unlock(&po->pg_vec_lock);
3661
3662         spin_lock(&po->bind_lock);
3663         if (was_running) {
3664                 po->num = num;
3665                 register_prot_hook(sk);
3666         }
3667         spin_unlock(&po->bind_lock);
3668         if (closing && (po->tp_version > TPACKET_V2)) {
3669                 /* Because we don't support block-based V3 on tx-ring */
3670                 if (!tx_ring)
3671                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3672         }
3673         release_sock(sk);
3674
3675         if (pg_vec)
3676                 free_pg_vec(pg_vec, order, req->tp_block_nr);
3677 out:
3678         return err;
3679 }
3680
3681 static int packet_mmap(struct file *file, struct socket *sock,
3682                 struct vm_area_struct *vma)
3683 {
3684         struct sock *sk = sock->sk;
3685         struct packet_sock *po = pkt_sk(sk);
3686         unsigned long size, expected_size;
3687         struct packet_ring_buffer *rb;
3688         unsigned long start;
3689         int err = -EINVAL;
3690         int i;
3691
3692         if (vma->vm_pgoff)
3693                 return -EINVAL;
3694
3695         mutex_lock(&po->pg_vec_lock);
3696
3697         expected_size = 0;
3698         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3699                 if (rb->pg_vec) {
3700                         expected_size += rb->pg_vec_len
3701                                                 * rb->pg_vec_pages
3702                                                 * PAGE_SIZE;
3703                 }
3704         }
3705
3706         if (expected_size == 0)
3707                 goto out;
3708
3709         size = vma->vm_end - vma->vm_start;
3710         if (size != expected_size)
3711                 goto out;
3712
3713         start = vma->vm_start;
3714         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3715                 if (rb->pg_vec == NULL)
3716                         continue;
3717
3718                 for (i = 0; i < rb->pg_vec_len; i++) {
3719                         struct page *page;
3720                         void *kaddr = rb->pg_vec[i].buffer;
3721                         int pg_num;
3722
3723                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3724                                 page = pgv_to_page(kaddr);
3725                                 err = vm_insert_page(vma, start, page);
3726                                 if (unlikely(err))
3727                                         goto out;
3728                                 start += PAGE_SIZE;
3729                                 kaddr += PAGE_SIZE;
3730                         }
3731                 }
3732         }
3733
3734         atomic_inc(&po->mapped);
3735         vma->vm_ops = &packet_mmap_ops;
3736         err = 0;
3737
3738 out:
3739         mutex_unlock(&po->pg_vec_lock);
3740         return err;
3741 }
3742
3743 static const struct proto_ops packet_ops_spkt = {
3744         .family =       PF_PACKET,
3745         .owner =        THIS_MODULE,
3746         .release =      packet_release,
3747         .bind =         packet_bind_spkt,
3748         .connect =      sock_no_connect,
3749         .socketpair =   sock_no_socketpair,
3750         .accept =       sock_no_accept,
3751         .getname =      packet_getname_spkt,
3752         .poll =         datagram_poll,
3753         .ioctl =        packet_ioctl,
3754         .listen =       sock_no_listen,
3755         .shutdown =     sock_no_shutdown,
3756         .setsockopt =   sock_no_setsockopt,
3757         .getsockopt =   sock_no_getsockopt,
3758         .sendmsg =      packet_sendmsg_spkt,
3759         .recvmsg =      packet_recvmsg,
3760         .mmap =         sock_no_mmap,
3761         .sendpage =     sock_no_sendpage,
3762 };
3763
3764 static const struct proto_ops packet_ops = {
3765         .family =       PF_PACKET,
3766         .owner =        THIS_MODULE,
3767         .release =      packet_release,
3768         .bind =         packet_bind,
3769         .connect =      sock_no_connect,
3770         .socketpair =   sock_no_socketpair,
3771         .accept =       sock_no_accept,
3772         .getname =      packet_getname,
3773         .poll =         packet_poll,
3774         .ioctl =        packet_ioctl,
3775         .listen =       sock_no_listen,
3776         .shutdown =     sock_no_shutdown,
