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
  97 #include "internal.h"
  98
  99 /*
 100    Assumptions:
 101    - if device has no dev->hard_header routine, it adds and removes ll header
 102      inside itself. In this case ll header is invisible outside of device,
 103      but higher levels still should reserve dev->hard_header_len.
 104      Some devices are enough clever to reallocate skb, when header
 105      will not fit to reserved space (tunnel), another ones are silly
 106      (PPP).
 107    - packet socket receives packets with pulled ll header,
 108      so that SOCK_RAW should push it back.
 109
 110 On receive:
 111 -----------
 112
 113 Incoming, dev->hard_header!=NULL
 114    mac_header -> ll header
 115    data       -> data
 116
 117 Outgoing, dev->hard_header!=NULL
 118    mac_header -> ll header
 119    data       -> ll header
 120
 121 Incoming, dev->hard_header==NULL
 122    mac_header -> UNKNOWN position. It is very likely, that it points to ll
 123                  header.  PPP makes it, that is wrong, because introduce
 124                  assymetry between rx and tx paths.
 125    data       -> data
 126
 127 Outgoing, dev->hard_header==NULL
 128    mac_header -> data. ll header is still not built!
 129    data       -> data
 130
 131 Resume
 132   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 133
 134
 135 On transmit:
 136 ------------
 137
 138 dev->hard_header != NULL
 139    mac_header -> ll header
 140    data       -> ll header
 141
 142 dev->hard_header == NULL (ll header is added by device, we cannot control it)
 143    mac_header -> data
 144    data       -> data
 145
 146    We should set nh.raw on output to correct posistion,
 147    packet classifier depends on it.
 148  */
 149
 150 /* Private packet socket structures. */
 151
 152 /* identical to struct packet_mreq except it has
 153  * a longer address field.
 154  */
 155 struct packet_mreq_max {
 156         int             mr_ifindex;
 157         unsigned short  mr_type;
 158         unsigned short  mr_alen;
 159         unsigned char   mr_address[MAX_ADDR_LEN];
 160 };
 161
 162 union tpacket_uhdr {
 163         struct tpacket_hdr  *h1;
 164         struct tpacket2_hdr *h2;
 165         struct tpacket3_hdr *h3;
 166         void *raw;
 167 };
 168
 169 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 170                 int closing, int tx_ring);
 171
 172 #define V3_ALIGNMENT    (8)
 173
 174 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 175
 176 #define BLK_PLUS_PRIV(sz_of_priv) \
 177         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 178
 179 #define PGV_FROM_VMALLOC 1
 180
 181 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
 182 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
 183 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
 184 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
 185 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
 186 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
 187 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
 188
 189 struct packet_sock;
 190 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
 191 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 192                        struct packet_type *pt, struct net_device *orig_dev);
 193
 194 static void *packet_previous_frame(struct packet_sock *po,
 195                 struct packet_ring_buffer *rb,
 196                 int status);
 197 static void packet_increment_head(struct packet_ring_buffer *buff);
 198 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
 199                         struct tpacket_block_desc *);
 200 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 201                         struct packet_sock *);
 202 static void prb_retire_current_block(struct tpacket_kbdq_core *,
 203                 struct packet_sock *, unsigned int status);
 204 static int prb_queue_frozen(struct tpacket_kbdq_core *);
 205 static void prb_open_block(struct tpacket_kbdq_core *,
 206                 struct tpacket_block_desc *);
 207 static void prb_retire_rx_blk_timer_expired(unsigned long);
 208 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 209 static void prb_init_blk_timer(struct packet_sock *,
 210                 struct tpacket_kbdq_core *,
 211                 void (*func) (unsigned long));
 212 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 213 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 214                 struct tpacket3_hdr *);
 215 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 216                 struct tpacket3_hdr *);
 217 static void packet_flush_mclist(struct sock *sk);
 218
 219 struct packet_skb_cb {
 220         union {
 221                 struct sockaddr_pkt pkt;
 222                 union {
 223                         /* Trick: alias skb original length with
 224                          * ll.sll_family and ll.protocol in order
 225                          * to save room.
 226                          */
 227                         unsigned int origlen;
 228                         struct sockaddr_ll ll;
 229                 };
 230         } sa;
 231 };
 232
 233 #define vio_le() virtio_legacy_is_little_endian()
 234
 235 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 236
 237 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 238 #define GET_PBLOCK_DESC(x, bid) \
 239         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 240 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 241         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 242 #define GET_NEXT_PRB_BLK_NUM(x) \
 243         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 244         ((x)->kactive_blk_num+1) : 0)
 245
 246 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 247 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 248
 249 static int packet_direct_xmit(struct sk_buff *skb)
 250 {
 251         struct net_device *dev = skb->dev;
 252         struct sk_buff *orig_skb = skb;
 253         struct netdev_queue *txq;
 254         int ret = NETDEV_TX_BUSY;
 255
 256         if (unlikely(!netif_running(dev) ||
 257                      !netif_carrier_ok(dev)))
 258                 goto drop;
 259
 260         skb = validate_xmit_skb_list(skb, dev);
 261         if (skb != orig_skb)
 262                 goto drop;
 263
 264         txq = skb_get_tx_queue(dev, skb);
 265
 266         local_bh_disable();
 267
 268         HARD_TX_LOCK(dev, txq, smp_processor_id());
 269         if (!netif_xmit_frozen_or_drv_stopped(txq))
 270                 ret = netdev_start_xmit(skb, dev, txq, false);
 271         HARD_TX_UNLOCK(dev, txq);
 272
 273         local_bh_enable();
 274
 275         if (!dev_xmit_complete(ret))
 276                 kfree_skb(skb);
 277
 278         return ret;
 279 drop:
 280         atomic_long_inc(&dev->tx_dropped);
 281         kfree_skb_list(skb);
 282         return NET_XMIT_DROP;
 283 }
 284
 285 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
 286 {
 287         struct net_device *dev;
 288
 289         rcu_read_lock();
 290         dev = rcu_dereference(po->cached_dev);
 291         if (likely(dev))
 292                 dev_hold(dev);
 293         rcu_read_unlock();
 294
 295         return dev;
 296 }
 297
 298 static void packet_cached_dev_assign(struct packet_sock *po,
 299                                      struct net_device *dev)
 300 {
 301         rcu_assign_pointer(po->cached_dev, dev);
 302 }
 303
 304 static void packet_cached_dev_reset(struct packet_sock *po)
 305 {
 306         RCU_INIT_POINTER(po->cached_dev, NULL);
 307 }
 308
 309 static bool packet_use_direct_xmit(const struct packet_sock *po)
 310 {
 311         return po->xmit == packet_direct_xmit;
 312 }
 313
 314 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
 315 {
 316         return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
 317 }
 318
 319 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
 320 {
 321         const struct net_device_ops *ops = dev->netdev_ops;
 322         u16 queue_index;
 323
 324         if (ops->ndo_select_queue) {
 325                 queue_index = ops->ndo_select_queue(dev, skb, NULL,
 326                                                     __packet_pick_tx_queue);
 327                 queue_index = netdev_cap_txqueue(dev, queue_index);
 328         } else {
 329                 queue_index = __packet_pick_tx_queue(dev, skb);
 330         }
 331
 332         skb_set_queue_mapping(skb, queue_index);
 333 }
 334
 335 /* __register_prot_hook must be invoked through register_prot_hook
 336  * or from a context in which asynchronous accesses to the packet
 337  * socket is not possible (packet_create()).
 338  */
 339 static void __register_prot_hook(struct sock *sk)
 340 {
 341         struct packet_sock *po = pkt_sk(sk);
 342
 343         if (!po->running) {
 344                 if (po->fanout)
 345                         __fanout_link(sk, po);
 346                 else
 347                         dev_add_pack(&po->prot_hook);
 348
 349                 sock_hold(sk);
 350                 po->running = 1;
 351         }
 352 }
 353
 354 static void register_prot_hook(struct sock *sk)
 355 {
 356         lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
 357         __register_prot_hook(sk);
 358 }
 359
 360 /* If the sync parameter is true, we will temporarily drop
 361  * the po->bind_lock and do a synchronize_net to make sure no
 362  * asynchronous packet processing paths still refer to the elements
 363  * of po->prot_hook.  If the sync parameter is false, it is the
 364  * callers responsibility to take care of this.
