Pileus Git - ~andy/linux/blob - 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 <net/flow_keys.h>
  92
  93 #ifdef CONFIG_INET
  94 #include <net/inet_common.h>
  95 #endif
  96
  97 #include "internal.h"
  98
  99 /*
 100    Assumptions:
 101    - if device has no dev->hard_header routine, it adds and removes ll header
 102      inside itself. In this case ll header is invisible outside of device,
 103      but higher levels still should reserve dev->hard_header_len.
 104      Some devices are enough clever to reallocate skb, when header
 105      will not fit to reserved space (tunnel), another ones are silly
 106      (PPP).
 107    - packet socket receives packets with pulled ll header,
 108      so that SOCK_RAW should push it back.
 109
 110 On receive:
 111 -----------
 112
 113 Incoming, dev->hard_header!=NULL
 114    mac_header -> ll header
 115    data       -> data
 116
 117 Outgoing, dev->hard_header!=NULL
 118    mac_header -> ll header
 119    data       -> ll header
 120
 121 Incoming, dev->hard_header==NULL
 122    mac_header -> UNKNOWN position. It is very likely, that it points to ll
 123                  header.  PPP makes it, that is wrong, because introduce
 124                  assymetry between rx and tx paths.
 125    data       -> data
 126
 127 Outgoing, dev->hard_header==NULL
 128    mac_header -> data. ll header is still not built!
 129    data       -> data
 130
 131 Resume
 132   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 133
 134
 135 On transmit:
 136 ------------
 137
 138 dev->hard_header != NULL
 139    mac_header -> ll header
 140    data       -> ll header
 141
 142 dev->hard_header == NULL (ll header is added by device, we cannot control it)
 143    mac_header -> data
 144    data       -> data
 145
 146    We should set nh.raw on output to correct posistion,
 147    packet classifier depends on it.
 148  */
 149
 150 /* Private packet socket structures. */
 151
 152 /* identical to struct packet_mreq except it has
 153  * a longer address field.
 154  */
 155 struct packet_mreq_max {
 156         int             mr_ifindex;
 157         unsigned short  mr_type;
 158         unsigned short  mr_alen;
 159         unsigned char   mr_address[MAX_ADDR_LEN];
 160 };
 161
 162 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 163                 int closing, int tx_ring);
 164
 165
 166 #define V3_ALIGNMENT    (8)
 167
 168 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 169
 170 #define BLK_PLUS_PRIV(sz_of_priv) \
 171         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 172
 173 #define PGV_FROM_VMALLOC 1
 174
 175 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
 176 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
 177 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
 178 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
 179 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
 180 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
 181 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
 182
 183 struct packet_sock;
 184 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
 185 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 186                        struct packet_type *pt, struct net_device *orig_dev);
 187
 188 static void *packet_previous_frame(struct packet_sock *po,
 189                 struct packet_ring_buffer *rb,
 190                 int status);
 191 static void packet_increment_head(struct packet_ring_buffer *buff);
 192 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
 193                         struct tpacket_block_desc *);
 194 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 195                         struct packet_sock *);
 196 static void prb_retire_current_block(struct tpacket_kbdq_core *,
 197                 struct packet_sock *, unsigned int status);
 198 static int prb_queue_frozen(struct tpacket_kbdq_core *);
 199 static void prb_open_block(struct tpacket_kbdq_core *,
 200                 struct tpacket_block_desc *);
 201 static void prb_retire_rx_blk_timer_expired(unsigned long);
 202 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 203 static void prb_init_blk_timer(struct packet_sock *,
 204                 struct tpacket_kbdq_core *,
 205                 void (*func) (unsigned long));
 206 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 207 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 208                 struct tpacket3_hdr *);
 209 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 210                 struct tpacket3_hdr *);
 211 static void packet_flush_mclist(struct sock *sk);
 212
 213 struct packet_skb_cb {
 214         unsigned int origlen;
 215         union {
 216                 struct sockaddr_pkt pkt;
 217                 struct sockaddr_ll ll;
 218         } sa;
 219 };
 220
 221 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 222
 223 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 224 #define GET_PBLOCK_DESC(x, bid) \
 225         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 226 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 227         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 228 #define GET_NEXT_PRB_BLK_NUM(x) \
 229         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 230         ((x)->kactive_blk_num+1) : 0)
 231
 232 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 233 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 234
 235 /* register_prot_hook must be invoked with the po->bind_lock held,
 236  * or from a context in which asynchronous accesses to the packet
 237  * socket is not possible (packet_create()).
 238  */
 239 static void register_prot_hook(struct sock *sk)
 240 {
 241         struct packet_sock *po = pkt_sk(sk);
 242         if (!po->running) {
 243                 if (po->fanout)
 244                         __fanout_link(sk, po);
 245                 else
 246                         dev_add_pack(&po->prot_hook);
 247                 sock_hold(sk);
 248                 po->running = 1;
 249         }
 250 }
 251
 252 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
 253  * held.   If the sync parameter is true, we will temporarily drop
 254  * the po->bind_lock and do a synchronize_net to make sure no
 255  * asynchronous packet processing paths still refer to the elements
 256  * of po->prot_hook.  If the sync parameter is false, it is the
 257  * callers responsibility to take care of this.
 258  */
 259 static void __unregister_prot_hook(struct sock *sk, bool sync)
 260 {
 261         struct packet_sock *po = pkt_sk(sk);
 262
 263         po->running = 0;
 264         if (po->fanout)
 265                 __fanout_unlink(sk, po);
 266         else
 267                 __dev_remove_pack(&po->prot_hook);
 268         __sock_put(sk);
 269
 270         if (sync) {
 271                 spin_unlock(&po->bind_lock);
 272                 synchronize_net();
 273                 spin_lock(&po->bind_lock);
 274         }
 275 }
 276
 277 static void unregister_prot_hook(struct sock *sk, bool sync)
 278 {
 279         struct packet_sock *po = pkt_sk(sk);
 280
 281         if (po->running)
 282                 __unregister_prot_hook(sk, sync);
 283 }
 284
 285 static inline __pure struct page *pgv_to_page(void *addr)
 286 {
 287         if (is_vmalloc_addr(addr))
 288                 return vmalloc_to_page(addr);
 289         return virt_to_page(addr);
 290 }
 291
 292 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 293 {
 294         union {
 295                 struct tpacket_hdr *h1;
 296                 struct tpacket2_hdr *h2;
 297                 void *raw;
 298         } h;
 299
 300         h.raw = frame;
 301         switch (po->tp_version) {
 302         case TPACKET_V1:
 303                 h.h1->tp_status = status;
 304                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 305                 break;
 306         case TPACKET_V2:
 307                 h.h2->tp_status = status;
 308                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 309                 break;
 310         case TPACKET_V3:
 311         default:
 312                 WARN(1, "TPACKET version not supported.\n");
 313                 BUG();
 314         }
 315
 316         smp_wmb();
 317 }
 318
 319 static int __packet_get_status(struct packet_sock *po, void *frame)
 320 {
 321         union {
 322                 struct tpacket_hdr *h1;
 323                 struct tpacket2_hdr *h2;
 324                 void *raw;
 325         } h;
 326
 327         smp_rmb();
 328
 329         h.raw = frame;
 330         switch (po->tp_version) {
 331         case TPACKET_V1:
 332                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 333                 return h.h1->tp_status;
 334         case TPACKET_V2:
 335                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 336                 return h.h2->tp_status;
 337         case TPACKET_V3:
 338         default:
 339                 WARN(1, "TPACKET version not supported.\n");
 340                 BUG();
 341                 return 0;
 342         }
 343 }
 344
 345 static void *packet_lookup_frame(struct packet_sock *po,
 346                 struct packet_ring_buffer *rb,
 347                 unsigned int position,
 348                 int status)
 349 {
 350         unsigned int pg_vec_pos, frame_offset;
 351         union {
 352                 struct tpacket_hdr *h1;
 353                 struct tpacket2_hdr *h2;
 354                 void *raw;
 355         } h;
 356
 357         pg_vec_pos = position / rb->frames_per_block;
 358         frame_offset = position % rb->frames_per_block;
 359
 360         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 361                 (frame_offset * rb->frame_size);
 362
 363         if (status != __packet_get_status(po, h.raw))
 364                 return NULL;
 365
 366         return h.raw;
 367 }
 368
 369 static void *packet_current_frame(struct packet_sock *po,
 370                 struct packet_ring_buffer *rb,
 371                 int status)
 372 {
 373         return packet_lookup_frame(po, rb, rb->head, status);
 374 }
 375
 376 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 377 {
 378         del_timer_sync(&pkc->retire_blk_timer);
 379 }
 380
 381 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 382                 int tx_ring,
 383                 struct sk_buff_head *rb_queue)
 384 {
 385         struct tpacket_kbdq_core *pkc;
 386
 387         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 388
 389         spin_lock(&rb_queue->lock);
 390         pkc->delete_blk_timer = 1;
 391         spin_unlock(&rb_queue->lock);
 392
 393         prb_del_retire_blk_timer(pkc);
 394 }
 395
 396 static void prb_init_blk_timer(struct packet_sock *po,
 397                 struct tpacket_kbdq_core *pkc,
 398                 void (*func) (unsigned long))
 399 {
 400         init_timer(&pkc->retire_blk_timer);
 401         pkc->retire_blk_timer.data = (long)po;
 402         pkc->retire_blk_timer.function = func;
 403         pkc->retire_blk_timer.expires = jiffies;
 404 }
 405
 406 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
 407 {
 408         struct tpacket_kbdq_core *pkc;
 409
 410         if (tx_ring)
 411                 BUG();
 412
 413         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 414         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
 415 }
 416
 417 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 418                                 int blk_size_in_bytes)
 419 {
 420         struct net_device *dev;
 421         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 422         struct ethtool_cmd ecmd;
 423         int err;
 424         u32 speed;
 425
 426         rtnl_lock();
 427         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 428         if (unlikely(!dev)) {
 429                 rtnl_unlock();
 430                 return DEFAULT_PRB_RETIRE_TOV;
 431         }
 432         err = __ethtool_get_settings(dev, &ecmd);
 433         speed = ethtool_cmd_speed(&ecmd);
 434         rtnl_unlock();
 435         if (!err) {
 436                 /*
 437                  * If the link speed is so slow you don't really
 438                  * need to worry about perf anyways
 439                  */
 440                 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
 441                         return DEFAULT_PRB_RETIRE_TOV;
 442                 } else {
 443                         msec = 1;
 444                         div = speed / 1000;
 445                 }
 446         }
 447
 448         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 449
 450         if (div)
 451                 mbits /= div;
 452
 453         tmo = mbits * msec;
 454
 455         if (div)
 456                 return tmo+1;
 457         return tmo;
 458 }
 459
 460 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 461                         union tpacket_req_u *req_u)
 462 {
 463         p1->feature_req_word = req_u->req3.tp_feature_req_word;
 464 }
 465
 466 static void init_prb_bdqc(struct packet_sock *po,
 467                         struct packet_ring_buffer *rb,
 468                         struct pgv *pg_vec,
 469                         union tpacket_req_u *req_u, int tx_ring)
 470 {
 471         struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
 472         struct tpacket_block_desc *pbd;
 473
 474         memset(p1, 0x0, sizeof(*p1));
 475
 476         p1->knxt_seq_num = 1;
 477         p1->pkbdq = pg_vec;
 478         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 479         p1->pkblk_start = pg_vec[0].buffer;
 480         p1->kblk_size = req_u->req3.tp_block_size;
 481         p1->knum_blocks = req_u->req3.tp_block_nr;
 482         p1->hdrlen = po->tp_hdrlen;
 483         p1->version = po->tp_version;
 484         p1->last_kactive_blk_num = 0;
 485         po->stats_u.stats3.tp_freeze_q_cnt = 0;
 486         if (req_u->req3.tp_retire_blk_tov)
 487                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 488         else
 489                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 490                                                 req_u->req3.tp_block_size);
 491         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 492         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 493
 494         prb_init_ft_ops(p1, req_u);
 495         prb_setup_retire_blk_timer(po, tx_ring);
 496         prb_open_block(p1, pbd);
 497 }
 498
 499 /*  Do NOT update the last_blk_num first.
 500  *  Assumes sk_buff_head lock is held.
 501  */
 502 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 503 {
 504         mod_timer(&pkc->retire_blk_timer,
 505                         jiffies + pkc->tov_in_jiffies);
 506         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 507 }
 508
 509 /*
 510  * Timer logic:
 511  * 1) We refresh the timer only when we open a block.
 512  *    By doing this we don't waste cycles refreshing the timer
 513  *        on packet-by-packet basis.
 514  *
 515  * With a 1MB block-size, on a 1Gbps line, it will take
 516  * i) ~8 ms to fill a block + ii) memcpy etc.
 517  * In this cut we are not accounting for the memcpy time.
 518  *
 519  * So, if the user sets the 'tmo' to 10ms then the timer
 520  * will never fire while the block is still getting filled
 521  * (which is what we want). However, the user could choose
 522  * to close a block early and that's fine.
 523  *
 524  * But when the timer does fire, we check whether or not to refresh it.
 525  * Since the tmo granularity is in msecs, it is not too expensive
 526  * to refresh the timer, lets say every '8' msecs.
 527  * Either the user can set the 'tmo' or we can derive it based on
 528  * a) line-speed and b) block-size.
 529  * prb_calc_retire_blk_tmo() calculates the tmo.
 530  *
 531  */
 532 static void prb_retire_rx_blk_timer_expired(unsigned long data)
 533 {
 534         struct packet_sock *po = (struct packet_sock *)data;
 535         struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
 536         unsigned int frozen;
 537         struct tpacket_block_desc *pbd;
 538
 539         spin_lock(&po->sk.sk_receive_queue.lock);
 540
 541         frozen = prb_queue_frozen(pkc);
 542         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 543
 544         if (unlikely(pkc->delete_blk_timer))
 545                 goto out;
 546
 547         /* We only need to plug the race when the block is partially filled.
 548          * tpacket_rcv:
 549          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 550          *              copy_bits() is in progress ...
 551          *              timer fires on other cpu:
 552          *              we can't retire the current block because copy_bits
 553          *              is in progress.
 554          *
 555          */
 556         if (BLOCK_NUM_PKTS(pbd)) {
 557                 while (atomic_read(&pkc->blk_fill_in_prog)) {
 558                         /* Waiting for skb_copy_bits to finish... */
 559                         cpu_relax();
 560                 }
 561         }
 562
 563         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 564                 if (!frozen) {
 565                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 566                         if (!prb_dispatch_next_block(pkc, po))
 567                                 goto refresh_timer;
 568                         else
 569                                 goto out;
 570                 } else {
 571                         /* Case 1. Queue was frozen because user-space was
 572                          *         lagging behind.