3777         .setsockopt =   packet_setsockopt,
3778         .getsockopt =   packet_getsockopt,
3779         .sendmsg =      packet_sendmsg,
3780         .recvmsg =      packet_recvmsg,
3781         .mmap =         packet_mmap,
3782         .sendpage =     sock_no_sendpage,
3783 };
3784
3785 static const struct net_proto_family packet_family_ops = {
3786         .family =       PF_PACKET,
3787         .create =       packet_create,
3788         .owner  =       THIS_MODULE,
3789 };
3790
3791 static struct notifier_block packet_netdev_notifier = {
3792         .notifier_call =        packet_notifier,
3793 };
3794
3795 #ifdef CONFIG_PROC_FS
3796
3797 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3798         __acquires(RCU)
3799 {
3800         struct net *net = seq_file_net(seq);
3801
3802         rcu_read_lock();
3803         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3804 }
3805
3806 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3807 {
3808         struct net *net = seq_file_net(seq);
3809         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3810 }
3811
3812 static void packet_seq_stop(struct seq_file *seq, void *v)
3813         __releases(RCU)
3814 {
3815         rcu_read_unlock();
3816 }
3817
3818 static int packet_seq_show(struct seq_file *seq, void *v)
3819 {
3820         if (v == SEQ_START_TOKEN)
3821                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
3822         else {
3823                 struct sock *s = sk_entry(v);
3824                 const struct packet_sock *po = pkt_sk(s);
3825
3826                 seq_printf(seq,
3827                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
3828                            s,
3829                            atomic_read(&s->sk_refcnt),
3830                            s->sk_type,
3831                            ntohs(po->num),
3832                            po->ifindex,
3833                            po->running,
3834                            atomic_read(&s->sk_rmem_alloc),
3835                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
3836                            sock_i_ino(s));
3837         }
3838
3839         return 0;
3840 }
3841
3842 static const struct seq_operations packet_seq_ops = {
3843         .start  = packet_seq_start,
3844         .next   = packet_seq_next,
3845         .stop   = packet_seq_stop,
3846         .show   = packet_seq_show,
3847 };
3848
3849 static int packet_seq_open(struct inode *inode, struct file *file)
3850 {
3851         return seq_open_net(inode, file, &packet_seq_ops,
3852                             sizeof(struct seq_net_private));
3853 }
3854
3855 static const struct file_operations packet_seq_fops = {
3856         .owner          = THIS_MODULE,
3857         .open           = packet_seq_open,
3858         .read           = seq_read,
3859         .llseek         = seq_lseek,
3860         .release        = seq_release_net,
3861 };
3862
3863 #endif
3864
3865 static int __net_init packet_net_init(struct net *net)
3866 {
3867         mutex_init(&net->packet.sklist_lock);
3868         INIT_HLIST_HEAD(&net->packet.sklist);
3869
3870         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
3871                 return -ENOMEM;
3872
3873         return 0;
3874 }
3875
3876 static void __net_exit packet_net_exit(struct net *net)
3877 {
3878         remove_proc_entry("packet", net->proc_net);
3879 }
3880
3881 static struct pernet_operations packet_net_ops = {
3882         .init = packet_net_init,
3883         .exit = packet_net_exit,
3884 };
3885
3886
3887 static void __exit packet_exit(void)
3888 {
3889         unregister_netdevice_notifier(&packet_netdev_notifier);
3890         unregister_pernet_subsys(&packet_net_ops);
3891         sock_unregister(PF_PACKET);
3892         proto_unregister(&packet_proto);
3893 }
3894
3895 static int __init packet_init(void)
3896 {
3897         int rc = proto_register(&packet_proto, 0);
3898
3899         if (rc != 0)
3900                 goto out;
3901
3902         sock_register(&packet_family_ops);
3903         register_pernet_subsys(&packet_net_ops);
3904         register_netdevice_notifier(&packet_netdev_notifier);
3905 out:
3906         return rc;
3907 }
3908
3909 module_init(packet_init);
3910 module_exit(packet_exit);
3911 MODULE_LICENSE("GPL");
3912 MODULE_ALIAS_NETPROTO(PF_PACKET);