net/packet/af_packet.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              PACKET - implements raw packet sockets.
   7  *
   8  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  11  *
  12  * Fixes:
  13  *              Alan Cox        :       verify_area() now used correctly
  14  *              Alan Cox        :       new skbuff lists, look ma no backlogs!
  15  *              Alan Cox        :       tidied skbuff lists.
  16  *              Alan Cox        :       Now uses generic datagram routines I
  17  *                                      added. Also fixed the peek/read crash
  18  *                                      from all old Linux datagram code.
  19  *              Alan Cox        :       Uses the improved datagram code.
  20  *              Alan Cox        :       Added NULL's for socket options.
  21  *              Alan Cox        :       Re-commented the code.
  22  *              Alan Cox        :       Use new kernel side addressing
  23  *              Rob Janssen     :       Correct MTU usage.
  24  *              Dave Platt      :       Counter leaks caused by incorrect
  25  *                                      interrupt locking and some slightly
  26  *                                      dubious gcc output. Can you read
  27  *                                      compiler: it said _VOLATILE_
  28  *      Richard Kooijman        :       Timestamp fixes.
  29  *              Alan Cox        :       New buffers. Use sk->mac.raw.
  30  *              Alan Cox        :       sendmsg/recvmsg support.
  31  *              Alan Cox        :       Protocol setting support
  32  *      Alexey Kuznetsov        :       Untied from IPv4 stack.
  33  *      Cyrus Durgin            :       Fixed kerneld for kmod.
  34  *      Michal Ostrowski        :       Module initialization cleanup.
  35  *         Ulises Alonso        :       Frame number limit removal and
  36  *                                      packet_set_ring memory leak.
  37  *              Eric Biederman  :       Allow for > 8 byte hardware addresses.
  38  *                                      The convention is that longer addresses
  39  *                                      will simply extend the hardware address
  40  *                                      byte arrays at the end of sockaddr_ll
  41  *                                      and packet_mreq.
  42  *              Johann Baudy    :       Added TX RING.
  43  *              Chetan Loke     :       Implemented TPACKET_V3 block abstraction
  44  *                                      layer.
  45  *                                      Copyright (C) 2011, <lokec@ccs.neu.edu>
  46  *
  47  *
  48  *              This program is free software; you can redistribute it and/or
  49  *              modify it under the terms of the GNU General Public License
  50  *              as published by the Free Software Foundation; either version
  51  *              2 of the License, or (at your option) any later version.
  52  *
  53  */
  54
  55 #include <linux/types.h>
  56 #include <linux/mm.h>
  57 #include <linux/capability.h>
  58 #include <linux/fcntl.h>
  59 #include <linux/socket.h>
  60 #include <linux/in.h>
  61 #include <linux/inet.h>
  62 #include <linux/netdevice.h>
  63 #include <linux/if_packet.h>
  64 #include <linux/wireless.h>
  65 #include <linux/kernel.h>
  66 #include <linux/kmod.h>
  67 #include <linux/slab.h>
  68 #include <linux/vmalloc.h>
  69 #include <net/net_namespace.h>
  70 #include <net/ip.h>
  71 #include <net/protocol.h>
  72 #include <linux/skbuff.h>
  73 #include <net/sock.h>
  74 #include <linux/errno.h>
  75 #include <linux/timer.h>
  76 #include <asm/uaccess.h>
  77 #include <asm/ioctls.h>
  78 #include <asm/page.h>
  79 #include <asm/cacheflush.h>
  80 #include <asm/io.h>
  81 #include <linux/proc_fs.h>
  82 #include <linux/seq_file.h>
  83 #include <linux/poll.h>
  84 #include <linux/module.h>
  85 #include <linux/init.h>
  86 #include <linux/mutex.h>
  87 #include <linux/if_vlan.h>
  88 #include <linux/virtio_net.h>
  89 #include <linux/errqueue.h>
  90 #include <linux/net_tstamp.h>
  91 #include <linux/percpu.h>
  92 #ifdef CONFIG_INET
  93 #include <net/inet_common.h>
  94 #endif
  95 #include <linux/bpf.h>
  96 #include <net/compat.h>
  97
  98 #include "internal.h"
  99
 100 /*
 101    Assumptions:
 102    - if device has no dev->hard_header routine, it adds and removes ll header
 103      inside itself. In this case ll header is invisible outside of device,
 104      but higher levels still should reserve dev->hard_header_len.
 105      Some devices are enough clever to reallocate skb, when header
 106      will not fit to reserved space (tunnel), another ones are silly
 107      (PPP).
 108    - packet socket receives packets with pulled ll header,
 109      so that SOCK_RAW should push it back.
 110
 111 On receive:
 112 -----------
 113
 114 Incoming, dev->hard_header!=NULL
 115    mac_header -> ll header
 116    data       -> data
 117
 118 Outgoing, dev->hard_header!=NULL
 119    mac_header -> ll header
 120    data       -> ll header
 121
 122 Incoming, dev->hard_header==NULL
 123    mac_header -> UNKNOWN position. It is very likely, that it points to ll
 124                  header.  PPP makes it, that is wrong, because introduce
 125                  assymetry between rx and tx paths.
 126    data       -> data
 127
 128 Outgoing, dev->hard_header==NULL
 129    mac_header -> data. ll header is still not built!
 130    data       -> data
 131
 132 Resume
 133   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 134
 135
 136 On transmit:
 137 ------------
 138
 139 dev->hard_header != NULL
 140    mac_header -> ll header
 141    data       -> ll header
 142
 143 dev->hard_header == NULL (ll header is added by device, we cannot control it)
 144    mac_header -> data
 145    data       -> data
 146
 147    We should set nh.raw on output to correct posistion,
 148    packet classifier depends on it.
 149  */
 150
 151 /* Private packet socket structures. */
 152
 153 /* identical to struct packet_mreq except it has
 154  * a longer address field.
 155  */
 156 struct packet_mreq_max {
 157         int             mr_ifindex;
 158         unsigned short  mr_type;
 159         unsigned short  mr_alen;
 160         unsigned char   mr_address[MAX_ADDR_LEN];
 161 };
 162
 163 union tpacket_uhdr {
 164         struct tpacket_hdr  *h1;
 165         struct tpacket2_hdr *h2;
 166         struct tpacket3_hdr *h3;
 167         void *raw;
 168 };
 169
 170 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 171                 int closing, int tx_ring);
 172
 173 #define V3_ALIGNMENT    (8)
 174
 175 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 176
 177 #define BLK_PLUS_PRIV(sz_of_priv) \
 178         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 179
 180 #define PGV_FROM_VMALLOC 1
 181
 182 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
 183 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
 184 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
 185 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
 186 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
 187 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
 188 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
 189
 190 struct packet_sock;
 191 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
 192 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 193                        struct packet_type *pt, struct net_device *orig_dev);
 194
 195 static void *packet_previous_frame(struct packet_sock *po,
 196                 struct packet_ring_buffer *rb,
 197                 int status);
 198 static void packet_increment_head(struct packet_ring_buffer *buff);
 199 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
 200                         struct tpacket_block_desc *);
 201 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 202                         struct packet_sock *);
 203 static void prb_retire_current_block(struct tpacket_kbdq_core *,
 204                 struct packet_sock *, unsigned int status);
 205 static int prb_queue_frozen(struct tpacket_kbdq_core *);
 206 static void prb_open_block(struct tpacket_kbdq_core *,
 207                 struct tpacket_block_desc *);
 208 static void prb_retire_rx_blk_timer_expired(unsigned long);
 209 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 210 static void prb_init_blk_timer(struct packet_sock *,
 211                 struct tpacket_kbdq_core *,
 212                 void (*func) (unsigned long));
 213 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 214 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 215                 struct tpacket3_hdr *);
 216 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 217                 struct tpacket3_hdr *);
 218 static void packet_flush_mclist(struct sock *sk);
 219
 220 struct packet_skb_cb {
 221         union {
 222                 struct sockaddr_pkt pkt;
 223                 union {
 224                         /* Trick: alias skb original length with
 225                          * ll.sll_family and ll.protocol in order
 226                          * to save room.
 227                          */
 228                         unsigned int origlen;
 229                         struct sockaddr_ll ll;
 230                 };
 231         } sa;
 232 };
 233
 234 #define vio_le() virtio_legacy_is_little_endian()
 235
 236 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 237
 238 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 239 #define GET_PBLOCK_DESC(x, bid) \
 240         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 241 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 242         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 243 #define GET_NEXT_PRB_BLK_NUM(x) \
 244         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 245         ((x)->kactive_blk_num+1) : 0)
 246
 247 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 248 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 249
 250 static int packet_direct_xmit(struct sk_buff *skb)
 251 {
 252         struct net_device *dev = skb->dev;
 253         struct sk_buff *orig_skb = skb;
 254         struct netdev_queue *txq;
 255         int ret = NETDEV_TX_BUSY;
 256
 257         if (unlikely(!netif_running(dev) ||
 258                      !netif_carrier_ok(dev)))
 259                 goto drop;
 260
 261         skb = validate_xmit_skb_list(skb, dev);
 262         if (skb != orig_skb)
 263                 goto drop;
 264
 265         txq = skb_get_tx_queue(dev, skb);
 266
 267         local_bh_disable();
 268
 269         HARD_TX_LOCK(dev, txq, smp_processor_id());
 270         if (!netif_xmit_frozen_or_drv_stopped(txq))
 271                 ret = netdev_start_xmit(skb, dev, txq, false);
 272         HARD_TX_UNLOCK(dev, txq);
 273
 274         local_bh_enable();
 275
 276         if (!dev_xmit_complete(ret))
 277                 kfree_skb(skb);
 278
 279         return ret;
 280 drop:
 281         atomic_long_inc(&dev->tx_dropped);
 282         kfree_skb_list(skb);
 283         return NET_XMIT_DROP;
 284 }
 285
 286 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
 287 {
 288         struct net_device *dev;
 289
 290         rcu_read_lock();
 291         dev = rcu_dereference(po->cached_dev);
 292         if (likely(dev))
 293                 dev_hold(dev);
 294         rcu_read_unlock();
 295
 296         return dev;
 297 }
 298
 299 static void packet_cached_dev_assign(struct packet_sock *po,
 300                                      struct net_device *dev)
 301 {
 302         rcu_assign_pointer(po->cached_dev, dev);
 303 }
 304
 305 static void packet_cached_dev_reset(struct packet_sock *po)
 306 {
 307         RCU_INIT_POINTER(po->cached_dev, NULL);
 308 }
 309
 310 static bool packet_use_direct_xmit(const struct packet_sock *po)
 311 {
 312         return po->xmit == packet_direct_xmit;
 313 }
 314
 315 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
 316 {
 317         return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
 318 }
 319
 320 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
 321 {
 322         const struct net_device_ops *ops = dev->netdev_ops;
 323         u16 queue_index;
 324
 325         if (ops->ndo_select_queue) {
 326                 queue_index = ops->ndo_select_queue(dev, skb, NULL,
 327                                                     __packet_pick_tx_queue);
 328                 queue_index = netdev_cap_txqueue(dev, queue_index);
 329         } else {
 330                 queue_index = __packet_pick_tx_queue(dev, skb);
 331         }
 332
 333         skb_set_queue_mapping(skb, queue_index);
 334 }
 335
 336 /* register_prot_hook must be invoked with the po->bind_lock held,
 337  * or from a context in which asynchronous accesses to the packet
 338  * socket is not possible (packet_create()).
 339  */
 340 static void register_prot_hook(struct sock *sk)
 341 {
 342         struct packet_sock *po = pkt_sk(sk);
 343
 344         if (!po->running) {
 345                 if (po->fanout)
 346                         __fanout_link(sk, po);
 347                 else
 348                         dev_add_pack(&po->prot_hook);
 349
 350                 sock_hold(sk);
 351                 po->running = 1;
 352         }
 353 }
 354
 355 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
 356  * held.   If the sync parameter is true, we will temporarily drop
 357  * the po->bind_lock and do a synchronize_net to make sure no
 358  * asynchronous packet processing paths still refer to the elements
 359  * of po->prot_hook.  If the sync parameter is false, it is the
 360  * callers responsibility to take care of this.
 361  */
 362 static void __unregister_prot_hook(struct sock *sk, bool sync)
 363 {
 364         struct packet_sock *po = pkt_sk(sk);
 365
 366         po->running = 0;
 367
 368         if (po->fanout)
 369                 __fanout_unlink(sk, po);
 370         else
 371                 __dev_remove_pack(&po->prot_hook);
 372
 373         __sock_put(sk);
 374
 375         if (sync) {
 376                 spin_unlock(&po->bind_lock);
 377                 synchronize_net();
 378                 spin_lock(&po->bind_lock);
 379         }
 380 }
 381
 382 static void unregister_prot_hook(struct sock *sk, bool sync)
 383 {
 384         struct packet_sock *po = pkt_sk(sk);
 385
 386         if (po->running)
 387                 __unregister_prot_hook(sk, sync);
 388 }
 389
 390 static inline struct page * __pure pgv_to_page(void *addr)
 391 {
 392         if (is_vmalloc_addr(addr))
 393                 return vmalloc_to_page(addr);
 394         return virt_to_page(addr);
 395 }
 396
 397 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 398 {
 399         union tpacket_uhdr h;
 400
 401         h.raw = frame;
 402         switch (po->tp_version) {
 403         case TPACKET_V1:
 404                 h.h1->tp_status = status;
 405                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 406                 break;
 407         case TPACKET_V2:
 408                 h.h2->tp_status = status;
 409                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 410                 break;
 411         case TPACKET_V3:
 412         default:
 413                 WARN(1, "TPACKET version not supported.\n");
 414                 BUG();
 415         }
 416
 417         smp_wmb();
 418 }
 419
 420 static int __packet_get_status(struct packet_sock *po, void *frame)
 421 {
 422         union tpacket_uhdr h;
 423
 424         smp_rmb();
 425
 426         h.raw = frame;
 427         switch (po->tp_version) {
 428         case TPACKET_V1:
 429                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 430                 return h.h1->tp_status;
 431         case TPACKET_V2:
 432                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 433                 return h.h2->tp_status;
 434         case TPACKET_V3:
 435         default:
 436                 WARN(1, "TPACKET version not supported.\n");
 437                 BUG();
 438                 return 0;
 439         }
 440 }
 441
 442 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
 443                                    unsigned int flags)
 444 {
 445         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 446
 447         if (shhwtstamps &&
 448             (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
 449             ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
 450                 return TP_STATUS_TS_RAW_HARDWARE;
 451
 452         if (ktime_to_timespec_cond(skb->tstamp, ts))
 453                 return TP_STATUS_TS_SOFTWARE;
 454
 455         return 0;
 456 }
 457
 458 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
 459                                     struct sk_buff *skb)
 460 {
 461         union tpacket_uhdr h;
 462         struct timespec ts;
 463         __u32 ts_status;
 464
 465         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
 466                 return 0;
 467
 468         h.raw = frame;
 469         switch (po->tp_version) {
 470         case TPACKET_V1:
 471                 h.h1->tp_sec = ts.tv_sec;
 472                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
 473                 break;
 474         case TPACKET_V2:
 475                 h.h2->tp_sec = ts.tv_sec;
 476                 h.h2->tp_nsec = ts.tv_nsec;
 477                 break;
 478         case TPACKET_V3:
 479         default:
 480                 WARN(1, "TPACKET version not supported.\n");
 481                 BUG();
 482         }
 483
 484         /* one flush is safe, as both fields always lie on the same cacheline */
 485         flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
 486         smp_wmb();
 487
 488         return ts_status;
 489 }
 490
 491 static void *packet_lookup_frame(struct packet_sock *po,
 492                 struct packet_ring_buffer *rb,
 493                 unsigned int position,
 494                 int status)
 495 {
 496         unsigned int pg_vec_pos, frame_offset;
 497         union tpacket_uhdr h;
 498
 499         pg_vec_pos = position / rb->frames_per_block;
 500         frame_offset = position % rb->frames_per_block;
 501
 502         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 503                 (frame_offset * rb->frame_size);
 504
 505         if (status != __packet_get_status(po, h.raw))
 506                 return NULL;
 507
 508         return h.raw;
 509 }
 510
 511 static void *packet_current_frame(struct packet_sock *po,
 512                 struct packet_ring_buffer *rb,
 513                 int status)
 514 {
 515         return packet_lookup_frame(po, rb, rb->head, status);
 516 }
 517
 518 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 519 {
 520         del_timer_sync(&pkc->retire_blk_timer);
 521 }
 522
 523 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 524                 struct sk_buff_head *rb_queue)
 525 {
 526         struct tpacket_kbdq_core *pkc;
 527
 528         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 529
 530         spin_lock_bh(&rb_queue->lock);
 531         pkc->delete_blk_timer = 1;
 532         spin_unlock_bh(&rb_queue->lock);
 533
 534         prb_del_retire_blk_timer(pkc);
 535 }
 536
 537 static void prb_init_blk_timer(struct packet_sock *po,
 538                 struct tpacket_kbdq_core *pkc,
 539                 void (*func) (unsigned long))
 540 {
 541         init_timer(&pkc->retire_blk_timer);
 542         pkc->retire_blk_timer.data = (long)po;
 543         pkc->retire_blk_timer.function = func;
 544         pkc->retire_blk_timer.expires = jiffies;
 545 }
 546
 547 static void prb_setup_retire_blk_timer(struct packet_sock *po)
 548 {
 549         struct tpacket_kbdq_core *pkc;
 550
 551         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 552         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
 553 }
 554
 555 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 556                                 int blk_size_in_bytes)
 557 {
 558         struct net_device *dev;
 559         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 560         struct ethtool_link_ksettings ecmd;
 561         int err;
 562
 563         rtnl_lock();
 564         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 565         if (unlikely(!dev)) {
 566                 rtnl_unlock();
 567                 return DEFAULT_PRB_RETIRE_TOV;
 568         }
 569         err = __ethtool_get_link_ksettings(dev, &ecmd);
 570         rtnl_unlock();
 571         if (!err) {
 572                 /*
 573                  * If the link speed is so slow you don't really
 574                  * need to worry about perf anyways
 575                  */
 576                 if (ecmd.base.speed < SPEED_1000 ||
 577                     ecmd.base.speed == SPEED_UNKNOWN) {
 578                         return DEFAULT_PRB_RETIRE_TOV;
 579                 } else {
 580                         msec = 1;
 581                         div = ecmd.base.speed / 1000;
 582                 }
 583         }
 584
 585         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 586
 587         if (div)
 588                 mbits /= div;
 589
 590         tmo = mbits * msec;
 591
 592         if (div)
 593                 return tmo+1;
 594         return tmo;
 595 }
 596
 597 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 598                         union tpacket_req_u *req_u)
 599 {
 600         p1->feature_req_word = req_u->req3.tp_feature_req_word;
 601 }
 602
 603 static void init_prb_bdqc(struct packet_sock *po,
 604                         struct packet_ring_buffer *rb,
 605                         struct pgv *pg_vec,
 606                         union tpacket_req_u *req_u)
 607 {
 608         struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
 609         struct tpacket_block_desc *pbd;
 610
 611         memset(p1, 0x0, sizeof(*p1));
 612
 613         p1->knxt_seq_num = 1;
 614         p1->pkbdq = pg_vec;
 615         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 616         p1->pkblk_start = pg_vec[0].buffer;
 617         p1->kblk_size = req_u->req3.tp_block_size;
 618         p1->knum_blocks = req_u->req3.tp_block_nr;
 619         p1->hdrlen = po->tp_hdrlen;
 620         p1->version = po->tp_version;
 621         p1->last_kactive_blk_num = 0;
 622         po->stats.stats3.tp_freeze_q_cnt = 0;
 623         if (req_u->req3.tp_retire_blk_tov)
 624                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 625         else
 626                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 627                                                 req_u->req3.tp_block_size);
 628         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 629         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 630
 631         p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
 632         prb_init_ft_ops(p1, req_u);
 633         prb_setup_retire_blk_timer(po);
 634         prb_open_block(p1, pbd);
 635 }
 636
 637 /*  Do NOT update the last_blk_num first.