 573                          */
 574                         if (prb_curr_blk_in_use(pkc, pbd)) {
 575                                 /*
 576                                  * Ok, user-space is still behind.
 577                                  * So just refresh the timer.
 578                                  */
 579                                 goto refresh_timer;
 580                         } else {
 581                                /* Case 2. queue was frozen,user-space caught up,
 582                                 * now the link went idle && the timer fired.
 583                                 * We don't have a block to close.So we open this
 584                                 * block and restart the timer.
 585                                 * opening a block thaws the queue,restarts timer
 586                                 * Thawing/timer-refresh is a side effect.
 587                                 */
 588                                 prb_open_block(pkc, pbd);
 589                                 goto out;
 590                         }
 591                 }
 592         }
 593
 594 refresh_timer:
 595         _prb_refresh_rx_retire_blk_timer(pkc);
 596
 597 out:
 598         spin_unlock(&po->sk.sk_receive_queue.lock);
 599 }
 600
 601 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 602                 struct tpacket_block_desc *pbd1, __u32 status)
 603 {
 604         /* Flush everything minus the block header */
 605
 606 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 607         u8 *start, *end;
 608
 609         start = (u8 *)pbd1;
 610
 611         /* Skip the block header(we know header WILL fit in 4K) */
 612         start += PAGE_SIZE;
 613
 614         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 615         for (; start < end; start += PAGE_SIZE)
 616                 flush_dcache_page(pgv_to_page(start));
 617
 618         smp_wmb();
 619 #endif
 620
 621         /* Now update the block status. */
 622
 623         BLOCK_STATUS(pbd1) = status;
 624
 625         /* Flush the block header */
 626
 627 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 628         start = (u8 *)pbd1;
 629         flush_dcache_page(pgv_to_page(start));
 630
 631         smp_wmb();
 632 #endif
 633 }
 634
 635 /*
 636  * Side effect:
 637  *
 638  * 1) flush the block
 639  * 2) Increment active_blk_num
 640  *
 641  * Note:We DONT refresh the timer on purpose.
 642  *      Because almost always the next block will be opened.
 643  */
 644 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 645                 struct tpacket_block_desc *pbd1,
 646                 struct packet_sock *po, unsigned int stat)
 647 {
 648         __u32 status = TP_STATUS_USER | stat;
 649
 650         struct tpacket3_hdr *last_pkt;
 651         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 652
 653         if (po->stats.tp_drops)
 654                 status |= TP_STATUS_LOSING;
 655
 656         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 657         last_pkt->tp_next_offset = 0;
 658
 659         /* Get the ts of the last pkt */
 660         if (BLOCK_NUM_PKTS(pbd1)) {
 661                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 662                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 663         } else {
 664                 /* Ok, we tmo'd - so get the current time */
 665                 struct timespec ts;
 666                 getnstimeofday(&ts);
 667                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 668                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 669         }
 670
 671         smp_wmb();
 672
 673         /* Flush the block */
 674         prb_flush_block(pkc1, pbd1, status);
 675
 676         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 677 }
 678
 679 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 680 {
 681         pkc->reset_pending_on_curr_blk = 0;
 682 }
 683
 684 /*
 685  * Side effect of opening a block:
 686  *
 687  * 1) prb_queue is thawed.
 688  * 2) retire_blk_timer is refreshed.
 689  *
 690  */
 691 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 692         struct tpacket_block_desc *pbd1)
 693 {
 694         struct timespec ts;
 695         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 696
 697         smp_rmb();
 698
 699         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd1))) {
 700
 701                 /* We could have just memset this but we will lose the
 702                  * flexibility of making the priv area sticky
 703                  */
 704                 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 705                 BLOCK_NUM_PKTS(pbd1) = 0;
 706                 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 707                 getnstimeofday(&ts);
 708                 h1->ts_first_pkt.ts_sec = ts.tv_sec;
 709                 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 710                 pkc1->pkblk_start = (char *)pbd1;
 711                 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 712                 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 713                 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 714                 pbd1->version = pkc1->version;
 715                 pkc1->prev = pkc1->nxt_offset;
 716                 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 717                 prb_thaw_queue(pkc1);
 718                 _prb_refresh_rx_retire_blk_timer(pkc1);
 719
 720                 smp_wmb();
 721
 722                 return;
 723         }
 724
 725         WARN(1, "ERROR block:%p is NOT FREE status:%d kactive_blk_num:%d\n",
 726                 pbd1, BLOCK_STATUS(pbd1), pkc1->kactive_blk_num);
 727         dump_stack();
 728         BUG();
 729 }
 730
 731 /*
 732  * Queue freeze logic:
 733  * 1) Assume tp_block_nr = 8 blocks.
 734  * 2) At time 't0', user opens Rx ring.
 735  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 736  * 4) user-space is either sleeping or processing block '0'.
 737  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 738  *    it will close block-7,loop around and try to fill block '0'.
 739  *    call-flow:
 740  *    __packet_lookup_frame_in_block
 741  *      prb_retire_current_block()
 742  *      prb_dispatch_next_block()
 743  *        |->(BLOCK_STATUS == USER) evaluates to true
 744  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 745  * 6) Now there are two cases:
 746  *    6.1) Link goes idle right after the queue is frozen.
 747  *         But remember, the last open_block() refreshed the timer.
 748  *         When this timer expires,it will refresh itself so that we can
 749  *         re-open block-0 in near future.
 750  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 751  *         case and __packet_lookup_frame_in_block will check if block-0
 752  *         is free and can now be re-used.
 753  */
 754 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 755                                   struct packet_sock *po)
 756 {
 757         pkc->reset_pending_on_curr_blk = 1;
 758         po->stats_u.stats3.tp_freeze_q_cnt++;
 759 }
 760
 761 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 762
 763 /*
 764  * If the next block is free then we will dispatch it
 765  * and return a good offset.
 766  * Else, we will freeze the queue.
 767  * So, caller must check the return value.
 768  */
 769 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 770                 struct packet_sock *po)
 771 {
 772         struct tpacket_block_desc *pbd;
 773
 774         smp_rmb();
 775
 776         /* 1. Get current block num */
 777         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 778
 779         /* 2. If this block is currently in_use then freeze the queue */
 780         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 781                 prb_freeze_queue(pkc, po);
 782                 return NULL;
 783         }
 784
 785         /*
 786          * 3.
 787          * open this block and return the offset where the first packet
 788          * needs to get stored.
 789          */
 790         prb_open_block(pkc, pbd);
 791         return (void *)pkc->nxt_offset;
 792 }
 793
 794 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 795                 struct packet_sock *po, unsigned int status)
 796 {
 797         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 798
 799         /* retire/close the current block */
 800         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 801                 /*
 802                  * Plug the case where copy_bits() is in progress on
 803                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 804                  * have space to copy the pkt in the current block and
 805                  * called prb_retire_current_block()
 806                  *
 807                  * We don't need to worry about the TMO case because
 808                  * the timer-handler already handled this case.
 809                  */
 810                 if (!(status & TP_STATUS_BLK_TMO)) {
 811                         while (atomic_read(&pkc->blk_fill_in_prog)) {
 812                                 /* Waiting for skb_copy_bits to finish... */
 813                                 cpu_relax();
 814                         }
 815                 }
 816                 prb_close_block(pkc, pbd, po, status);
 817                 return;
 818         }
 819
 820         WARN(1, "ERROR-pbd[%d]:%p\n", pkc->kactive_blk_num, pbd);
 821         dump_stack();
 822         BUG();
 823 }
 824
 825 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
 826                                       struct tpacket_block_desc *pbd)
 827 {
 828         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 829 }
 830
 831 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 832 {
 833         return pkc->reset_pending_on_curr_blk;
 834 }
 835
 836 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 837 {
 838         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 839         atomic_dec(&pkc->blk_fill_in_prog);
 840 }
 841
 842 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 843                         struct tpacket3_hdr *ppd)
 844 {
 845         ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
 846 }
 847
 848 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 849                         struct tpacket3_hdr *ppd)
 850 {
 851         ppd->hv1.tp_rxhash = 0;
 852 }
 853
 854 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
 855                         struct tpacket3_hdr *ppd)
 856 {
 857         if (vlan_tx_tag_present(pkc->skb)) {
 858                 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
 859                 ppd->tp_status = TP_STATUS_VLAN_VALID;
 860         } else {
 861                 ppd->hv1.tp_vlan_tci = 0;
 862                 ppd->tp_status = TP_STATUS_AVAILABLE;
 863         }
 864 }
 865
 866 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
 867                         struct tpacket3_hdr *ppd)
 868 {
 869         prb_fill_vlan_info(pkc, ppd);
 870
 871         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
 872                 prb_fill_rxhash(pkc, ppd);
 873         else
 874                 prb_clear_rxhash(pkc, ppd);
 875 }
 876
 877 static void prb_fill_curr_block(char *curr,
 878                                 struct tpacket_kbdq_core *pkc,
 879                                 struct tpacket_block_desc *pbd,
 880                                 unsigned int len)
 881 {
 882         struct tpacket3_hdr *ppd;
 883
 884         ppd  = (struct tpacket3_hdr *)curr;
 885         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
 886         pkc->prev = curr;
 887         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
 888         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
 889         BLOCK_NUM_PKTS(pbd) += 1;
 890         atomic_inc(&pkc->blk_fill_in_prog);
 891         prb_run_all_ft_ops(pkc, ppd);
 892 }
 893
 894 /* Assumes caller has the sk->rx_queue.lock */
 895 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
 896                                             struct sk_buff *skb,
 897                                                 int status,
 898                                             unsigned int len
 899                                             )
 900 {
 901         struct tpacket_kbdq_core *pkc;
 902         struct tpacket_block_desc *pbd;
 903         char *curr, *end;
 904
 905         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 906         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 907
 908         /* Queue is frozen when user space is lagging behind */
 909         if (prb_queue_frozen(pkc)) {
 910                 /*
 911                  * Check if that last block which caused the queue to freeze,
 912                  * is still in_use by user-space.
 913                  */
 914                 if (prb_curr_blk_in_use(pkc, pbd)) {
 915                         /* Can't record this packet */
 916                         return NULL;
 917                 } else {
 918                         /*
 919                          * Ok, the block was released by user-space.
 920                          * Now let's open that block.
 921                          * opening a block also thaws the queue.
 922                          * Thawing is a side effect.
 923                          */
 924                         prb_open_block(pkc, pbd);
 925                 }
 926         }
 927
 928         smp_mb();
 929         curr = pkc->nxt_offset;
 930         pkc->skb = skb;
 931         end = (char *)pbd + pkc->kblk_size;
 932
 933         /* first try the current block */
 934         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
 935                 prb_fill_curr_block(curr, pkc, pbd, len);
 936                 return (void *)curr;
 937         }
 938
 939         /* Ok, close the current block */
 940         prb_retire_current_block(pkc, po, 0);
 941
 942         /* Now, try to dispatch the next block */
 943         curr = (char *)prb_dispatch_next_block(pkc, po);
 944         if (curr) {
 945                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 946                 prb_fill_curr_block(curr, pkc, pbd, len);
 947                 return (void *)curr;
 948         }
 949
 950         /*
 951          * No free blocks are available.user_space hasn't caught up yet.
 952          * Queue was just frozen and now this packet will get dropped.