 365  */
 366 static void __unregister_prot_hook(struct sock *sk, bool sync)
 367 {
 368         struct packet_sock *po = pkt_sk(sk);
 369
 370         lockdep_assert_held_once(&po->bind_lock);
 371
 372         po->running = 0;
 373
 374         if (po->fanout)
 375                 __fanout_unlink(sk, po);
 376         else
 377                 __dev_remove_pack(&po->prot_hook);
 378
 379         __sock_put(sk);
 380
 381         if (sync) {
 382                 spin_unlock(&po->bind_lock);
 383                 synchronize_net();
 384                 spin_lock(&po->bind_lock);
 385         }
 386 }
 387
 388 static void unregister_prot_hook(struct sock *sk, bool sync)
 389 {
 390         struct packet_sock *po = pkt_sk(sk);
 391
 392         if (po->running)
 393                 __unregister_prot_hook(sk, sync);
 394 }
 395
 396 static inline struct page * __pure pgv_to_page(void *addr)
 397 {
 398         if (is_vmalloc_addr(addr))
 399                 return vmalloc_to_page(addr);
 400         return virt_to_page(addr);
 401 }
 402
 403 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 404 {
 405         union tpacket_uhdr h;
 406
 407         h.raw = frame;
 408         switch (po->tp_version) {
 409         case TPACKET_V1:
 410                 h.h1->tp_status = status;
 411                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 412                 break;
 413         case TPACKET_V2:
 414                 h.h2->tp_status = status;
 415                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 416                 break;
 417         case TPACKET_V3:
 418         default:
 419                 WARN(1, "TPACKET version not supported.\n");
 420                 BUG();
 421         }
 422
 423         smp_wmb();
 424 }
 425
 426 static int __packet_get_status(struct packet_sock *po, void *frame)
 427 {
 428         union tpacket_uhdr h;
 429
 430         smp_rmb();
 431
 432         h.raw = frame;
 433         switch (po->tp_version) {
 434         case TPACKET_V1:
 435                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 436                 return h.h1->tp_status;
 437         case TPACKET_V2:
 438                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 439                 return h.h2->tp_status;
 440         case TPACKET_V3:
 441         default:
 442                 WARN(1, "TPACKET version not supported.\n");
 443                 BUG();
 444                 return 0;
 445         }
 446 }
 447
 448 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
 449                                    unsigned int flags)
 450 {
 451         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 452
 453         if (shhwtstamps &&
 454             (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
 455             ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
 456                 return TP_STATUS_TS_RAW_HARDWARE;
 457
 458         if (ktime_to_timespec_cond(skb->tstamp, ts))
 459                 return TP_STATUS_TS_SOFTWARE;
 460
 461         return 0;
 462 }
 463
 464 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
 465                                     struct sk_buff *skb)
 466 {
 467         union tpacket_uhdr h;
 468         struct timespec ts;
 469         __u32 ts_status;
 470
 471         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
 472                 return 0;
 473
 474         h.raw = frame;
 475         switch (po->tp_version) {
 476         case TPACKET_V1:
 477                 h.h1->tp_sec = ts.tv_sec;
 478                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
 479                 break;
 480         case TPACKET_V2:
 481                 h.h2->tp_sec = ts.tv_sec;
 482                 h.h2->tp_nsec = ts.tv_nsec;
 483                 break;
 484         case TPACKET_V3:
 485         default:
 486                 WARN(1, "TPACKET version not supported.\n");
 487                 BUG();
 488         }
 489
 490         /* one flush is safe, as both fields always lie on the same cacheline */
 491         flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
 492         smp_wmb();
 493
 494         return ts_status;
 495 }
 496
 497 static void *packet_lookup_frame(struct packet_sock *po,
 498                 struct packet_ring_buffer *rb,
 499                 unsigned int position,
 500                 int status)
 501 {
 502         unsigned int pg_vec_pos, frame_offset;
 503         union tpacket_uhdr h;
 504
 505         pg_vec_pos = position / rb->frames_per_block;
 506         frame_offset = position % rb->frames_per_block;
 507
 508         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 509                 (frame_offset * rb->frame_size);
 510
 511         if (status != __packet_get_status(po, h.raw))
 512                 return NULL;
 513
 514         return h.raw;
 515 }
 516
 517 static void *packet_current_frame(struct packet_sock *po,
 518                 struct packet_ring_buffer *rb,
 519                 int status)
 520 {
 521         return packet_lookup_frame(po, rb, rb->head, status);
 522 }
 523
 524 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 525 {
 526         del_timer_sync(&pkc->retire_blk_timer);
 527 }
 528
 529 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 530                 struct sk_buff_head *rb_queue)
 531 {
 532         struct tpacket_kbdq_core *pkc;
 533
 534         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 535
 536         spin_lock_bh(&rb_queue->lock);
 537         pkc->delete_blk_timer = 1;
 538         spin_unlock_bh(&rb_queue->lock);
 539
 540         prb_del_retire_blk_timer(pkc);
 541 }
 542
 543 static void prb_init_blk_timer(struct packet_sock *po,
 544                 struct tpacket_kbdq_core *pkc,
 545                 void (*func) (unsigned long))
 546 {
 547         init_timer(&pkc->retire_blk_timer);
 548         pkc->retire_blk_timer.data = (long)po;
 549         pkc->retire_blk_timer.function = func;
 550         pkc->retire_blk_timer.expires = jiffies;
 551 }
 552
 553 static void prb_setup_retire_blk_timer(struct packet_sock *po)
 554 {
 555         struct tpacket_kbdq_core *pkc;
 556
 557         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 558         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
 559 }
 560
 561 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 562                                 int blk_size_in_bytes)
 563 {
 564         struct net_device *dev;
 565         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 566         struct ethtool_cmd ecmd;
 567         int err;
 568         u32 speed;
 569
 570         rtnl_lock();
 571         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 572         if (unlikely(!dev)) {
 573                 rtnl_unlock();
 574                 return DEFAULT_PRB_RETIRE_TOV;
 575         }
 576         err = __ethtool_get_settings(dev, &ecmd);
 577         speed = ethtool_cmd_speed(&ecmd);
 578         rtnl_unlock();
 579         if (!err) {
 580                 /*
 581                  * If the link speed is so slow you don't really
 582                  * need to worry about perf anyways
 583                  */
 584                 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
 585                         return DEFAULT_PRB_RETIRE_TOV;
 586                 } else {
 587                         msec = 1;
 588                         div = speed / 1000;
 589                 }
 590         }
 591
 592         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 593
 594         if (div)
 595                 mbits /= div;
 596
 597         tmo = mbits * msec;
 598
 599         if (div)
 600                 return tmo+1;
 601         return tmo;
 602 }
 603
 604 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 605                         union tpacket_req_u *req_u)
 606 {
 607         p1->feature_req_word = req_u->req3.tp_feature_req_word;
 608 }
 609
 610 static void init_prb_bdqc(struct packet_sock *po,
 611                         struct packet_ring_buffer *rb,
 612                         struct pgv *pg_vec,
 613                         union tpacket_req_u *req_u)
 614 {
 615         struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
 616         struct tpacket_block_desc *pbd;
 617
 618         memset(p1, 0x0, sizeof(*p1));
 619
 620         p1->knxt_seq_num = 1;
 621         p1->pkbdq = pg_vec;
 622         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 623         p1->pkblk_start = pg_vec[0].buffer;
 624         p1->kblk_size = req_u->req3.tp_block_size;
 625         p1->knum_blocks = req_u->req3.tp_block_nr;
 626         p1->hdrlen = po->tp_hdrlen;
 627         p1->version = po->tp_version;
 628         p1->last_kactive_blk_num = 0;
 629         po->stats.stats3.tp_freeze_q_cnt = 0;
 630         if (req_u->req3.tp_retire_blk_tov)
 631                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 632         else
 633                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 634                                                 req_u->req3.tp_block_size);
 635         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 636         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 637
 638         p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
 639         prb_init_ft_ops(p1, req_u);
 640         prb_setup_retire_blk_timer(po);
 641         prb_open_block(p1, pbd);
 642 }
 643
 644 /*  Do NOT update the last_blk_num first.
 645  *  Assumes sk_buff_head lock is held.
 646  */
 647 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 648 {
 649         mod_timer(&pkc->retire_blk_timer,
 650                         jiffies + pkc->tov_in_jiffies);
 651         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 652 }
 653
 654 /*
 655  * Timer logic:
 656  * 1) We refresh the timer only when we open a block.
 657  *    By doing this we don't waste cycles refreshing the timer
 658  *        on packet-by-packet basis.
 659  *
 660  * With a 1MB block-size, on a 1Gbps line, it will take
 661  * i) ~8 ms to fill a block + ii) memcpy etc.
 662  * In this cut we are not accounting for the memcpy time.
 663  *
 664  * So, if the user sets the 'tmo' to 10ms then the timer
 665  * will never fire while the block is still getting filled
 666  * (which is what we want). However, the user could choose
 667  * to close a block early and that's fine.
 668  *
 669  * But when the timer does fire, we check whether or not to refresh it.
 670  * Since the tmo granularity is in msecs, it is not too expensive
 671  * to refresh the timer, lets say every '8' msecs.
 672  * Either the user can set the 'tmo' or we can derive it based on
 673  * a) line-speed and b) block-size.
 674  * prb_calc_retire_blk_tmo() calculates the tmo.
 675  *
 676  */
 677 static void prb_retire_rx_blk_timer_expired(unsigned long data)
 678 {
 679         struct packet_sock *po = (struct packet_sock *)data;
 680         struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 681         unsigned int frozen;
 682         struct tpacket_block_desc *pbd;
 683
 684         spin_lock(&po->sk.sk_receive_queue.lock);
 685
 686         frozen = prb_queue_frozen(pkc);
 687         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 688
 689         if (unlikely(pkc->delete_blk_timer))
 690                 goto out;
 691
 692         /* We only need to plug the race when the block is partially filled.
 693          * tpacket_rcv:
 694          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 695          *              copy_bits() is in progress ...
 696          *              timer fires on other cpu:
 697          *              we can't retire the current block because copy_bits
 698          *              is in progress.
 699          *
 700          */
 701         if (BLOCK_NUM_PKTS(pbd)) {
 702                 while (atomic_read(&pkc->blk_fill_in_prog)) {
 703                         /* Waiting for skb_copy_bits to finish... */
 704                         cpu_relax();
 705                 }
 706         }
 707
 708         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 709                 if (!frozen) {
 710                         if (!BLOCK_NUM_PKTS(pbd)) {
 711                                 /* An empty block. Just refresh the timer. */
 712                                 goto refresh_timer;
 713                         }
 714                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 715                         if (!prb_dispatch_next_block(pkc, po))
 716                                 goto refresh_timer;
 717                         else
 718                                 goto out;
 719                 } else {
 720                         /* Case 1. Queue was frozen because user-space was
 721                          *         lagging behind.
 722                          */
 723                         if (prb_curr_blk_in_use(pkc, pbd)) {
 724                                 /*
 725                                  * Ok, user-space is still behind.
 726                                  * So just refresh the timer.
 727                                  */
 728                                 goto refresh_timer;
 729                         } else {
 730                                /* Case 2. queue was frozen,user-space caught up,
 731                                 * now the link went idle && the timer fired.
 732                                 * We don't have a block to close.So we open this
 733                                 * block and restart the timer.
 734                                 * opening a block thaws the queue,restarts timer
 735                                 * Thawing/timer-refresh is a side effect.
 736                                 */
 737                                 prb_open_block(pkc, pbd);
 738                                 goto out;
 739                         }
 740                 }
 741         }
 742
 743 refresh_timer:
 744         _prb_refresh_rx_retire_blk_timer(pkc);
 745
 746 out:
 747         spin_unlock(&po->sk.sk_receive_queue.lock);
 748 }
 749
 750 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 751                 struct tpacket_block_desc *pbd1, __u32 status)
 752 {
 753         /* Flush everything minus the block header */
 754
 755 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 756         u8 *start, *end;
 757
 758         start = (u8 *)pbd1;
 759
 760         /* Skip the block header(we know header WILL fit in 4K) */
 761         start += PAGE_SIZE;
 762
 763         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 764         for (; start < end; start += PAGE_SIZE)
 765                 flush_dcache_page(pgv_to_page(start));
 766
 767         smp_wmb();
 768 #endif
 769
 770         /* Now update the block status. */
 771
 772         BLOCK_STATUS(pbd1) = status;
 773
 774         /* Flush the block header */
 775
 776 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 777         start = (u8 *)pbd1;
 778         flush_dcache_page(pgv_to_page(start));
 779
 780         smp_wmb();
 781 #endif
 782 }
 783
 784 /*
 785  * Side effect:
 786  *
 787  * 1) flush the block
 788  * 2) Increment active_blk_num
 789  *
 790  * Note:We DONT refresh the timer on purpose.
 791  *      Because almost always the next block will be opened.