 638  *  Assumes sk_buff_head lock is held.
 639  */
 640 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 641 {
 642         mod_timer(&pkc->retire_blk_timer,
 643                         jiffies + pkc->tov_in_jiffies);
 644         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 645 }
 646
 647 /*
 648  * Timer logic:
 649  * 1) We refresh the timer only when we open a block.
 650  *    By doing this we don't waste cycles refreshing the timer
 651  *        on packet-by-packet basis.
 652  *
 653  * With a 1MB block-size, on a 1Gbps line, it will take
 654  * i) ~8 ms to fill a block + ii) memcpy etc.
 655  * In this cut we are not accounting for the memcpy time.
 656  *
 657  * So, if the user sets the 'tmo' to 10ms then the timer
 658  * will never fire while the block is still getting filled
 659  * (which is what we want). However, the user could choose
 660  * to close a block early and that's fine.
 661  *
 662  * But when the timer does fire, we check whether or not to refresh it.
 663  * Since the tmo granularity is in msecs, it is not too expensive
 664  * to refresh the timer, lets say every '8' msecs.
 665  * Either the user can set the 'tmo' or we can derive it based on
 666  * a) line-speed and b) block-size.
 667  * prb_calc_retire_blk_tmo() calculates the tmo.
 668  *
 669  */
 670 static void prb_retire_rx_blk_timer_expired(unsigned long data)
 671 {
 672         struct packet_sock *po = (struct packet_sock *)data;
 673         struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 674         unsigned int frozen;
 675         struct tpacket_block_desc *pbd;
 676
 677         spin_lock(&po->sk.sk_receive_queue.lock);
 678
 679         frozen = prb_queue_frozen(pkc);
 680         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 681
 682         if (unlikely(pkc->delete_blk_timer))
 683                 goto out;
 684
 685         /* We only need to plug the race when the block is partially filled.
 686          * tpacket_rcv:
 687          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 688          *              copy_bits() is in progress ...
 689          *              timer fires on other cpu:
 690          *              we can't retire the current block because copy_bits
 691          *              is in progress.
 692          *
 693          */
 694         if (BLOCK_NUM_PKTS(pbd)) {
 695                 while (atomic_read(&pkc->blk_fill_in_prog)) {
 696                         /* Waiting for skb_copy_bits to finish... */
 697                         cpu_relax();
 698                 }
 699         }
 700
 701         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 702                 if (!frozen) {
 703                         if (!BLOCK_NUM_PKTS(pbd)) {
 704                                 /* An empty block. Just refresh the timer. */
 705                                 goto refresh_timer;
 706                         }
 707                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 708                         if (!prb_dispatch_next_block(pkc, po))
 709                                 goto refresh_timer;
 710                         else
 711                                 goto out;
 712                 } else {
 713                         /* Case 1. Queue was frozen because user-space was
 714                          *         lagging behind.
 715                          */
 716                         if (prb_curr_blk_in_use(pkc, pbd)) {
 717                                 /*
 718                                  * Ok, user-space is still behind.
 719                                  * So just refresh the timer.
 720                                  */
 721                                 goto refresh_timer;
 722                         } else {
 723                                /* Case 2. queue was frozen,user-space caught up,
 724                                 * now the link went idle && the timer fired.
 725                                 * We don't have a block to close.So we open this
 726                                 * block and restart the timer.
 727                                 * opening a block thaws the queue,restarts timer
 728                                 * Thawing/timer-refresh is a side effect.
 729                                 */
 730                                 prb_open_block(pkc, pbd);
 731                                 goto out;
 732                         }
 733                 }
 734         }
 735
 736 refresh_timer:
 737         _prb_refresh_rx_retire_blk_timer(pkc);
 738
 739 out:
 740         spin_unlock(&po->sk.sk_receive_queue.lock);
 741 }
 742
 743 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 744                 struct tpacket_block_desc *pbd1, __u32 status)
 745 {
 746         /* Flush everything minus the block header */
 747
 748 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 749         u8 *start, *end;
 750
 751         start = (u8 *)pbd1;
 752
 753         /* Skip the block header(we know header WILL fit in 4K) */
 754         start += PAGE_SIZE;
 755
 756         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 757         for (; start < end; start += PAGE_SIZE)
 758                 flush_dcache_page(pgv_to_page(start));
 759
 760         smp_wmb();
 761 #endif
 762
 763         /* Now update the block status. */
 764
 765         BLOCK_STATUS(pbd1) = status;
 766
 767         /* Flush the block header */
 768
 769 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 770         start = (u8 *)pbd1;
 771         flush_dcache_page(pgv_to_page(start));
 772
 773         smp_wmb();
 774 #endif
 775 }
 776
 777 /*
 778  * Side effect:
 779  *
 780  * 1) flush the block
 781  * 2) Increment active_blk_num
 782  *
 783  * Note:We DONT refresh the timer on purpose.
 784  *      Because almost always the next block will be opened.
 785  */
 786 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 787                 struct tpacket_block_desc *pbd1,
 788                 struct packet_sock *po, unsigned int stat)
 789 {
 790         __u32 status = TP_STATUS_USER | stat;
 791
 792         struct tpacket3_hdr *last_pkt;
 793         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 794         struct sock *sk = &po->sk;
 795
 796         if (po->stats.stats3.tp_drops)
 797                 status |= TP_STATUS_LOSING;
 798
 799         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 800         last_pkt->tp_next_offset = 0;
 801
 802         /* Get the ts of the last pkt */
 803         if (BLOCK_NUM_PKTS(pbd1)) {
 804                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 805                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 806         } else {
 807                 /* Ok, we tmo'd - so get the current time.
 808                  *
 809                  * It shouldn't really happen as we don't close empty
 810                  * blocks. See prb_retire_rx_blk_timer_expired().
 811                  */
 812                 struct timespec ts;
 813                 getnstimeofday(&ts);
 814                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 815                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 816         }
 817
 818         smp_wmb();
 819
 820         /* Flush the block */
 821         prb_flush_block(pkc1, pbd1, status);
 822
 823         sk->sk_data_ready(sk);
 824
 825         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 826 }
 827
 828 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 829 {
 830         pkc->reset_pending_on_curr_blk = 0;
 831 }
 832
 833 /*
 834  * Side effect of opening a block:
 835  *
 836  * 1) prb_queue is thawed.
 837  * 2) retire_blk_timer is refreshed.
 838  *
 839  */
 840 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 841         struct tpacket_block_desc *pbd1)
 842 {
 843         struct timespec ts;
 844         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 845
 846         smp_rmb();
 847
 848         /* We could have just memset this but we will lose the
 849          * flexibility of making the priv area sticky
 850          */
 851
 852         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 853         BLOCK_NUM_PKTS(pbd1) = 0;
 854         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 855
 856         getnstimeofday(&ts);
 857
 858         h1->ts_first_pkt.ts_sec = ts.tv_sec;
 859         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 860
 861         pkc1->pkblk_start = (char *)pbd1;
 862         pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 863
 864         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 865         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 866
 867         pbd1->version = pkc1->version;
 868         pkc1->prev = pkc1->nxt_offset;
 869         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 870
 871         prb_thaw_queue(pkc1);
 872         _prb_refresh_rx_retire_blk_timer(pkc1);
 873
 874         smp_wmb();
 875 }
 876
 877 /*
 878  * Queue freeze logic:
 879  * 1) Assume tp_block_nr = 8 blocks.
 880  * 2) At time 't0', user opens Rx ring.
 881  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 882  * 4) user-space is either sleeping or processing block '0'.
 883  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 884  *    it will close block-7,loop around and try to fill block '0'.
 885  *    call-flow:
 886  *    __packet_lookup_frame_in_block
 887  *      prb_retire_current_block()
 888  *      prb_dispatch_next_block()
 889  *        |->(BLOCK_STATUS == USER) evaluates to true
 890  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 891  * 6) Now there are two cases:
 892  *    6.1) Link goes idle right after the queue is frozen.
 893  *         But remember, the last open_block() refreshed the timer.
 894  *         When this timer expires,it will refresh itself so that we can
 895  *         re-open block-0 in near future.
 896  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 897  *         case and __packet_lookup_frame_in_block will check if block-0
 898  *         is free and can now be re-used.
 899  */
 900 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 901                                   struct packet_sock *po)
 902 {
 903         pkc->reset_pending_on_curr_blk = 1;
 904         po->stats.stats3.tp_freeze_q_cnt++;
 905 }
 906
 907 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 908
 909 /*
 910  * If the next block is free then we will dispatch it
 911  * and return a good offset.
 912  * Else, we will freeze the queue.
 913  * So, caller must check the return value.
 914  */
 915 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 916                 struct packet_sock *po)
 917 {
 918         struct tpacket_block_desc *pbd;
 919
 920         smp_rmb();
 921
 922         /* 1. Get current block num */
 923         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 924
 925         /* 2. If this block is currently in_use then freeze the queue */
 926         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 927                 prb_freeze_queue(pkc, po);
 928                 return NULL;
 929         }
 930
 931         /*
 932          * 3.
 933          * open this block and return the offset where the first packet
 934          * needs to get stored.
 935          */
 936         prb_open_block(pkc, pbd);
 937         return (void *)pkc->nxt_offset;
 938 }
 939
 940 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 941                 struct packet_sock *po, unsigned int status)
 942 {
 943         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 944
 945         /* retire/close the current block */
 946         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 947                 /*
 948                  * Plug the case where copy_bits() is in progress on
 949                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 950                  * have space to copy the pkt in the current block and
 951                  * called prb_retire_current_block()
 952                  *
 953                  * We don't need to worry about the TMO case because
 954                  * the timer-handler already handled this case.
 955                  */
 956                 if (!(status & TP_STATUS_BLK_TMO)) {
 957                         while (atomic_read(&pkc->blk_fill_in_prog)) {
 958                                 /* Waiting for skb_copy_bits to finish... */
 959                                 cpu_relax();
 960                         }
 961                 }
 962                 prb_close_block(pkc, pbd, po, status);
 963                 return;
 964         }
 965 }
 966
 967 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
 968                                       struct tpacket_block_desc *pbd)
 969 {
 970         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 971 }
 972
 973 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 974 {
 975         return pkc->reset_pending_on_curr_blk;
 976 }
 977
 978 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 979 {
 980         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 981         atomic_dec(&pkc->blk_fill_in_prog);
 982 }
 983
 984 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 985                         struct tpacket3_hdr *ppd)
 986 {
 987         ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
 988 }
 989
 990 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 991                         struct tpacket3_hdr *ppd)
 992 {
 993         ppd->hv1.tp_rxhash = 0;
 994 }
 995
 996 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
 997                         struct tpacket3_hdr *ppd)
 998 {
 999         if (skb_vlan_tag_present(pkc->skb)) {
1000                 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1001                 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1002                 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1003         } else {
1004                 ppd->hv1.tp_vlan_tci = 0;
1005                 ppd->hv1.tp_vlan_tpid = 0;
1006                 ppd->tp_status = TP_STATUS_AVAILABLE;
1007         }
1008 }
1009
1010 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1011                         struct tpacket3_hdr *ppd)
1012 {
1013         ppd->hv1.tp_padding = 0;
1014         prb_fill_vlan_info(pkc, ppd);
1015
1016         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1017                 prb_fill_rxhash(pkc, ppd);
1018         else
1019                 prb_clear_rxhash(pkc, ppd);
1020 }
1021
1022 static void prb_fill_curr_block(char *curr,
1023                                 struct tpacket_kbdq_core *pkc,
1024                                 struct tpacket_block_desc *pbd,
1025                                 unsigned int len)
1026 {
1027         struct tpacket3_hdr *ppd;
1028
1029         ppd  = (struct tpacket3_hdr *)curr;
1030         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1031         pkc->prev = curr;
1032         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1033         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1034         BLOCK_NUM_PKTS(pbd) += 1;
1035         atomic_inc(&pkc->blk_fill_in_prog);
1036         prb_run_all_ft_ops(pkc, ppd);
1037 }
1038
1039 /* Assumes caller has the sk->rx_queue.lock */
1040 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1041                                             struct sk_buff *skb,
1042                                                 int status,
1043                                             unsigned int len
1044                                             )
1045 {
1046         struct tpacket_kbdq_core *pkc;
1047         struct tpacket_block_desc *pbd;
1048         char *curr, *end;
1049
1050         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1051         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1052
1053         /* Queue is frozen when user space is lagging behind */
1054         if (prb_queue_frozen(pkc)) {
1055                 /*
1056                  * Check if that last block which caused the queue to freeze,
1057                  * is still in_use by user-space.
1058                  */
1059                 if (prb_curr_blk_in_use(pkc, pbd)) {
1060                         /* Can't record this packet */
1061                         return NULL;
1062                 } else {
1063                         /*
1064                          * Ok, the block was released by user-space.
1065                          * Now let's open that block.
1066                          * opening a block also thaws the queue.
1067                          * Thawing is a side effect.
1068                          */
1069                         prb_open_block(pkc, pbd);
1070                 }
1071         }
1072
1073         smp_mb();
1074         curr = pkc->nxt_offset;
1075         pkc->skb = skb;
1076         end = (char *)pbd + pkc->kblk_size;
1077
1078         /* first try the current block */
1079         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1080                 prb_fill_curr_block(curr, pkc, pbd, len);
1081                 return (void *)curr;
1082         }
1083
1084         /* Ok, close the current block */
1085         prb_retire_current_block(pkc, po, 0);
1086
1087         /* Now, try to dispatch the next block */
1088         curr = (char *)prb_dispatch_next_block(pkc, po);
1089         if (curr) {
1090                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1091                 prb_fill_curr_block(curr, pkc, pbd, len);
1092                 return (void *)curr;
1093         }
1094
1095         /*
1096          * No free blocks are available.user_space hasn't caught up yet.
1097          * Queue was just frozen and now this packet will get dropped.