 953          */
 954         return NULL;
 955 }
 956
 957 static void *packet_current_rx_frame(struct packet_sock *po,
 958                                             struct sk_buff *skb,
 959                                             int status, unsigned int len)
 960 {
 961         char *curr = NULL;
 962         switch (po->tp_version) {
 963         case TPACKET_V1:
 964         case TPACKET_V2:
 965                 curr = packet_lookup_frame(po, &po->rx_ring,
 966                                         po->rx_ring.head, status);
 967                 return curr;
 968         case TPACKET_V3:
 969                 return __packet_lookup_frame_in_block(po, skb, status, len);
 970         default:
 971                 WARN(1, "TPACKET version not supported\n");
 972                 BUG();
 973                 return NULL;
 974         }
 975 }
 976
 977 static void *prb_lookup_block(struct packet_sock *po,
 978                                      struct packet_ring_buffer *rb,
 979                                      unsigned int idx,
 980                                      int status)
 981 {
 982         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 983         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
 984
 985         if (status != BLOCK_STATUS(pbd))
 986                 return NULL;
 987         return pbd;
 988 }
 989
 990 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
 991 {
 992         unsigned int prev;
 993         if (rb->prb_bdqc.kactive_blk_num)
 994                 prev = rb->prb_bdqc.kactive_blk_num-1;
 995         else
 996                 prev = rb->prb_bdqc.knum_blocks-1;
 997         return prev;
 998 }
 999
1000 /* Assumes caller has held the rx_queue.lock */
1001 static void *__prb_previous_block(struct packet_sock *po,
1002                                          struct packet_ring_buffer *rb,
1003                                          int status)
1004 {
1005         unsigned int previous = prb_previous_blk_num(rb);
1006         return prb_lookup_block(po, rb, previous, status);
1007 }
1008
1009 static void *packet_previous_rx_frame(struct packet_sock *po,
1010                                              struct packet_ring_buffer *rb,
1011                                              int status)
1012 {
1013         if (po->tp_version <= TPACKET_V2)
1014                 return packet_previous_frame(po, rb, status);
1015
1016         return __prb_previous_block(po, rb, status);
1017 }
1018
1019 static void packet_increment_rx_head(struct packet_sock *po,
1020                                             struct packet_ring_buffer *rb)
1021 {
1022         switch (po->tp_version) {
1023         case TPACKET_V1:
1024         case TPACKET_V2:
1025                 return packet_increment_head(rb);
1026         case TPACKET_V3:
1027         default:
1028                 WARN(1, "TPACKET version not supported.\n");
1029                 BUG();
1030                 return;
1031         }
1032 }
1033
1034 static void *packet_previous_frame(struct packet_sock *po,
1035                 struct packet_ring_buffer *rb,
1036                 int status)
1037 {
1038         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1039         return packet_lookup_frame(po, rb, previous, status);
1040 }
1041
1042 static void packet_increment_head(struct packet_ring_buffer *buff)
1043 {
1044         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1045 }
1046
1047 static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1048 {
1049         struct sock *sk = &po->sk;
1050         bool has_room;
1051
1052         if (po->prot_hook.func != tpacket_rcv)
1053                 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1054                         <= sk->sk_rcvbuf;
1055
1056         spin_lock(&sk->sk_receive_queue.lock);
1057         if (po->tp_version == TPACKET_V3)
1058                 has_room = prb_lookup_block(po, &po->rx_ring,
1059                                             po->rx_ring.prb_bdqc.kactive_blk_num,
1060                                             TP_STATUS_KERNEL);
1061         else
1062                 has_room = packet_lookup_frame(po, &po->rx_ring,
1063                                                po->rx_ring.head,
1064                                                TP_STATUS_KERNEL);
1065         spin_unlock(&sk->sk_receive_queue.lock);
1066
1067         return has_room;
1068 }
1069
1070 static void packet_sock_destruct(struct sock *sk)
1071 {
1072         skb_queue_purge(&sk->sk_error_queue);
1073
1074         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1075         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1076
1077         if (!sock_flag(sk, SOCK_DEAD)) {
1078                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1079                 return;
1080         }
1081
1082         sk_refcnt_debug_dec(sk);
1083 }
1084
1085 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1086 {
1087         int x = atomic_read(&f->rr_cur) + 1;
1088
1089         if (x >= num)
1090                 x = 0;
1091
1092         return x;
1093 }
1094
1095 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1096                                       struct sk_buff *skb,
1097                                       unsigned int num)
1098 {
1099         return (((u64)skb->rxhash) * num) >> 32;
1100 }
1101
1102 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1103                                     struct sk_buff *skb,
1104                                     unsigned int num)
1105 {
1106         int cur, old;
1107
1108         cur = atomic_read(&f->rr_cur);
1109         while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1110                                      fanout_rr_next(f, num))) != cur)
1111                 cur = old;
1112         return cur;
1113 }
1114
1115 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1116                                      struct sk_buff *skb,
1117                                      unsigned int num)
1118 {
1119         return smp_processor_id() % num;
1120 }
1121
1122 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1123                                           struct sk_buff *skb,
1124                                           unsigned int idx, unsigned int skip,
1125                                           unsigned int num)
1126 {
1127         unsigned int i, j;
1128
1129         i = j = min_t(int, f->next[idx], num - 1);
1130         do {
1131                 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1132                         if (i != j)
1133                                 f->next[idx] = i;
1134                         return i;
1135                 }
1136                 if (++i == num)
1137                         i = 0;
1138         } while (i != j);
1139
1140         return idx;
1141 }
1142
1143 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1144 {
1145         return f->flags & (flag >> 8);
1146 }
1147
1148 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1149                              struct packet_type *pt, struct net_device *orig_dev)
1150 {
1151         struct packet_fanout *f = pt->af_packet_priv;
1152         unsigned int num = f->num_members;
1153         struct packet_sock *po;
1154         unsigned int idx;
1155
1156         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1157             !num) {
1158                 kfree_skb(skb);
1159                 return 0;
1160         }
1161
1162         switch (f->type) {
1163         case PACKET_FANOUT_HASH:
1164         default:
1165                 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1166                         skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1167                         if (!skb)
1168                                 return 0;
1169                 }
1170                 skb_get_rxhash(skb);
1171                 idx = fanout_demux_hash(f, skb, num);
1172                 break;
1173         case PACKET_FANOUT_LB:
1174                 idx = fanout_demux_lb(f, skb, num);
1175                 break;
1176         case PACKET_FANOUT_CPU:
1177                 idx = fanout_demux_cpu(f, skb, num);
1178                 break;
1179         case PACKET_FANOUT_ROLLOVER:
1180                 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1181                 break;
1182         }
1183
1184         po = pkt_sk(f->arr[idx]);
1185         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1186             unlikely(!packet_rcv_has_room(po, skb))) {
1187                 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1188                 po = pkt_sk(f->arr[idx]);
1189         }
1190
1191         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1192 }
1193
1194 DEFINE_MUTEX(fanout_mutex);
1195 EXPORT_SYMBOL_GPL(fanout_mutex);
1196 static LIST_HEAD(fanout_list);
1197
1198 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1199 {
1200         struct packet_fanout *f = po->fanout;
1201
1202         spin_lock(&f->lock);
1203         f->arr[f->num_members] = sk;
1204         smp_wmb();
1205         f->num_members++;
1206         spin_unlock(&f->lock);
1207 }
1208
1209 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1210 {
1211         struct packet_fanout *f = po->fanout;
1212         int i;
1213
1214         spin_lock(&f->lock);
1215         for (i = 0; i < f->num_members; i++) {
1216                 if (f->arr[i] == sk)
1217                         break;
1218         }
1219         BUG_ON(i >= f->num_members);
1220         f->arr[i] = f->arr[f->num_members - 1];
1221         f->num_members--;
1222         spin_unlock(&f->lock);
1223 }
1224
1225 static bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1226 {
1227         if (ptype->af_packet_priv == (void*)((struct packet_sock *)sk)->fanout)
1228                 return true;
1229
1230         return false;
1231 }
1232
1233 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1234 {
1235         struct packet_sock *po = pkt_sk(sk);
1236         struct packet_fanout *f, *match;
1237         u8 type = type_flags & 0xff;
1238         u8 flags = type_flags >> 8;
1239         int err;
1240
1241         switch (type) {
1242         case PACKET_FANOUT_ROLLOVER:
1243                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1244                         return -EINVAL;
1245         case PACKET_FANOUT_HASH:
1246         case PACKET_FANOUT_LB:
1247         case PACKET_FANOUT_CPU:
1248                 break;
1249         default:
1250                 return -EINVAL;
1251         }
1252
1253         if (!po->running)
1254                 return -EINVAL;
1255
1256         if (po->fanout)
1257                 return -EALREADY;
1258
1259         mutex_lock(&fanout_mutex);
1260         match = NULL;
1261         list_for_each_entry(f, &fanout_list, list) {
1262                 if (f->id == id &&
1263                     read_pnet(&f->net) == sock_net(sk)) {
1264                         match = f;
1265                         break;
1266                 }
1267         }
1268         err = -EINVAL;
1269         if (match && match->flags != flags)
1270                 goto out;
1271         if (!match) {
1272                 err = -ENOMEM;
1273                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1274                 if (!match)
1275                         goto out;
1276                 write_pnet(&match->net, sock_net(sk));
1277                 match->id = id;
1278                 match->type = type;
1279                 match->flags = flags;
1280                 atomic_set(&match->rr_cur, 0);
1281                 INIT_LIST_HEAD(&match->list);
1282                 spin_lock_init(&match->lock);
1283                 atomic_set(&match->sk_ref, 0);
1284                 match->prot_hook.type = po->prot_hook.type;
1285                 match->prot_hook.dev = po->prot_hook.dev;
1286                 match->prot_hook.func = packet_rcv_fanout;
1287                 match->prot_hook.af_packet_priv = match;
1288                 match->prot_hook.id_match = match_fanout_group;
1289                 dev_add_pack(&match->prot_hook);
1290                 list_add(&match->list, &fanout_list);
1291         }
1292         err = -EINVAL;
1293         if (match->type == type &&
1294             match->prot_hook.type == po->prot_hook.type &&
1295             match->prot_hook.dev == po->prot_hook.dev) {
1296                 err = -ENOSPC;
1297                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1298                         __dev_remove_pack(&po->prot_hook);
1299                         po->fanout = match;
1300                         atomic_inc(&match->sk_ref);
1301                         __fanout_link(sk, po);
1302                         err = 0;
1303                 }
1304         }
1305 out:
1306         mutex_unlock(&fanout_mutex);
1307         return err;
1308 }
1309
1310 static void fanout_release(struct sock *sk)
1311 {
1312         struct packet_sock *po = pkt_sk(sk);
1313         struct packet_fanout *f;
1314
1315         f = po->fanout;
1316         if (!f)
1317                 return;
1318
1319         mutex_lock(&fanout_mutex);
1320         po->fanout = NULL;
1321
1322         if (atomic_dec_and_test(&f->sk_ref)) {
1323                 list_del(&f->list);
1324                 dev_remove_pack(&f->prot_hook);
1325                 kfree(f);
1326         }
1327         mutex_unlock(&fanout_mutex);
1328 }
1329
1330 static const struct proto_ops packet_ops;
1331
1332 static const struct proto_ops packet_ops_spkt;
1333
1334 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1335                            struct packet_type *pt, struct net_device *orig_dev)
1336 {
1337         struct sock *sk;
1338         struct sockaddr_pkt *spkt;
1339
1340         /*
1341          *      When we registered the protocol we saved the socket in the data
1342          *      field for just this event.
1343          */
1344
1345         sk = pt->af_packet_priv;
1346
1347         /*
1348          *      Yank back the headers [hope the device set this
1349          *      right or kerboom...]
1350          *
1351          *      Incoming packets have ll header pulled,
1352          *      push it back.
1353          *
1354          *      For outgoing ones skb->data == skb_mac_header(skb)
1355          *      so that this procedure is noop.
1356          */
1357
1358         if (skb->pkt_type == PACKET_LOOPBACK)
1359                 goto out;
1360
1361         if (!net_eq(dev_net(dev), sock_net(sk)))
1362                 goto out;
1363
1364         skb = skb_share_check(skb, GFP_ATOMIC);
1365         if (skb == NULL)
1366                 goto oom;
1367
1368         /* drop any routing info */
1369         skb_dst_drop(skb);
1370
1371         /* drop conntrack reference */
1372         nf_reset(skb);
1373
1374         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1375
1376         skb_push(skb, skb->data - skb_mac_header(skb));
1377
1378         /*
1379          *      The SOCK_PACKET socket receives _all_ frames.
1380          */
1381
1382         spkt->spkt_family = dev->type;
1383         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1384         spkt->spkt_protocol = skb->protocol;
1385
1386         /*
1387          *      Charge the memory to the socket. This is done specifically
1388          *      to prevent sockets using all the memory up.
1389          */
1390
1391         if (sock_queue_rcv_skb(sk, skb) == 0)
1392                 return 0;
1393
1394 out:
1395         kfree_skb(skb);
1396 oom:
1397         return 0;
1398 }
1399
1400
1401 /*
1402  *      Output a raw packet to a device layer. This bypasses all the other
1403  *      protocol layers and you must therefore supply it with a complete frame
1404  */
1405
1406 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1407                                struct msghdr *msg, size_t len)
1408 {
1409         struct sock *sk = sock->sk;
1410         struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1411         struct sk_buff *skb = NULL;
1412         struct net_device *dev;
1413         __be16 proto = 0;
1414         int err;
1415         int extra_len = 0;
1416         struct flow_keys keys;
1417
1418         /*
1419          *      Get and verify the address.
1420          */
1421
1422         if (saddr) {
1423                 if (msg->msg_namelen < sizeof(struct sockaddr))
1424                         return -EINVAL;
1425                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1426                         proto = saddr->spkt_protocol;
1427         } else
1428                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1429
1430         /*
1431          *      Find the device first to size check it
1432          */
1433
1434         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1435 retry:
1436         rcu_read_lock();
1437         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1438         err = -ENODEV;
1439         if (dev == NULL)
1440                 goto out_unlock;
1441
1442         err = -ENETDOWN;
1443         if (!(dev->flags & IFF_UP))
1444                 goto out_unlock;
1445
1446         /*
1447          * You may not queue a frame bigger than the mtu. This is the lowest level
1448          * raw protocol and you must do your own fragmentation at this level.
1449          */
1450
1451         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1452                 if (!netif_supports_nofcs(dev)) {
1453                         err = -EPROTONOSUPPORT;
1454                         goto out_unlock;
1455                 }
1456                 extra_len = 4; /* We're doing our own CRC */
1457         }
1458
1459         err = -EMSGSIZE;
1460         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1461                 goto out_unlock;
1462
1463         if (!skb) {
1464                 size_t reserved = LL_RESERVED_SPACE(dev);
1465                 int tlen = dev->needed_tailroom;
1466                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1467
1468                 rcu_read_unlock();
1469                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1470                 if (skb == NULL)
1471                         return -ENOBUFS;
1472                 /* FIXME: Save some space for broken drivers that write a hard
1473                  * header at transmission time by themselves. PPP is the notable
1474                  * one here. This should really be fixed at the driver level.
1475                  */
1476                 skb_reserve(skb, reserved);
1477                 skb_reset_network_header(skb);
1478
1479                 /* Try to align data part correctly */
1480                 if (hhlen) {
1481                         skb->data -= hhlen;
1482                         skb->tail -= hhlen;
1483                         if (len < hhlen)
1484                                 skb_reset_network_header(skb);
1485                 }
1486                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1487                 if (err)
1488                         goto out_free;
1489                 goto retry;
1490         }
1491
1492         if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1493                 /* Earlier code assumed this would be a VLAN pkt,
1494                  * double-check this now that we have the actual
1495                  * packet in hand.
1496                  */
1497                 struct ethhdr *ehdr;
1498                 skb_reset_mac_header(skb);
1499                 ehdr = eth_hdr(skb);
1500                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1501                         err = -EMSGSIZE;
1502                         goto out_unlock;
1503                 }
1504         }
1505
1506         skb->protocol = proto;
1507         skb->dev = dev;
1508         skb->priority = sk->sk_priority;
1509         skb->mark = sk->sk_mark;
1510         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1511         if (err < 0)
1512                 goto out_unlock;
1513
1514         if (unlikely(extra_len == 4))
1515                 skb->no_fcs = 1;
1516
1517         if (skb_flow_dissect(skb, &keys))
1518                 skb_set_transport_header(skb, keys.thoff);
1519         else
1520                 skb_reset_transport_header(skb);
1521
1522         dev_queue_xmit(skb);
1523         rcu_read_unlock();
1524         return len;
1525
1526 out_unlock:
1527         rcu_read_unlock();
1528 out_free:
1529         kfree_skb(skb);
1530         return err;
1531 }
1532
1533 static unsigned int run_filter(const struct sk_buff *skb,
1534                                       const struct sock *sk,
1535                                       unsigned int res)
1536 {
1537         struct sk_filter *filter;
1538
1539         rcu_read_lock();
1540         filter = rcu_dereference(sk->sk_filter);
1541         if (filter != NULL)
1542                 res = SK_RUN_FILTER(filter, skb);
1543         rcu_read_unlock();
1544
1545         return res;
1546 }
1547
1548 /*
1549  * This function makes lazy skb cloning in hope that most of packets
1550  * are discarded by BPF.
1551  *
1552  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1553  * and skb->cb are mangled. It works because (and until) packets
1554  * falling here are owned by current CPU. Output packets are cloned
1555  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1556  * sequencially, so that if we return skb to original state on exit,
1557  * we will not harm anyone.
1558  */
1559
1560 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1561                       struct packet_type *pt, struct net_device *orig_dev)
1562 {
1563         struct sock *sk;
1564         struct sockaddr_ll *sll;
1565         struct packet_sock *po;
1566         u8 *skb_head = skb->data;
1567         int skb_len = skb->len;
1568         unsigned int snaplen, res;
1569
1570         if (skb->pkt_type == PACKET_LOOPBACK)
1571                 goto drop;
1572
1573         sk = pt->af_packet_priv;
1574         po = pkt_sk(sk);
1575
1576         if (!net_eq(dev_net(dev), sock_net(sk)))
1577                 goto drop;
1578
1579         skb->dev = dev;
1580
1581         if (dev->header_ops) {
1582                 /* The device has an explicit notion of ll header,
1583                  * exported to higher levels.