 792  */
 793 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 794                 struct tpacket_block_desc *pbd1,
 795                 struct packet_sock *po, unsigned int stat)
 796 {
 797         __u32 status = TP_STATUS_USER | stat;
 798
 799         struct tpacket3_hdr *last_pkt;
 800         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 801         struct sock *sk = &po->sk;
 802
 803         if (po->stats.stats3.tp_drops)
 804                 status |= TP_STATUS_LOSING;
 805
 806         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 807         last_pkt->tp_next_offset = 0;
 808
 809         /* Get the ts of the last pkt */
 810         if (BLOCK_NUM_PKTS(pbd1)) {
 811                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 812                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 813         } else {
 814                 /* Ok, we tmo'd - so get the current time.
 815                  *
 816                  * It shouldn't really happen as we don't close empty
 817                  * blocks. See prb_retire_rx_blk_timer_expired().
 818                  */
 819                 struct timespec ts;
 820                 getnstimeofday(&ts);
 821                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 822                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 823         }
 824
 825         smp_wmb();
 826
 827         /* Flush the block */
 828         prb_flush_block(pkc1, pbd1, status);
 829
 830         sk->sk_data_ready(sk);
 831
 832         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 833 }
 834
 835 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 836 {
 837         pkc->reset_pending_on_curr_blk = 0;
 838 }
 839
 840 /*
 841  * Side effect of opening a block:
 842  *
 843  * 1) prb_queue is thawed.
 844  * 2) retire_blk_timer is refreshed.
 845  *
 846  */
 847 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 848         struct tpacket_block_desc *pbd1)
 849 {
 850         struct timespec ts;
 851         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 852
 853         smp_rmb();
 854
 855         /* We could have just memset this but we will lose the
 856          * flexibility of making the priv area sticky
 857          */
 858
 859         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 860         BLOCK_NUM_PKTS(pbd1) = 0;
 861         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 862
 863         getnstimeofday(&ts);
 864
 865         h1->ts_first_pkt.ts_sec = ts.tv_sec;
 866         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 867
 868         pkc1->pkblk_start = (char *)pbd1;
 869         pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 870
 871         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 872         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 873
 874         pbd1->version = pkc1->version;
 875         pkc1->prev = pkc1->nxt_offset;
 876         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 877
 878         prb_thaw_queue(pkc1);
 879         _prb_refresh_rx_retire_blk_timer(pkc1);
 880
 881         smp_wmb();
 882 }
 883
 884 /*
 885  * Queue freeze logic:
 886  * 1) Assume tp_block_nr = 8 blocks.
 887  * 2) At time 't0', user opens Rx ring.
 888  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 889  * 4) user-space is either sleeping or processing block '0'.
 890  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 891  *    it will close block-7,loop around and try to fill block '0'.
 892  *    call-flow:
 893  *    __packet_lookup_frame_in_block
 894  *      prb_retire_current_block()
 895  *      prb_dispatch_next_block()
 896  *        |->(BLOCK_STATUS == USER) evaluates to true
 897  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 898  * 6) Now there are two cases:
 899  *    6.1) Link goes idle right after the queue is frozen.
 900  *         But remember, the last open_block() refreshed the timer.
 901  *         When this timer expires,it will refresh itself so that we can
 902  *         re-open block-0 in near future.
 903  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 904  *         case and __packet_lookup_frame_in_block will check if block-0
 905  *         is free and can now be re-used.
 906  */
 907 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 908                                   struct packet_sock *po)
 909 {
 910         pkc->reset_pending_on_curr_blk = 1;
 911         po->stats.stats3.tp_freeze_q_cnt++;
 912 }
 913
 914 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 915
 916 /*
 917  * If the next block is free then we will dispatch it
 918  * and return a good offset.
 919  * Else, we will freeze the queue.
 920  * So, caller must check the return value.
 921  */
 922 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 923                 struct packet_sock *po)
 924 {
 925         struct tpacket_block_desc *pbd;
 926
 927         smp_rmb();
 928
 929         /* 1. Get current block num */
 930         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 931
 932         /* 2. If this block is currently in_use then freeze the queue */
 933         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 934                 prb_freeze_queue(pkc, po);
 935                 return NULL;
 936         }
 937
 938         /*
 939          * 3.
 940          * open this block and return the offset where the first packet
 941          * needs to get stored.
 942          */
 943         prb_open_block(pkc, pbd);
 944         return (void *)pkc->nxt_offset;
 945 }
 946
 947 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 948                 struct packet_sock *po, unsigned int status)
 949 {
 950         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 951
 952         /* retire/close the current block */
 953         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 954                 /*
 955                  * Plug the case where copy_bits() is in progress on
 956                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 957                  * have space to copy the pkt in the current block and
 958                  * called prb_retire_current_block()
 959                  *
 960                  * We don't need to worry about the TMO case because
 961                  * the timer-handler already handled this case.
 962                  */
 963                 if (!(status & TP_STATUS_BLK_TMO)) {
 964                         while (atomic_read(&pkc->blk_fill_in_prog)) {
 965                                 /* Waiting for skb_copy_bits to finish... */
 966                                 cpu_relax();
 967                         }
 968                 }
 969                 prb_close_block(pkc, pbd, po, status);
 970                 return;
 971         }
 972 }
 973
 974 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
 975                                       struct tpacket_block_desc *pbd)
 976 {
 977         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 978 }
 979
 980 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 981 {
 982         return pkc->reset_pending_on_curr_blk;
 983 }
 984
 985 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 986 {
 987         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 988         atomic_dec(&pkc->blk_fill_in_prog);
 989 }
 990
 991 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 992                         struct tpacket3_hdr *ppd)
 993 {
 994         ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
 995 }
 996
 997 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 998                         struct tpacket3_hdr *ppd)
 999 {
1000         ppd->hv1.tp_rxhash = 0;
1001 }
1002
1003 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1004                         struct tpacket3_hdr *ppd)
1005 {
1006         if (skb_vlan_tag_present(pkc->skb)) {
1007                 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1008                 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1009                 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1010         } else {
1011                 ppd->hv1.tp_vlan_tci = 0;
1012                 ppd->hv1.tp_vlan_tpid = 0;
1013                 ppd->tp_status = TP_STATUS_AVAILABLE;
1014         }
1015 }
1016
1017 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1018                         struct tpacket3_hdr *ppd)
1019 {
1020         ppd->hv1.tp_padding = 0;
1021         prb_fill_vlan_info(pkc, ppd);
1022
1023         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1024                 prb_fill_rxhash(pkc, ppd);
1025         else
1026                 prb_clear_rxhash(pkc, ppd);
1027 }
1028
1029 static void prb_fill_curr_block(char *curr,
1030                                 struct tpacket_kbdq_core *pkc,
1031                                 struct tpacket_block_desc *pbd,
1032                                 unsigned int len)
1033 {
1034         struct tpacket3_hdr *ppd;
1035
1036         ppd  = (struct tpacket3_hdr *)curr;
1037         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1038         pkc->prev = curr;
1039         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1040         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1041         BLOCK_NUM_PKTS(pbd) += 1;
1042         atomic_inc(&pkc->blk_fill_in_prog);
1043         prb_run_all_ft_ops(pkc, ppd);
1044 }
1045
1046 /* Assumes caller has the sk->rx_queue.lock */
1047 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1048                                             struct sk_buff *skb,
1049                                                 int status,
1050                                             unsigned int len
1051                                             )
1052 {
1053         struct tpacket_kbdq_core *pkc;
1054         struct tpacket_block_desc *pbd;
1055         char *curr, *end;
1056
1057         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1058         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1059
1060         /* Queue is frozen when user space is lagging behind */
1061         if (prb_queue_frozen(pkc)) {
1062                 /*
1063                  * Check if that last block which caused the queue to freeze,
1064                  * is still in_use by user-space.
1065                  */
1066                 if (prb_curr_blk_in_use(pkc, pbd)) {
1067                         /* Can't record this packet */
1068                         return NULL;
1069                 } else {
1070                         /*
1071                          * Ok, the block was released by user-space.
1072                          * Now let's open that block.
1073                          * opening a block also thaws the queue.
1074                          * Thawing is a side effect.
1075                          */
1076                         prb_open_block(pkc, pbd);
1077                 }
1078         }
1079
1080         smp_mb();
1081         curr = pkc->nxt_offset;
1082         pkc->skb = skb;
1083         end = (char *)pbd + pkc->kblk_size;
1084
1085         /* first try the current block */
1086         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1087                 prb_fill_curr_block(curr, pkc, pbd, len);
1088                 return (void *)curr;
1089         }
1090
1091         /* Ok, close the current block */
1092         prb_retire_current_block(pkc, po, 0);
1093
1094         /* Now, try to dispatch the next block */
1095         curr = (char *)prb_dispatch_next_block(pkc, po);
1096         if (curr) {
1097                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1098                 prb_fill_curr_block(curr, pkc, pbd, len);
1099                 return (void *)curr;
1100         }
1101
1102         /*
1103          * No free blocks are available.user_space hasn't caught up yet.
1104          * Queue was just frozen and now this packet will get dropped.