1098          */
1099         return NULL;
1100 }
1101
1102 static void *packet_current_rx_frame(struct packet_sock *po,
1103                                             struct sk_buff *skb,
1104                                             int status, unsigned int len)
1105 {
1106         char *curr = NULL;
1107         switch (po->tp_version) {
1108         case TPACKET_V1:
1109         case TPACKET_V2:
1110                 curr = packet_lookup_frame(po, &po->rx_ring,
1111                                         po->rx_ring.head, status);
1112                 return curr;
1113         case TPACKET_V3:
1114                 return __packet_lookup_frame_in_block(po, skb, status, len);
1115         default:
1116                 WARN(1, "TPACKET version not supported\n");
1117                 BUG();
1118                 return NULL;
1119         }
1120 }
1121
1122 static void *prb_lookup_block(struct packet_sock *po,
1123                                      struct packet_ring_buffer *rb,
1124                                      unsigned int idx,
1125                                      int status)
1126 {
1127         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1128         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1129
1130         if (status != BLOCK_STATUS(pbd))
1131                 return NULL;
1132         return pbd;
1133 }
1134
1135 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1136 {
1137         unsigned int prev;
1138         if (rb->prb_bdqc.kactive_blk_num)
1139                 prev = rb->prb_bdqc.kactive_blk_num-1;
1140         else
1141                 prev = rb->prb_bdqc.knum_blocks-1;
1142         return prev;
1143 }
1144
1145 /* Assumes caller has held the rx_queue.lock */
1146 static void *__prb_previous_block(struct packet_sock *po,
1147                                          struct packet_ring_buffer *rb,
1148                                          int status)
1149 {
1150         unsigned int previous = prb_previous_blk_num(rb);
1151         return prb_lookup_block(po, rb, previous, status);
1152 }
1153
1154 static void *packet_previous_rx_frame(struct packet_sock *po,
1155                                              struct packet_ring_buffer *rb,
1156                                              int status)
1157 {
1158         if (po->tp_version <= TPACKET_V2)
1159                 return packet_previous_frame(po, rb, status);
1160
1161         return __prb_previous_block(po, rb, status);
1162 }
1163
1164 static void packet_increment_rx_head(struct packet_sock *po,
1165                                             struct packet_ring_buffer *rb)
1166 {
1167         switch (po->tp_version) {
1168         case TPACKET_V1:
1169         case TPACKET_V2:
1170                 return packet_increment_head(rb);
1171         case TPACKET_V3:
1172         default:
1173                 WARN(1, "TPACKET version not supported.\n");
1174                 BUG();
1175                 return;
1176         }
1177 }
1178
1179 static void *packet_previous_frame(struct packet_sock *po,
1180                 struct packet_ring_buffer *rb,
1181                 int status)
1182 {
1183         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1184         return packet_lookup_frame(po, rb, previous, status);
1185 }
1186
1187 static void packet_increment_head(struct packet_ring_buffer *buff)
1188 {
1189         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1190 }
1191
1192 static void packet_inc_pending(struct packet_ring_buffer *rb)
1193 {
1194         this_cpu_inc(*rb->pending_refcnt);
1195 }
1196
1197 static void packet_dec_pending(struct packet_ring_buffer *rb)
1198 {
1199         this_cpu_dec(*rb->pending_refcnt);
1200 }
1201
1202 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1203 {
1204         unsigned int refcnt = 0;
1205         int cpu;
1206
1207         /* We don't use pending refcount in rx_ring. */
1208         if (rb->pending_refcnt == NULL)
1209                 return 0;
1210
1211         for_each_possible_cpu(cpu)
1212                 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1213
1214         return refcnt;
1215 }
1216
1217 static int packet_alloc_pending(struct packet_sock *po)
1218 {
1219         po->rx_ring.pending_refcnt = NULL;
1220
1221         po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1222         if (unlikely(po->tx_ring.pending_refcnt == NULL))
1223                 return -ENOBUFS;
1224
1225         return 0;
1226 }
1227
1228 static void packet_free_pending(struct packet_sock *po)
1229 {
1230         free_percpu(po->tx_ring.pending_refcnt);
1231 }
1232
1233 #define ROOM_POW_OFF    2
1234 #define ROOM_NONE       0x0
1235 #define ROOM_LOW        0x1
1236 #define ROOM_NORMAL     0x2
1237
1238 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1239 {
1240         int idx, len;
1241
1242         len = po->rx_ring.frame_max + 1;
1243         idx = po->rx_ring.head;
1244         if (pow_off)
1245                 idx += len >> pow_off;
1246         if (idx >= len)
1247                 idx -= len;
1248         return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1249 }
1250
1251 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1252 {
1253         int idx, len;
1254
1255         len = po->rx_ring.prb_bdqc.knum_blocks;
1256         idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1257         if (pow_off)
1258                 idx += len >> pow_off;
1259         if (idx >= len)
1260                 idx -= len;
1261         return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1262 }
1263
1264 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1265 {
1266         struct sock *sk = &po->sk;
1267         int ret = ROOM_NONE;
1268
1269         if (po->prot_hook.func != tpacket_rcv) {
1270                 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1271                                           - (skb ? skb->truesize : 0);
1272                 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1273                         return ROOM_NORMAL;
1274                 else if (avail > 0)
1275                         return ROOM_LOW;
1276                 else
1277                         return ROOM_NONE;
1278         }
1279
1280         if (po->tp_version == TPACKET_V3) {
1281                 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1282                         ret = ROOM_NORMAL;
1283                 else if (__tpacket_v3_has_room(po, 0))
1284                         ret = ROOM_LOW;
1285         } else {
1286                 if (__tpacket_has_room(po, ROOM_POW_OFF))
1287                         ret = ROOM_NORMAL;
1288                 else if (__tpacket_has_room(po, 0))
1289                         ret = ROOM_LOW;
1290         }
1291
1292         return ret;
1293 }
1294
1295 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1296 {
1297         int ret;
1298         bool has_room;
1299
1300         spin_lock_bh(&po->sk.sk_receive_queue.lock);
1301         ret = __packet_rcv_has_room(po, skb);
1302         has_room = ret == ROOM_NORMAL;
1303         if (po->pressure == has_room)
1304                 po->pressure = !has_room;
1305         spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1306
1307         return ret;
1308 }
1309
1310 static void packet_sock_destruct(struct sock *sk)
1311 {
1312         skb_queue_purge(&sk->sk_error_queue);
1313
1314         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1315         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1316
1317         if (!sock_flag(sk, SOCK_DEAD)) {
1318                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1319                 return;
1320         }
1321
1322         sk_refcnt_debug_dec(sk);
1323 }
1324
1325 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1326 {
1327         u32 rxhash;
1328         int i, count = 0;
1329
1330         rxhash = skb_get_hash(skb);
1331         for (i = 0; i < ROLLOVER_HLEN; i++)
1332                 if (po->rollover->history[i] == rxhash)
1333                         count++;
1334
1335         po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1336         return count > (ROLLOVER_HLEN >> 1);
1337 }
1338
1339 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1340                                       struct sk_buff *skb,
1341                                       unsigned int num)
1342 {
1343         return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1344 }
1345
1346 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1347                                     struct sk_buff *skb,
1348                                     unsigned int num)
1349 {
1350         unsigned int val = atomic_inc_return(&f->rr_cur);
1351
1352         return val % num;
1353 }
1354
1355 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1356                                      struct sk_buff *skb,
1357                                      unsigned int num)
1358 {
1359         return smp_processor_id() % num;
1360 }
1361
1362 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1363                                      struct sk_buff *skb,
1364                                      unsigned int num)
1365 {
1366         return prandom_u32_max(num);
1367 }
1368
1369 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1370                                           struct sk_buff *skb,
1371                                           unsigned int idx, bool try_self,
1372                                           unsigned int num)
1373 {
1374         struct packet_sock *po, *po_next, *po_skip = NULL;
1375         unsigned int i, j, room = ROOM_NONE;
1376
1377         po = pkt_sk(f->arr[idx]);
1378
1379         if (try_self) {
1380                 room = packet_rcv_has_room(po, skb);
1381                 if (room == ROOM_NORMAL ||
1382                     (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1383                         return idx;
1384                 po_skip = po;
1385         }
1386
1387         i = j = min_t(int, po->rollover->sock, num - 1);
1388         do {
1389                 po_next = pkt_sk(f->arr[i]);
1390                 if (po_next != po_skip && !po_next->pressure &&
1391                     packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1392                         if (i != j)
1393                                 po->rollover->sock = i;
1394                         atomic_long_inc(&po->rollover->num);
1395                         if (room == ROOM_LOW)
1396                                 atomic_long_inc(&po->rollover->num_huge);
1397                         return i;
1398                 }
1399
1400                 if (++i == num)
1401                         i = 0;
1402         } while (i != j);
1403
1404         atomic_long_inc(&po->rollover->num_failed);
1405         return idx;
1406 }
1407
1408 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1409                                     struct sk_buff *skb,
1410                                     unsigned int num)
1411 {
1412         return skb_get_queue_mapping(skb) % num;
1413 }
1414
1415 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1416                                      struct sk_buff *skb,
1417                                      unsigned int num)
1418 {
1419         struct bpf_prog *prog;
1420         unsigned int ret = 0;
1421
1422         rcu_read_lock();
1423         prog = rcu_dereference(f->bpf_prog);
1424         if (prog)
1425                 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1426         rcu_read_unlock();
1427
1428         return ret;
1429 }
1430
1431 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1432 {
1433         return f->flags & (flag >> 8);
1434 }
1435
1436 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1437                              struct packet_type *pt, struct net_device *orig_dev)
1438 {
1439         struct packet_fanout *f = pt->af_packet_priv;
1440         unsigned int num = READ_ONCE(f->num_members);
1441         struct net *net = read_pnet(&f->net);
1442         struct packet_sock *po;
1443         unsigned int idx;
1444
1445         if (!net_eq(dev_net(dev), net) || !num) {
1446                 kfree_skb(skb);
1447                 return 0;
1448         }
1449
1450         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1451                 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1452                 if (!skb)
1453                         return 0;
1454         }
1455         switch (f->type) {
1456         case PACKET_FANOUT_HASH:
1457         default:
1458                 idx = fanout_demux_hash(f, skb, num);
1459                 break;
1460         case PACKET_FANOUT_LB:
1461                 idx = fanout_demux_lb(f, skb, num);
1462                 break;
1463         case PACKET_FANOUT_CPU:
1464                 idx = fanout_demux_cpu(f, skb, num);
1465                 break;
1466         case PACKET_FANOUT_RND:
1467                 idx = fanout_demux_rnd(f, skb, num);
1468                 break;
1469         case PACKET_FANOUT_QM:
1470                 idx = fanout_demux_qm(f, skb, num);
1471                 break;
1472         case PACKET_FANOUT_ROLLOVER:
1473                 idx = fanout_demux_rollover(f, skb, 0, false, num);
1474                 break;
1475         case PACKET_FANOUT_CBPF:
1476         case PACKET_FANOUT_EBPF:
1477                 idx = fanout_demux_bpf(f, skb, num);
1478                 break;
1479         }
1480
1481         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1482                 idx = fanout_demux_rollover(f, skb, idx, true, num);
1483
1484         po = pkt_sk(f->arr[idx]);
1485         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1486 }
1487
1488 DEFINE_MUTEX(fanout_mutex);
1489 EXPORT_SYMBOL_GPL(fanout_mutex);
1490 static LIST_HEAD(fanout_list);
1491
1492 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1493 {
1494         struct packet_fanout *f = po->fanout;
1495
1496         spin_lock(&f->lock);
1497         f->arr[f->num_members] = sk;
1498         smp_wmb();
1499         f->num_members++;
1500         spin_unlock(&f->lock);
1501 }
1502
1503 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1504 {
1505         struct packet_fanout *f = po->fanout;
1506         int i;
1507
1508         spin_lock(&f->lock);
1509         for (i = 0; i < f->num_members; i++) {
1510                 if (f->arr[i] == sk)
1511                         break;
1512         }
1513         BUG_ON(i >= f->num_members);
1514         f->arr[i] = f->arr[f->num_members - 1];
1515         f->num_members--;
1516         spin_unlock(&f->lock);
1517 }
1518
1519 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1520 {
1521         if (sk->sk_family != PF_PACKET)
1522                 return false;
1523
1524         return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1525 }
1526
1527 static void fanout_init_data(struct packet_fanout *f)
1528 {
1529         switch (f->type) {
1530         case PACKET_FANOUT_LB:
1531                 atomic_set(&f->rr_cur, 0);
1532                 break;
1533         case PACKET_FANOUT_CBPF:
1534         case PACKET_FANOUT_EBPF:
1535                 RCU_INIT_POINTER(f->bpf_prog, NULL);
1536                 break;
1537         }
1538 }
1539
1540 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1541 {
1542         struct bpf_prog *old;
1543
1544         spin_lock(&f->lock);
1545         old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1546         rcu_assign_pointer(f->bpf_prog, new);
1547         spin_unlock(&f->lock);
1548
1549         if (old) {
1550                 synchronize_net();
1551                 bpf_prog_destroy(old);
1552         }
1553 }
1554
1555 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1556                                 unsigned int len)
1557 {
1558         struct bpf_prog *new;
1559         struct sock_fprog fprog;
1560         int ret;
1561
1562         if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1563                 return -EPERM;
1564         if (len != sizeof(fprog))
1565                 return -EINVAL;
1566         if (copy_from_user(&fprog, data, len))
1567                 return -EFAULT;
1568
1569         ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1570         if (ret)
1571                 return ret;
1572
1573         __fanout_set_data_bpf(po->fanout, new);
1574         return 0;
1575 }
1576
1577 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1578                                 unsigned int len)
1579 {
1580         struct bpf_prog *new;
1581         u32 fd;
1582
1583         if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1584                 return -EPERM;
1585         if (len != sizeof(fd))
1586                 return -EINVAL;
1587         if (copy_from_user(&fd, data, len))
1588                 return -EFAULT;
1589
1590         new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1591         if (IS_ERR(new))
1592                 return PTR_ERR(new);
1593
1594         __fanout_set_data_bpf(po->fanout, new);
1595         return 0;
1596 }
1597
1598 static int fanout_set_data(struct packet_sock *po, char __user *data,
1599                            unsigned int len)
1600 {
1601         switch (po->fanout->type) {
1602         case PACKET_FANOUT_CBPF:
1603                 return fanout_set_data_cbpf(po, data, len);
1604         case PACKET_FANOUT_EBPF:
1605                 return fanout_set_data_ebpf(po, data, len);
1606         default:
1607                 return -EINVAL;
1608         };
1609 }
1610
1611 static void fanout_release_data(struct packet_fanout *f)
1612 {
1613         switch (f->type) {
1614         case PACKET_FANOUT_CBPF:
1615         case PACKET_FANOUT_EBPF:
1616                 __fanout_set_data_bpf(f, NULL);
1617         };
1618 }
1619
1620 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1621 {
1622         struct packet_sock *po = pkt_sk(sk);
1623         struct packet_fanout *f, *match;
1624         u8 type = type_flags & 0xff;
1625         u8 flags = type_flags >> 8;
1626         int err;
1627
1628         switch (type) {
1629         case PACKET_FANOUT_ROLLOVER:
1630                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1631                         return -EINVAL;
1632         case PACKET_FANOUT_HASH:
1633         case PACKET_FANOUT_LB:
1634         case PACKET_FANOUT_CPU:
1635         case PACKET_FANOUT_RND:
1636         case PACKET_FANOUT_QM:
1637         case PACKET_FANOUT_CBPF:
1638         case PACKET_FANOUT_EBPF:
1639                 break;
1640         default:
1641                 return -EINVAL;
1642         }
1643
1644         if (!po->running)
1645                 return -EINVAL;
1646
1647         if (po->fanout)
1648                 return -EALREADY;
1649
1650         if (type == PACKET_FANOUT_ROLLOVER ||
1651             (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1652                 po->rollover = kzalloc(sizeof(*po->rollover), GFP_KERNEL);
1653                 if (!po->rollover)
1654                         return -ENOMEM;
1655                 atomic_long_set(&po->rollover->num, 0);
1656                 atomic_long_set(&po->rollover->num_huge, 0);
1657                 atomic_long_set(&po->rollover->num_failed, 0);
1658         }
1659
1660         mutex_lock(&fanout_mutex);
1661         match = NULL;
1662         list_for_each_entry(f, &fanout_list, list) {
1663                 if (f->id == id &&
1664                     read_pnet(&f->net) == sock_net(sk)) {
1665                         match = f;
1666                         break;
1667                 }
1668         }
1669         err = -EINVAL;
1670         if (match && match->flags != flags)
1671                 goto out;
1672         if (!match) {
1673                 err = -ENOMEM;
1674                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1675                 if (!match)
1676                         goto out;
1677                 write_pnet(&match->net, sock_net(sk));
1678                 match->id = id;
1679                 match->type = type;
1680                 match->flags = flags;
1681                 INIT_LIST_HEAD(&match->list);
1682                 spin_lock_init(&match->lock);
1683                 atomic_set(&match->sk_ref, 0);
1684                 fanout_init_data(match);
1685                 match->prot_hook.type = po->prot_hook.type;
1686                 match->prot_hook.dev = po->prot_hook.dev;
1687                 match->prot_hook.func = packet_rcv_fanout;
1688                 match->prot_hook.af_packet_priv = match;
1689                 match->prot_hook.id_match = match_fanout_group;
1690                 dev_add_pack(&match->prot_hook);
1691                 list_add(&match->list, &fanout_list);
1692         }
1693         err = -EINVAL;
1694         if (match->type == type &&
1695             match->prot_hook.type == po->prot_hook.type &&
1696             match->prot_hook.dev == po->prot_hook.dev) {
1697                 err = -ENOSPC;
1698                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1699                         __dev_remove_pack(&po->prot_hook);
1700                         po->fanout = match;
1701                         atomic_inc(&match->sk_ref);
1702                         __fanout_link(sk, po);
1703                         err = 0;
1704                 }
1705         }
1706 out:
1707         mutex_unlock(&fanout_mutex);
1708         if (err) {
1709                 kfree(po->rollover);
1710                 po->rollover = NULL;
1711         }
1712         return err;
1713 }
1714
1715 static void fanout_release(struct sock *sk)
1716 {
1717         struct packet_sock *po = pkt_sk(sk);
1718         struct packet_fanout *f;
1719
1720         f = po->fanout;
1721         if (!f)
1722                 return;
1723
1724         mutex_lock(&fanout_mutex);
1725         po->fanout = NULL;
1726
1727         if (atomic_dec_and_test(&f->sk_ref)) {
1728                 list_del(&f->list);
1729                 dev_remove_pack(&f->prot_hook);
1730                 fanout_release_data(f);
1731                 kfree(f);
1732         }
1733         mutex_unlock(&fanout_mutex);
1734
1735         if (po->rollover)
1736                 kfree_rcu(po->rollover, rcu);
1737 }
1738
1739 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1740                                           struct sk_buff *skb)
1741 {
1742         /* Earlier code assumed this would be a VLAN pkt, double-check
1743          * this now that we have the actual packet in hand. We can only
1744          * do this check on Ethernet devices.