1584                  *
1585                  * Otherwise, the device hides details of its frame
1586                  * structure, so that corresponding packet head is
1587                  * never delivered to user.
1588                  */
1589                 if (sk->sk_type != SOCK_DGRAM)
1590                         skb_push(skb, skb->data - skb_mac_header(skb));
1591                 else if (skb->pkt_type == PACKET_OUTGOING) {
1592                         /* Special case: outgoing packets have ll header at head */
1593                         skb_pull(skb, skb_network_offset(skb));
1594                 }
1595         }
1596
1597         snaplen = skb->len;
1598
1599         res = run_filter(skb, sk, snaplen);
1600         if (!res)
1601                 goto drop_n_restore;
1602         if (snaplen > res)
1603                 snaplen = res;
1604
1605         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1606                 goto drop_n_acct;
1607
1608         if (skb_shared(skb)) {
1609                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1610                 if (nskb == NULL)
1611                         goto drop_n_acct;
1612
1613                 if (skb_head != skb->data) {
1614                         skb->data = skb_head;
1615                         skb->len = skb_len;
1616                 }
1617                 consume_skb(skb);
1618                 skb = nskb;
1619         }
1620
1621         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1622                      sizeof(skb->cb));
1623
1624         sll = &PACKET_SKB_CB(skb)->sa.ll;
1625         sll->sll_family = AF_PACKET;
1626         sll->sll_hatype = dev->type;
1627         sll->sll_protocol = skb->protocol;
1628         sll->sll_pkttype = skb->pkt_type;
1629         if (unlikely(po->origdev))
1630                 sll->sll_ifindex = orig_dev->ifindex;
1631         else
1632                 sll->sll_ifindex = dev->ifindex;
1633
1634         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1635
1636         PACKET_SKB_CB(skb)->origlen = skb->len;
1637
1638         if (pskb_trim(skb, snaplen))
1639                 goto drop_n_acct;
1640
1641         skb_set_owner_r(skb, sk);
1642         skb->dev = NULL;
1643         skb_dst_drop(skb);
1644
1645         /* drop conntrack reference */
1646         nf_reset(skb);
1647
1648         spin_lock(&sk->sk_receive_queue.lock);
1649         po->stats.tp_packets++;
1650         skb->dropcount = atomic_read(&sk->sk_drops);
1651         __skb_queue_tail(&sk->sk_receive_queue, skb);
1652         spin_unlock(&sk->sk_receive_queue.lock);
1653         sk->sk_data_ready(sk, skb->len);
1654         return 0;
1655
1656 drop_n_acct:
1657         spin_lock(&sk->sk_receive_queue.lock);
1658         po->stats.tp_drops++;
1659         atomic_inc(&sk->sk_drops);
1660         spin_unlock(&sk->sk_receive_queue.lock);
1661
1662 drop_n_restore:
1663         if (skb_head != skb->data && skb_shared(skb)) {
1664                 skb->data = skb_head;
1665                 skb->len = skb_len;
1666         }
1667 drop:
1668         consume_skb(skb);
1669         return 0;
1670 }
1671
1672 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1673                        struct packet_type *pt, struct net_device *orig_dev)
1674 {
1675         struct sock *sk;
1676         struct packet_sock *po;
1677         struct sockaddr_ll *sll;
1678         union {
1679                 struct tpacket_hdr *h1;
1680                 struct tpacket2_hdr *h2;
1681                 struct tpacket3_hdr *h3;
1682                 void *raw;
1683         } h;
1684         u8 *skb_head = skb->data;
1685         int skb_len = skb->len;
1686         unsigned int snaplen, res;
1687         unsigned long status = TP_STATUS_USER;
1688         unsigned short macoff, netoff, hdrlen;
1689         struct sk_buff *copy_skb = NULL;
1690         struct timeval tv;
1691         struct timespec ts;
1692         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1693
1694         if (skb->pkt_type == PACKET_LOOPBACK)
1695                 goto drop;
1696
1697         sk = pt->af_packet_priv;
1698         po = pkt_sk(sk);
1699
1700         if (!net_eq(dev_net(dev), sock_net(sk)))
1701                 goto drop;
1702
1703         if (dev->header_ops) {
1704                 if (sk->sk_type != SOCK_DGRAM)
1705                         skb_push(skb, skb->data - skb_mac_header(skb));
1706                 else if (skb->pkt_type == PACKET_OUTGOING) {
1707                         /* Special case: outgoing packets have ll header at head */
1708                         skb_pull(skb, skb_network_offset(skb));
1709                 }
1710         }
1711
1712         if (skb->ip_summed == CHECKSUM_PARTIAL)
1713                 status |= TP_STATUS_CSUMNOTREADY;
1714
1715         snaplen = skb->len;
1716
1717         res = run_filter(skb, sk, snaplen);
1718         if (!res)
1719                 goto drop_n_restore;
1720         if (snaplen > res)
1721                 snaplen = res;
1722
1723         if (sk->sk_type == SOCK_DGRAM) {
1724                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1725                                   po->tp_reserve;
1726         } else {
1727                 unsigned int maclen = skb_network_offset(skb);
1728                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1729                                        (maclen < 16 ? 16 : maclen)) +
1730                         po->tp_reserve;
1731                 macoff = netoff - maclen;
1732         }
1733         if (po->tp_version <= TPACKET_V2) {
1734                 if (macoff + snaplen > po->rx_ring.frame_size) {
1735                         if (po->copy_thresh &&
1736                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1737                                 if (skb_shared(skb)) {
1738                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1739                                 } else {
1740                                         copy_skb = skb_get(skb);
1741                                         skb_head = skb->data;
1742                                 }
1743                                 if (copy_skb)
1744                                         skb_set_owner_r(copy_skb, sk);
1745                         }
1746                         snaplen = po->rx_ring.frame_size - macoff;
1747                         if ((int)snaplen < 0)
1748                                 snaplen = 0;
1749                 }
1750         }
1751         spin_lock(&sk->sk_receive_queue.lock);
1752         h.raw = packet_current_rx_frame(po, skb,
1753                                         TP_STATUS_KERNEL, (macoff+snaplen));
1754         if (!h.raw)
1755                 goto ring_is_full;
1756         if (po->tp_version <= TPACKET_V2) {
1757                 packet_increment_rx_head(po, &po->rx_ring);
1758         /*
1759          * LOSING will be reported till you read the stats,
1760          * because it's COR - Clear On Read.
1761          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1762          * at packet level.
1763          */
1764                 if (po->stats.tp_drops)
1765                         status |= TP_STATUS_LOSING;
1766         }
1767         po->stats.tp_packets++;
1768         if (copy_skb) {
1769                 status |= TP_STATUS_COPY;
1770                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1771         }
1772         spin_unlock(&sk->sk_receive_queue.lock);
1773
1774         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1775
1776         switch (po->tp_version) {
1777         case TPACKET_V1:
1778                 h.h1->tp_len = skb->len;
1779                 h.h1->tp_snaplen = snaplen;
1780                 h.h1->tp_mac = macoff;
1781                 h.h1->tp_net = netoff;
1782                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1783                                 && shhwtstamps->syststamp.tv64)
1784                         tv = ktime_to_timeval(shhwtstamps->syststamp);
1785                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1786                                 && shhwtstamps->hwtstamp.tv64)
1787                         tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1788                 else if (skb->tstamp.tv64)
1789                         tv = ktime_to_timeval(skb->tstamp);
1790                 else
1791                         do_gettimeofday(&tv);
1792                 h.h1->tp_sec = tv.tv_sec;
1793                 h.h1->tp_usec = tv.tv_usec;
1794                 hdrlen = sizeof(*h.h1);
1795                 break;
1796         case TPACKET_V2:
1797                 h.h2->tp_len = skb->len;
1798                 h.h2->tp_snaplen = snaplen;
1799                 h.h2->tp_mac = macoff;
1800                 h.h2->tp_net = netoff;
1801                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1802                                 && shhwtstamps->syststamp.tv64)
1803                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1804                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1805                                 && shhwtstamps->hwtstamp.tv64)
1806                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1807                 else if (skb->tstamp.tv64)
1808                         ts = ktime_to_timespec(skb->tstamp);
1809                 else
1810                         getnstimeofday(&ts);
1811                 h.h2->tp_sec = ts.tv_sec;
1812                 h.h2->tp_nsec = ts.tv_nsec;
1813                 if (vlan_tx_tag_present(skb)) {
1814                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1815                         status |= TP_STATUS_VLAN_VALID;
1816                 } else {
1817                         h.h2->tp_vlan_tci = 0;
1818                 }
1819                 h.h2->tp_padding = 0;
1820                 hdrlen = sizeof(*h.h2);
1821                 break;
1822         case TPACKET_V3:
1823                 /* tp_nxt_offset,vlan are already populated above.
1824                  * So DONT clear those fields here
1825                  */
1826                 h.h3->tp_status |= status;
1827                 h.h3->tp_len = skb->len;
1828                 h.h3->tp_snaplen = snaplen;
1829                 h.h3->tp_mac = macoff;
1830                 h.h3->tp_net = netoff;
1831                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1832                                 && shhwtstamps->syststamp.tv64)
1833                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1834                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1835                                 && shhwtstamps->hwtstamp.tv64)
1836                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1837                 else if (skb->tstamp.tv64)
1838                         ts = ktime_to_timespec(skb->tstamp);
1839                 else
1840                         getnstimeofday(&ts);
1841                 h.h3->tp_sec  = ts.tv_sec;
1842                 h.h3->tp_nsec = ts.tv_nsec;
1843                 hdrlen = sizeof(*h.h3);
1844                 break;
1845         default:
1846                 BUG();
1847         }
1848
1849         sll = h.raw + TPACKET_ALIGN(hdrlen);
1850         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1851         sll->sll_family = AF_PACKET;
1852         sll->sll_hatype = dev->type;
1853         sll->sll_protocol = skb->protocol;
1854         sll->sll_pkttype = skb->pkt_type;
1855         if (unlikely(po->origdev))
1856                 sll->sll_ifindex = orig_dev->ifindex;
1857         else
1858                 sll->sll_ifindex = dev->ifindex;
1859
1860         smp_mb();
1861 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1862         {
1863                 u8 *start, *end;
1864
1865                 if (po->tp_version <= TPACKET_V2) {
1866                         end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1867                                 + macoff + snaplen);
1868                         for (start = h.raw; start < end; start += PAGE_SIZE)
1869                                 flush_dcache_page(pgv_to_page(start));
1870                 }
1871                 smp_wmb();
1872         }
1873 #endif
1874         if (po->tp_version <= TPACKET_V2)
1875                 __packet_set_status(po, h.raw, status);
1876         else
1877                 prb_clear_blk_fill_status(&po->rx_ring);
1878
1879         sk->sk_data_ready(sk, 0);
1880
1881 drop_n_restore:
1882         if (skb_head != skb->data && skb_shared(skb)) {
1883                 skb->data = skb_head;
1884                 skb->len = skb_len;
1885         }
1886 drop:
1887         kfree_skb(skb);
1888         return 0;
1889
1890 ring_is_full:
1891         po->stats.tp_drops++;
1892         spin_unlock(&sk->sk_receive_queue.lock);
1893
1894         sk->sk_data_ready(sk, 0);
1895         kfree_skb(copy_skb);
1896         goto drop_n_restore;
1897 }
1898
1899 static void tpacket_destruct_skb(struct sk_buff *skb)
1900 {
1901         struct packet_sock *po = pkt_sk(skb->sk);
1902         void *ph;
1903
1904         if (likely(po->tx_ring.pg_vec)) {
1905                 ph = skb_shinfo(skb)->destructor_arg;
1906                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1907                 atomic_dec(&po->tx_ring.pending);
1908                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1909         }
1910
1911         sock_wfree(skb);
1912 }
1913
1914 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1915                 void *frame, struct net_device *dev, int size_max,
1916                 __be16 proto, unsigned char *addr, int hlen)
1917 {
1918         union {
1919                 struct tpacket_hdr *h1;
1920                 struct tpacket2_hdr *h2;
1921                 void *raw;
1922         } ph;
1923         int to_write, offset, len, tp_len, nr_frags, len_max;
1924         struct socket *sock = po->sk.sk_socket;
1925         struct page *page;
1926         void *data;
1927         int err;
1928         struct flow_keys keys;
1929
1930         ph.raw = frame;
1931
1932         skb->protocol = proto;
1933         skb->dev = dev;
1934         skb->priority = po->sk.sk_priority;
1935         skb->mark = po->sk.sk_mark;
1936         skb_shinfo(skb)->destructor_arg = ph.raw;
1937
1938         switch (po->tp_version) {
1939         case TPACKET_V2:
1940                 tp_len = ph.h2->tp_len;
1941                 break;
1942         default:
1943                 tp_len = ph.h1->tp_len;
1944                 break;
1945         }
1946         if (unlikely(tp_len > size_max)) {
1947                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1948                 return -EMSGSIZE;
1949         }
1950
1951         skb_reserve(skb, hlen);
1952         skb_reset_network_header(skb);
1953
1954         if (skb_flow_dissect(skb, &keys))
1955                 skb_set_transport_header(skb, keys.thoff);
1956         else
1957                 skb_reset_transport_header(skb);
1958
1959         if (po->tp_tx_has_off) {
1960                 int off_min, off_max, off;
1961                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
1962                 off_max = po->tx_ring.frame_size - tp_len;
1963                 if (sock->type == SOCK_DGRAM) {
1964                         switch (po->tp_version) {
1965                         case TPACKET_V2:
1966                                 off = ph.h2->tp_net;
1967                                 break;
1968                         default:
1969                                 off = ph.h1->tp_net;
1970                                 break;
1971                         }
1972                 } else {
1973                         switch (po->tp_version) {
1974                         case TPACKET_V2:
1975                                 off = ph.h2->tp_mac;
1976                                 break;
1977                         default:
1978                                 off = ph.h1->tp_mac;
1979                                 break;
1980                         }
1981                 }
1982                 if (unlikely((off < off_min) || (off_max < off)))
1983                         return -EINVAL;
1984                 data = ph.raw + off;
1985         } else {
1986                 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1987         }
1988         to_write = tp_len;
1989
1990         if (sock->type == SOCK_DGRAM) {
1991                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
1992                                 NULL, tp_len);
1993                 if (unlikely(err < 0))
1994                         return -EINVAL;
1995         } else if (dev->hard_header_len) {
1996                 /* net device doesn't like empty head */
1997                 if (unlikely(tp_len <= dev->hard_header_len)) {
1998                         pr_err("packet size is too short (%d < %d)\n",
1999                                tp_len, dev->hard_header_len);
2000                         return -EINVAL;
2001                 }
2002
2003                 skb_push(skb, dev->hard_header_len);
2004                 err = skb_store_bits(skb, 0, data,
2005                                 dev->hard_header_len);
2006                 if (unlikely(err))