1105          */
1106         return NULL;
1107 }
1108
1109 static void *packet_current_rx_frame(struct packet_sock *po,
1110                                             struct sk_buff *skb,
1111                                             int status, unsigned int len)
1112 {
1113         char *curr = NULL;
1114         switch (po->tp_version) {
1115         case TPACKET_V1:
1116         case TPACKET_V2:
1117                 curr = packet_lookup_frame(po, &po->rx_ring,
1118                                         po->rx_ring.head, status);
1119                 return curr;
1120         case TPACKET_V3:
1121                 return __packet_lookup_frame_in_block(po, skb, status, len);
1122         default:
1123                 WARN(1, "TPACKET version not supported\n");
1124                 BUG();
1125                 return NULL;
1126         }
1127 }
1128
1129 static void *prb_lookup_block(struct packet_sock *po,
1130                                      struct packet_ring_buffer *rb,
1131                                      unsigned int idx,
1132                                      int status)
1133 {
1134         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1135         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1136
1137         if (status != BLOCK_STATUS(pbd))
1138                 return NULL;
1139         return pbd;
1140 }
1141
1142 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1143 {
1144         unsigned int prev;
1145         if (rb->prb_bdqc.kactive_blk_num)
1146                 prev = rb->prb_bdqc.kactive_blk_num-1;
1147         else
1148                 prev = rb->prb_bdqc.knum_blocks-1;
1149         return prev;
1150 }
1151
1152 /* Assumes caller has held the rx_queue.lock */
1153 static void *__prb_previous_block(struct packet_sock *po,
1154                                          struct packet_ring_buffer *rb,
1155                                          int status)
1156 {
1157         unsigned int previous = prb_previous_blk_num(rb);
1158         return prb_lookup_block(po, rb, previous, status);
1159 }
1160
1161 static void *packet_previous_rx_frame(struct packet_sock *po,
1162                                              struct packet_ring_buffer *rb,
1163                                              int status)
1164 {
1165         if (po->tp_version <= TPACKET_V2)
1166                 return packet_previous_frame(po, rb, status);
1167
1168         return __prb_previous_block(po, rb, status);
1169 }
1170
1171 static void packet_increment_rx_head(struct packet_sock *po,
1172                                             struct packet_ring_buffer *rb)
1173 {
1174         switch (po->tp_version) {
1175         case TPACKET_V1:
1176         case TPACKET_V2:
1177                 return packet_increment_head(rb);
1178         case TPACKET_V3:
1179         default:
1180                 WARN(1, "TPACKET version not supported.\n");
1181                 BUG();
1182                 return;
1183         }
1184 }
1185
1186 static void *packet_previous_frame(struct packet_sock *po,
1187                 struct packet_ring_buffer *rb,
1188                 int status)
1189 {
1190         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1191         return packet_lookup_frame(po, rb, previous, status);
1192 }
1193
1194 static void packet_increment_head(struct packet_ring_buffer *buff)
1195 {
1196         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1197 }
1198
1199 static void packet_inc_pending(struct packet_ring_buffer *rb)
1200 {
1201         this_cpu_inc(*rb->pending_refcnt);
1202 }
1203
1204 static void packet_dec_pending(struct packet_ring_buffer *rb)
1205 {
1206         this_cpu_dec(*rb->pending_refcnt);
1207 }
1208
1209 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1210 {
1211         unsigned int refcnt = 0;
1212         int cpu;
1213
1214         /* We don't use pending refcount in rx_ring. */
1215         if (rb->pending_refcnt == NULL)
1216                 return 0;
1217
1218         for_each_possible_cpu(cpu)
1219                 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1220
1221         return refcnt;
1222 }
1223
1224 static int packet_alloc_pending(struct packet_sock *po)
1225 {
1226         po->rx_ring.pending_refcnt = NULL;
1227
1228         po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1229         if (unlikely(po->tx_ring.pending_refcnt == NULL))
1230                 return -ENOBUFS;
1231
1232         return 0;
1233 }
1234
1235 static void packet_free_pending(struct packet_sock *po)
1236 {
1237         free_percpu(po->tx_ring.pending_refcnt);
1238 }
1239
1240 #define ROOM_POW_OFF    2
1241 #define ROOM_NONE       0x0
1242 #define ROOM_LOW        0x1
1243 #define ROOM_NORMAL     0x2
1244
1245 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1246 {
1247         int idx, len;
1248
1249         len = po->rx_ring.frame_max + 1;
1250         idx = po->rx_ring.head;
1251         if (pow_off)
1252                 idx += len >> pow_off;
1253         if (idx >= len)
1254                 idx -= len;
1255         return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1256 }
1257
1258 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1259 {
1260         int idx, len;
1261
1262         len = po->rx_ring.prb_bdqc.knum_blocks;
1263         idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1264         if (pow_off)
1265                 idx += len >> pow_off;
1266         if (idx >= len)
1267                 idx -= len;
1268         return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1269 }
1270
1271 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1272 {
1273         struct sock *sk = &po->sk;
1274         int ret = ROOM_NONE;
1275
1276         if (po->prot_hook.func != tpacket_rcv) {
1277                 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1278                                           - (skb ? skb->truesize : 0);
1279                 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1280                         return ROOM_NORMAL;
1281                 else if (avail > 0)
1282                         return ROOM_LOW;
1283                 else
1284                         return ROOM_NONE;
1285         }
1286
1287         if (po->tp_version == TPACKET_V3) {
1288                 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1289                         ret = ROOM_NORMAL;
1290                 else if (__tpacket_v3_has_room(po, 0))
1291                         ret = ROOM_LOW;
1292         } else {
1293                 if (__tpacket_has_room(po, ROOM_POW_OFF))
1294                         ret = ROOM_NORMAL;
1295                 else if (__tpacket_has_room(po, 0))
1296                         ret = ROOM_LOW;
1297         }
1298
1299         return ret;
1300 }
1301
1302 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1303 {
1304         int ret;
1305         bool has_room;
1306
1307         spin_lock_bh(&po->sk.sk_receive_queue.lock);
1308         ret = __packet_rcv_has_room(po, skb);
1309         has_room = ret == ROOM_NORMAL;
1310         if (po->pressure == has_room)
1311                 po->pressure = !has_room;
1312         spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1313
1314         return ret;
1315 }
1316
1317 static void packet_sock_destruct(struct sock *sk)
1318 {
1319         skb_queue_purge(&sk->sk_error_queue);
1320
1321         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1322         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1323
1324         if (!sock_flag(sk, SOCK_DEAD)) {
1325                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1326                 return;
1327         }
1328
1329         sk_refcnt_debug_dec(sk);
1330 }
1331
1332 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1333 {
1334         u32 rxhash;
1335         int i, count = 0;
1336
1337         rxhash = skb_get_hash(skb);
1338         for (i = 0; i < ROLLOVER_HLEN; i++)
1339                 if (po->rollover->history[i] == rxhash)
1340                         count++;
1341
1342         po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1343         return count > (ROLLOVER_HLEN >> 1);
1344 }
1345
1346 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1347                                       struct sk_buff *skb,
1348                                       unsigned int num)
1349 {
1350         return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1351 }
1352
1353 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1354                                     struct sk_buff *skb,
1355                                     unsigned int num)
1356 {
1357         unsigned int val = atomic_inc_return(&f->rr_cur);
1358
1359         return val % num;
1360 }
1361
1362 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1363                                      struct sk_buff *skb,
1364                                      unsigned int num)
1365 {
1366         return smp_processor_id() % num;
1367 }
1368
1369 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1370                                      struct sk_buff *skb,
1371                                      unsigned int num)
1372 {
1373         return prandom_u32_max(num);
1374 }
1375
1376 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1377                                           struct sk_buff *skb,
1378                                           unsigned int idx, bool try_self,
1379                                           unsigned int num)
1380 {
1381         struct packet_sock *po, *po_next, *po_skip = NULL;
1382         unsigned int i, j, room = ROOM_NONE;
1383
1384         po = pkt_sk(f->arr[idx]);
1385
1386         if (try_self) {
1387                 room = packet_rcv_has_room(po, skb);
1388                 if (room == ROOM_NORMAL ||
1389                     (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1390                         return idx;
1391                 po_skip = po;
1392         }
1393
1394         i = j = min_t(int, po->rollover->sock, num - 1);
1395         do {
1396                 po_next = pkt_sk(f->arr[i]);
1397                 if (po_next != po_skip && !po_next->pressure &&
1398                     packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1399                         if (i != j)
1400                                 po->rollover->sock = i;
1401                         atomic_long_inc(&po->rollover->num);
1402                         if (room == ROOM_LOW)
1403                                 atomic_long_inc(&po->rollover->num_huge);
1404                         return i;
1405                 }
1406
1407                 if (++i == num)
1408                         i = 0;
1409         } while (i != j);
1410
1411         atomic_long_inc(&po->rollover->num_failed);
1412         return idx;
1413 }
1414
1415 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1416                                     struct sk_buff *skb,
1417                                     unsigned int num)
1418 {
1419         return skb_get_queue_mapping(skb) % num;
1420 }
1421
1422 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1423                                      struct sk_buff *skb,
1424                                      unsigned int num)
1425 {
1426         struct bpf_prog *prog;
1427         unsigned int ret = 0;
1428
1429         rcu_read_lock();
1430         prog = rcu_dereference(f->bpf_prog);
1431         if (prog)
1432                 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1433         rcu_read_unlock();
1434
1435         return ret;
1436 }
1437
1438 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1439 {
1440         return f->flags & (flag >> 8);
1441 }
1442
1443 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1444                              struct packet_type *pt, struct net_device *orig_dev)
1445 {
1446         struct packet_fanout *f = pt->af_packet_priv;
1447         unsigned int num = READ_ONCE(f->num_members);
1448         struct net *net = read_pnet(&f->net);
1449         struct packet_sock *po;
1450         unsigned int idx;
1451
1452         if (!net_eq(dev_net(dev), net) || !num) {
1453                 kfree_skb(skb);
1454                 return 0;
1455         }
1456
1457         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1458                 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1459                 if (!skb)
1460                         return 0;
1461         }
1462         switch (f->type) {
1463         case PACKET_FANOUT_HASH:
1464         default:
1465                 idx = fanout_demux_hash(f, skb, num);
1466                 break;
1467         case PACKET_FANOUT_LB:
1468                 idx = fanout_demux_lb(f, skb, num);
1469                 break;
1470         case PACKET_FANOUT_CPU:
1471                 idx = fanout_demux_cpu(f, skb, num);
1472                 break;
1473         case PACKET_FANOUT_RND:
1474                 idx = fanout_demux_rnd(f, skb, num);
1475                 break;
1476         case PACKET_FANOUT_QM:
1477                 idx = fanout_demux_qm(f, skb, num);
1478                 break;
1479         case PACKET_FANOUT_ROLLOVER:
1480                 idx = fanout_demux_rollover(f, skb, 0, false, num);
1481                 break;
1482         case PACKET_FANOUT_CBPF:
1483         case PACKET_FANOUT_EBPF:
1484                 idx = fanout_demux_bpf(f, skb, num);
1485                 break;
1486         }
1487
1488         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1489                 idx = fanout_demux_rollover(f, skb, idx, true, num);
1490
1491         po = pkt_sk(f->arr[idx]);
1492         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1493 }