1745          */
1746         if (unlikely(dev->type != ARPHRD_ETHER))
1747                 return false;
1748
1749         skb_reset_mac_header(skb);
1750         return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1751 }
1752
1753 static const struct proto_ops packet_ops;
1754
1755 static const struct proto_ops packet_ops_spkt;
1756
1757 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1758                            struct packet_type *pt, struct net_device *orig_dev)
1759 {
1760         struct sock *sk;
1761         struct sockaddr_pkt *spkt;
1762
1763         /*
1764          *      When we registered the protocol we saved the socket in the data
1765          *      field for just this event.
1766          */
1767
1768         sk = pt->af_packet_priv;
1769
1770         /*
1771          *      Yank back the headers [hope the device set this
1772          *      right or kerboom...]
1773          *
1774          *      Incoming packets have ll header pulled,
1775          *      push it back.
1776          *
1777          *      For outgoing ones skb->data == skb_mac_header(skb)
1778          *      so that this procedure is noop.
1779          */
1780
1781         if (skb->pkt_type == PACKET_LOOPBACK)
1782                 goto out;
1783
1784         if (!net_eq(dev_net(dev), sock_net(sk)))
1785                 goto out;
1786
1787         skb = skb_share_check(skb, GFP_ATOMIC);
1788         if (skb == NULL)
1789                 goto oom;
1790
1791         /* drop any routing info */
1792         skb_dst_drop(skb);
1793
1794         /* drop conntrack reference */
1795         nf_reset(skb);
1796
1797         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1798
1799         skb_push(skb, skb->data - skb_mac_header(skb));
1800
1801         /*
1802          *      The SOCK_PACKET socket receives _all_ frames.
1803          */
1804
1805         spkt->spkt_family = dev->type;
1806         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1807         spkt->spkt_protocol = skb->protocol;
1808
1809         /*
1810          *      Charge the memory to the socket. This is done specifically
1811          *      to prevent sockets using all the memory up.
1812          */
1813
1814         if (sock_queue_rcv_skb(sk, skb) == 0)
1815                 return 0;
1816
1817 out:
1818         kfree_skb(skb);
1819 oom:
1820         return 0;
1821 }
1822
1823
1824 /*
1825  *      Output a raw packet to a device layer. This bypasses all the other
1826  *      protocol layers and you must therefore supply it with a complete frame
1827  */
1828
1829 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1830                                size_t len)
1831 {
1832         struct sock *sk = sock->sk;
1833         DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1834         struct sk_buff *skb = NULL;
1835         struct net_device *dev;
1836         struct sockcm_cookie sockc;
1837         __be16 proto = 0;
1838         int err;
1839         int extra_len = 0;
1840
1841         /*
1842          *      Get and verify the address.
1843          */
1844
1845         if (saddr) {
1846                 if (msg->msg_namelen < sizeof(struct sockaddr))
1847                         return -EINVAL;
1848                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1849                         proto = saddr->spkt_protocol;
1850         } else
1851                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1852
1853         /*
1854          *      Find the device first to size check it
1855          */
1856
1857         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1858 retry:
1859         rcu_read_lock();
1860         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1861         err = -ENODEV;
1862         if (dev == NULL)
1863                 goto out_unlock;
1864
1865         err = -ENETDOWN;
1866         if (!(dev->flags & IFF_UP))
1867                 goto out_unlock;
1868
1869         /*
1870          * You may not queue a frame bigger than the mtu. This is the lowest level
1871          * raw protocol and you must do your own fragmentation at this level.
1872          */
1873
1874         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1875                 if (!netif_supports_nofcs(dev)) {
1876                         err = -EPROTONOSUPPORT;
1877                         goto out_unlock;
1878                 }
1879                 extra_len = 4; /* We're doing our own CRC */
1880         }
1881
1882         err = -EMSGSIZE;
1883         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1884                 goto out_unlock;
1885
1886         if (!skb) {
1887                 size_t reserved = LL_RESERVED_SPACE(dev);
1888                 int tlen = dev->needed_tailroom;
1889                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1890
1891                 rcu_read_unlock();
1892                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1893                 if (skb == NULL)
1894                         return -ENOBUFS;
1895                 /* FIXME: Save some space for broken drivers that write a hard
1896                  * header at transmission time by themselves. PPP is the notable
1897                  * one here. This should really be fixed at the driver level.
1898                  */
1899                 skb_reserve(skb, reserved);
1900                 skb_reset_network_header(skb);
1901
1902                 /* Try to align data part correctly */
1903                 if (hhlen) {
1904                         skb->data -= hhlen;
1905                         skb->tail -= hhlen;
1906                         if (len < hhlen)
1907                                 skb_reset_network_header(skb);
1908                 }
1909                 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1910                 if (err)
1911                         goto out_free;
1912                 goto retry;
1913         }
1914
1915         if (!dev_validate_header(dev, skb->data, len)) {
1916                 err = -EINVAL;
1917                 goto out_unlock;
1918         }
1919         if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1920             !packet_extra_vlan_len_allowed(dev, skb)) {
1921                 err = -EMSGSIZE;
1922                 goto out_unlock;
1923         }
1924
1925         sockc.tsflags = sk->sk_tsflags;
1926         if (msg->msg_controllen) {
1927                 err = sock_cmsg_send(sk, msg, &sockc);
1928                 if (unlikely(err))
1929                         goto out_unlock;
1930         }
1931
1932         skb->protocol = proto;
1933         skb->dev = dev;
1934         skb->priority = sk->sk_priority;
1935         skb->mark = sk->sk_mark;
1936
1937         sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
1938
1939         if (unlikely(extra_len == 4))
1940                 skb->no_fcs = 1;
1941
1942         skb_probe_transport_header(skb, 0);
1943
1944         dev_queue_xmit(skb);
1945         rcu_read_unlock();
1946         return len;
1947
1948 out_unlock:
1949         rcu_read_unlock();
1950 out_free:
1951         kfree_skb(skb);
1952         return err;
1953 }
1954
1955 static unsigned int run_filter(struct sk_buff *skb,
1956                                const struct sock *sk,
1957                                unsigned int res)
1958 {
1959         struct sk_filter *filter;
1960
1961         rcu_read_lock();
1962         filter = rcu_dereference(sk->sk_filter);
1963         if (filter != NULL)
1964                 res = bpf_prog_run_clear_cb(filter->prog, skb);
1965         rcu_read_unlock();
1966
1967         return res;
1968 }
1969
1970 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
1971                            size_t *len)
1972 {
1973         struct virtio_net_hdr vnet_hdr;
1974
1975         if (*len < sizeof(vnet_hdr))
1976                 return -EINVAL;
1977         *len -= sizeof(vnet_hdr);
1978
1979         if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le()))
1980                 return -EINVAL;
1981
1982         return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
1983 }
1984
1985 /*
1986  * This function makes lazy skb cloning in hope that most of packets
1987  * are discarded by BPF.
1988  *
1989  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1990  * and skb->cb are mangled. It works because (and until) packets
1991  * falling here are owned by current CPU. Output packets are cloned
1992  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1993  * sequencially, so that if we return skb to original state on exit,
1994  * we will not harm anyone.
1995  */
1996
1997 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1998                       struct packet_type *pt, struct net_device *orig_dev)
1999 {
2000         struct sock *sk;
2001         struct sockaddr_ll *sll;
2002         struct packet_sock *po;
2003         u8 *skb_head = skb->data;
2004         int skb_len = skb->len;
2005         unsigned int snaplen, res;
2006         bool is_drop_n_account = false;
2007
2008         if (skb->pkt_type == PACKET_LOOPBACK)
2009                 goto drop;
2010
2011         sk = pt->af_packet_priv;
2012         po = pkt_sk(sk);
2013
2014         if (!net_eq(dev_net(dev), sock_net(sk)))
2015                 goto drop;
2016
2017         skb->dev = dev;
2018
2019         if (dev->header_ops) {
2020                 /* The device has an explicit notion of ll header,
2021                  * exported to higher levels.
2022                  *
2023                  * Otherwise, the device hides details of its frame
2024                  * structure, so that corresponding packet head is
2025                  * never delivered to user.
2026                  */
2027                 if (sk->sk_type != SOCK_DGRAM)
2028                         skb_push(skb, skb->data - skb_mac_header(skb));
2029                 else if (skb->pkt_type == PACKET_OUTGOING) {
2030                         /* Special case: outgoing packets have ll header at head */
2031                         skb_pull(skb, skb_network_offset(skb));
2032                 }
2033         }
2034
2035         snaplen = skb->len;
2036
2037         res = run_filter(skb, sk, snaplen);
2038         if (!res)
2039                 goto drop_n_restore;
2040         if (snaplen > res)
2041                 snaplen = res;
2042
2043         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2044                 goto drop_n_acct;
2045
2046         if (skb_shared(skb)) {
2047                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2048                 if (nskb == NULL)
2049                         goto drop_n_acct;
2050
2051                 if (skb_head != skb->data) {
2052                         skb->data = skb_head;
2053                         skb->len = skb_len;
2054                 }
2055                 consume_skb(skb);
2056                 skb = nskb;
2057         }
2058
2059         sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2060
2061         sll = &PACKET_SKB_CB(skb)->sa.ll;
2062         sll->sll_hatype = dev->type;
2063         sll->sll_pkttype = skb->pkt_type;
2064         if (unlikely(po->origdev))
2065                 sll->sll_ifindex = orig_dev->ifindex;
2066         else
2067                 sll->sll_ifindex = dev->ifindex;
2068
2069         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2070
2071         /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2072          * Use their space for storing the original skb length.
2073          */
2074         PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2075
2076         if (pskb_trim(skb, snaplen))
2077                 goto drop_n_acct;
2078
2079         skb_set_owner_r(skb, sk);
2080         skb->dev = NULL;
2081         skb_dst_drop(skb);
2082
2083         /* drop conntrack reference */
2084         nf_reset(skb);
2085
2086         spin_lock(&sk->sk_receive_queue.lock);
2087         po->stats.stats1.tp_packets++;
2088         sock_skb_set_dropcount(sk, skb);
2089         __skb_queue_tail(&sk->sk_receive_queue, skb);
2090         spin_unlock(&sk->sk_receive_queue.lock);
2091         sk->sk_data_ready(sk);
2092         return 0;
2093
2094 drop_n_acct:
2095         is_drop_n_account = true;
2096         spin_lock(&sk->sk_receive_queue.lock);
2097         po->stats.stats1.tp_drops++;
2098         atomic_inc(&sk->sk_drops);
2099         spin_unlock(&sk->sk_receive_queue.lock);
2100
2101 drop_n_restore:
2102         if (skb_head != skb->data && skb_shared(skb)) {
2103                 skb->data = skb_head;
2104                 skb->len = skb_len;
2105         }
2106 drop:
2107         if (!is_drop_n_account)
2108                 consume_skb(skb);
2109         else
2110                 kfree_skb(skb);
2111         return 0;
2112 }
2113
2114 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2115                        struct packet_type *pt, struct net_device *orig_dev)
2116 {
2117         struct sock *sk;
2118         struct packet_sock *po;
2119         struct sockaddr_ll *sll;
2120         union tpacket_uhdr h;
2121         u8 *skb_head = skb->data;
2122         int skb_len = skb->len;
2123         unsigned int snaplen, res;
2124         unsigned long status = TP_STATUS_USER;
2125         unsigned short macoff, netoff, hdrlen;
2126         struct sk_buff *copy_skb = NULL;
2127         struct timespec ts;
2128         __u32 ts_status;
2129         bool is_drop_n_account = false;
2130
2131         /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2132          * We may add members to them until current aligned size without forcing
2133          * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2134          */
2135         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2136         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2137
2138         if (skb->pkt_type == PACKET_LOOPBACK)
2139                 goto drop;
2140
2141         sk = pt->af_packet_priv;
2142         po = pkt_sk(sk);
2143
2144         if (!net_eq(dev_net(dev), sock_net(sk)))
2145                 goto drop;
2146
2147         if (dev->header_ops) {
2148                 if (sk->sk_type != SOCK_DGRAM)
2149                         skb_push(skb, skb->data - skb_mac_header(skb));
2150                 else if (skb->pkt_type == PACKET_OUTGOING) {
2151                         /* Special case: outgoing packets have ll header at head */
2152                         skb_pull(skb, skb_network_offset(skb));
2153                 }
2154         }
2155
2156         snaplen = skb->len;
2157
2158         res = run_filter(skb, sk, snaplen);
2159         if (!res)
2160                 goto drop_n_restore;
2161
2162         if (skb->ip_summed == CHECKSUM_PARTIAL)
2163                 status |= TP_STATUS_CSUMNOTREADY;
2164         else if (skb->pkt_type != PACKET_OUTGOING &&
2165                  (skb->ip_summed == CHECKSUM_COMPLETE ||
2166                   skb_csum_unnecessary(skb)))
2167                 status |= TP_STATUS_CSUM_VALID;
2168
2169         if (snaplen > res)
2170                 snaplen = res;
2171
2172         if (sk->sk_type == SOCK_DGRAM) {
2173                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2174                                   po->tp_reserve;
2175         } else {
2176                 unsigned int maclen = skb_network_offset(skb);
2177                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2178                                        (maclen < 16 ? 16 : maclen)) +
2179                                        po->tp_reserve;
2180                 if (po->has_vnet_hdr)
2181                         netoff += sizeof(struct virtio_net_hdr);
2182                 macoff = netoff - maclen;
2183         }
2184         if (po->tp_version <= TPACKET_V2) {
2185                 if (macoff + snaplen > po->rx_ring.frame_size) {
2186                         if (po->copy_thresh &&
2187                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2188                                 if (skb_shared(skb)) {
2189                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
2190                                 } else {
2191                                         copy_skb = skb_get(skb);
2192                                         skb_head = skb->data;
2193                                 }
2194                                 if (copy_skb)
2195                                         skb_set_owner_r(copy_skb, sk);
2196                         }
2197                         snaplen = po->rx_ring.frame_size - macoff;
2198                         if ((int)snaplen < 0)
2199                                 snaplen = 0;
2200                 }
2201         } else if (unlikely(macoff + snaplen >
2202                             GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2203                 u32 nval;
2204
2205                 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2206                 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2207                             snaplen, nval, macoff);
2208                 snaplen = nval;
2209                 if (unlikely((int)snaplen < 0)) {
2210                         snaplen = 0;
2211                         macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2212                 }
2213         }
2214         spin_lock(&sk->sk_receive_queue.lock);
2215         h.raw = packet_current_rx_frame(po, skb,
2216                                         TP_STATUS_KERNEL, (macoff+snaplen));
2217         if (!h.raw)
2218                 goto drop_n_account;
2219         if (po->tp_version <= TPACKET_V2) {
2220                 packet_increment_rx_head(po, &po->rx_ring);
2221         /*
2222          * LOSING will be reported till you read the stats,
2223          * because it's COR - Clear On Read.
2224          * Anyways, moving it for V1/V2 only as V3 doesn't need this
2225          * at packet level.
2226          */
2227                 if (po->stats.stats1.tp_drops)
2228                         status |= TP_STATUS_LOSING;
2229         }
2230         po->stats.stats1.tp_packets++;
2231         if (copy_skb) {
2232                 status |= TP_STATUS_COPY;
2233                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2234         }
2235         spin_unlock(&sk->sk_receive_queue.lock);
2236
2237         if (po->has_vnet_hdr) {
2238                 if (virtio_net_hdr_from_skb(skb, h.raw + macoff -
2239                                             sizeof(struct virtio_net_hdr),
2240                                             vio_le())) {
2241                         spin_lock(&sk->sk_receive_queue.lock);
2242                         goto drop_n_account;
2243                 }
2244         }
2245
2246         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2247
2248         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2249                 getnstimeofday(&ts);
2250
2251         status |= ts_status;
2252
2253         switch (po->tp_version) {
2254         case TPACKET_V1:
2255                 h.h1->tp_len = skb->len;
2256                 h.h1->tp_snaplen = snaplen;
2257                 h.h1->tp_mac = macoff;
2258                 h.h1->tp_net = netoff;
2259                 h.h1->tp_sec = ts.tv_sec;
2260                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2261                 hdrlen = sizeof(*h.h1);
2262                 break;
2263         case TPACKET_V2:
2264                 h.h2->tp_len = skb->len;
2265                 h.h2->tp_snaplen = snaplen;
2266                 h.h2->tp_mac = macoff;
2267                 h.h2->tp_net = netoff;
2268                 h.h2->tp_sec = ts.tv_sec;
2269                 h.h2->tp_nsec = ts.tv_nsec;
2270                 if (skb_vlan_tag_present(skb)) {
2271                         h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2272                         h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2273                         status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2274                 } else {
2275                         h.h2->tp_vlan_tci = 0;
2276                         h.h2->tp_vlan_tpid = 0;
2277                 }
2278                 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2279                 hdrlen = sizeof(*h.h2);
2280                 break;
2281         case TPACKET_V3:
2282                 /* tp_nxt_offset,vlan are already populated above.