2007                         return err;
2008
2009                 data += dev->hard_header_len;
2010                 to_write -= dev->hard_header_len;
2011         }
2012
2013         offset = offset_in_page(data);
2014         len_max = PAGE_SIZE - offset;
2015         len = ((to_write > len_max) ? len_max : to_write);
2016
2017         skb->data_len = to_write;
2018         skb->len += to_write;
2019         skb->truesize += to_write;
2020         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2021
2022         while (likely(to_write)) {
2023                 nr_frags = skb_shinfo(skb)->nr_frags;
2024
2025                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2026                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2027                                MAX_SKB_FRAGS);
2028                         return -EFAULT;
2029                 }
2030
2031                 page = pgv_to_page(data);
2032                 data += len;
2033                 flush_dcache_page(page);
2034                 get_page(page);
2035                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2036                 to_write -= len;
2037                 offset = 0;
2038                 len_max = PAGE_SIZE;
2039                 len = ((to_write > len_max) ? len_max : to_write);
2040         }
2041
2042         return tp_len;
2043 }
2044
2045 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2046 {
2047         struct sk_buff *skb;
2048         struct net_device *dev;
2049         __be16 proto;
2050         bool need_rls_dev = false;
2051         int err, reserve = 0;
2052         void *ph;
2053         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2054         int tp_len, size_max;
2055         unsigned char *addr;
2056         int len_sum = 0;
2057         int status = TP_STATUS_AVAILABLE;
2058         int hlen, tlen;
2059
2060         mutex_lock(&po->pg_vec_lock);
2061
2062         if (saddr == NULL) {
2063                 dev = po->prot_hook.dev;
2064                 proto   = po->num;
2065                 addr    = NULL;
2066         } else {
2067                 err = -EINVAL;
2068                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2069                         goto out;
2070                 if (msg->msg_namelen < (saddr->sll_halen
2071                                         + offsetof(struct sockaddr_ll,
2072                                                 sll_addr)))
2073                         goto out;
2074                 proto   = saddr->sll_protocol;
2075                 addr    = saddr->sll_addr;
2076                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2077                 need_rls_dev = true;
2078         }
2079
2080         err = -ENXIO;
2081         if (unlikely(dev == NULL))
2082                 goto out;
2083
2084         reserve = dev->hard_header_len;
2085
2086         err = -ENETDOWN;
2087         if (unlikely(!(dev->flags & IFF_UP)))
2088                 goto out_put;
2089
2090         size_max = po->tx_ring.frame_size
2091                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2092
2093         if (size_max > dev->mtu + reserve)
2094                 size_max = dev->mtu + reserve;
2095
2096         do {
2097                 ph = packet_current_frame(po, &po->tx_ring,
2098                                 TP_STATUS_SEND_REQUEST);
2099
2100                 if (unlikely(ph == NULL)) {
2101                         schedule();
2102                         continue;
2103                 }
2104
2105                 status = TP_STATUS_SEND_REQUEST;
2106                 hlen = LL_RESERVED_SPACE(dev);
2107                 tlen = dev->needed_tailroom;
2108                 skb = sock_alloc_send_skb(&po->sk,
2109                                 hlen + tlen + sizeof(struct sockaddr_ll),
2110                                 0, &err);
2111
2112                 if (unlikely(skb == NULL))
2113                         goto out_status;
2114
2115                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2116                                 addr, hlen);
2117
2118                 if (unlikely(tp_len < 0)) {
2119                         if (po->tp_loss) {
2120                                 __packet_set_status(po, ph,
2121                                                 TP_STATUS_AVAILABLE);
2122                                 packet_increment_head(&po->tx_ring);
2123                                 kfree_skb(skb);
2124                                 continue;
2125                         } else {
2126                                 status = TP_STATUS_WRONG_FORMAT;
2127                                 err = tp_len;
2128                                 goto out_status;
2129                         }
2130                 }
2131
2132                 skb->destructor = tpacket_destruct_skb;
2133                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2134                 atomic_inc(&po->tx_ring.pending);
2135
2136                 status = TP_STATUS_SEND_REQUEST;
2137                 err = dev_queue_xmit(skb);
2138                 if (unlikely(err > 0)) {
2139                         err = net_xmit_errno(err);
2140                         if (err && __packet_get_status(po, ph) ==
2141                                    TP_STATUS_AVAILABLE) {
2142                                 /* skb was destructed already */
2143                                 skb = NULL;
2144                                 goto out_status;
2145                         }
2146                         /*
2147                          * skb was dropped but not destructed yet;
2148                          * let's treat it like congestion or err < 0
2149                          */
2150                         err = 0;
2151                 }
2152                 packet_increment_head(&po->tx_ring);
2153                 len_sum += tp_len;
2154         } while (likely((ph != NULL) ||
2155                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2156                          (atomic_read(&po->tx_ring.pending))))
2157                 );
2158
2159         err = len_sum;
2160         goto out_put;
2161
2162 out_status:
2163         __packet_set_status(po, ph, status);
2164         kfree_skb(skb);
2165 out_put:
2166         if (need_rls_dev)
2167                 dev_put(dev);
2168 out:
2169         mutex_unlock(&po->pg_vec_lock);
2170         return err;
2171 }
2172
2173 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2174                                         size_t reserve, size_t len,
2175                                         size_t linear, int noblock,
2176                                         int *err)
2177 {
2178         struct sk_buff *skb;
2179
2180         /* Under a page?  Don't bother with paged skb. */
2181         if (prepad + len < PAGE_SIZE || !linear)
2182                 linear = len;
2183
2184         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2185                                    err);
2186         if (!skb)
2187                 return NULL;
2188
2189         skb_reserve(skb, reserve);
2190         skb_put(skb, linear);
2191         skb->data_len = len - linear;
2192         skb->len += len - linear;
2193
2194         return skb;
2195 }
2196
2197 static int packet_snd(struct socket *sock,
2198                           struct msghdr *msg, size_t len)
2199 {
2200         struct sock *sk = sock->sk;
2201         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2202         struct sk_buff *skb;
2203         struct net_device *dev;
2204         __be16 proto;
2205         bool need_rls_dev = false;
2206         unsigned char *addr;
2207         int err, reserve = 0;
2208         struct virtio_net_hdr vnet_hdr = { 0 };
2209         int offset = 0;
2210         int vnet_hdr_len;
2211         struct packet_sock *po = pkt_sk(sk);
2212         unsigned short gso_type = 0;
2213         int hlen, tlen;
2214         int extra_len = 0;
2215         struct flow_keys keys;
2216
2217         /*
2218          *      Get and verify the address.
2219          */
2220
2221         if (saddr == NULL) {
2222                 dev = po->prot_hook.dev;
2223                 proto   = po->num;
2224                 addr    = NULL;
2225         } else {
2226                 err = -EINVAL;
2227                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2228                         goto out;
2229                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2230                         goto out;
2231                 proto   = saddr->sll_protocol;
2232                 addr    = saddr->sll_addr;
2233                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2234                 need_rls_dev = true;
2235         }
2236
2237         err = -ENXIO;
2238         if (dev == NULL)
2239                 goto out_unlock;
2240         if (sock->type == SOCK_RAW)
2241                 reserve = dev->hard_header_len;
2242
2243         err = -ENETDOWN;
2244         if (!(dev->flags & IFF_UP))
2245                 goto out_unlock;
2246
2247         if (po->has_vnet_hdr) {
2248                 vnet_hdr_len = sizeof(vnet_hdr);
2249
2250                 err = -EINVAL;
2251                 if (len < vnet_hdr_len)
2252                         goto out_unlock;
2253
2254                 len -= vnet_hdr_len;
2255
2256                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2257                                        vnet_hdr_len);
2258                 if (err < 0)
2259                         goto out_unlock;
2260
2261                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2262                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2263                       vnet_hdr.hdr_len))
2264                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
2265                                                  vnet_hdr.csum_offset + 2;
2266
2267                 err = -EINVAL;
2268                 if (vnet_hdr.hdr_len > len)
2269                         goto out_unlock;
2270
2271                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2272                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2273                         case VIRTIO_NET_HDR_GSO_TCPV4:
2274                                 gso_type = SKB_GSO_TCPV4;
2275                                 break;
2276                         case VIRTIO_NET_HDR_GSO_TCPV6:
2277                                 gso_type = SKB_GSO_TCPV6;
2278                                 break;
2279                         case VIRTIO_NET_HDR_GSO_UDP:
2280                                 gso_type = SKB_GSO_UDP;
2281                                 break;
2282                         default:
2283                                 goto out_unlock;
2284                         }
2285
2286                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2287                                 gso_type |= SKB_GSO_TCP_ECN;
2288
2289                         if (vnet_hdr.gso_size == 0)
2290                                 goto out_unlock;
2291
2292                 }
2293         }
2294
2295         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2296                 if (!netif_supports_nofcs(dev)) {
2297                         err = -EPROTONOSUPPORT;
2298                         goto out_unlock;
2299                 }
2300                 extra_len = 4; /* We're doing our own CRC */
2301         }
2302
2303         err = -EMSGSIZE;
2304         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2305                 goto out_unlock;
2306
2307         err = -ENOBUFS;
2308         hlen = LL_RESERVED_SPACE(dev);
2309         tlen = dev->needed_tailroom;
2310         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2311                                msg->msg_flags & MSG_DONTWAIT, &err);
2312         if (skb == NULL)
2313                 goto out_unlock;
2314
2315         skb_set_network_header(skb, reserve);
2316
2317         err = -EINVAL;
2318         if (sock->type == SOCK_DGRAM &&
2319             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2320                 goto out_free;
2321
2322         /* Returns -EFAULT on error */
2323         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2324         if (err)
2325                 goto out_free;
2326         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2327         if (err < 0)
2328                 goto out_free;
2329
2330         if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2331                 /* Earlier code assumed this would be a VLAN pkt,
2332                  * double-check this now that we have the actual
2333                  * packet in hand.
2334                  */
2335                 struct ethhdr *ehdr;
2336                 skb_reset_mac_header(skb);
2337                 ehdr = eth_hdr(skb);
2338                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2339                         err = -EMSGSIZE;
2340                         goto out_free;
2341                 }
2342         }
2343
2344         skb->protocol = proto;
2345         skb->dev = dev;
2346         skb->priority = sk->sk_priority;
2347         skb->mark = sk->sk_mark;
2348
2349         if (po->has_vnet_hdr) {
2350                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2351                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2352                                                   vnet_hdr.csum_offset)) {
2353                                 err = -EINVAL;
2354                                 goto out_free;
2355                         }
2356                 }
2357
2358                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2359                 skb_shinfo(skb)->gso_type = gso_type;
2360
2361                 /* Header must be checked, and gso_segs computed. */
2362                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2363                 skb_shinfo(skb)->gso_segs = 0;
2364
2365                 len += vnet_hdr_len;
2366         }
2367
2368         if (skb->ip_summed == CHECKSUM_PARTIAL)
2369                 skb_set_transport_header(skb, skb_checksum_start_offset(skb));
2370         else if (skb_flow_dissect(skb, &keys))
2371                 skb_set_transport_header(skb, keys.thoff);
2372         else
2373                 skb_set_transport_header(skb, reserve);
2374
2375         if (unlikely(extra_len == 4))
2376                 skb->no_fcs = 1;
2377
2378         /*
2379          *      Now send it
2380          */
2381
2382         err = dev_queue_xmit(skb);
2383         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2384                 goto out_unlock;
2385
2386         if (need_rls_dev)
2387                 dev_put(dev);
2388
2389         return len;
2390
2391 out_free:
2392         kfree_skb(skb);
2393 out_unlock:
2394         if (dev && need_rls_dev)
2395                 dev_put(dev);
2396 out:
2397         return err;
2398 }
2399
2400 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2401                 struct msghdr *msg, size_t len)
2402 {
2403         struct sock *sk = sock->sk;
2404         struct packet_sock *po = pkt_sk(sk);
2405         if (po->tx_ring.pg_vec)
2406                 return tpacket_snd(po, msg);
2407         else
2408                 return packet_snd(sock, msg, len);
2409 }
2410
2411 /*
2412  *      Close a PACKET socket. This is fairly simple. We immediately go
2413  *      to 'closed' state and remove our protocol entry in the device list.
2414  */
2415
2416 static int packet_release(struct socket *sock)
2417 {
2418         struct sock *sk = sock->sk;
2419         struct packet_sock *po;
2420         struct net *net;
2421         union tpacket_req_u req_u;
2422
2423         if (!sk)
2424                 return 0;
2425
2426         net = sock_net(sk);
2427         po = pkt_sk(sk);
2428
2429         mutex_lock(&net->packet.sklist_lock);
2430         sk_del_node_init_rcu(sk);
2431         mutex_unlock(&net->packet.sklist_lock);
2432
2433         preempt_disable();
2434         sock_prot_inuse_add(net, sk->sk_prot, -1);
2435         preempt_enable();
2436
2437         spin_lock(&po->bind_lock);
2438         unregister_prot_hook(sk, false);
2439         if (po->prot_hook.dev) {
2440                 dev_put(po->prot_hook.dev);
2441                 po->prot_hook.dev = NULL;
2442         }
2443         spin_unlock(&po->bind_lock);
2444
2445         packet_flush_mclist(sk);
2446
2447         if (po->rx_ring.pg_vec) {
2448                 memset(&req_u, 0, sizeof(req_u));
2449                 packet_set_ring(sk, &req_u, 1, 0);
2450         }
2451
2452         if (po->tx_ring.pg_vec) {
2453                 memset(&req_u, 0, sizeof(req_u));
2454                 packet_set_ring(sk, &req_u, 1, 1);
2455         }
2456
2457         fanout_release(sk);
2458
2459         synchronize_net();
2460         /*
2461          *      Now the socket is dead. No more input will appear.
2462          */
2463         sock_orphan(sk);
2464         sock->sk = NULL;
2465
2466         /* Purge queues */
2467
2468         skb_queue_purge(&sk->sk_receive_queue);
2469         sk_refcnt_debug_release(sk);
2470
2471         sock_put(sk);
2472         return 0;
2473 }
2474
2475 /*
2476  *      Attach a packet hook.