1494
1495 DEFINE_MUTEX(fanout_mutex);
1496 EXPORT_SYMBOL_GPL(fanout_mutex);
1497 static LIST_HEAD(fanout_list);
1498
1499 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1500 {
1501         struct packet_fanout *f = po->fanout;
1502
1503         spin_lock(&f->lock);
1504         f->arr[f->num_members] = sk;
1505         smp_wmb();
1506         f->num_members++;
1507         if (f->num_members == 1)
1508                 dev_add_pack(&f->prot_hook);
1509         spin_unlock(&f->lock);
1510 }
1511
1512 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1513 {
1514         struct packet_fanout *f = po->fanout;
1515         int i;
1516
1517         spin_lock(&f->lock);
1518         for (i = 0; i < f->num_members; i++) {
1519                 if (f->arr[i] == sk)
1520                         break;
1521         }
1522         BUG_ON(i >= f->num_members);
1523         f->arr[i] = f->arr[f->num_members - 1];
1524         f->num_members--;
1525         if (f->num_members == 0)
1526                 __dev_remove_pack(&f->prot_hook);
1527         spin_unlock(&f->lock);
1528 }
1529
1530 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1531 {
1532         if (sk->sk_family != PF_PACKET)
1533                 return false;
1534
1535         return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1536 }
1537
1538 static void fanout_init_data(struct packet_fanout *f)
1539 {
1540         switch (f->type) {
1541         case PACKET_FANOUT_LB:
1542                 atomic_set(&f->rr_cur, 0);
1543                 break;
1544         case PACKET_FANOUT_CBPF:
1545         case PACKET_FANOUT_EBPF:
1546                 RCU_INIT_POINTER(f->bpf_prog, NULL);
1547                 break;
1548         }
1549 }
1550
1551 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1552 {
1553         struct bpf_prog *old;
1554
1555         spin_lock(&f->lock);
1556         old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1557         rcu_assign_pointer(f->bpf_prog, new);
1558         spin_unlock(&f->lock);
1559
1560         if (old) {
1561                 synchronize_net();
1562                 bpf_prog_destroy(old);
1563         }
1564 }
1565
1566 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1567                                 unsigned int len)
1568 {
1569         struct bpf_prog *new;
1570         struct sock_fprog fprog;
1571         int ret;
1572
1573         if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1574                 return -EPERM;
1575         if (len != sizeof(fprog))
1576                 return -EINVAL;
1577         if (copy_from_user(&fprog, data, len))
1578                 return -EFAULT;
1579
1580         ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1581         if (ret)
1582                 return ret;
1583
1584         __fanout_set_data_bpf(po->fanout, new);
1585         return 0;
1586 }
1587
1588 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1589                                 unsigned int len)
1590 {
1591         struct bpf_prog *new;
1592         u32 fd;
1593
1594         if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1595                 return -EPERM;
1596         if (len != sizeof(fd))
1597                 return -EINVAL;
1598         if (copy_from_user(&fd, data, len))
1599                 return -EFAULT;
1600
1601         new = bpf_prog_get(fd);
1602         if (IS_ERR(new))
1603                 return PTR_ERR(new);
1604         if (new->type != BPF_PROG_TYPE_SOCKET_FILTER) {
1605                 bpf_prog_put(new);
1606                 return -EINVAL;
1607         }
1608
1609         __fanout_set_data_bpf(po->fanout, new);
1610         return 0;
1611 }
1612
1613 static int fanout_set_data(struct packet_sock *po, char __user *data,
1614                            unsigned int len)
1615 {
1616         switch (po->fanout->type) {
1617         case PACKET_FANOUT_CBPF:
1618                 return fanout_set_data_cbpf(po, data, len);
1619         case PACKET_FANOUT_EBPF:
1620                 return fanout_set_data_ebpf(po, data, len);
1621         default:
1622                 return -EINVAL;
1623         };
1624 }
1625
1626 static void fanout_release_data(struct packet_fanout *f)
1627 {
1628         switch (f->type) {
1629         case PACKET_FANOUT_CBPF:
1630         case PACKET_FANOUT_EBPF:
1631                 __fanout_set_data_bpf(f, NULL);
1632         };
1633 }
1634
1635 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1636 {
1637         struct packet_rollover *rollover = NULL;
1638         struct packet_sock *po = pkt_sk(sk);
1639         struct packet_fanout *f, *match;
1640         u8 type = type_flags & 0xff;
1641         u8 flags = type_flags >> 8;
1642         int err;
1643
1644         switch (type) {
1645         case PACKET_FANOUT_ROLLOVER:
1646                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1647                         return -EINVAL;
1648         case PACKET_FANOUT_HASH:
1649         case PACKET_FANOUT_LB:
1650         case PACKET_FANOUT_CPU:
1651         case PACKET_FANOUT_RND:
1652         case PACKET_FANOUT_QM:
1653         case PACKET_FANOUT_CBPF:
1654         case PACKET_FANOUT_EBPF:
1655                 break;
1656         default:
1657                 return -EINVAL;
1658         }
1659
1660         mutex_lock(&fanout_mutex);
1661
1662         err = -EALREADY;
1663         if (po->fanout)
1664                 goto out;
1665
1666         if (type == PACKET_FANOUT_ROLLOVER ||
1667             (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1668                 err = -ENOMEM;
1669                 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1670                 if (!rollover)
1671                         goto out;
1672                 atomic_long_set(&rollover->num, 0);
1673                 atomic_long_set(&rollover->num_huge, 0);
1674                 atomic_long_set(&rollover->num_failed, 0);
1675         }
1676
1677         match = NULL;
1678         list_for_each_entry(f, &fanout_list, list) {
1679                 if (f->id == id &&
1680                     read_pnet(&f->net) == sock_net(sk)) {
1681                         match = f;
1682                         break;
1683                 }
1684         }
1685         err = -EINVAL;
1686         if (match && match->flags != flags)
1687                 goto out;
1688         if (!match) {
1689                 err = -ENOMEM;
1690                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1691                 if (!match)
1692                         goto out;
1693                 write_pnet(&match->net, sock_net(sk));
1694                 match->id = id;
1695                 match->type = type;
1696                 match->flags = flags;
1697                 INIT_LIST_HEAD(&match->list);
1698                 spin_lock_init(&match->lock);
1699                 atomic_set(&match->sk_ref, 0);
1700                 fanout_init_data(match);
1701                 match->prot_hook.type = po->prot_hook.type;
1702                 match->prot_hook.dev = po->prot_hook.dev;
1703                 match->prot_hook.func = packet_rcv_fanout;
1704                 match->prot_hook.af_packet_priv = match;
1705                 match->prot_hook.id_match = match_fanout_group;
1706                 list_add(&match->list, &fanout_list);
1707         }
1708         err = -EINVAL;
1709
1710         spin_lock(&po->bind_lock);
1711         if (po->running &&
1712             match->type == type &&
1713             match->prot_hook.type == po->prot_hook.type &&
1714             match->prot_hook.dev == po->prot_hook.dev) {
1715                 err = -ENOSPC;
1716                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1717                         __dev_remove_pack(&po->prot_hook);
1718                         po->fanout = match;
1719                         po->rollover = rollover;
1720                         rollover = NULL;
1721                         atomic_inc(&match->sk_ref);
1722                         __fanout_link(sk, po);
1723                         err = 0;
1724                 }
1725         }
1726         spin_unlock(&po->bind_lock);
1727
1728         if (err && !atomic_read(&match->sk_ref)) {
1729                 list_del(&match->list);
1730                 kfree(match);
1731         }
1732
1733 out:
1734         kfree(rollover);
1735         mutex_unlock(&fanout_mutex);
1736         return err;
1737 }
1738
1739 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1740  * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1741  * It is the responsibility of the caller to call fanout_release_data() and
1742  * free the returned packet_fanout (after synchronize_net())
1743  */
1744 static struct packet_fanout *fanout_release(struct sock *sk)
1745 {
1746         struct packet_sock *po = pkt_sk(sk);
1747         struct packet_fanout *f;
1748
1749         mutex_lock(&fanout_mutex);
1750         f = po->fanout;
1751         if (f) {
1752                 po->fanout = NULL;
1753
1754                 if (atomic_dec_and_test(&f->sk_ref))
1755                         list_del(&f->list);
1756                 else
1757                         f = NULL;
1758         }
1759         mutex_unlock(&fanout_mutex);
1760
1761         return f;
1762 }
1763
1764 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1765                                           struct sk_buff *skb)
1766 {
1767         /* Earlier code assumed this would be a VLAN pkt, double-check
1768          * this now that we have the actual packet in hand. We can only
1769          * do this check on Ethernet devices.
1770          */
1771         if (unlikely(dev->type != ARPHRD_ETHER))
1772                 return false;
1773
1774         skb_reset_mac_header(skb);
1775         return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1776 }
1777
1778 static const struct proto_ops packet_ops;
1779
1780 static const struct proto_ops packet_ops_spkt;
1781
1782 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1783                            struct packet_type *pt, struct net_device *orig_dev)
1784 {
1785         struct sock *sk;
1786         struct sockaddr_pkt *spkt;
1787
1788         /*
1789          *      When we registered the protocol we saved the socket in the data
1790          *      field for just this event.
1791          */
1792
1793         sk = pt->af_packet_priv;
1794
1795         /*
1796          *      Yank back the headers [hope the device set this
1797          *      right or kerboom...]
1798          *
1799          *      Incoming packets have ll header pulled,
1800          *      push it back.
1801          *
1802          *      For outgoing ones skb->data == skb_mac_header(skb)
1803          *      so that this procedure is noop.
1804          */
1805
1806         if (skb->pkt_type == PACKET_LOOPBACK)
1807                 goto out;
1808
1809         if (!net_eq(dev_net(dev), sock_net(sk)))
1810                 goto out;
1811
1812         skb = skb_share_check(skb, GFP_ATOMIC);
1813         if (skb == NULL)
1814                 goto oom;
1815
1816         /* drop any routing info */
1817         skb_dst_drop(skb);
1818
1819         /* drop conntrack reference */
1820         nf_reset(skb);
1821
1822         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1823
1824         skb_push(skb, skb->data - skb_mac_header(skb));
1825
1826         /*
1827          *      The SOCK_PACKET socket receives _all_ frames.
1828          */
1829
1830         spkt->spkt_family = dev->type;
1831         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1832         spkt->spkt_protocol = skb->protocol;
1833
1834         /*
1835          *      Charge the memory to the socket. This is done specifically
1836          *      to prevent sockets using all the memory up.
1837          */
1838
1839         if (sock_queue_rcv_skb(sk, skb) == 0)
1840                 return 0;
1841
1842 out:
1843         kfree_skb(skb);
1844 oom:
1845         return 0;
1846 }
1847
1848
1849 /*
1850  *      Output a raw packet to a device layer. This bypasses all the other
1851  *      protocol layers and you must therefore supply it with a complete frame
1852  */
1853
1854 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1855                                size_t len)
1856 {
1857         struct sock *sk = sock->sk;
1858         DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1859         struct sk_buff *skb = NULL;
1860         struct net_device *dev;
1861         __be16 proto = 0;
1862         int err;
1863         int extra_len = 0;
1864
1865         /*
1866          *      Get and verify the address.
1867          */
1868
1869         if (saddr) {
1870                 if (msg->msg_namelen < sizeof(struct sockaddr))
1871                         return -EINVAL;
1872                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1873                         proto = saddr->spkt_protocol;
1874         } else
1875                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1876
1877         /*
1878          *      Find the device first to size check it
1879          */
1880
1881         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1882 retry:
1883         rcu_read_lock();
1884         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1885         err = -ENODEV;
1886         if (dev == NULL)
1887                 goto out_unlock;
1888
1889         err = -ENETDOWN;
1890         if (!(dev->flags & IFF_UP))
1891                 goto out_unlock;
1892
1893         /*
1894          * You may not queue a frame bigger than the mtu. This is the lowest level
1895          * raw protocol and you must do your own fragmentation at this level.