2283                  * So DONT clear those fields here
2284                  */
2285                 h.h3->tp_status |= status;
2286                 h.h3->tp_len = skb->len;
2287                 h.h3->tp_snaplen = snaplen;
2288                 h.h3->tp_mac = macoff;
2289                 h.h3->tp_net = netoff;
2290                 h.h3->tp_sec  = ts.tv_sec;
2291                 h.h3->tp_nsec = ts.tv_nsec;
2292                 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2293                 hdrlen = sizeof(*h.h3);
2294                 break;
2295         default:
2296                 BUG();
2297         }
2298
2299         sll = h.raw + TPACKET_ALIGN(hdrlen);
2300         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2301         sll->sll_family = AF_PACKET;
2302         sll->sll_hatype = dev->type;
2303         sll->sll_protocol = skb->protocol;
2304         sll->sll_pkttype = skb->pkt_type;
2305         if (unlikely(po->origdev))
2306                 sll->sll_ifindex = orig_dev->ifindex;
2307         else
2308                 sll->sll_ifindex = dev->ifindex;
2309
2310         smp_mb();
2311
2312 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2313         if (po->tp_version <= TPACKET_V2) {
2314                 u8 *start, *end;
2315
2316                 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2317                                         macoff + snaplen);
2318
2319                 for (start = h.raw; start < end; start += PAGE_SIZE)
2320                         flush_dcache_page(pgv_to_page(start));
2321         }
2322         smp_wmb();
2323 #endif
2324
2325         if (po->tp_version <= TPACKET_V2) {
2326                 __packet_set_status(po, h.raw, status);
2327                 sk->sk_data_ready(sk);
2328         } else {
2329                 prb_clear_blk_fill_status(&po->rx_ring);
2330         }
2331
2332 drop_n_restore:
2333         if (skb_head != skb->data && skb_shared(skb)) {
2334                 skb->data = skb_head;
2335                 skb->len = skb_len;
2336         }
2337 drop:
2338         if (!is_drop_n_account)
2339                 consume_skb(skb);
2340         else
2341                 kfree_skb(skb);
2342         return 0;
2343
2344 drop_n_account:
2345         is_drop_n_account = true;
2346         po->stats.stats1.tp_drops++;
2347         spin_unlock(&sk->sk_receive_queue.lock);
2348
2349         sk->sk_data_ready(sk);
2350         kfree_skb(copy_skb);
2351         goto drop_n_restore;
2352 }
2353
2354 static void tpacket_destruct_skb(struct sk_buff *skb)
2355 {
2356         struct packet_sock *po = pkt_sk(skb->sk);
2357
2358         if (likely(po->tx_ring.pg_vec)) {
2359                 void *ph;
2360                 __u32 ts;
2361
2362                 ph = skb_shinfo(skb)->destructor_arg;
2363                 packet_dec_pending(&po->tx_ring);
2364
2365                 ts = __packet_set_timestamp(po, ph, skb);
2366                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2367         }
2368
2369         sock_wfree(skb);
2370 }
2371
2372 static void tpacket_set_protocol(const struct net_device *dev,
2373                                  struct sk_buff *skb)
2374 {
2375         if (dev->type == ARPHRD_ETHER) {
2376                 skb_reset_mac_header(skb);
2377                 skb->protocol = eth_hdr(skb)->h_proto;
2378         }
2379 }
2380
2381 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2382 {
2383         if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2384             (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2385              __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2386               __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2387                 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2388                          __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2389                         __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2390
2391         if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2392                 return -EINVAL;
2393
2394         return 0;
2395 }
2396
2397 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2398                                  struct virtio_net_hdr *vnet_hdr)
2399 {
2400         int n;
2401
2402         if (*len < sizeof(*vnet_hdr))
2403                 return -EINVAL;
2404         *len -= sizeof(*vnet_hdr);
2405
2406         n = copy_from_iter(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter);
2407         if (n != sizeof(*vnet_hdr))
2408                 return -EFAULT;
2409
2410         return __packet_snd_vnet_parse(vnet_hdr, *len);
2411 }
2412
2413 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2414                 void *frame, struct net_device *dev, void *data, int tp_len,
2415                 __be16 proto, unsigned char *addr, int hlen, int copylen,
2416                 const struct sockcm_cookie *sockc)
2417 {
2418         union tpacket_uhdr ph;
2419         int to_write, offset, len, nr_frags, len_max;
2420         struct socket *sock = po->sk.sk_socket;
2421         struct page *page;
2422         int err;
2423
2424         ph.raw = frame;
2425
2426         skb->protocol = proto;
2427         skb->dev = dev;
2428         skb->priority = po->sk.sk_priority;
2429         skb->mark = po->sk.sk_mark;
2430         sock_tx_timestamp(&po->sk, sockc->tsflags, &skb_shinfo(skb)->tx_flags);
2431         skb_shinfo(skb)->destructor_arg = ph.raw;
2432
2433         skb_reserve(skb, hlen);
2434         skb_reset_network_header(skb);
2435
2436         to_write = tp_len;
2437
2438         if (sock->type == SOCK_DGRAM) {
2439                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2440                                 NULL, tp_len);
2441                 if (unlikely(err < 0))
2442                         return -EINVAL;
2443         } else if (copylen) {
2444                 int hdrlen = min_t(int, copylen, tp_len);
2445
2446                 skb_push(skb, dev->hard_header_len);
2447                 skb_put(skb, copylen - dev->hard_header_len);
2448                 err = skb_store_bits(skb, 0, data, hdrlen);
2449                 if (unlikely(err))
2450                         return err;
2451                 if (!dev_validate_header(dev, skb->data, hdrlen))
2452                         return -EINVAL;
2453                 if (!skb->protocol)
2454                         tpacket_set_protocol(dev, skb);
2455
2456                 data += hdrlen;
2457                 to_write -= hdrlen;
2458         }
2459
2460         offset = offset_in_page(data);
2461         len_max = PAGE_SIZE - offset;
2462         len = ((to_write > len_max) ? len_max : to_write);
2463
2464         skb->data_len = to_write;
2465         skb->len += to_write;
2466         skb->truesize += to_write;
2467         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2468
2469         while (likely(to_write)) {
2470                 nr_frags = skb_shinfo(skb)->nr_frags;
2471
2472                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2473                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2474                                MAX_SKB_FRAGS);
2475                         return -EFAULT;
2476                 }
2477
2478                 page = pgv_to_page(data);
2479                 data += len;
2480                 flush_dcache_page(page);
2481                 get_page(page);
2482                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2483                 to_write -= len;
2484                 offset = 0;
2485                 len_max = PAGE_SIZE;
2486                 len = ((to_write > len_max) ? len_max : to_write);
2487         }
2488
2489         skb_probe_transport_header(skb, 0);
2490
2491         return tp_len;
2492 }
2493
2494 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2495                                 int size_max, void **data)
2496 {
2497         union tpacket_uhdr ph;
2498         int tp_len, off;
2499
2500         ph.raw = frame;
2501
2502         switch (po->tp_version) {
2503         case TPACKET_V2:
2504                 tp_len = ph.h2->tp_len;
2505                 break;
2506         default:
2507                 tp_len = ph.h1->tp_len;
2508                 break;
2509         }
2510         if (unlikely(tp_len > size_max)) {
2511                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2512                 return -EMSGSIZE;
2513         }
2514
2515         if (unlikely(po->tp_tx_has_off)) {
2516                 int off_min, off_max;
2517
2518                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2519                 off_max = po->tx_ring.frame_size - tp_len;
2520                 if (po->sk.sk_type == SOCK_DGRAM) {
2521                         switch (po->tp_version) {
2522                         case TPACKET_V2:
2523                                 off = ph.h2->tp_net;
2524                                 break;
2525                         default:
2526                                 off = ph.h1->tp_net;
2527                                 break;
2528                         }
2529                 } else {
2530                         switch (po->tp_version) {
2531                         case TPACKET_V2:
2532                                 off = ph.h2->tp_mac;
2533                                 break;
2534                         default:
2535                                 off = ph.h1->tp_mac;
2536                                 break;
2537                         }
2538                 }
2539                 if (unlikely((off < off_min) || (off_max < off)))
2540                         return -EINVAL;
2541         } else {
2542                 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2543         }
2544
2545         *data = frame + off;
2546         return tp_len;
2547 }
2548
2549 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2550 {
2551         struct sk_buff *skb;
2552         struct net_device *dev;
2553         struct virtio_net_hdr *vnet_hdr = NULL;
2554         struct sockcm_cookie sockc;
2555         __be16 proto;
2556         int err, reserve = 0;
2557         void *ph;
2558         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2559         bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2560         int tp_len, size_max;
2561         unsigned char *addr;
2562         void *data;
2563         int len_sum = 0;
2564         int status = TP_STATUS_AVAILABLE;
2565         int hlen, tlen, copylen = 0;
2566
2567         mutex_lock(&po->pg_vec_lock);
2568
2569         if (likely(saddr == NULL)) {
2570                 dev     = packet_cached_dev_get(po);
2571                 proto   = po->num;
2572                 addr    = NULL;
2573         } else {
2574                 err = -EINVAL;
2575                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2576                         goto out;
2577                 if (msg->msg_namelen < (saddr->sll_halen
2578                                         + offsetof(struct sockaddr_ll,
2579                                                 sll_addr)))
2580                         goto out;
2581                 proto   = saddr->sll_protocol;
2582                 addr    = saddr->sll_addr;
2583                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2584         }
2585
2586         sockc.tsflags = po->sk.sk_tsflags;
2587         if (msg->msg_controllen) {
2588                 err = sock_cmsg_send(&po->sk, msg, &sockc);
2589                 if (unlikely(err))
2590                         goto out;
2591         }
2592
2593         err = -ENXIO;
2594         if (unlikely(dev == NULL))
2595                 goto out;
2596         err = -ENETDOWN;
2597         if (unlikely(!(dev->flags & IFF_UP)))
2598                 goto out_put;
2599
2600         if (po->sk.sk_socket->type == SOCK_RAW)
2601                 reserve = dev->hard_header_len;
2602         size_max = po->tx_ring.frame_size
2603                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2604
2605         if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2606                 size_max = dev->mtu + reserve + VLAN_HLEN;
2607
2608         do {
2609                 ph = packet_current_frame(po, &po->tx_ring,
2610                                           TP_STATUS_SEND_REQUEST);
2611                 if (unlikely(ph == NULL)) {
2612                         if (need_wait && need_resched())
2613                                 schedule();
2614                         continue;
2615                 }
2616
2617                 skb = NULL;
2618                 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2619                 if (tp_len < 0)
2620                         goto tpacket_error;
2621
2622                 status = TP_STATUS_SEND_REQUEST;
2623                 hlen = LL_RESERVED_SPACE(dev);
2624                 tlen = dev->needed_tailroom;
2625                 if (po->has_vnet_hdr) {
2626                         vnet_hdr = data;
2627                         data += sizeof(*vnet_hdr);
2628                         tp_len -= sizeof(*vnet_hdr);
2629                         if (tp_len < 0 ||
2630                             __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2631                                 tp_len = -EINVAL;
2632                                 goto tpacket_error;
2633                         }
2634                         copylen = __virtio16_to_cpu(vio_le(),
2635                                                     vnet_hdr->hdr_len);
2636                 }
2637                 copylen = max_t(int, copylen, dev->hard_header_len);
2638                 skb = sock_alloc_send_skb(&po->sk,
2639                                 hlen + tlen + sizeof(struct sockaddr_ll) +
2640                                 (copylen - dev->hard_header_len),
2641                                 !need_wait, &err);
2642
2643                 if (unlikely(skb == NULL)) {
2644                         /* we assume the socket was initially writeable ... */
2645                         if (likely(len_sum > 0))
2646                                 err = len_sum;
2647                         goto out_status;
2648                 }
2649                 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2650                                           addr, hlen, copylen, &sockc);
2651                 if (likely(tp_len >= 0) &&
2652                     tp_len > dev->mtu + reserve &&
2653                     !po->has_vnet_hdr &&
2654                     !packet_extra_vlan_len_allowed(dev, skb))
2655                         tp_len = -EMSGSIZE;
2656
2657                 if (unlikely(tp_len < 0)) {
2658 tpacket_error:
2659                         if (po->tp_loss) {
2660                                 __packet_set_status(po, ph,
2661                                                 TP_STATUS_AVAILABLE);
2662                                 packet_increment_head(&po->tx_ring);
2663                                 kfree_skb(skb);
2664                                 continue;
2665                         } else {
2666                                 status = TP_STATUS_WRONG_FORMAT;
2667                                 err = tp_len;
2668                                 goto out_status;
2669                         }
2670                 }
2671
2672                 if (po->has_vnet_hdr && virtio_net_hdr_to_skb(skb, vnet_hdr,
2673                                                               vio_le())) {
2674                         tp_len = -EINVAL;
2675                         goto tpacket_error;
2676                 }
2677
2678                 packet_pick_tx_queue(dev, skb);
2679
2680                 skb->destructor = tpacket_destruct_skb;
2681                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2682                 packet_inc_pending(&po->tx_ring);
2683
2684                 status = TP_STATUS_SEND_REQUEST;
2685                 err = po->xmit(skb);
2686                 if (unlikely(err > 0)) {
2687                         err = net_xmit_errno(err);
2688                         if (err && __packet_get_status(po, ph) ==
2689                                    TP_STATUS_AVAILABLE) {
2690                                 /* skb was destructed already */
2691                                 skb = NULL;
2692                                 goto out_status;
2693                         }
2694                         /*
2695                          * skb was dropped but not destructed yet;
2696                          * let's treat it like congestion or err < 0
2697                          */
2698                         err = 0;
2699                 }
2700                 packet_increment_head(&po->tx_ring);
2701                 len_sum += tp_len;
2702         } while (likely((ph != NULL) ||
2703                 /* Note: packet_read_pending() might be slow if we have
2704                  * to call it as it's per_cpu variable, but in fast-path
2705                  * we already short-circuit the loop with the first
2706                  * condition, and luckily don't have to go that path
2707                  * anyway.
2708                  */
2709                  (need_wait && packet_read_pending(&po->tx_ring))));
2710
2711         err = len_sum;
2712         goto out_put;
2713
2714 out_status:
2715         __packet_set_status(po, ph, status);
2716         kfree_skb(skb);
2717 out_put:
2718         dev_put(dev);
2719 out:
2720         mutex_unlock(&po->pg_vec_lock);
2721         return err;
2722 }
2723
2724 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2725                                         size_t reserve, size_t len,
2726                                         size_t linear, int noblock,
2727                                         int *err)
2728 {
2729         struct sk_buff *skb;
2730
2731         /* Under a page?  Don't bother with paged skb. */
2732         if (prepad + len < PAGE_SIZE || !linear)
2733                 linear = len;
2734
2735         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2736                                    err, 0);
2737         if (!skb)
2738                 return NULL;
2739
2740         skb_reserve(skb, reserve);
2741         skb_put(skb, linear);
2742         skb->data_len = len - linear;
2743         skb->len += len - linear;
2744
2745         return skb;
2746 }
2747
2748 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2749 {
2750         struct sock *sk = sock->sk;
2751         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2752         struct sk_buff *skb;
2753         struct net_device *dev;
2754         __be16 proto;
2755         unsigned char *addr;
2756         int err, reserve = 0;
2757         struct sockcm_cookie sockc;
2758         struct virtio_net_hdr vnet_hdr = { 0 };
2759         int offset = 0;
2760         struct packet_sock *po = pkt_sk(sk);
2761         int hlen, tlen;
2762         int extra_len = 0;
2763
2764         /*
2765          *      Get and verify the address.