2477  */
2478
2479 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2480 {
2481         struct packet_sock *po = pkt_sk(sk);
2482
2483         if (po->fanout) {
2484                 if (dev)
2485                         dev_put(dev);
2486
2487                 return -EINVAL;
2488         }
2489
2490         lock_sock(sk);
2491
2492         spin_lock(&po->bind_lock);
2493         unregister_prot_hook(sk, true);
2494         po->num = protocol;
2495         po->prot_hook.type = protocol;
2496         if (po->prot_hook.dev)
2497                 dev_put(po->prot_hook.dev);
2498         po->prot_hook.dev = dev;
2499
2500         po->ifindex = dev ? dev->ifindex : 0;
2501
2502         if (protocol == 0)
2503                 goto out_unlock;
2504
2505         if (!dev || (dev->flags & IFF_UP)) {
2506                 register_prot_hook(sk);
2507         } else {
2508                 sk->sk_err = ENETDOWN;
2509                 if (!sock_flag(sk, SOCK_DEAD))
2510                         sk->sk_error_report(sk);
2511         }
2512
2513 out_unlock:
2514         spin_unlock(&po->bind_lock);
2515         release_sock(sk);
2516         return 0;
2517 }
2518
2519 /*
2520  *      Bind a packet socket to a device
2521  */
2522
2523 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2524                             int addr_len)
2525 {
2526         struct sock *sk = sock->sk;
2527         char name[15];
2528         struct net_device *dev;
2529         int err = -ENODEV;
2530
2531         /*
2532          *      Check legality
2533          */
2534
2535         if (addr_len != sizeof(struct sockaddr))
2536                 return -EINVAL;
2537         strlcpy(name, uaddr->sa_data, sizeof(name));
2538
2539         dev = dev_get_by_name(sock_net(sk), name);
2540         if (dev)
2541                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2542         return err;
2543 }
2544
2545 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2546 {
2547         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2548         struct sock *sk = sock->sk;
2549         struct net_device *dev = NULL;
2550         int err;
2551
2552
2553         /*
2554          *      Check legality
2555          */
2556
2557         if (addr_len < sizeof(struct sockaddr_ll))
2558                 return -EINVAL;
2559         if (sll->sll_family != AF_PACKET)
2560                 return -EINVAL;
2561
2562         if (sll->sll_ifindex) {
2563                 err = -ENODEV;
2564                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2565                 if (dev == NULL)
2566                         goto out;
2567         }
2568         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2569
2570 out:
2571         return err;
2572 }
2573
2574 static struct proto packet_proto = {
2575         .name     = "PACKET",
2576         .owner    = THIS_MODULE,
2577         .obj_size = sizeof(struct packet_sock),
2578 };
2579
2580 /*
2581  *      Create a packet of type SOCK_PACKET.
2582  */
2583
2584 static int packet_create(struct net *net, struct socket *sock, int protocol,
2585                          int kern)
2586 {
2587         struct sock *sk;
2588         struct packet_sock *po;
2589         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2590         int err;
2591
2592         if (!ns_capable(net->user_ns, CAP_NET_RAW))
2593                 return -EPERM;
2594         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2595             sock->type != SOCK_PACKET)
2596                 return -ESOCKTNOSUPPORT;
2597
2598         sock->state = SS_UNCONNECTED;
2599
2600         err = -ENOBUFS;
2601         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2602         if (sk == NULL)
2603                 goto out;
2604
2605         sock->ops = &packet_ops;
2606         if (sock->type == SOCK_PACKET)
2607                 sock->ops = &packet_ops_spkt;
2608
2609         sock_init_data(sock, sk);
2610
2611         po = pkt_sk(sk);
2612         sk->sk_family = PF_PACKET;
2613         po->num = proto;
2614
2615         sk->sk_destruct = packet_sock_destruct;
2616         sk_refcnt_debug_inc(sk);
2617
2618         /*
2619          *      Attach a protocol block
2620          */
2621
2622         spin_lock_init(&po->bind_lock);
2623         mutex_init(&po->pg_vec_lock);
2624         po->prot_hook.func = packet_rcv;
2625
2626         if (sock->type == SOCK_PACKET)
2627                 po->prot_hook.func = packet_rcv_spkt;
2628
2629         po->prot_hook.af_packet_priv = sk;
2630
2631         if (proto) {
2632                 po->prot_hook.type = proto;
2633                 register_prot_hook(sk);
2634         }
2635
2636         mutex_lock(&net->packet.sklist_lock);
2637         sk_add_node_rcu(sk, &net->packet.sklist);
2638         mutex_unlock(&net->packet.sklist_lock);
2639
2640         preempt_disable();
2641         sock_prot_inuse_add(net, &packet_proto, 1);
2642         preempt_enable();
2643
2644         return 0;
2645 out:
2646         return err;
2647 }
2648
2649 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2650 {
2651         struct sock_exterr_skb *serr;
2652         struct sk_buff *skb, *skb2;
2653         int copied, err;
2654
2655         err = -EAGAIN;
2656         skb = skb_dequeue(&sk->sk_error_queue);
2657         if (skb == NULL)
2658                 goto out;
2659
2660         copied = skb->len;
2661         if (copied > len) {
2662                 msg->msg_flags |= MSG_TRUNC;
2663                 copied = len;
2664         }
2665         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2666         if (err)
2667                 goto out_free_skb;
2668
2669         sock_recv_timestamp(msg, sk, skb);
2670
2671         serr = SKB_EXT_ERR(skb);
2672         put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2673                  sizeof(serr->ee), &serr->ee);
2674
2675         msg->msg_flags |= MSG_ERRQUEUE;
2676         err = copied;
2677
2678         /* Reset and regenerate socket error */
2679         spin_lock_bh(&sk->sk_error_queue.lock);
2680         sk->sk_err = 0;
2681         if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2682                 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2683                 spin_unlock_bh(&sk->sk_error_queue.lock);
2684                 sk->sk_error_report(sk);
2685         } else
2686                 spin_unlock_bh(&sk->sk_error_queue.lock);
2687
2688 out_free_skb:
2689         kfree_skb(skb);
2690 out:
2691         return err;
2692 }
2693
2694 /*
2695  *      Pull a packet from our receive queue and hand it to the user.
2696  *      If necessary we block.
2697  */
2698
2699 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2700                           struct msghdr *msg, size_t len, int flags)
2701 {
2702         struct sock *sk = sock->sk;
2703         struct sk_buff *skb;
2704         int copied, err;
2705         struct sockaddr_ll *sll;
2706         int vnet_hdr_len = 0;
2707
2708         err = -EINVAL;
2709         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2710                 goto out;
2711
2712 #if 0
2713         /* What error should we return now? EUNATTACH? */
2714         if (pkt_sk(sk)->ifindex < 0)
2715                 return -ENODEV;
2716 #endif
2717
2718         if (flags & MSG_ERRQUEUE) {
2719                 err = packet_recv_error(sk, msg, len);
2720                 goto out;
2721         }
2722
2723         /*
2724          *      Call the generic datagram receiver. This handles all sorts
2725          *      of horrible races and re-entrancy so we can forget about it
2726          *      in the protocol layers.
2727          *
2728          *      Now it will return ENETDOWN, if device have just gone down,
2729          *      but then it will block.
2730          */
2731
2732         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2733
2734         /*
2735          *      An error occurred so return it. Because skb_recv_datagram()
2736          *      handles the blocking we don't see and worry about blocking
2737          *      retries.
2738          */
2739
2740         if (skb == NULL)
2741                 goto out;
2742
2743         if (pkt_sk(sk)->has_vnet_hdr) {
2744                 struct virtio_net_hdr vnet_hdr = { 0 };
2745
2746                 err = -EINVAL;
2747                 vnet_hdr_len = sizeof(vnet_hdr);
2748                 if (len < vnet_hdr_len)
2749                         goto out_free;
2750
2751                 len -= vnet_hdr_len;
2752
2753                 if (skb_is_gso(skb)) {
2754                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2755
2756                         /* This is a hint as to how much should be linear. */
2757                         vnet_hdr.hdr_len = skb_headlen(skb);
2758                         vnet_hdr.gso_size = sinfo->gso_size;
2759                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2760                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2761                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2762                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2763                         else if (sinfo->gso_type & SKB_GSO_UDP)
2764                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2765                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2766                                 goto out_free;
2767                         else
2768                                 BUG();
2769                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2770                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2771                 } else
2772                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2773
2774                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2775                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2776                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2777                         vnet_hdr.csum_offset = skb->csum_offset;
2778                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2779                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2780                 } /* else everything is zero */
2781
2782                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2783                                      vnet_hdr_len);
2784                 if (err < 0)
2785                         goto out_free;
2786         }
2787
2788         /*
2789          *      If the address length field is there to be filled in, we fill
2790          *      it in now.
2791          */
2792
2793         sll = &PACKET_SKB_CB(skb)->sa.ll;
2794         if (sock->type == SOCK_PACKET)
2795                 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2796         else
2797                 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2798
2799         /*
2800          *      You lose any data beyond the buffer you gave. If it worries a
2801          *      user program they can ask the device for its MTU anyway.
2802          */
2803
2804         copied = skb->len;
2805         if (copied > len) {
2806                 copied = len;
2807                 msg->msg_flags |= MSG_TRUNC;
2808         }
2809
2810         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2811         if (err)
2812                 goto out_free;
2813
2814         sock_recv_ts_and_drops(msg, sk, skb);
2815
2816         if (msg->msg_name)
2817                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2818                        msg->msg_namelen);
2819
2820         if (pkt_sk(sk)->auxdata) {
2821                 struct tpacket_auxdata aux;
2822
2823                 aux.tp_status = TP_STATUS_USER;
2824                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2825                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2826                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2827                 aux.tp_snaplen = skb->len;
2828                 aux.tp_mac = 0;
2829                 aux.tp_net = skb_network_offset(skb);
2830                 if (vlan_tx_tag_present(skb)) {
2831                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2832                         aux.tp_status |= TP_STATUS_VLAN_VALID;
2833                 } else {
2834                         aux.tp_vlan_tci = 0;
2835                 }
2836                 aux.tp_padding = 0;
2837                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2838         }
2839
2840         /*
2841          *      Free or return the buffer as appropriate. Again this
2842          *      hides all the races and re-entrancy issues from us.
2843          */
2844         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2845
2846 out_free:
2847         skb_free_datagram(sk, skb);
2848 out:
2849         return err;
2850 }
2851
2852 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2853                                int *uaddr_len, int peer)
2854 {
2855         struct net_device *dev;
2856         struct sock *sk = sock->sk;
2857
2858         if (peer)
2859                 return -EOPNOTSUPP;
2860
2861         uaddr->sa_family = AF_PACKET;
2862         rcu_read_lock();
2863         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2864         if (dev)
2865                 strncpy(uaddr->sa_data, dev->name, 14);
2866         else
2867                 memset(uaddr->sa_data, 0, 14);
2868         rcu_read_unlock();
2869         *uaddr_len = sizeof(*uaddr);
2870
2871         return 0;
2872 }
2873
2874 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2875                           int *uaddr_len, int peer)
2876 {
2877         struct net_device *dev;
2878         struct sock *sk = sock->sk;
2879         struct packet_sock *po = pkt_sk(sk);
2880         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2881
2882         if (peer)
2883                 return -EOPNOTSUPP;
2884
2885         sll->sll_family = AF_PACKET;
2886         sll->sll_ifindex = po->ifindex;
2887         sll->sll_protocol = po->num;
2888         sll->sll_pkttype = 0;
2889         rcu_read_lock();
2890         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2891         if (dev) {
2892                 sll->sll_hatype = dev->type;
2893                 sll->sll_halen = dev->addr_len;
2894                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2895         } else {
2896                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
2897                 sll->sll_halen = 0;
2898         }
2899         rcu_read_unlock();
2900         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2901
2902         return 0;
2903 }
2904
2905 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2906                          int what)
2907 {
2908         switch (i->type) {
2909         case PACKET_MR_MULTICAST:
2910                 if (i->alen != dev->addr_len)
2911                         return -EINVAL;
2912                 if (what > 0)
2913                         return dev_mc_add(dev, i->addr);
2914                 else
2915                         return dev_mc_del(dev, i->addr);
2916                 break;
2917         case PACKET_MR_PROMISC:
2918                 return dev_set_promiscuity(dev, what);
2919                 break;
2920         case PACKET_MR_ALLMULTI:
2921                 return dev_set_allmulti(dev, what);
2922                 break;
2923         case PACKET_MR_UNICAST:
2924                 if (i->alen != dev->addr_len)
2925                         return -EINVAL;
2926                 if (what > 0)
2927                         return dev_uc_add(dev, i->addr);
2928                 else
2929                         return dev_uc_del(dev, i->addr);
2930                 break;
2931         default:
2932                 break;
2933         }
2934         return 0;
2935 }
2936
2937 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2938 {
2939         for ( ; i; i = i->next) {
2940                 if (i->ifindex == dev->ifindex)
2941                         packet_dev_mc(dev, i, what);
2942         }
2943 }
2944
2945 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2946 {
2947         struct packet_sock *po = pkt_sk(sk);
2948         struct packet_mclist *ml, *i;
2949         struct net_device *dev;
2950         int err;
2951
2952         rtnl_lock();
2953
2954         err = -ENODEV;
2955         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2956         if (!dev)
2957                 goto done;
2958
2959         err = -EINVAL;
2960         if (mreq->mr_alen > dev->addr_len)
2961                 goto done;
2962
2963         err = -ENOBUFS;
2964         i = kmalloc(sizeof(*i), GFP_KERNEL);
2965         if (i == NULL)
2966                 goto done;
2967
2968         err = 0;
2969         for (ml = po->mclist; ml; ml = ml->next) {
2970                 if (ml->ifindex == mreq->mr_ifindex &&
2971                     ml->type == mreq->mr_type &&
2972                     ml->alen == mreq->mr_alen &&
2973                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2974                         ml->count++;
2975                         /* Free the new element ... */
2976                         kfree(i);
2977                         goto done;
2978                 }
2979         }
2980
2981         i->type = mreq->mr_type;
2982         i->ifindex = mreq->mr_ifindex;
2983         i->alen = mreq->mr_alen;
2984         memcpy(i->addr, mreq->mr_address, i->alen);
2985         i->count = 1;
2986         i->next = po->mclist;
2987         po->mclist = i;
2988         err = packet_dev_mc(dev, i, 1);
2989         if (err) {
2990                 po->mclist = i->next;
2991                 kfree(i);
2992         }
2993
2994 done:
2995         rtnl_unlock();
2996         return err;
2997 }
2998