1896          */
1897
1898         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1899                 if (!netif_supports_nofcs(dev)) {
1900                         err = -EPROTONOSUPPORT;
1901                         goto out_unlock;
1902                 }
1903                 extra_len = 4; /* We're doing our own CRC */
1904         }
1905
1906         err = -EMSGSIZE;
1907         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1908                 goto out_unlock;
1909
1910         if (!skb) {
1911                 size_t reserved = LL_RESERVED_SPACE(dev);
1912                 int tlen = dev->needed_tailroom;
1913                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1914
1915                 rcu_read_unlock();
1916                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1917                 if (skb == NULL)
1918                         return -ENOBUFS;
1919                 /* FIXME: Save some space for broken drivers that write a hard
1920                  * header at transmission time by themselves. PPP is the notable
1921                  * one here. This should really be fixed at the driver level.
1922                  */
1923                 skb_reserve(skb, reserved);
1924                 skb_reset_network_header(skb);
1925
1926                 /* Try to align data part correctly */
1927                 if (hhlen) {
1928                         skb->data -= hhlen;
1929                         skb->tail -= hhlen;
1930                         if (len < hhlen)
1931                                 skb_reset_network_header(skb);
1932                 }
1933                 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1934                 if (err)
1935                         goto out_free;
1936                 goto retry;
1937         }
1938
1939         if (!dev_validate_header(dev, skb->data, len)) {
1940                 err = -EINVAL;
1941                 goto out_unlock;
1942         }
1943         if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1944             !packet_extra_vlan_len_allowed(dev, skb)) {
1945                 err = -EMSGSIZE;
1946                 goto out_unlock;
1947         }
1948
1949         skb->protocol = proto;
1950         skb->dev = dev;
1951         skb->priority = sk->sk_priority;
1952         skb->mark = sk->sk_mark;
1953
1954         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1955
1956         if (unlikely(extra_len == 4))
1957                 skb->no_fcs = 1;
1958
1959         skb_probe_transport_header(skb, 0);
1960
1961         dev_queue_xmit(skb);
1962         rcu_read_unlock();
1963         return len;
1964
1965 out_unlock:
1966         rcu_read_unlock();
1967 out_free:
1968         kfree_skb(skb);
1969         return err;
1970 }
1971
1972 static unsigned int run_filter(struct sk_buff *skb,
1973                                const struct sock *sk,
1974                                unsigned int res)
1975 {
1976         struct sk_filter *filter;
1977
1978         rcu_read_lock();
1979         filter = rcu_dereference(sk->sk_filter);
1980         if (filter != NULL)
1981                 res = bpf_prog_run_clear_cb(filter->prog, skb);
1982         rcu_read_unlock();
1983
1984         return res;
1985 }
1986
1987 /*
1988  * This function makes lazy skb cloning in hope that most of packets
1989  * are discarded by BPF.
1990  *
1991  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1992  * and skb->cb are mangled. It works because (and until) packets
1993  * falling here are owned by current CPU. Output packets are cloned
1994  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1995  * sequencially, so that if we return skb to original state on exit,
1996  * we will not harm anyone.
1997  */
1998
1999 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2000                       struct packet_type *pt, struct net_device *orig_dev)
2001 {
2002         struct sock *sk;
2003         struct sockaddr_ll *sll;
2004         struct packet_sock *po;
2005         u8 *skb_head = skb->data;
2006         int skb_len = skb->len;
2007         unsigned int snaplen, res;
2008
2009         if (skb->pkt_type == PACKET_LOOPBACK)
2010                 goto drop;
2011
2012         sk = pt->af_packet_priv;
2013         po = pkt_sk(sk);
2014
2015         if (!net_eq(dev_net(dev), sock_net(sk)))
2016                 goto drop;
2017
2018         skb->dev = dev;
2019
2020         if (dev->header_ops) {
2021                 /* The device has an explicit notion of ll header,
2022                  * exported to higher levels.
2023                  *
2024                  * Otherwise, the device hides details of its frame
2025                  * structure, so that corresponding packet head is
2026                  * never delivered to user.
2027                  */
2028                 if (sk->sk_type != SOCK_DGRAM)
2029                         skb_push(skb, skb->data - skb_mac_header(skb));
2030                 else if (skb->pkt_type == PACKET_OUTGOING) {
2031                         /* Special case: outgoing packets have ll header at head */
2032                         skb_pull(skb, skb_network_offset(skb));
2033                 }
2034         }
2035
2036         snaplen = skb->len;
2037
2038         res = run_filter(skb, sk, snaplen);
2039         if (!res)
2040                 goto drop_n_restore;
2041         if (snaplen > res)
2042                 snaplen = res;
2043
2044         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2045                 goto drop_n_acct;
2046
2047         if (skb_shared(skb)) {
2048                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2049                 if (nskb == NULL)
2050                         goto drop_n_acct;
2051
2052                 if (skb_head != skb->data) {
2053                         skb->data = skb_head;
2054                         skb->len = skb_len;
2055                 }
2056                 consume_skb(skb);
2057                 skb = nskb;
2058         }
2059
2060         sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2061
2062         sll = &PACKET_SKB_CB(skb)->sa.ll;
2063         sll->sll_hatype = dev->type;
2064         sll->sll_pkttype = skb->pkt_type;
2065         if (unlikely(po->origdev))
2066                 sll->sll_ifindex = orig_dev->ifindex;
2067         else
2068                 sll->sll_ifindex = dev->ifindex;
2069
2070         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2071
2072         /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2073          * Use their space for storing the original skb length.
2074          */
2075         PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2076
2077         if (pskb_trim(skb, snaplen))
2078                 goto drop_n_acct;
2079
2080         skb_set_owner_r(skb, sk);
2081         skb->dev = NULL;
2082         skb_dst_drop(skb);
2083
2084         /* drop conntrack reference */
2085         nf_reset(skb);
2086
2087         spin_lock(&sk->sk_receive_queue.lock);
2088         po->stats.stats1.tp_packets++;
2089         sock_skb_set_dropcount(sk, skb);
2090         __skb_queue_tail(&sk->sk_receive_queue, skb);
2091         spin_unlock(&sk->sk_receive_queue.lock);
2092         sk->sk_data_ready(sk);
2093         return 0;
2094
2095 drop_n_acct:
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         consume_skb(skb);
2108         return 0;
2109 }
2110
2111 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2112                        struct packet_type *pt, struct net_device *orig_dev)
2113 {
2114         struct sock *sk;
2115         struct packet_sock *po;
2116         struct sockaddr_ll *sll;
2117         union tpacket_uhdr h;
2118         u8 *skb_head = skb->data;
2119         int skb_len = skb->len;
2120         unsigned int snaplen, res;
2121         unsigned long status = TP_STATUS_USER;
2122         unsigned short macoff, netoff, hdrlen;
2123         struct sk_buff *copy_skb = NULL;
2124         struct timespec ts;
2125         __u32 ts_status;
2126
2127         /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2128          * We may add members to them until current aligned size without forcing
2129          * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2130          */
2131         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2132         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2133
2134         if (skb->pkt_type == PACKET_LOOPBACK)
2135                 goto drop;
2136
2137         sk = pt->af_packet_priv;
2138         po = pkt_sk(sk);
2139
2140         if (!net_eq(dev_net(dev), sock_net(sk)))
2141                 goto drop;
2142
2143         if (dev->header_ops) {
2144                 if (sk->sk_type != SOCK_DGRAM)
2145                         skb_push(skb, skb->data - skb_mac_header(skb));
2146                 else if (skb->pkt_type == PACKET_OUTGOING) {
2147                         /* Special case: outgoing packets have ll header at head */
2148                         skb_pull(skb, skb_network_offset(skb));
2149                 }
2150         }
2151
2152         snaplen = skb->len;
2153
2154         res = run_filter(skb, sk, snaplen);
2155         if (!res)
2156                 goto drop_n_restore;
2157
2158         if (skb->ip_summed == CHECKSUM_PARTIAL)
2159                 status |= TP_STATUS_CSUMNOTREADY;
2160         else if (skb->pkt_type != PACKET_OUTGOING &&
2161                  (skb->ip_summed == CHECKSUM_COMPLETE ||
2162                   skb_csum_unnecessary(skb)))
2163                 status |= TP_STATUS_CSUM_VALID;
2164
2165         if (snaplen > res)
2166                 snaplen = res;
2167
2168         if (sk->sk_type == SOCK_DGRAM) {
2169                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2170                                   po->tp_reserve;
2171         } else {
2172                 unsigned int maclen = skb_network_offset(skb);
2173                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2174                                        (maclen < 16 ? 16 : maclen)) +
2175                         po->tp_reserve;
2176                 macoff = netoff - maclen;
2177         }
2178         if (po->tp_version <= TPACKET_V2) {
2179                 if (macoff + snaplen > po->rx_ring.frame_size) {
2180                         if (po->copy_thresh &&
2181                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2182                                 if (skb_shared(skb)) {
2183                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
2184                                 } else {
2185                                         copy_skb = skb_get(skb);
2186                                         skb_head = skb->data;
2187                                 }
2188                                 if (copy_skb)
2189                                         skb_set_owner_r(copy_skb, sk);
2190                         }
2191                         snaplen = po->rx_ring.frame_size - macoff;
2192                         if ((int)snaplen < 0)
2193                                 snaplen = 0;
2194                 }
2195         } else if (unlikely(macoff + snaplen >
2196                             GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2197                 u32 nval;
2198
2199                 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2200                 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2201                             snaplen, nval, macoff);
2202                 snaplen = nval;
2203                 if (unlikely((int)snaplen < 0)) {
2204                         snaplen = 0;
2205                         macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2206                 }
2207         }
2208         spin_lock(&sk->sk_receive_queue.lock);
2209         h.raw = packet_current_rx_frame(po, skb,
2210                                         TP_STATUS_KERNEL, (macoff+snaplen));
2211         if (!h.raw)
2212                 goto ring_is_full;
2213         if (po->tp_version <= TPACKET_V2) {
2214                 packet_increment_rx_head(po, &po->rx_ring);
2215         /*
2216          * LOSING will be reported till you read the stats,
2217          * because it's COR - Clear On Read.
2218          * Anyways, moving it for V1/V2 only as V3 doesn't need this
2219          * at packet level.