2766          */
2767
2768         if (likely(saddr == NULL)) {
2769                 dev     = packet_cached_dev_get(po);
2770                 proto   = po->num;
2771                 addr    = NULL;
2772         } else {
2773                 err = -EINVAL;
2774                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2775                         goto out;
2776                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2777                         goto out;
2778                 proto   = saddr->sll_protocol;
2779                 addr    = saddr->sll_addr;
2780                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2781         }
2782
2783         err = -ENXIO;
2784         if (unlikely(dev == NULL))
2785                 goto out_unlock;
2786         err = -ENETDOWN;
2787         if (unlikely(!(dev->flags & IFF_UP)))
2788                 goto out_unlock;
2789
2790         sockc.tsflags = sk->sk_tsflags;
2791         sockc.mark = sk->sk_mark;
2792         if (msg->msg_controllen) {
2793                 err = sock_cmsg_send(sk, msg, &sockc);
2794                 if (unlikely(err))
2795                         goto out_unlock;
2796         }
2797
2798         if (sock->type == SOCK_RAW)
2799                 reserve = dev->hard_header_len;
2800         if (po->has_vnet_hdr) {
2801                 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2802                 if (err)
2803                         goto out_unlock;
2804         }
2805
2806         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2807                 if (!netif_supports_nofcs(dev)) {
2808                         err = -EPROTONOSUPPORT;
2809                         goto out_unlock;
2810                 }
2811                 extra_len = 4; /* We're doing our own CRC */
2812         }
2813
2814         err = -EMSGSIZE;
2815         if (!vnet_hdr.gso_type &&
2816             (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2817                 goto out_unlock;
2818
2819         err = -ENOBUFS;
2820         hlen = LL_RESERVED_SPACE(dev);
2821         tlen = dev->needed_tailroom;
2822         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
2823                                __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len),
2824                                msg->msg_flags & MSG_DONTWAIT, &err);
2825         if (skb == NULL)
2826                 goto out_unlock;
2827
2828         skb_set_network_header(skb, reserve);
2829
2830         err = -EINVAL;
2831         if (sock->type == SOCK_DGRAM) {
2832                 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2833                 if (unlikely(offset < 0))
2834                         goto out_free;
2835         }
2836
2837         /* Returns -EFAULT on error */
2838         err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2839         if (err)
2840                 goto out_free;
2841
2842         if (sock->type == SOCK_RAW &&
2843             !dev_validate_header(dev, skb->data, len)) {
2844                 err = -EINVAL;
2845                 goto out_free;
2846         }
2847
2848         sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
2849
2850         if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2851             !packet_extra_vlan_len_allowed(dev, skb)) {
2852                 err = -EMSGSIZE;
2853                 goto out_free;
2854         }
2855
2856         skb->protocol = proto;
2857         skb->dev = dev;
2858         skb->priority = sk->sk_priority;
2859         skb->mark = sockc.mark;
2860
2861         packet_pick_tx_queue(dev, skb);
2862
2863         if (po->has_vnet_hdr) {
2864                 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2865                 if (err)
2866                         goto out_free;
2867                 len += sizeof(vnet_hdr);
2868         }
2869
2870         skb_probe_transport_header(skb, reserve);
2871
2872         if (unlikely(extra_len == 4))
2873                 skb->no_fcs = 1;
2874
2875         err = po->xmit(skb);
2876         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2877                 goto out_unlock;
2878
2879         dev_put(dev);
2880
2881         return len;
2882
2883 out_free:
2884         kfree_skb(skb);
2885 out_unlock:
2886         if (dev)
2887                 dev_put(dev);
2888 out:
2889         return err;
2890 }
2891
2892 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2893 {
2894         struct sock *sk = sock->sk;
2895         struct packet_sock *po = pkt_sk(sk);
2896
2897         if (po->tx_ring.pg_vec)
2898                 return tpacket_snd(po, msg);
2899         else
2900                 return packet_snd(sock, msg, len);
2901 }
2902
2903 /*
2904  *      Close a PACKET socket. This is fairly simple. We immediately go
2905  *      to 'closed' state and remove our protocol entry in the device list.
2906  */
2907
2908 static int packet_release(struct socket *sock)
2909 {
2910         struct sock *sk = sock->sk;
2911         struct packet_sock *po;
2912         struct net *net;
2913         union tpacket_req_u req_u;
2914
2915         if (!sk)
2916                 return 0;
2917
2918         net = sock_net(sk);
2919         po = pkt_sk(sk);
2920
2921         mutex_lock(&net->packet.sklist_lock);
2922         sk_del_node_init_rcu(sk);
2923         mutex_unlock(&net->packet.sklist_lock);
2924
2925         preempt_disable();
2926         sock_prot_inuse_add(net, sk->sk_prot, -1);
2927         preempt_enable();
2928
2929         spin_lock(&po->bind_lock);
2930         unregister_prot_hook(sk, false);
2931         packet_cached_dev_reset(po);
2932
2933         if (po->prot_hook.dev) {
2934                 dev_put(po->prot_hook.dev);
2935                 po->prot_hook.dev = NULL;
2936         }
2937         spin_unlock(&po->bind_lock);
2938
2939         packet_flush_mclist(sk);
2940
2941         if (po->rx_ring.pg_vec) {
2942                 memset(&req_u, 0, sizeof(req_u));
2943                 packet_set_ring(sk, &req_u, 1, 0);
2944         }
2945
2946         if (po->tx_ring.pg_vec) {
2947                 memset(&req_u, 0, sizeof(req_u));
2948                 packet_set_ring(sk, &req_u, 1, 1);
2949         }
2950
2951         fanout_release(sk);
2952
2953         synchronize_net();
2954         /*
2955          *      Now the socket is dead. No more input will appear.
2956          */
2957         sock_orphan(sk);
2958         sock->sk = NULL;
2959
2960         /* Purge queues */
2961
2962         skb_queue_purge(&sk->sk_receive_queue);
2963         packet_free_pending(po);
2964         sk_refcnt_debug_release(sk);
2965
2966         sock_put(sk);
2967         return 0;
2968 }
2969
2970 /*
2971  *      Attach a packet hook.
2972  */
2973
2974 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
2975                           __be16 proto)
2976 {
2977         struct packet_sock *po = pkt_sk(sk);
2978         struct net_device *dev_curr;
2979         __be16 proto_curr;
2980         bool need_rehook;
2981         struct net_device *dev = NULL;
2982         int ret = 0;
2983         bool unlisted = false;
2984
2985         if (po->fanout)
2986                 return -EINVAL;
2987
2988         lock_sock(sk);
2989         spin_lock(&po->bind_lock);
2990         rcu_read_lock();
2991
2992         if (name) {
2993                 dev = dev_get_by_name_rcu(sock_net(sk), name);
2994                 if (!dev) {
2995                         ret = -ENODEV;
2996                         goto out_unlock;
2997                 }
2998         } else if (ifindex) {
2999                 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3000                 if (!dev) {
3001                         ret = -ENODEV;
3002                         goto out_unlock;
3003                 }
3004         }
3005
3006         if (dev)
3007                 dev_hold(dev);
3008
3009         proto_curr = po->prot_hook.type;
3010         dev_curr = po->prot_hook.dev;
3011
3012         need_rehook = proto_curr != proto || dev_curr != dev;
3013
3014         if (need_rehook) {
3015                 if (po->running) {
3016                         rcu_read_unlock();
3017                         __unregister_prot_hook(sk, true);
3018                         rcu_read_lock();
3019                         dev_curr = po->prot_hook.dev;
3020                         if (dev)
3021                                 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3022                                                                  dev->ifindex);
3023                 }
3024
3025                 po->num = proto;
3026                 po->prot_hook.type = proto;
3027
3028                 if (unlikely(unlisted)) {
3029                         dev_put(dev);
3030                         po->prot_hook.dev = NULL;
3031                         po->ifindex = -1;
3032                         packet_cached_dev_reset(po);
3033                 } else {
3034                         po->prot_hook.dev = dev;
3035                         po->ifindex = dev ? dev->ifindex : 0;
3036                         packet_cached_dev_assign(po, dev);
3037                 }
3038         }
3039         if (dev_curr)
3040                 dev_put(dev_curr);
3041
3042         if (proto == 0 || !need_rehook)
3043                 goto out_unlock;
3044
3045         if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3046                 register_prot_hook(sk);
3047         } else {
3048                 sk->sk_err = ENETDOWN;
3049                 if (!sock_flag(sk, SOCK_DEAD))
3050                         sk->sk_error_report(sk);
3051         }
3052
3053 out_unlock:
3054         rcu_read_unlock();
3055         spin_unlock(&po->bind_lock);
3056         release_sock(sk);
3057         return ret;
3058 }
3059
3060 /*
3061  *      Bind a packet socket to a device
3062  */
3063
3064 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3065                             int addr_len)
3066 {
3067         struct sock *sk = sock->sk;
3068         char name[15];
3069
3070         /*
3071          *      Check legality
3072          */
3073
3074         if (addr_len != sizeof(struct sockaddr))
3075                 return -EINVAL;
3076         strlcpy(name, uaddr->sa_data, sizeof(name));
3077
3078         return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3079 }
3080
3081 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3082 {
3083         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3084         struct sock *sk = sock->sk;
3085
3086         /*
3087          *      Check legality
3088          */
3089
3090         if (addr_len < sizeof(struct sockaddr_ll))
3091                 return -EINVAL;
3092         if (sll->sll_family != AF_PACKET)
3093                 return -EINVAL;
3094
3095         return packet_do_bind(sk, NULL, sll->sll_ifindex,
3096                               sll->sll_protocol ? : pkt_sk(sk)->num);
3097 }
3098
3099 static struct proto packet_proto = {
3100         .name     = "PACKET",
3101         .owner    = THIS_MODULE,
3102         .obj_size = sizeof(struct packet_sock),
3103 };
3104
3105 /*
3106  *      Create a packet of type SOCK_PACKET.
3107  */
3108
3109 static int packet_create(struct net *net, struct socket *sock, int protocol,
3110                          int kern)
3111 {
3112         struct sock *sk;
3113         struct packet_sock *po;
3114         __be16 proto = (__force __be16)protocol; /* weird, but documented */
3115         int err;
3116
3117         if (!ns_capable(net->user_ns, CAP_NET_RAW))
3118                 return -EPERM;
3119         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3120             sock->type != SOCK_PACKET)
3121                 return -ESOCKTNOSUPPORT;
3122
3123         sock->state = SS_UNCONNECTED;
3124
3125         err = -ENOBUFS;
3126         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3127         if (sk == NULL)
3128                 goto out;
3129
3130         sock->ops = &packet_ops;
3131         if (sock->type == SOCK_PACKET)
3132                 sock->ops = &packet_ops_spkt;
3133
3134         sock_init_data(sock, sk);
3135
3136         po = pkt_sk(sk);
3137         sk->sk_family = PF_PACKET;
3138         po->num = proto;
3139         po->xmit = dev_queue_xmit;
3140
3141         err = packet_alloc_pending(po);
3142         if (err)
3143                 goto out2;
3144
3145         packet_cached_dev_reset(po);
3146
3147         sk->sk_destruct = packet_sock_destruct;
3148         sk_refcnt_debug_inc(sk);
3149
3150         /*
3151          *      Attach a protocol block
3152          */
3153
3154         spin_lock_init(&po->bind_lock);
3155         mutex_init(&po->pg_vec_lock);
3156         po->rollover = NULL;
3157         po->prot_hook.func = packet_rcv;
3158
3159         if (sock->type == SOCK_PACKET)
3160                 po->prot_hook.func = packet_rcv_spkt;
3161
3162         po->prot_hook.af_packet_priv = sk;
3163
3164         if (proto) {
3165                 po->prot_hook.type = proto;
3166                 register_prot_hook(sk);
3167         }
3168
3169         mutex_lock(&net->packet.sklist_lock);
3170         sk_add_node_rcu(sk, &net->packet.sklist);
3171         mutex_unlock(&net->packet.sklist_lock);
3172
3173         preempt_disable();
3174         sock_prot_inuse_add(net, &packet_proto, 1);
3175         preempt_enable();
3176
3177         return 0;
3178 out2:
3179         sk_free(sk);
3180 out:
3181         return err;
3182 }
3183
3184 /*
3185  *      Pull a packet from our receive queue and hand it to the user.
3186  *      If necessary we block.
3187  */
3188
3189 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3190                           int flags)
3191 {
3192         struct sock *sk = sock->sk;
3193         struct sk_buff *skb;
3194         int copied, err;
3195         int vnet_hdr_len = 0;
3196         unsigned int origlen = 0;
3197
3198         err = -EINVAL;
3199         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3200                 goto out;
3201
3202 #if 0
3203         /* What error should we return now? EUNATTACH? */
3204         if (pkt_sk(sk)->ifindex < 0)
3205                 return -ENODEV;
3206 #endif
3207
3208         if (flags & MSG_ERRQUEUE) {
3209                 err = sock_recv_errqueue(sk, msg, len,
3210                                          SOL_PACKET, PACKET_TX_TIMESTAMP);
3211                 goto out;
3212         }
3213
3214         /*
3215          *      Call the generic datagram receiver. This handles all sorts
3216          *      of horrible races and re-entrancy so we can forget about it
3217          *      in the protocol layers.
3218          *
3219          *      Now it will return ENETDOWN, if device have just gone down,
3220          *      but then it will block.
3221          */
3222
3223         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3224
3225         /*
3226          *      An error occurred so return it. Because skb_recv_datagram()
3227          *      handles the blocking we don't see and worry about blocking
3228          *      retries.
3229          */
3230
3231         if (skb == NULL)
3232                 goto out;
3233
3234         if (pkt_sk(sk)->pressure)
3235                 packet_rcv_has_room(pkt_sk(sk), NULL);
3236
3237         if (pkt_sk(sk)->has_vnet_hdr) {
3238                 err = packet_rcv_vnet(msg, skb, &len);
3239                 if (err)
3240                         goto out_free;
3241                 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3242         }
3243
3244         /* You lose any data beyond the buffer you gave. If it worries
3245          * a user program they can ask the device for its MTU
3246          * anyway.
3247          */
3248         copied = skb->len;
3249         if (copied > len) {
3250                 copied = len;
3251                 msg->msg_flags |= MSG_TRUNC;
3252         }
3253
3254         err = skb_copy_datagram_msg(skb, 0, msg, copied);
3255         if (err)
3256                 goto out_free;
3257
3258         if (sock->type != SOCK_PACKET) {
3259                 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3260
3261                 /* Original length was stored in sockaddr_ll fields */
3262                 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3263                 sll->sll_family = AF_PACKET;
3264                 sll->sll_protocol = skb->protocol;
3265         }
3266
3267         sock_recv_ts_and_drops(msg, sk, skb);
3268
3269         if (msg->msg_name) {
3270                 /* If the address length field is there to be filled
3271                  * in, we fill it in now.
3272                  */
3273                 if (sock->type == SOCK_PACKET) {
3274                         __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3275                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
3276                 } else {
3277                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3278
3279                         msg->msg_namelen = sll->sll_halen +
3280                                 offsetof(struct sockaddr_ll, sll_addr);
3281                 }
3282                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3283                        msg->msg_namelen);
3284         }
3285
3286         if (pkt_sk(sk)->auxdata) {
3287                 struct tpacket_auxdata aux;
3288
3289                 aux.tp_status = TP_STATUS_USER;
3290                 if (skb->ip_summed == CHECKSUM_PARTIAL)
3291                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3292                 else if (skb->pkt_type != PACKET_OUTGOING &&
3293                          (skb->ip_summed == CHECKSUM_COMPLETE ||
3294                           skb_csum_unnecessary(skb)))
3295                         aux.tp_status |= TP_STATUS_CSUM_VALID;
3296
3297                 aux.tp_len = origlen;
3298                 aux.tp_snaplen = skb->len;
3299                 aux.tp_mac = 0;
3300                 aux.tp_net = skb_network_offset(skb);
3301                 if (skb_vlan_tag_present(skb)) {
3302                         aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3303                         aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3304                         aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3305                 } else {
3306                         aux.tp_vlan_tci = 0;
3307                         aux.tp_vlan_tpid = 0;
3308                 }
3309                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3310         }
3311
3312         /*
3313          *      Free or return the buffer as appropriate. Again this
3314          *      hides all the races and re-entrancy issues from us.