2999 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3000 {
3001         struct packet_mclist *ml, **mlp;
3002
3003         rtnl_lock();
3004
3005         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3006                 if (ml->ifindex == mreq->mr_ifindex &&
3007                     ml->type == mreq->mr_type &&
3008                     ml->alen == mreq->mr_alen &&
3009                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3010                         if (--ml->count == 0) {
3011                                 struct net_device *dev;
3012                                 *mlp = ml->next;
3013                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3014                                 if (dev)
3015                                         packet_dev_mc(dev, ml, -1);
3016                                 kfree(ml);
3017                         }
3018                         rtnl_unlock();
3019                         return 0;
3020                 }
3021         }
3022         rtnl_unlock();
3023         return -EADDRNOTAVAIL;
3024 }
3025
3026 static void packet_flush_mclist(struct sock *sk)
3027 {
3028         struct packet_sock *po = pkt_sk(sk);
3029         struct packet_mclist *ml;
3030
3031         if (!po->mclist)
3032                 return;
3033
3034         rtnl_lock();
3035         while ((ml = po->mclist) != NULL) {
3036                 struct net_device *dev;
3037
3038                 po->mclist = ml->next;
3039                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3040                 if (dev != NULL)
3041                         packet_dev_mc(dev, ml, -1);
3042                 kfree(ml);
3043         }
3044         rtnl_unlock();
3045 }
3046
3047 static int
3048 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3049 {
3050         struct sock *sk = sock->sk;
3051         struct packet_sock *po = pkt_sk(sk);
3052         int ret;
3053
3054         if (level != SOL_PACKET)
3055                 return -ENOPROTOOPT;
3056
3057         switch (optname) {
3058         case PACKET_ADD_MEMBERSHIP:
3059         case PACKET_DROP_MEMBERSHIP:
3060         {
3061                 struct packet_mreq_max mreq;
3062                 int len = optlen;
3063                 memset(&mreq, 0, sizeof(mreq));
3064                 if (len < sizeof(struct packet_mreq))
3065                         return -EINVAL;
3066                 if (len > sizeof(mreq))
3067                         len = sizeof(mreq);
3068                 if (copy_from_user(&mreq, optval, len))
3069                         return -EFAULT;
3070                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3071                         return -EINVAL;
3072                 if (optname == PACKET_ADD_MEMBERSHIP)
3073                         ret = packet_mc_add(sk, &mreq);
3074                 else
3075                         ret = packet_mc_drop(sk, &mreq);
3076                 return ret;
3077         }
3078
3079         case PACKET_RX_RING:
3080         case PACKET_TX_RING:
3081         {
3082                 union tpacket_req_u req_u;
3083                 int len;
3084
3085                 switch (po->tp_version) {
3086                 case TPACKET_V1:
3087                 case TPACKET_V2:
3088                         len = sizeof(req_u.req);
3089                         break;
3090                 case TPACKET_V3:
3091                 default:
3092                         len = sizeof(req_u.req3);
3093                         break;
3094                 }
3095                 if (optlen < len)
3096                         return -EINVAL;
3097                 if (pkt_sk(sk)->has_vnet_hdr)
3098                         return -EINVAL;
3099                 if (copy_from_user(&req_u.req, optval, len))
3100                         return -EFAULT;
3101                 return packet_set_ring(sk, &req_u, 0,
3102                         optname == PACKET_TX_RING);
3103         }
3104         case PACKET_COPY_THRESH:
3105         {
3106                 int val;
3107
3108                 if (optlen != sizeof(val))
3109                         return -EINVAL;
3110                 if (copy_from_user(&val, optval, sizeof(val)))
3111                         return -EFAULT;
3112
3113                 pkt_sk(sk)->copy_thresh = val;
3114                 return 0;
3115         }
3116         case PACKET_VERSION:
3117         {
3118                 int val;
3119
3120                 if (optlen != sizeof(val))
3121                         return -EINVAL;
3122                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3123                         return -EBUSY;
3124                 if (copy_from_user(&val, optval, sizeof(val)))
3125                         return -EFAULT;
3126                 switch (val) {
3127                 case TPACKET_V1:
3128                 case TPACKET_V2:
3129                 case TPACKET_V3:
3130                         po->tp_version = val;
3131                         return 0;
3132                 default:
3133                         return -EINVAL;
3134                 }
3135         }
3136         case PACKET_RESERVE:
3137         {
3138                 unsigned int val;
3139
3140                 if (optlen != sizeof(val))
3141                         return -EINVAL;
3142                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3143                         return -EBUSY;
3144                 if (copy_from_user(&val, optval, sizeof(val)))
3145                         return -EFAULT;
3146                 po->tp_reserve = val;
3147                 return 0;
3148         }
3149         case PACKET_LOSS:
3150         {
3151                 unsigned int val;
3152
3153                 if (optlen != sizeof(val))
3154                         return -EINVAL;
3155                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3156                         return -EBUSY;
3157                 if (copy_from_user(&val, optval, sizeof(val)))
3158                         return -EFAULT;
3159                 po->tp_loss = !!val;
3160                 return 0;
3161         }
3162         case PACKET_AUXDATA:
3163         {
3164                 int val;
3165
3166                 if (optlen < sizeof(val))
3167                         return -EINVAL;
3168                 if (copy_from_user(&val, optval, sizeof(val)))
3169                         return -EFAULT;
3170
3171                 po->auxdata = !!val;
3172                 return 0;
3173         }
3174         case PACKET_ORIGDEV:
3175         {
3176                 int val;
3177
3178                 if (optlen < sizeof(val))
3179                         return -EINVAL;
3180                 if (copy_from_user(&val, optval, sizeof(val)))
3181                         return -EFAULT;
3182
3183                 po->origdev = !!val;
3184                 return 0;
3185         }
3186         case PACKET_VNET_HDR:
3187         {
3188                 int val;
3189
3190                 if (sock->type != SOCK_RAW)
3191                         return -EINVAL;
3192                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3193                         return -EBUSY;
3194                 if (optlen < sizeof(val))
3195                         return -EINVAL;
3196                 if (copy_from_user(&val, optval, sizeof(val)))
3197                         return -EFAULT;
3198
3199                 po->has_vnet_hdr = !!val;
3200                 return 0;
3201         }
3202         case PACKET_TIMESTAMP:
3203         {
3204                 int val;
3205
3206                 if (optlen != sizeof(val))
3207                         return -EINVAL;
3208                 if (copy_from_user(&val, optval, sizeof(val)))
3209                         return -EFAULT;
3210
3211                 po->tp_tstamp = val;
3212                 return 0;
3213         }
3214         case PACKET_FANOUT:
3215         {
3216                 int val;
3217
3218                 if (optlen != sizeof(val))
3219                         return -EINVAL;
3220                 if (copy_from_user(&val, optval, sizeof(val)))
3221                         return -EFAULT;
3222
3223                 return fanout_add(sk, val & 0xffff, val >> 16);
3224         }
3225         case PACKET_TX_HAS_OFF:
3226         {
3227                 unsigned int val;
3228
3229                 if (optlen != sizeof(val))
3230                         return -EINVAL;
3231                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3232                         return -EBUSY;
3233                 if (copy_from_user(&val, optval, sizeof(val)))
3234                         return -EFAULT;
3235                 po->tp_tx_has_off = !!val;
3236                 return 0;
3237         }
3238         default:
3239                 return -ENOPROTOOPT;
3240         }
3241 }
3242
3243 static int packet_getsockopt(struct socket *sock, int level, int optname,
3244                              char __user *optval, int __user *optlen)
3245 {
3246         int len;
3247         int val, lv = sizeof(val);
3248         struct sock *sk = sock->sk;
3249         struct packet_sock *po = pkt_sk(sk);
3250         void *data = &val;
3251         struct tpacket_stats st;
3252         union tpacket_stats_u st_u;
3253
3254         if (level != SOL_PACKET)
3255                 return -ENOPROTOOPT;
3256
3257         if (get_user(len, optlen))
3258                 return -EFAULT;
3259
3260         if (len < 0)
3261                 return -EINVAL;
3262
3263         switch (optname) {
3264         case PACKET_STATISTICS:
3265                 spin_lock_bh(&sk->sk_receive_queue.lock);
3266                 if (po->tp_version == TPACKET_V3) {
3267                         lv = sizeof(struct tpacket_stats_v3);
3268                         memcpy(&st_u.stats3, &po->stats,
3269                                sizeof(struct tpacket_stats));
3270                         st_u.stats3.tp_freeze_q_cnt =
3271                                         po->stats_u.stats3.tp_freeze_q_cnt;
3272                         st_u.stats3.tp_packets += po->stats.tp_drops;
3273                         data = &st_u.stats3;
3274                 } else {
3275                         lv = sizeof(struct tpacket_stats);
3276                         st = po->stats;
3277                         st.tp_packets += st.tp_drops;
3278                         data = &st;
3279                 }
3280                 memset(&po->stats, 0, sizeof(st));
3281                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3282                 break;
3283         case PACKET_AUXDATA:
3284                 val = po->auxdata;
3285                 break;
3286         case PACKET_ORIGDEV:
3287                 val = po->origdev;
3288                 break;
3289         case PACKET_VNET_HDR:
3290                 val = po->has_vnet_hdr;
3291                 break;
3292         case PACKET_VERSION:
3293                 val = po->tp_version;
3294                 break;
3295         case PACKET_HDRLEN:
3296                 if (len > sizeof(int))
3297                         len = sizeof(int);
3298                 if (copy_from_user(&val, optval, len))
3299                         return -EFAULT;
3300                 switch (val) {
3301                 case TPACKET_V1:
3302                         val = sizeof(struct tpacket_hdr);
3303                         break;
3304                 case TPACKET_V2:
3305                         val = sizeof(struct tpacket2_hdr);
3306                         break;
3307                 case TPACKET_V3:
3308                         val = sizeof(struct tpacket3_hdr);
3309                         break;
3310                 default:
3311                         return -EINVAL;
3312                 }
3313                 break;
3314         case PACKET_RESERVE:
3315                 val = po->tp_reserve;
3316                 break;
3317         case PACKET_LOSS:
3318                 val = po->tp_loss;
3319                 break;
3320         case PACKET_TIMESTAMP:
3321                 val = po->tp_tstamp;
3322                 break;
3323         case PACKET_FANOUT:
3324                 val = (po->fanout ?
3325                        ((u32)po->fanout->id |
3326                         ((u32)po->fanout->type << 16) |
3327                         ((u32)po->fanout->flags << 24)) :
3328                        0);
3329                 break;
3330         case PACKET_TX_HAS_OFF:
3331                 val = po->tp_tx_has_off;
3332                 break;
3333         default:
3334                 return -ENOPROTOOPT;
3335         }
3336
3337         if (len > lv)
3338                 len = lv;
3339         if (put_user(len, optlen))
3340                 return -EFAULT;
3341         if (copy_to_user(optval, data, len))
3342                 return -EFAULT;
3343         return 0;
3344 }
3345
3346
3347 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3348 {
3349         struct sock *sk;
3350         struct net_device *dev = data;
3351         struct net *net = dev_net(dev);
3352
3353         rcu_read_lock();
3354         sk_for_each_rcu(sk, &net->packet.sklist) {
3355                 struct packet_sock *po = pkt_sk(sk);
3356
3357                 switch (msg) {
3358                 case NETDEV_UNREGISTER:
3359                         if (po->mclist)
3360                                 packet_dev_mclist(dev, po->mclist, -1);
3361                         /* fallthrough */
3362
3363                 case NETDEV_DOWN:
3364                         if (dev->ifindex == po->ifindex) {
3365                                 spin_lock(&po->bind_lock);
3366                                 if (po->running) {
3367                                         __unregister_prot_hook(sk, false);
3368                                         sk->sk_err = ENETDOWN;
3369                                         if (!sock_flag(sk, SOCK_DEAD))
3370                                                 sk->sk_error_report(sk);
3371                                 }
3372                                 if (msg == NETDEV_UNREGISTER) {
3373                                         po->ifindex = -1;
3374                                         if (po->prot_hook.dev)
3375                                                 dev_put(po->prot_hook.dev);
3376                                         po->prot_hook.dev = NULL;
3377                                 }
3378                                 spin_unlock(&po->bind_lock);
3379                         }
3380                         break;
3381                 case NETDEV_UP:
3382                         if (dev->ifindex == po->ifindex) {
3383                                 spin_lock(&po->bind_lock);
3384                                 if (po->num)
3385                                         register_prot_hook(sk);
3386                                 spin_unlock(&po->bind_lock);
3387                         }
3388                         break;
3389                 }
3390         }
3391         rcu_read_unlock();
3392         return NOTIFY_DONE;
3393 }
3394
3395
3396 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3397                         unsigned long arg)
3398 {
3399         struct sock *sk = sock->sk;
3400
3401         switch (cmd) {
3402         case SIOCOUTQ:
3403         {
3404                 int amount = sk_wmem_alloc_get(sk);
3405
3406                 return put_user(amount, (int __user *)arg);
3407         }
3408         case SIOCINQ:
3409         {
3410                 struct sk_buff *skb;
3411                 int amount = 0;
3412
3413                 spin_lock_bh(&sk->sk_receive_queue.lock);
3414                 skb = skb_peek(&sk->sk_receive_queue);
3415                 if (skb)
3416                         amount = skb->len;
3417                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3418                 return put_user(amount, (int __user *)arg);
3419         }
3420         case SIOCGSTAMP:
3421                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3422         case SIOCGSTAMPNS:
3423                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3424
3425 #ifdef CONFIG_INET
3426         case SIOCADDRT:
3427         case SIOCDELRT:
3428         case SIOCDARP:
3429         case SIOCGARP:
3430         case SIOCSARP:
3431         case SIOCGIFADDR:
3432         case SIOCSIFADDR:
3433         case SIOCGIFBRDADDR:
3434         case SIOCSIFBRDADDR:
3435         case SIOCGIFNETMASK:
3436         case SIOCSIFNETMASK:
3437         case SIOCGIFDSTADDR:
3438         case SIOCSIFDSTADDR:
3439         case SIOCSIFFLAGS:
3440                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3441 #endif
3442
3443         default:
3444                 return -ENOIOCTLCMD;
3445         }
3446         return 0;
3447 }
3448
3449 static unsigned int packet_poll(struct file *file, struct socket *sock,
3450                                 poll_table *wait)
3451 {
3452         struct sock *sk = sock->sk;
3453         struct packet_sock *po = pkt_sk(sk);
3454         unsigned int mask = datagram_poll(file, sock, wait);
3455
3456         spin_lock_bh(&sk->sk_receive_queue.lock);
3457         if (po->rx_ring.pg_vec) {
3458                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3459                         TP_STATUS_KERNEL))
3460                         mask |= POLLIN | POLLRDNORM;
3461         }
3462         spin_unlock_bh(&sk->sk_receive_queue.lock);
3463         spin_lock_bh(&sk->sk_write_queue.lock);
3464         if (po->tx_ring.pg_vec) {
3465                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3466                         mask |= POLLOUT | POLLWRNORM;
3467         }
3468         spin_unlock_bh(&sk->sk_write_queue.lock);
3469         return mask;
3470 }
3471
3472
3473 /* Dirty? Well, I still did not learn better way to account
3474  * for user mmaps.