2220          */
2221                 if (po->stats.stats1.tp_drops)
2222                         status |= TP_STATUS_LOSING;
2223         }
2224         po->stats.stats1.tp_packets++;
2225         if (copy_skb) {
2226                 status |= TP_STATUS_COPY;
2227                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2228         }
2229         spin_unlock(&sk->sk_receive_queue.lock);
2230
2231         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2232
2233         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2234                 getnstimeofday(&ts);
2235
2236         status |= ts_status;
2237
2238         switch (po->tp_version) {
2239         case TPACKET_V1:
2240                 h.h1->tp_len = skb->len;
2241                 h.h1->tp_snaplen = snaplen;
2242                 h.h1->tp_mac = macoff;
2243                 h.h1->tp_net = netoff;
2244                 h.h1->tp_sec = ts.tv_sec;
2245                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2246                 hdrlen = sizeof(*h.h1);
2247                 break;
2248         case TPACKET_V2:
2249                 h.h2->tp_len = skb->len;
2250                 h.h2->tp_snaplen = snaplen;
2251                 h.h2->tp_mac = macoff;
2252                 h.h2->tp_net = netoff;
2253                 h.h2->tp_sec = ts.tv_sec;
2254                 h.h2->tp_nsec = ts.tv_nsec;
2255                 if (skb_vlan_tag_present(skb)) {
2256                         h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2257                         h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2258                         status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2259                 } else {
2260                         h.h2->tp_vlan_tci = 0;
2261                         h.h2->tp_vlan_tpid = 0;
2262                 }
2263                 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2264                 hdrlen = sizeof(*h.h2);
2265                 break;
2266         case TPACKET_V3:
2267                 /* tp_nxt_offset,vlan are already populated above.
2268                  * So DONT clear those fields here
2269                  */
2270                 h.h3->tp_status |= status;
2271                 h.h3->tp_len = skb->len;
2272                 h.h3->tp_snaplen = snaplen;
2273                 h.h3->tp_mac = macoff;
2274                 h.h3->tp_net = netoff;
2275                 h.h3->tp_sec  = ts.tv_sec;
2276                 h.h3->tp_nsec = ts.tv_nsec;
2277                 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2278                 hdrlen = sizeof(*h.h3);
2279                 break;
2280         default:
2281                 BUG();
2282         }
2283
2284         sll = h.raw + TPACKET_ALIGN(hdrlen);
2285         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2286         sll->sll_family = AF_PACKET;
2287         sll->sll_hatype = dev->type;
2288         sll->sll_protocol = skb->protocol;
2289         sll->sll_pkttype = skb->pkt_type;
2290         if (unlikely(po->origdev))
2291                 sll->sll_ifindex = orig_dev->ifindex;
2292         else
2293                 sll->sll_ifindex = dev->ifindex;
2294
2295         smp_mb();
2296
2297 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2298         if (po->tp_version <= TPACKET_V2) {
2299                 u8 *start, *end;
2300
2301                 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2302                                         macoff + snaplen);
2303
2304                 for (start = h.raw; start < end; start += PAGE_SIZE)
2305                         flush_dcache_page(pgv_to_page(start));
2306         }
2307         smp_wmb();
2308 #endif
2309
2310         if (po->tp_version <= TPACKET_V2) {
2311                 __packet_set_status(po, h.raw, status);
2312                 sk->sk_data_ready(sk);
2313         } else {
2314                 prb_clear_blk_fill_status(&po->rx_ring);
2315         }
2316
2317 drop_n_restore:
2318         if (skb_head != skb->data && skb_shared(skb)) {
2319                 skb->data = skb_head;
2320                 skb->len = skb_len;
2321         }
2322 drop:
2323         kfree_skb(skb);
2324         return 0;
2325
2326 ring_is_full:
2327         po->stats.stats1.tp_drops++;
2328         spin_unlock(&sk->sk_receive_queue.lock);
2329
2330         sk->sk_data_ready(sk);
2331         kfree_skb(copy_skb);
2332         goto drop_n_restore;
2333 }
2334
2335 static void tpacket_destruct_skb(struct sk_buff *skb)
2336 {
2337         struct packet_sock *po = pkt_sk(skb->sk);
2338
2339         if (likely(po->tx_ring.pg_vec)) {
2340                 void *ph;
2341                 __u32 ts;
2342
2343                 ph = skb_shinfo(skb)->destructor_arg;
2344                 packet_dec_pending(&po->tx_ring);
2345
2346                 ts = __packet_set_timestamp(po, ph, skb);
2347                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2348         }
2349
2350         sock_wfree(skb);
2351 }
2352
2353 static void tpacket_set_protocol(const struct net_device *dev,
2354                                  struct sk_buff *skb)
2355 {
2356         if (dev->type == ARPHRD_ETHER) {
2357                 skb_reset_mac_header(skb);
2358                 skb->protocol = eth_hdr(skb)->h_proto;
2359         }
2360 }
2361
2362 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2363                 void *frame, struct net_device *dev, int size_max,
2364                 __be16 proto, unsigned char *addr, int hlen)
2365 {
2366         union tpacket_uhdr ph;
2367         int to_write, offset, len, tp_len, nr_frags, len_max;
2368         struct socket *sock = po->sk.sk_socket;
2369         struct page *page;
2370         void *data;
2371         int err;
2372
2373         ph.raw = frame;
2374
2375         skb->protocol = proto;
2376         skb->dev = dev;
2377         skb->priority = po->sk.sk_priority;
2378         skb->mark = po->sk.sk_mark;
2379         sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2380         skb_shinfo(skb)->destructor_arg = ph.raw;
2381
2382         switch (po->tp_version) {
2383         case TPACKET_V2:
2384                 tp_len = ph.h2->tp_len;
2385                 break;
2386         default:
2387                 tp_len = ph.h1->tp_len;
2388                 break;
2389         }
2390         if (unlikely(tp_len > size_max)) {
2391                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2392                 return -EMSGSIZE;
2393         }
2394
2395         skb_reserve(skb, hlen);
2396         skb_reset_network_header(skb);
2397
2398         if (unlikely(po->tp_tx_has_off)) {
2399                 int off_min, off_max, off;
2400                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2401                 off_max = po->tx_ring.frame_size - tp_len;
2402                 if (sock->type == SOCK_DGRAM) {
2403                         switch (po->tp_version) {
2404                         case TPACKET_V2:
2405                                 off = ph.h2->tp_net;
2406                                 break;
2407                         default:
2408                                 off = ph.h1->tp_net;
2409                                 break;
2410                         }
2411                 } else {
2412                         switch (po->tp_version) {
2413                         case TPACKET_V2:
2414                                 off = ph.h2->tp_mac;
2415                                 break;
2416                         default:
2417                                 off = ph.h1->tp_mac;
2418                                 break;
2419                         }
2420                 }
2421                 if (unlikely((off < off_min) || (off_max < off)))
2422                         return -EINVAL;
2423                 data = ph.raw + off;
2424         } else {
2425                 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2426         }
2427         to_write = tp_len;
2428
2429         if (sock->type == SOCK_DGRAM) {
2430                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2431                                 NULL, tp_len);
2432                 if (unlikely(err < 0))
2433                         return -EINVAL;
2434         } else if (dev->hard_header_len) {
2435                 int hdrlen = min_t(int, dev->hard_header_len, tp_len);
2436
2437                 skb_push(skb, dev->hard_header_len);
2438                 err = skb_store_bits(skb, 0, data, hdrlen);
2439                 if (unlikely(err))
2440                         return err;
2441                 if (!dev_validate_header(dev, skb->data, hdrlen))
2442                         return -EINVAL;
2443                 if (!skb->protocol)
2444                         tpacket_set_protocol(dev, skb);
2445
2446                 data += hdrlen;
2447                 to_write -= hdrlen;
2448         }
2449
2450         offset = offset_in_page(data);
2451         len_max = PAGE_SIZE - offset;
2452         len = ((to_write > len_max) ? len_max : to_write);
2453
2454         skb->data_len = to_write;
2455         skb->len += to_write;
2456         skb->truesize += to_write;
2457         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2458
2459         while (likely(to_write)) {
2460                 nr_frags = skb_shinfo(skb)->nr_frags;
2461
2462                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2463                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2464                                MAX_SKB_FRAGS);
2465                         return -EFAULT;
2466                 }
2467
2468                 page = pgv_to_page(data);
2469                 data += len;
2470                 flush_dcache_page(page);
2471                 get_page(page);
2472                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2473                 to_write -= len;
2474                 offset = 0;
2475                 len_max = PAGE_SIZE;
2476                 len = ((to_write > len_max) ? len_max : to_write);
2477         }
2478
2479         skb_probe_transport_header(skb, 0);
2480
2481         return tp_len;
2482 }
2483
2484 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2485 {
2486         struct sk_buff *skb;
2487         struct net_device *dev;
2488         __be16 proto;
2489         int err, reserve = 0;
2490         void *ph;
2491         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2492         bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2493         unsigned char *addr = NULL;
2494         int tp_len, size_max;
2495         int len_sum = 0;
2496         int status = TP_STATUS_AVAILABLE;
2497         int hlen, tlen;
2498
2499         mutex_lock(&po->pg_vec_lock);
2500
2501         if (likely(saddr == NULL)) {
2502                 dev     = packet_cached_dev_get(po);
2503                 proto   = po->num;
2504                 addr    = NULL;
2505         } else {
2506                 err = -EINVAL;
2507                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2508                         goto out;
2509                 if (msg->msg_namelen < (saddr->sll_halen
2510                                         + offsetof(struct sockaddr_ll,
2511                                                 sll_addr)))
2512                         goto out;
2513                 proto   = saddr->sll_protocol;
2514                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2515                 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2516                         if (dev && msg->msg_namelen < dev->addr_len +
2517                                    offsetof(struct sockaddr_ll, sll_addr))
2518                                 goto out_put;
2519                         addr = saddr->sll_addr;
2520                 }
2521         }
2522
2523         err = -ENXIO;
2524         if (unlikely(dev == NULL))
2525                 goto out;
2526         err = -ENETDOWN;
2527         if (unlikely(!(dev->flags & IFF_UP)))
2528                 goto out_put;
2529
2530         if (po->sk.sk_socket->type == SOCK_RAW)
2531                 reserve = dev->hard_header_len;
2532         size_max = po->tx_ring.frame_size
2533                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2534
2535         if (size_max > dev->mtu + reserve + VLAN_HLEN)
2536                 size_max = dev->mtu + reserve + VLAN_HLEN;
2537
2538         do {
2539                 ph = packet_current_frame(po, &po->tx_ring,
2540                                           TP_STATUS_SEND_REQUEST);
2541                 if (unlikely(ph == NULL)) {
2542                         if (need_wait && need_resched())
2543                                 schedule();
2544                         continue;
2545                 }
2546
2547                 status = TP_STATUS_SEND_REQUEST;
2548                 hlen = LL_RESERVED_SPACE(dev);
2549                 tlen = dev->needed_tailroom;
2550                 skb = sock_alloc_send_skb(&po->sk,
2551                                 hlen + tlen + sizeof(struct sockaddr_ll),
2552                                 !need_wait, &err);
2553
2554                 if (unlikely(skb == NULL)) {
2555                         /* we assume the socket was initially writeable ... */
2556                         if (likely(len_sum > 0))
2557                                 err = len_sum;
2558                         goto out_status;
2559                 }
2560                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2561                                           addr, hlen);
2562                 if (likely(tp_len >= 0) &&
2563                     tp_len > dev->mtu + reserve &&
2564                     !packet_extra_vlan_len_allowed(dev, skb))
2565                         tp_len = -EMSGSIZE;
2566
2567                 if (unlikely(tp_len < 0)) {
2568                         if (po->tp_loss) {
2569                                 __packet_set_status(po, ph,
2570                                                 TP_STATUS_AVAILABLE);
2571                                 packet_increment_head(&po->tx_ring);
2572                                 kfree_skb(skb);
2573                                 continue;
2574                         } else {
2575                                 status = TP_STATUS_WRONG_FORMAT;
2576                                 err = tp_len;
2577                                 goto out_status;
2578                         }
2579                 }
2580
2581                 packet_pick_tx_queue(dev, skb);
2582
2583                 skb->destructor = tpacket_destruct_skb;
2584                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2585                 packet_inc_pending(&po->tx_ring);
2586
2587                 status = TP_STATUS_SEND_REQUEST;
2588                 err = po->xmit(skb);
2589                 if (unlikely(err > 0)) {
2590                         err = net_xmit_errno(err);
2591                         if (err && __packet_get_status(po, ph) ==
2592                                    TP_STATUS_AVAILABLE) {
2593                                 /* skb was destructed already */
2594                                 skb = NULL;
2595                                 goto out_status;
2596                         }
2597                         /*
2598                          * skb was dropped but not destructed yet;
2599                          * let's treat it like congestion or err < 0
2600                          */
2601                         err = 0;
2602                 }
2603                 packet_increment_head(&po->tx_ring);
2604                 len_sum += tp_len;
2605         } while (likely((ph != NULL) ||
2606                 /* Note: packet_read_pending() might be slow if we have
2607                  * to call it as it's per_cpu variable, but in fast-path
2608                  * we already short-circuit the loop with the first
2609                  * condition, and luckily don't have to go that path
2610                  * anyway.