3315          */
3316         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3317
3318 out_free:
3319         skb_free_datagram(sk, skb);
3320 out:
3321         return err;
3322 }
3323
3324 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3325                                int *uaddr_len, int peer)
3326 {
3327         struct net_device *dev;
3328         struct sock *sk = sock->sk;
3329
3330         if (peer)
3331                 return -EOPNOTSUPP;
3332
3333         uaddr->sa_family = AF_PACKET;
3334         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3335         rcu_read_lock();
3336         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3337         if (dev)
3338                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3339         rcu_read_unlock();
3340         *uaddr_len = sizeof(*uaddr);
3341
3342         return 0;
3343 }
3344
3345 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3346                           int *uaddr_len, int peer)
3347 {
3348         struct net_device *dev;
3349         struct sock *sk = sock->sk;
3350         struct packet_sock *po = pkt_sk(sk);
3351         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3352
3353         if (peer)
3354                 return -EOPNOTSUPP;
3355
3356         sll->sll_family = AF_PACKET;
3357         sll->sll_ifindex = po->ifindex;
3358         sll->sll_protocol = po->num;
3359         sll->sll_pkttype = 0;
3360         rcu_read_lock();
3361         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3362         if (dev) {
3363                 sll->sll_hatype = dev->type;
3364                 sll->sll_halen = dev->addr_len;
3365                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3366         } else {
3367                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
3368                 sll->sll_halen = 0;
3369         }
3370         rcu_read_unlock();
3371         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3372
3373         return 0;
3374 }
3375
3376 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3377                          int what)
3378 {
3379         switch (i->type) {
3380         case PACKET_MR_MULTICAST:
3381                 if (i->alen != dev->addr_len)
3382                         return -EINVAL;
3383                 if (what > 0)
3384                         return dev_mc_add(dev, i->addr);
3385                 else
3386                         return dev_mc_del(dev, i->addr);
3387                 break;
3388         case PACKET_MR_PROMISC:
3389                 return dev_set_promiscuity(dev, what);
3390         case PACKET_MR_ALLMULTI:
3391                 return dev_set_allmulti(dev, what);
3392         case PACKET_MR_UNICAST:
3393                 if (i->alen != dev->addr_len)
3394                         return -EINVAL;
3395                 if (what > 0)
3396                         return dev_uc_add(dev, i->addr);
3397                 else
3398                         return dev_uc_del(dev, i->addr);
3399                 break;
3400         default:
3401                 break;
3402         }
3403         return 0;
3404 }
3405
3406 static void packet_dev_mclist_delete(struct net_device *dev,
3407                                      struct packet_mclist **mlp)
3408 {
3409         struct packet_mclist *ml;
3410
3411         while ((ml = *mlp) != NULL) {
3412                 if (ml->ifindex == dev->ifindex) {
3413                         packet_dev_mc(dev, ml, -1);
3414                         *mlp = ml->next;
3415                         kfree(ml);
3416                 } else
3417                         mlp = &ml->next;
3418         }
3419 }
3420
3421 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3422 {
3423         struct packet_sock *po = pkt_sk(sk);
3424         struct packet_mclist *ml, *i;
3425         struct net_device *dev;
3426         int err;
3427
3428         rtnl_lock();
3429
3430         err = -ENODEV;
3431         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3432         if (!dev)
3433                 goto done;
3434
3435         err = -EINVAL;
3436         if (mreq->mr_alen > dev->addr_len)
3437                 goto done;
3438
3439         err = -ENOBUFS;
3440         i = kmalloc(sizeof(*i), GFP_KERNEL);
3441         if (i == NULL)
3442                 goto done;
3443
3444         err = 0;
3445         for (ml = po->mclist; ml; ml = ml->next) {
3446                 if (ml->ifindex == mreq->mr_ifindex &&
3447                     ml->type == mreq->mr_type &&
3448                     ml->alen == mreq->mr_alen &&
3449                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3450                         ml->count++;
3451                         /* Free the new element ... */
3452                         kfree(i);
3453                         goto done;
3454                 }
3455         }
3456
3457         i->type = mreq->mr_type;
3458         i->ifindex = mreq->mr_ifindex;
3459         i->alen = mreq->mr_alen;
3460         memcpy(i->addr, mreq->mr_address, i->alen);
3461         memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3462         i->count = 1;
3463         i->next = po->mclist;
3464         po->mclist = i;
3465         err = packet_dev_mc(dev, i, 1);
3466         if (err) {
3467                 po->mclist = i->next;
3468                 kfree(i);
3469         }
3470
3471 done:
3472         rtnl_unlock();
3473         return err;
3474 }
3475
3476 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3477 {
3478         struct packet_mclist *ml, **mlp;
3479
3480         rtnl_lock();
3481
3482         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3483                 if (ml->ifindex == mreq->mr_ifindex &&
3484                     ml->type == mreq->mr_type &&
3485                     ml->alen == mreq->mr_alen &&
3486                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3487                         if (--ml->count == 0) {
3488                                 struct net_device *dev;
3489                                 *mlp = ml->next;
3490                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3491                                 if (dev)
3492                                         packet_dev_mc(dev, ml, -1);
3493                                 kfree(ml);
3494                         }
3495                         break;
3496                 }
3497         }
3498         rtnl_unlock();
3499         return 0;
3500 }
3501
3502 static void packet_flush_mclist(struct sock *sk)
3503 {
3504         struct packet_sock *po = pkt_sk(sk);
3505         struct packet_mclist *ml;
3506
3507         if (!po->mclist)
3508                 return;
3509
3510         rtnl_lock();
3511         while ((ml = po->mclist) != NULL) {
3512                 struct net_device *dev;
3513
3514                 po->mclist = ml->next;
3515                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3516                 if (dev != NULL)
3517                         packet_dev_mc(dev, ml, -1);
3518                 kfree(ml);
3519         }
3520         rtnl_unlock();
3521 }
3522
3523 static int
3524 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3525 {
3526         struct sock *sk = sock->sk;
3527         struct packet_sock *po = pkt_sk(sk);
3528         int ret;
3529
3530         if (level != SOL_PACKET)
3531                 return -ENOPROTOOPT;
3532
3533         switch (optname) {
3534         case PACKET_ADD_MEMBERSHIP:
3535         case PACKET_DROP_MEMBERSHIP:
3536         {
3537                 struct packet_mreq_max mreq;
3538                 int len = optlen;
3539                 memset(&mreq, 0, sizeof(mreq));
3540                 if (len < sizeof(struct packet_mreq))
3541                         return -EINVAL;
3542                 if (len > sizeof(mreq))
3543                         len = sizeof(mreq);
3544                 if (copy_from_user(&mreq, optval, len))
3545                         return -EFAULT;
3546                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3547                         return -EINVAL;
3548                 if (optname == PACKET_ADD_MEMBERSHIP)
3549                         ret = packet_mc_add(sk, &mreq);
3550                 else
3551                         ret = packet_mc_drop(sk, &mreq);
3552                 return ret;
3553         }
3554
3555         case PACKET_RX_RING:
3556         case PACKET_TX_RING:
3557         {
3558                 union tpacket_req_u req_u;
3559                 int len;
3560
3561                 switch (po->tp_version) {
3562                 case TPACKET_V1:
3563                 case TPACKET_V2:
3564                         len = sizeof(req_u.req);
3565                         break;
3566                 case TPACKET_V3:
3567                 default:
3568                         len = sizeof(req_u.req3);
3569                         break;
3570                 }
3571                 if (optlen < len)
3572                         return -EINVAL;
3573                 if (copy_from_user(&req_u.req, optval, len))
3574                         return -EFAULT;
3575                 return packet_set_ring(sk, &req_u, 0,
3576                         optname == PACKET_TX_RING);
3577         }
3578         case PACKET_COPY_THRESH:
3579         {
3580                 int val;
3581
3582                 if (optlen != sizeof(val))
3583                         return -EINVAL;
3584                 if (copy_from_user(&val, optval, sizeof(val)))
3585                         return -EFAULT;
3586
3587                 pkt_sk(sk)->copy_thresh = val;
3588                 return 0;
3589         }
3590         case PACKET_VERSION:
3591         {
3592                 int val;
3593
3594                 if (optlen != sizeof(val))
3595                         return -EINVAL;
3596                 if (copy_from_user(&val, optval, sizeof(val)))
3597                         return -EFAULT;
3598                 switch (val) {
3599                 case TPACKET_V1:
3600                 case TPACKET_V2:
3601                 case TPACKET_V3:
3602                         break;
3603                 default:
3604                         return -EINVAL;
3605                 }
3606                 lock_sock(sk);
3607                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3608                         ret = -EBUSY;
3609                 } else {
3610                         po->tp_version = val;
3611                         ret = 0;
3612                 }
3613                 release_sock(sk);
3614                 return ret;
3615         }
3616         case PACKET_RESERVE:
3617         {
3618                 unsigned int val;
3619
3620                 if (optlen != sizeof(val))
3621                         return -EINVAL;
3622                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3623                         return -EBUSY;
3624                 if (copy_from_user(&val, optval, sizeof(val)))
3625                         return -EFAULT;
3626                 po->tp_reserve = val;
3627                 return 0;
3628         }
3629         case PACKET_LOSS:
3630         {
3631                 unsigned int val;
3632
3633                 if (optlen != sizeof(val))
3634                         return -EINVAL;
3635                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3636                         return -EBUSY;
3637                 if (copy_from_user(&val, optval, sizeof(val)))
3638                         return -EFAULT;
3639                 po->tp_loss = !!val;
3640                 return 0;
3641         }
3642         case PACKET_AUXDATA:
3643         {
3644                 int val;
3645
3646                 if (optlen < sizeof(val))
3647                         return -EINVAL;
3648                 if (copy_from_user(&val, optval, sizeof(val)))
3649                         return -EFAULT;
3650
3651                 po->auxdata = !!val;
3652                 return 0;
3653         }
3654         case PACKET_ORIGDEV:
3655         {
3656                 int val;
3657
3658                 if (optlen < sizeof(val))
3659                         return -EINVAL;
3660                 if (copy_from_user(&val, optval, sizeof(val)))
3661                         return -EFAULT;
3662
3663                 po->origdev = !!val;
3664                 return 0;
3665         }
3666         case PACKET_VNET_HDR:
3667         {
3668                 int val;
3669
3670                 if (sock->type != SOCK_RAW)
3671                         return -EINVAL;
3672                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3673                         return -EBUSY;
3674                 if (optlen < sizeof(val))
3675                         return -EINVAL;
3676                 if (copy_from_user(&val, optval, sizeof(val)))
3677                         return -EFAULT;
3678
3679                 po->has_vnet_hdr = !!val;
3680                 return 0;
3681         }
3682         case PACKET_TIMESTAMP:
3683         {
3684                 int val;
3685
3686                 if (optlen != sizeof(val))
3687                         return -EINVAL;
3688                 if (copy_from_user(&val, optval, sizeof(val)))
3689                         return -EFAULT;
3690
3691                 po->tp_tstamp = val;
3692                 return 0;
3693         }
3694         case PACKET_FANOUT:
3695         {
3696                 int val;
3697
3698                 if (optlen != sizeof(val))
3699                         return -EINVAL;
3700                 if (copy_from_user(&val, optval, sizeof(val)))
3701                         return -EFAULT;
3702
3703                 return fanout_add(sk, val & 0xffff, val >> 16);
3704         }
3705         case PACKET_FANOUT_DATA:
3706         {
3707                 if (!po->fanout)
3708                         return -EINVAL;
3709
3710                 return fanout_set_data(po, optval, optlen);
3711         }
3712         case PACKET_TX_HAS_OFF:
3713         {
3714                 unsigned int val;
3715
3716                 if (optlen != sizeof(val))
3717                         return -EINVAL;
3718                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3719                         return -EBUSY;
3720                 if (copy_from_user(&val, optval, sizeof(val)))
3721                         return -EFAULT;
3722                 po->tp_tx_has_off = !!val;
3723                 return 0;
3724         }
3725         case PACKET_QDISC_BYPASS:
3726         {
3727                 int val;
3728
3729                 if (optlen != sizeof(val))
3730                         return -EINVAL;
3731                 if (copy_from_user(&val, optval, sizeof(val)))
3732                         return -EFAULT;
3733
3734                 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3735                 return 0;
3736         }
3737         default:
3738                 return -ENOPROTOOPT;
3739         }
3740 }
3741
3742 static int packet_getsockopt(struct socket *sock, int level, int optname,
3743                              char __user *optval, int __user *optlen)
3744 {
3745         int len;
3746         int val, lv = sizeof(val);
3747         struct sock *sk = sock->sk;
3748         struct packet_sock *po = pkt_sk(sk);
3749         void *data = &val;
3750         union tpacket_stats_u st;
3751         struct tpacket_rollover_stats rstats;
3752
3753         if (level != SOL_PACKET)
3754                 return -ENOPROTOOPT;
3755
3756         if (get_user(len, optlen))
3757                 return -EFAULT;
3758
3759         if (len < 0)
3760                 return -EINVAL;
3761
3762         switch (optname) {
3763         case PACKET_STATISTICS:
3764                 spin_lock_bh(&sk->sk_receive_queue.lock);
3765                 memcpy(&st, &po->stats, sizeof(st));
3766                 memset(&po->stats, 0, sizeof(po->stats));
3767                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3768
3769                 if (po->tp_version == TPACKET_V3) {
3770                         lv = sizeof(struct tpacket_stats_v3);
3771                         st.stats3.tp_packets += st.stats3.tp_drops;
3772                         data = &st.stats3;
3773                 } else {
3774                         lv = sizeof(struct tpacket_stats);
3775                         st.stats1.tp_packets += st.stats1.tp_drops;
3776                         data = &st.stats1;
3777                 }
3778
3779                 break;
3780         case PACKET_AUXDATA:
3781                 val = po->auxdata;
3782                 break;
3783         case PACKET_ORIGDEV:
3784                 val = po->origdev;
3785                 break;
3786         case PACKET_VNET_HDR:
3787                 val = po->has_vnet_hdr;
3788                 break;
3789         case PACKET_VERSION:
3790                 val = po->tp_version;
3791                 break;
3792         case PACKET_HDRLEN:
3793                 if (len > sizeof(int))
3794                         len = sizeof(int);
3795                 if (copy_from_user(&val, optval, len))
3796                         return -EFAULT;
3797                 switch (val) {
3798                 case TPACKET_V1:
3799                         val = sizeof(struct tpacket_hdr);
3800                         break;
3801                 case TPACKET_V2:
3802                         val = sizeof(struct tpacket2_hdr);
3803                         break;
3804                 case TPACKET_V3:
3805                         val = sizeof(struct tpacket3_hdr);
3806                         break;
3807                 default:
3808                         return -EINVAL;
3809                 }
3810                 break;
3811         case PACKET_RESERVE:
3812                 val = po->tp_reserve;
3813                 break;
3814         case PACKET_LOSS:
3815                 val = po->tp_loss;
3816                 break;
3817         case PACKET_TIMESTAMP:
3818                 val = po->tp_tstamp;
3819                 break;
3820         case PACKET_FANOUT:
3821                 val = (po->fanout ?
3822                        ((u32)po->fanout->id |
3823                         ((u32)po->fanout->type << 16) |
3824                         ((u32)po->fanout->flags << 24)) :
3825                        0);
3826                 break;
3827         case PACKET_ROLLOVER_STATS:
3828                 if (!po->rollover)
3829                         return -EINVAL;
3830                 rstats.tp_all = atomic_long_read(&po->rollover->num);
3831                 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3832                 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3833                 data = &rstats;
3834                 lv = sizeof(rstats);
3835                 break;
3836         case PACKET_TX_HAS_OFF:
3837                 val = po->tp_tx_has_off;
3838                 break;
3839         case PACKET_QDISC_BYPASS:
3840                 val = packet_use_direct_xmit(po);
3841                 break;
3842         default:
3843                 return -ENOPROTOOPT;
3844         }
3845
3846         if (len > lv)
3847                 len = lv;
3848         if (put_user(len, optlen))
3849                 return -EFAULT;
3850         if (copy_to_user(optval, data, len))
3851                 return -EFAULT;
3852         return 0;
3853 }
3854
3855
3856 #ifdef CONFIG_COMPAT
3857 static int compat_packet_setsockopt(struct socket *sock, int level, int optname,
3858                                     char __user *optval, unsigned int optlen)
3859 {
3860         struct packet_sock *po = pkt_sk(sock->sk);
3861
3862         if (level != SOL_PACKET)
3863                 return -ENOPROTOOPT;
3864
3865         if (optname == PACKET_FANOUT_DATA &&
3866             po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) {
3867                 optval = (char __user *)get_compat_bpf_fprog(optval);
3868                 if (!optval)
3869                         return -EFAULT;
3870                 optlen = sizeof(struct sock_fprog);
3871         }
3872
3873         return packet_setsockopt(sock, level, optname, optval, optlen);
3874 }
3875 #endif
3876
3877 static int packet_notifier(struct notifier_block *this,
3878                            unsigned long msg, void *ptr)
3879 {
3880         struct sock *sk;
3881         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3882         struct net *net = dev_net(dev);
3883
3884         rcu_read_lock();
3885         sk_for_each_rcu(sk, &net->packet.sklist) {
3886                 struct packet_sock *po = pkt_sk(sk);
3887
3888                 switch (msg) {
3889                 case NETDEV_UNREGISTER:
3890                         if (po->mclist)
3891                                 packet_dev_mclist_delete(dev, &po->mclist);
3892                         /* fallthrough */
3893
3894                 case NETDEV_DOWN:
3895                         if (dev->ifindex == po->ifindex) {
3896                                 spin_lock(&po->bind_lock);
3897                                 if (po->running) {
3898                                         __unregister_prot_hook(sk, false);
3899                                         sk->sk_err = ENETDOWN;
3900                                         if (!sock_flag(sk, SOCK_DEAD))
3901                                                 sk->sk_error_report(sk);
3902                                 }
3903                                 if (msg == NETDEV_UNREGISTER) {
3904                                         packet_cached_dev_reset(po);
3905                                         fanout_release(sk);
3906                                         po->ifindex = -1;
3907                                         if (po->prot_hook.dev)
3908                                                 dev_put(po->prot_hook.dev);
3909                                         po->prot_hook.dev = NULL;
3910                                 }
3911                                 spin_unlock(&po->bind_lock);
3912                         }
3913                         break;
3914                 case NETDEV_UP:
3915                         if (dev->ifindex == po->ifindex) {
3916                                 spin_lock(&po->bind_lock);
3917                                 if (po->num)
3918                                         register_prot_hook(sk);
3919                                 spin_unlock(&po->bind_lock);
3920                         }
3921                         break;
3922                 }
3923         }
3924         rcu_read_unlock();
3925         return NOTIFY_DONE;
3926 }
3927
3928
3929 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3930                         unsigned long arg)
3931 {
3932         struct sock *sk = sock->sk;
3933
3934         switch (cmd) {
3935         case SIOCOUTQ:
3936         {
3937                 int amount = sk_wmem_alloc_get(sk);
3938
3939                 return put_user(amount, (int __user *)arg);
3940         }
3941         case SIOCINQ:
3942         {
3943                 struct sk_buff *skb;
3944                 int amount = 0;
3945
3946                 spin_lock_bh(&sk->sk_receive_queue.lock);
3947                 skb = skb_peek(&sk->sk_receive_queue);
3948                 if (skb)
3949                         amount = skb->len;
3950                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3951                 return put_user(amount, (int __user *)arg);
3952         }
3953         case SIOCGSTAMP:
3954                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3955         case SIOCGSTAMPNS:
3956                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3957
3958 #ifdef CONFIG_INET
3959         case SIOCADDRT:
3960         case SIOCDELRT:
3961         case SIOCDARP:
3962         case SIOCGARP:
3963         case SIOCSARP:
3964         case SIOCGIFADDR:
3965         case SIOCSIFADDR:
3966         case SIOCGIFBRDADDR:
3967         case SIOCSIFBRDADDR:
3968         case SIOCGIFNETMASK:
3969         case SIOCSIFNETMASK:
3970         case SIOCGIFDSTADDR:
3971         case SIOCSIFDSTADDR:
3972         case SIOCSIFFLAGS:
3973                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3974 #endif
3975
3976         default:
3977                 return -ENOIOCTLCMD;
3978         }
3979         return 0;
3980 }
3981
3982 static unsigned int packet_poll(struct file *file, struct socket *sock,
3983                                 poll_table *wait)
3984 {
3985         struct sock *sk = sock->sk;
3986         struct packet_sock *po = pkt_sk(sk);
3987         unsigned int mask = datagram_poll(file, sock, wait);
3988
3989         spin_lock_bh(&sk->sk_receive_queue.lock);
3990         if (po->rx_ring.pg_vec) {
3991                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3992                         TP_STATUS_KERNEL))
3993                         mask |= POLLIN | POLLRDNORM;
3994         }
3995         if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
3996                 po->pressure = 0;
3997         spin_unlock_bh(&sk->sk_receive_queue.lock);
3998         spin_lock_bh(&sk->sk_write_queue.lock);
3999         if (po->tx_ring.pg_vec) {
4000                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4001                         mask |= POLLOUT | POLLWRNORM;
4002         }
4003         spin_unlock_bh(&sk->sk_write_queue.lock);
4004         return mask;
4005 }
4006
4007
4008 /* Dirty? Well, I still did not learn better way to account
4009  * for user mmaps.