3475  */
3476
3477 static void packet_mm_open(struct vm_area_struct *vma)
3478 {
3479         struct file *file = vma->vm_file;
3480         struct socket *sock = file->private_data;
3481         struct sock *sk = sock->sk;
3482
3483         if (sk)
3484                 atomic_inc(&pkt_sk(sk)->mapped);
3485 }
3486
3487 static void packet_mm_close(struct vm_area_struct *vma)
3488 {
3489         struct file *file = vma->vm_file;
3490         struct socket *sock = file->private_data;
3491         struct sock *sk = sock->sk;
3492
3493         if (sk)
3494                 atomic_dec(&pkt_sk(sk)->mapped);
3495 }
3496
3497 static const struct vm_operations_struct packet_mmap_ops = {
3498         .open   =       packet_mm_open,
3499         .close  =       packet_mm_close,
3500 };
3501
3502 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3503                         unsigned int len)
3504 {
3505         int i;
3506
3507         for (i = 0; i < len; i++) {
3508                 if (likely(pg_vec[i].buffer)) {
3509                         if (is_vmalloc_addr(pg_vec[i].buffer))
3510                                 vfree(pg_vec[i].buffer);
3511                         else
3512                                 free_pages((unsigned long)pg_vec[i].buffer,
3513                                            order);
3514                         pg_vec[i].buffer = NULL;
3515                 }
3516         }
3517         kfree(pg_vec);
3518 }
3519
3520 static char *alloc_one_pg_vec_page(unsigned long order)
3521 {
3522         char *buffer = NULL;
3523         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3524                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3525
3526         buffer = (char *) __get_free_pages(gfp_flags, order);
3527
3528         if (buffer)
3529                 return buffer;
3530
3531         /*
3532          * __get_free_pages failed, fall back to vmalloc
3533          */
3534         buffer = vzalloc((1 << order) * PAGE_SIZE);
3535
3536         if (buffer)
3537                 return buffer;
3538
3539         /*
3540          * vmalloc failed, lets dig into swap here
3541          */
3542         gfp_flags &= ~__GFP_NORETRY;
3543         buffer = (char *)__get_free_pages(gfp_flags, order);
3544         if (buffer)
3545                 return buffer;
3546
3547         /*
3548          * complete and utter failure
3549          */
3550         return NULL;
3551 }
3552
3553 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3554 {
3555         unsigned int block_nr = req->tp_block_nr;
3556         struct pgv *pg_vec;
3557         int i;
3558
3559         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3560         if (unlikely(!pg_vec))
3561                 goto out;
3562
3563         for (i = 0; i < block_nr; i++) {
3564                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3565                 if (unlikely(!pg_vec[i].buffer))
3566                         goto out_free_pgvec;
3567         }
3568
3569 out:
3570         return pg_vec;
3571
3572 out_free_pgvec:
3573         free_pg_vec(pg_vec, order, block_nr);
3574         pg_vec = NULL;
3575         goto out;
3576 }
3577
3578 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3579                 int closing, int tx_ring)
3580 {
3581         struct pgv *pg_vec = NULL;
3582         struct packet_sock *po = pkt_sk(sk);
3583         int was_running, order = 0;
3584         struct packet_ring_buffer *rb;
3585         struct sk_buff_head *rb_queue;
3586         __be16 num;
3587         int err = -EINVAL;
3588         /* Added to avoid minimal code churn */
3589         struct tpacket_req *req = &req_u->req;
3590
3591         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3592         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3593                 WARN(1, "Tx-ring is not supported.\n");
3594                 goto out;
3595         }
3596
3597         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3598         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3599
3600         err = -EBUSY;
3601         if (!closing) {
3602                 if (atomic_read(&po->mapped))
3603                         goto out;
3604                 if (atomic_read(&rb->pending))
3605                         goto out;
3606         }
3607
3608         if (req->tp_block_nr) {
3609                 /* Sanity tests and some calculations */
3610                 err = -EBUSY;
3611                 if (unlikely(rb->pg_vec))
3612                         goto out;
3613
3614                 switch (po->tp_version) {
3615                 case TPACKET_V1:
3616                         po->tp_hdrlen = TPACKET_HDRLEN;
3617                         break;
3618                 case TPACKET_V2:
3619                         po->tp_hdrlen = TPACKET2_HDRLEN;
3620                         break;
3621                 case TPACKET_V3:
3622                         po->tp_hdrlen = TPACKET3_HDRLEN;
3623                         break;
3624                 }
3625
3626                 err = -EINVAL;
3627                 if (unlikely((int)req->tp_block_size <= 0))
3628                         goto out;
3629                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3630                         goto out;
3631                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3632                                         po->tp_reserve))
3633                         goto out;
3634                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3635                         goto out;
3636
3637                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3638                 if (unlikely(rb->frames_per_block <= 0))
3639                         goto out;
3640                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3641                                         req->tp_frame_nr))
3642                         goto out;
3643
3644                 err = -ENOMEM;
3645                 order = get_order(req->tp_block_size);
3646                 pg_vec = alloc_pg_vec(req, order);
3647                 if (unlikely(!pg_vec))
3648                         goto out;
3649                 switch (po->tp_version) {
3650                 case TPACKET_V3:
3651                 /* Transmit path is not supported. We checked
3652                  * it above but just being paranoid
3653                  */
3654                         if (!tx_ring)
3655                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3656                                 break;
3657                 default:
3658                         break;
3659                 }
3660         }
3661         /* Done */
3662         else {
3663                 err = -EINVAL;
3664                 if (unlikely(req->tp_frame_nr))
3665                         goto out;
3666         }
3667
3668         lock_sock(sk);
3669
3670         /* Detach socket from network */
3671         spin_lock(&po->bind_lock);
3672         was_running = po->running;
3673         num = po->num;
3674         if (was_running) {
3675                 po->num = 0;
3676                 __unregister_prot_hook(sk, false);
3677         }
3678         spin_unlock(&po->bind_lock);
3679
3680         synchronize_net();
3681
3682         err = -EBUSY;
3683         mutex_lock(&po->pg_vec_lock);
3684         if (closing || atomic_read(&po->mapped) == 0) {
3685                 err = 0;
3686                 spin_lock_bh(&rb_queue->lock);
3687                 swap(rb->pg_vec, pg_vec);
3688                 rb->frame_max = (req->tp_frame_nr - 1);
3689                 rb->head = 0;
3690                 rb->frame_size = req->tp_frame_size;
3691                 spin_unlock_bh(&rb_queue->lock);
3692
3693                 swap(rb->pg_vec_order, order);
3694                 swap(rb->pg_vec_len, req->tp_block_nr);
3695
3696                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3697                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3698                                                 tpacket_rcv : packet_rcv;
3699                 skb_queue_purge(rb_queue);
3700                 if (atomic_read(&po->mapped))
3701                         pr_err("packet_mmap: vma is busy: %d\n",
3702                                atomic_read(&po->mapped));
3703         }
3704         mutex_unlock(&po->pg_vec_lock);
3705
3706         spin_lock(&po->bind_lock);
3707         if (was_running) {
3708                 po->num = num;
3709                 register_prot_hook(sk);
3710         }
3711         spin_unlock(&po->bind_lock);
3712         if (closing && (po->tp_version > TPACKET_V2)) {
3713                 /* Because we don't support block-based V3 on tx-ring */
3714                 if (!tx_ring)
3715                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3716         }
3717         release_sock(sk);
3718
3719         if (pg_vec)
3720                 free_pg_vec(pg_vec, order, req->tp_block_nr);
3721 out:
3722         return err;
3723 }
3724
3725 static int packet_mmap(struct file *file, struct socket *sock,
3726                 struct vm_area_struct *vma)
3727 {
3728         struct sock *sk = sock->sk;
3729         struct packet_sock *po = pkt_sk(sk);
3730         unsigned long size, expected_size;
3731         struct packet_ring_buffer *rb;
3732         unsigned long start;
3733         int err = -EINVAL;
3734         int i;
3735
3736         if (vma->vm_pgoff)
3737                 return -EINVAL;
3738
3739         mutex_lock(&po->pg_vec_lock);
3740
3741         expected_size = 0;
3742         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3743                 if (rb->pg_vec) {
3744                         expected_size += rb->pg_vec_len
3745                                                 * rb->pg_vec_pages
3746                                                 * PAGE_SIZE;
3747                 }
3748         }
3749
3750         if (expected_size == 0)
3751                 goto out;
3752
3753         size = vma->vm_end - vma->vm_start;
3754         if (size != expected_size)
3755                 goto out;
3756
3757         start = vma->vm_start;
3758         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3759                 if (rb->pg_vec == NULL)
3760                         continue;
3761
3762                 for (i = 0; i < rb->pg_vec_len; i++) {
3763                         struct page *page;
3764                         void *kaddr = rb->pg_vec[i].buffer;
3765                         int pg_num;
3766
3767                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3768                                 page = pgv_to_page(kaddr);
3769                                 err = vm_insert_page(vma, start, page);
3770                                 if (unlikely(err))
3771                                         goto out;
3772                                 start += PAGE_SIZE;
3773                                 kaddr += PAGE_SIZE;
3774                         }
3775                 }
3776         }
3777
3778         atomic_inc(&po->mapped);
3779         vma->vm_ops = &packet_mmap_ops;
3780         err = 0;
3781
3782 out:
3783         mutex_unlock(&po->pg_vec_lock);
3784         return err;
3785 }
3786
3787 static const struct proto_ops packet_ops_spkt = {
3788         .family =       PF_PACKET,
3789         .owner =        THIS_MODULE,
3790         .release =      packet_release,
3791         .bind =         packet_bind_spkt,
3792         .connect =      sock_no_connect,
3793         .socketpair =   sock_no_socketpair,
3794         .accept =       sock_no_accept,
3795         .getname =      packet_getname_spkt,
3796         .poll =         datagram_poll,
3797         .ioctl =        packet_ioctl,
3798         .listen =       sock_no_listen,
3799         .shutdown =     sock_no_shutdown,
3800         .setsockopt =   sock_no_setsockopt,
3801         .getsockopt =   sock_no_getsockopt,
3802         .sendmsg =      packet_sendmsg_spkt,
3803         .recvmsg =      packet_recvmsg,
3804         .mmap =         sock_no_mmap,
3805         .sendpage =     sock_no_sendpage,
3806 };
3807
3808 static const struct proto_ops packet_ops = {
3809         .family =       PF_PACKET,
3810         .owner =        THIS_MODULE,
3811         .release =      packet_release,
3812         .bind =         packet_bind,
3813         .connect =      sock_no_connect,
3814         .socketpair =   sock_no_socketpair,
3815         .accept =       sock_no_accept,
3816         .getname =      packet_getname,
3817         .poll =         packet_poll,
3818         .ioctl =        packet_ioctl,
3819         .listen =       sock_no_listen,
3820         .shutdown =     sock_no_shutdown,
3821         .setsockopt =   packet_setsockopt,
3822         .getsockopt =   packet_getsockopt,
3823         .sendmsg =      packet_sendmsg,
3824         .recvmsg =      packet_recvmsg,
3825         .mmap =         packet_mmap,
3826         .sendpage =     sock_no_sendpage,
3827 };
3828
3829 static const struct net_proto_family packet_family_ops = {
3830         .family =       PF_PACKET,
3831         .create =       packet_create,
3832         .owner  =       THIS_MODULE,
3833 };
3834
3835 static struct notifier_block packet_netdev_notifier = {
3836         .notifier_call =        packet_notifier,
3837 };
3838
3839 #ifdef CONFIG_PROC_FS
3840
3841 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3842         __acquires(RCU)
3843 {
3844         struct net *net = seq_file_net(seq);
3845
3846         rcu_read_lock();
3847         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3848 }
3849
3850 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3851 {
3852         struct net *net = seq_file_net(seq);
3853         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3854 }
3855
3856 static void packet_seq_stop(struct seq_file *seq, void *v)
3857         __releases(RCU)
3858 {
3859         rcu_read_unlock();
3860 }
3861
3862 static int packet_seq_show(struct seq_file *seq, void *v)
3863 {
3864         if (v == SEQ_START_TOKEN)
3865                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
3866         else {
3867                 struct sock *s = sk_entry(v);
3868                 const struct packet_sock *po = pkt_sk(s);
3869
3870                 seq_printf(seq,
3871                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
3872                            s,
3873                            atomic_read(&s->sk_refcnt),
3874                            s->sk_type,
3875                            ntohs(po->num),
3876                            po->ifindex,
3877                            po->running,
3878                            atomic_read(&s->sk_rmem_alloc),
3879                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
3880                            sock_i_ino(s));
3881         }
3882
3883         return 0;
3884 }
3885
3886 static const struct seq_operations packet_seq_ops = {
3887         .start  = packet_seq_start,
3888         .next   = packet_seq_next,
3889         .stop   = packet_seq_stop,
3890         .show   = packet_seq_show,
3891 };
3892
3893 static int packet_seq_open(struct inode *inode, struct file *file)
3894 {
3895         return seq_open_net(inode, file, &packet_seq_ops,
3896                             sizeof(struct seq_net_private));
3897 }
3898
3899 static const struct file_operations packet_seq_fops = {
3900         .owner          = THIS_MODULE,
3901         .open           = packet_seq_open,
3902         .read           = seq_read,
3903         .llseek         = seq_lseek,
3904         .release        = seq_release_net,
3905 };
3906
3907 #endif
3908
3909 static int __net_init packet_net_init(struct net *net)
3910 {
3911         mutex_init(&net->packet.sklist_lock);
3912         INIT_HLIST_HEAD(&net->packet.sklist);
3913
3914         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
3915                 return -ENOMEM;
3916
3917         return 0;
3918 }
3919
3920 static void __net_exit packet_net_exit(struct net *net)
3921 {
3922         remove_proc_entry("packet", net->proc_net);
3923 }
3924
3925 static struct pernet_operations packet_net_ops = {
3926         .init = packet_net_init,
3927         .exit = packet_net_exit,
3928 };
3929
3930
3931 static void __exit packet_exit(void)
3932 {
3933         unregister_netdevice_notifier(&packet_netdev_notifier);
3934         unregister_pernet_subsys(&packet_net_ops);
3935         sock_unregister(PF_PACKET);
3936         proto_unregister(&packet_proto);
3937 }
3938
3939 static int __init packet_init(void)
3940 {
3941         int rc = proto_register(&packet_proto, 0);
3942
3943         if (rc != 0)
3944                 goto out;
3945
3946         sock_register(&packet_family_ops);
3947         register_pernet_subsys(&packet_net_ops);
3948         register_netdevice_notifier(&packet_netdev_notifier);
3949 out:
3950         return rc;
3951 }
3952
3953 module_init(packet_init);
3954 module_exit(packet_exit);
3955 MODULE_LICENSE("GPL");
3956 MODULE_ALIAS_NETPROTO(PF_PACKET);