4010  */
4011
4012 static void packet_mm_open(struct vm_area_struct *vma)
4013 {
4014         struct file *file = vma->vm_file;
4015         struct socket *sock = file->private_data;
4016         struct sock *sk = sock->sk;
4017
4018         if (sk)
4019                 atomic_inc(&pkt_sk(sk)->mapped);
4020 }
4021
4022 static void packet_mm_close(struct vm_area_struct *vma)
4023 {
4024         struct file *file = vma->vm_file;
4025         struct socket *sock = file->private_data;
4026         struct sock *sk = sock->sk;
4027
4028         if (sk)
4029                 atomic_dec(&pkt_sk(sk)->mapped);
4030 }
4031
4032 static const struct vm_operations_struct packet_mmap_ops = {
4033         .open   =       packet_mm_open,
4034         .close  =       packet_mm_close,
4035 };
4036
4037 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4038                         unsigned int len)
4039 {
4040         int i;
4041
4042         for (i = 0; i < len; i++) {
4043                 if (likely(pg_vec[i].buffer)) {
4044                         if (is_vmalloc_addr(pg_vec[i].buffer))
4045                                 vfree(pg_vec[i].buffer);
4046                         else
4047                                 free_pages((unsigned long)pg_vec[i].buffer,
4048                                            order);
4049                         pg_vec[i].buffer = NULL;
4050                 }
4051         }
4052         kfree(pg_vec);
4053 }
4054
4055 static char *alloc_one_pg_vec_page(unsigned long order)
4056 {
4057         char *buffer;
4058         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4059                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4060
4061         buffer = (char *) __get_free_pages(gfp_flags, order);
4062         if (buffer)
4063                 return buffer;
4064
4065         /* __get_free_pages failed, fall back to vmalloc */
4066         buffer = vzalloc((1 << order) * PAGE_SIZE);
4067         if (buffer)
4068                 return buffer;
4069
4070         /* vmalloc failed, lets dig into swap here */
4071         gfp_flags &= ~__GFP_NORETRY;
4072         buffer = (char *) __get_free_pages(gfp_flags, order);
4073         if (buffer)
4074                 return buffer;
4075
4076         /* complete and utter failure */
4077         return NULL;
4078 }
4079
4080 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4081 {
4082         unsigned int block_nr = req->tp_block_nr;
4083         struct pgv *pg_vec;
4084         int i;
4085
4086         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4087         if (unlikely(!pg_vec))
4088                 goto out;
4089
4090         for (i = 0; i < block_nr; i++) {
4091                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4092                 if (unlikely(!pg_vec[i].buffer))
4093                         goto out_free_pgvec;
4094         }
4095
4096 out:
4097         return pg_vec;
4098
4099 out_free_pgvec:
4100         free_pg_vec(pg_vec, order, block_nr);
4101         pg_vec = NULL;
4102         goto out;
4103 }
4104
4105 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4106                 int closing, int tx_ring)
4107 {
4108         struct pgv *pg_vec = NULL;
4109         struct packet_sock *po = pkt_sk(sk);
4110         int was_running, order = 0;
4111         struct packet_ring_buffer *rb;
4112         struct sk_buff_head *rb_queue;
4113         __be16 num;
4114         int err = -EINVAL;
4115         /* Added to avoid minimal code churn */
4116         struct tpacket_req *req = &req_u->req;
4117
4118         lock_sock(sk);
4119         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
4120         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
4121                 net_warn_ratelimited("Tx-ring is not supported.\n");
4122                 goto out;
4123         }
4124
4125         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4126         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4127
4128         err = -EBUSY;
4129         if (!closing) {
4130                 if (atomic_read(&po->mapped))
4131                         goto out;
4132                 if (packet_read_pending(rb))
4133                         goto out;
4134         }
4135
4136         if (req->tp_block_nr) {
4137                 /* Sanity tests and some calculations */
4138                 err = -EBUSY;
4139                 if (unlikely(rb->pg_vec))
4140                         goto out;
4141
4142                 switch (po->tp_version) {
4143                 case TPACKET_V1:
4144                         po->tp_hdrlen = TPACKET_HDRLEN;
4145                         break;
4146                 case TPACKET_V2:
4147                         po->tp_hdrlen = TPACKET2_HDRLEN;
4148                         break;
4149                 case TPACKET_V3:
4150                         po->tp_hdrlen = TPACKET3_HDRLEN;
4151                         break;
4152                 }
4153
4154                 err = -EINVAL;
4155                 if (unlikely((int)req->tp_block_size <= 0))
4156                         goto out;
4157                 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4158                         goto out;
4159                 if (po->tp_version >= TPACKET_V3 &&
4160                     (int)(req->tp_block_size -
4161                           BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
4162                         goto out;
4163                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
4164                                         po->tp_reserve))
4165                         goto out;
4166                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4167                         goto out;
4168
4169                 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4170                 if (unlikely(rb->frames_per_block == 0))
4171                         goto out;
4172                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4173                                         req->tp_frame_nr))
4174                         goto out;
4175
4176                 err = -ENOMEM;
4177                 order = get_order(req->tp_block_size);
4178                 pg_vec = alloc_pg_vec(req, order);
4179                 if (unlikely(!pg_vec))
4180                         goto out;
4181                 switch (po->tp_version) {
4182                 case TPACKET_V3:
4183                 /* Transmit path is not supported. We checked
4184                  * it above but just being paranoid
4185                  */
4186                         if (!tx_ring)
4187                                 init_prb_bdqc(po, rb, pg_vec, req_u);
4188                         break;
4189                 default:
4190                         break;
4191                 }
4192         }
4193         /* Done */
4194         else {
4195                 err = -EINVAL;
4196                 if (unlikely(req->tp_frame_nr))
4197                         goto out;
4198         }
4199
4200
4201         /* Detach socket from network */
4202         spin_lock(&po->bind_lock);
4203         was_running = po->running;
4204         num = po->num;
4205         if (was_running) {
4206                 po->num = 0;
4207                 __unregister_prot_hook(sk, false);
4208         }
4209         spin_unlock(&po->bind_lock);
4210
4211         synchronize_net();
4212
4213         err = -EBUSY;
4214         mutex_lock(&po->pg_vec_lock);
4215         if (closing || atomic_read(&po->mapped) == 0) {
4216                 err = 0;
4217                 spin_lock_bh(&rb_queue->lock);
4218                 swap(rb->pg_vec, pg_vec);
4219                 rb->frame_max = (req->tp_frame_nr - 1);
4220                 rb->head = 0;
4221                 rb->frame_size = req->tp_frame_size;
4222                 spin_unlock_bh(&rb_queue->lock);
4223
4224                 swap(rb->pg_vec_order, order);
4225                 swap(rb->pg_vec_len, req->tp_block_nr);
4226
4227                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4228                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4229                                                 tpacket_rcv : packet_rcv;
4230                 skb_queue_purge(rb_queue);
4231                 if (atomic_read(&po->mapped))
4232                         pr_err("packet_mmap: vma is busy: %d\n",
4233                                atomic_read(&po->mapped));
4234         }
4235         mutex_unlock(&po->pg_vec_lock);
4236
4237         spin_lock(&po->bind_lock);
4238         if (was_running) {
4239                 po->num = num;
4240                 register_prot_hook(sk);
4241         }
4242         spin_unlock(&po->bind_lock);
4243         if (closing && (po->tp_version > TPACKET_V2)) {
4244                 /* Because we don't support block-based V3 on tx-ring */
4245                 if (!tx_ring)
4246                         prb_shutdown_retire_blk_timer(po, rb_queue);
4247         }
4248
4249         if (pg_vec)
4250                 free_pg_vec(pg_vec, order, req->tp_block_nr);
4251 out:
4252         release_sock(sk);
4253         return err;
4254 }
4255
4256 static int packet_mmap(struct file *file, struct socket *sock,
4257                 struct vm_area_struct *vma)
4258 {
4259         struct sock *sk = sock->sk;
4260         struct packet_sock *po = pkt_sk(sk);
4261         unsigned long size, expected_size;
4262         struct packet_ring_buffer *rb;
4263         unsigned long start;
4264         int err = -EINVAL;
4265         int i;
4266
4267         if (vma->vm_pgoff)
4268                 return -EINVAL;
4269
4270         mutex_lock(&po->pg_vec_lock);
4271
4272         expected_size = 0;
4273         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4274                 if (rb->pg_vec) {
4275                         expected_size += rb->pg_vec_len
4276                                                 * rb->pg_vec_pages
4277                                                 * PAGE_SIZE;
4278                 }
4279         }
4280
4281         if (expected_size == 0)
4282                 goto out;
4283
4284         size = vma->vm_end - vma->vm_start;
4285         if (size != expected_size)
4286                 goto out;
4287
4288         start = vma->vm_start;
4289         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4290                 if (rb->pg_vec == NULL)
4291                         continue;
4292
4293                 for (i = 0; i < rb->pg_vec_len; i++) {
4294                         struct page *page;
4295                         void *kaddr = rb->pg_vec[i].buffer;
4296                         int pg_num;
4297
4298                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4299                                 page = pgv_to_page(kaddr);
4300                                 err = vm_insert_page(vma, start, page);
4301                                 if (unlikely(err))
4302                                         goto out;
4303                                 start += PAGE_SIZE;
4304                                 kaddr += PAGE_SIZE;
4305                         }
4306                 }
4307         }
4308
4309         atomic_inc(&po->mapped);
4310         vma->vm_ops = &packet_mmap_ops;
4311         err = 0;
4312
4313 out:
4314         mutex_unlock(&po->pg_vec_lock);
4315         return err;
4316 }
4317
4318 static const struct proto_ops packet_ops_spkt = {
4319         .family =       PF_PACKET,
4320         .owner =        THIS_MODULE,
4321         .release =      packet_release,
4322         .bind =         packet_bind_spkt,
4323         .connect =      sock_no_connect,
4324         .socketpair =   sock_no_socketpair,
4325         .accept =       sock_no_accept,
4326         .getname =      packet_getname_spkt,
4327         .poll =         datagram_poll,
4328         .ioctl =        packet_ioctl,
4329         .listen =       sock_no_listen,
4330         .shutdown =     sock_no_shutdown,
4331         .setsockopt =   sock_no_setsockopt,
4332         .getsockopt =   sock_no_getsockopt,
4333         .sendmsg =      packet_sendmsg_spkt,
4334         .recvmsg =      packet_recvmsg,
4335         .mmap =         sock_no_mmap,
4336         .sendpage =     sock_no_sendpage,
4337 };
4338
4339 static const struct proto_ops packet_ops = {
4340         .family =       PF_PACKET,
4341         .owner =        THIS_MODULE,
4342         .release =      packet_release,
4343         .bind =         packet_bind,
4344         .connect =      sock_no_connect,
4345         .socketpair =   sock_no_socketpair,
4346         .accept =       sock_no_accept,
4347         .getname =      packet_getname,
4348         .poll =         packet_poll,
4349         .ioctl =        packet_ioctl,
4350         .listen =       sock_no_listen,
4351         .shutdown =     sock_no_shutdown,
4352         .setsockopt =   packet_setsockopt,
4353         .getsockopt =   packet_getsockopt,
4354 #ifdef CONFIG_COMPAT
4355         .compat_setsockopt = compat_packet_setsockopt,
4356 #endif
4357         .sendmsg =      packet_sendmsg,
4358         .recvmsg =      packet_recvmsg,
4359         .mmap =         packet_mmap,
4360         .sendpage =     sock_no_sendpage,
4361 };
4362
4363 static const struct net_proto_family packet_family_ops = {
4364         .family =       PF_PACKET,
4365         .create =       packet_create,
4366         .owner  =       THIS_MODULE,
4367 };
4368
4369 static struct notifier_block packet_netdev_notifier = {
4370         .notifier_call =        packet_notifier,
4371 };
4372
4373 #ifdef CONFIG_PROC_FS
4374
4375 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4376         __acquires(RCU)
4377 {
4378         struct net *net = seq_file_net(seq);
4379
4380         rcu_read_lock();
4381         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4382 }
4383
4384 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4385 {
4386         struct net *net = seq_file_net(seq);
4387         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4388 }
4389
4390 static void packet_seq_stop(struct seq_file *seq, void *v)
4391         __releases(RCU)
4392 {
4393         rcu_read_unlock();
4394 }
4395
4396 static int packet_seq_show(struct seq_file *seq, void *v)
4397 {
4398         if (v == SEQ_START_TOKEN)
4399                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
4400         else {
4401                 struct sock *s = sk_entry(v);
4402                 const struct packet_sock *po = pkt_sk(s);
4403
4404                 seq_printf(seq,
4405                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4406                            s,
4407                            atomic_read(&s->sk_refcnt),
4408                            s->sk_type,
4409                            ntohs(po->num),
4410                            po->ifindex,
4411                            po->running,
4412                            atomic_read(&s->sk_rmem_alloc),
4413                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4414                            sock_i_ino(s));
4415         }
4416
4417         return 0;
4418 }
4419
4420 static const struct seq_operations packet_seq_ops = {
4421         .start  = packet_seq_start,
4422         .next   = packet_seq_next,
4423         .stop   = packet_seq_stop,
4424         .show   = packet_seq_show,
4425 };
4426
4427 static int packet_seq_open(struct inode *inode, struct file *file)
4428 {
4429         return seq_open_net(inode, file, &packet_seq_ops,
4430                             sizeof(struct seq_net_private));
4431 }
4432
4433 static const struct file_operations packet_seq_fops = {
4434         .owner          = THIS_MODULE,
4435         .open           = packet_seq_open,
4436         .read           = seq_read,
4437         .llseek         = seq_lseek,
4438         .release        = seq_release_net,
4439 };
4440
4441 #endif
4442
4443 static int __net_init packet_net_init(struct net *net)
4444 {
4445         mutex_init(&net->packet.sklist_lock);
4446         INIT_HLIST_HEAD(&net->packet.sklist);
4447
4448         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4449                 return -ENOMEM;
4450
4451         return 0;
4452 }
4453
4454 static void __net_exit packet_net_exit(struct net *net)
4455 {
4456         remove_proc_entry("packet", net->proc_net);
4457 }
4458
4459 static struct pernet_operations packet_net_ops = {
4460         .init = packet_net_init,
4461         .exit = packet_net_exit,
4462 };
4463
4464
4465 static void __exit packet_exit(void)
4466 {
4467         unregister_netdevice_notifier(&packet_netdev_notifier);
4468         unregister_pernet_subsys(&packet_net_ops);
4469         sock_unregister(PF_PACKET);
4470         proto_unregister(&packet_proto);
4471 }
4472
4473 static int __init packet_init(void)
4474 {
4475         int rc = proto_register(&packet_proto, 0);
4476
4477         if (rc != 0)
4478                 goto out;
4479
4480         sock_register(&packet_family_ops);
4481         register_pernet_subsys(&packet_net_ops);
4482         register_netdevice_notifier(&packet_netdev_notifier);
4483 out:
4484         return rc;
4485 }
4486
4487 module_init(packet_init);
4488 module_exit(packet_exit);
4489 MODULE_LICENSE("GPL");
4490 MODULE_ALIAS_NETPROTO(PF_PACKET);