2611                  */
2612                  (need_wait && packet_read_pending(&po->tx_ring))));
2613
2614         err = len_sum;
2615         goto out_put;
2616
2617 out_status:
2618         __packet_set_status(po, ph, status);
2619         kfree_skb(skb);
2620 out_put:
2621         dev_put(dev);
2622 out:
2623         mutex_unlock(&po->pg_vec_lock);
2624         return err;
2625 }
2626
2627 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2628                                         size_t reserve, size_t len,
2629                                         size_t linear, int noblock,
2630                                         int *err)
2631 {
2632         struct sk_buff *skb;
2633
2634         /* Under a page?  Don't bother with paged skb. */
2635         if (prepad + len < PAGE_SIZE || !linear)
2636                 linear = len;
2637
2638         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2639                                    err, 0);
2640         if (!skb)
2641                 return NULL;
2642
2643         skb_reserve(skb, reserve);
2644         skb_put(skb, linear);
2645         skb->data_len = len - linear;
2646         skb->len += len - linear;
2647
2648         return skb;
2649 }
2650
2651 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2652 {
2653         struct sock *sk = sock->sk;
2654         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2655         struct sk_buff *skb;
2656         struct net_device *dev;
2657         __be16 proto;
2658         unsigned char *addr = NULL;
2659         int err, reserve = 0;
2660         struct sockcm_cookie sockc;
2661         struct virtio_net_hdr vnet_hdr = { 0 };
2662         int offset = 0;
2663         int vnet_hdr_len;
2664         struct packet_sock *po = pkt_sk(sk);
2665         unsigned short gso_type = 0;
2666         bool has_vnet_hdr = false;
2667         int hlen, tlen, linear;
2668         int extra_len = 0;
2669         ssize_t n;
2670
2671         /*
2672          *      Get and verify the address.
2673          */
2674
2675         if (likely(saddr == NULL)) {
2676                 dev     = packet_cached_dev_get(po);
2677                 proto   = po->num;
2678         } else {
2679                 err = -EINVAL;
2680                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2681                         goto out;
2682                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2683                         goto out;
2684                 proto   = saddr->sll_protocol;
2685                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2686                 if (sock->type == SOCK_DGRAM) {
2687                         if (dev && msg->msg_namelen < dev->addr_len +
2688                                    offsetof(struct sockaddr_ll, sll_addr))
2689                                 goto out_unlock;
2690                         addr = saddr->sll_addr;
2691                 }
2692         }
2693
2694         err = -ENXIO;
2695         if (unlikely(dev == NULL))
2696                 goto out_unlock;
2697         err = -ENETDOWN;
2698         if (unlikely(!(dev->flags & IFF_UP)))
2699                 goto out_unlock;
2700
2701         sockc.mark = sk->sk_mark;
2702         if (msg->msg_controllen) {
2703                 err = sock_cmsg_send(sk, msg, &sockc);
2704                 if (unlikely(err))
2705                         goto out_unlock;
2706         }
2707
2708         if (sock->type == SOCK_RAW)
2709                 reserve = dev->hard_header_len;
2710         if (po->has_vnet_hdr) {
2711                 vnet_hdr_len = sizeof(vnet_hdr);
2712
2713                 err = -EINVAL;
2714                 if (len < vnet_hdr_len)
2715                         goto out_unlock;
2716
2717                 len -= vnet_hdr_len;
2718
2719                 err = -EFAULT;
2720                 n = copy_from_iter(&vnet_hdr, vnet_hdr_len, &msg->msg_iter);
2721                 if (n != vnet_hdr_len)
2722                         goto out_unlock;
2723
2724                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2725                     (__virtio16_to_cpu(vio_le(), vnet_hdr.csum_start) +
2726                      __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset) + 2 >
2727                       __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len)))
2728                         vnet_hdr.hdr_len = __cpu_to_virtio16(vio_le(),
2729                                  __virtio16_to_cpu(vio_le(), vnet_hdr.csum_start) +
2730                                 __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset) + 2);
2731
2732                 err = -EINVAL;
2733                 if (__virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len) > len)
2734                         goto out_unlock;
2735
2736                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2737                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2738                         case VIRTIO_NET_HDR_GSO_TCPV4:
2739                                 gso_type = SKB_GSO_TCPV4;
2740                                 break;
2741                         case VIRTIO_NET_HDR_GSO_TCPV6:
2742                                 gso_type = SKB_GSO_TCPV6;
2743                                 break;
2744                         case VIRTIO_NET_HDR_GSO_UDP:
2745                                 gso_type = SKB_GSO_UDP;
2746                                 break;
2747                         default:
2748                                 goto out_unlock;
2749                         }
2750
2751                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2752                                 gso_type |= SKB_GSO_TCP_ECN;
2753
2754                         if (vnet_hdr.gso_size == 0)
2755                                 goto out_unlock;
2756
2757                 }
2758                 has_vnet_hdr = true;
2759         }
2760
2761         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2762                 if (!netif_supports_nofcs(dev)) {
2763                         err = -EPROTONOSUPPORT;
2764                         goto out_unlock;
2765                 }
2766                 extra_len = 4; /* We're doing our own CRC */
2767         }
2768
2769         err = -EMSGSIZE;
2770         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2771                 goto out_unlock;
2772
2773         err = -ENOBUFS;
2774         hlen = LL_RESERVED_SPACE(dev);
2775         tlen = dev->needed_tailroom;
2776         linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2777         linear = max(linear, min_t(int, len, dev->hard_header_len));
2778         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2779                                msg->msg_flags & MSG_DONTWAIT, &err);
2780         if (skb == NULL)
2781                 goto out_unlock;
2782
2783         skb_reset_network_header(skb);
2784
2785         err = -EINVAL;
2786         if (sock->type == SOCK_DGRAM) {
2787                 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2788                 if (unlikely(offset < 0))
2789                         goto out_free;
2790         } else if (reserve) {
2791                 skb_reserve(skb, -reserve);
2792                 if (len < reserve)
2793                         skb_reset_network_header(skb);
2794         }
2795
2796         /* Returns -EFAULT on error */
2797         err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2798         if (err)
2799                 goto out_free;
2800
2801         if (sock->type == SOCK_RAW &&
2802             !dev_validate_header(dev, skb->data, len)) {
2803                 err = -EINVAL;
2804                 goto out_free;
2805         }
2806
2807         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2808
2809         if (!gso_type && (len > dev->mtu + reserve + extra_len) &&
2810             !packet_extra_vlan_len_allowed(dev, skb)) {
2811                 err = -EMSGSIZE;
2812                 goto out_free;
2813         }
2814
2815         skb->protocol = proto;
2816         skb->dev = dev;
2817         skb->priority = sk->sk_priority;
2818         skb->mark = sockc.mark;
2819
2820         packet_pick_tx_queue(dev, skb);
2821
2822         if (has_vnet_hdr) {
2823                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2824                         u16 s = __virtio16_to_cpu(vio_le(), vnet_hdr.csum_start);
2825                         u16 o = __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset);
2826                         if (!skb_partial_csum_set(skb, s, o)) {
2827                                 err = -EINVAL;
2828                                 goto out_free;
2829                         }
2830                 }
2831
2832                 skb_shinfo(skb)->gso_size =
2833                         __virtio16_to_cpu(vio_le(), vnet_hdr.gso_size);
2834                 skb_shinfo(skb)->gso_type = gso_type;
2835
2836                 /* Header must be checked, and gso_segs computed. */
2837                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2838                 skb_shinfo(skb)->gso_segs = 0;
2839
2840                 len += vnet_hdr_len;
2841         }
2842
2843         skb_probe_transport_header(skb, reserve);
2844
2845         if (unlikely(extra_len == 4))
2846                 skb->no_fcs = 1;
2847
2848         err = po->xmit(skb);
2849         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2850                 goto out_unlock;
2851
2852         dev_put(dev);
2853
2854         return len;
2855
2856 out_free:
2857         kfree_skb(skb);
2858 out_unlock:
2859         if (dev)
2860                 dev_put(dev);
2861 out:
2862         return err;
2863 }
2864
2865 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2866 {
2867         struct sock *sk = sock->sk;
2868         struct packet_sock *po = pkt_sk(sk);
2869
2870         if (po->tx_ring.pg_vec)
2871                 return tpacket_snd(po, msg);
2872         else
2873                 return packet_snd(sock, msg, len);
2874 }
2875
2876 /*
2877  *      Close a PACKET socket. This is fairly simple. We immediately go
2878  *      to 'closed' state and remove our protocol entry in the device list.
2879  */
2880
2881 static int packet_release(struct socket *sock)
2882 {
2883         struct sock *sk = sock->sk;
2884         struct packet_sock *po;
2885         struct packet_fanout *f;
2886         struct net *net;
2887         union tpacket_req_u req_u;
2888
2889         if (!sk)
2890                 return 0;
2891
2892         net = sock_net(sk);
2893         po = pkt_sk(sk);
2894
2895         mutex_lock(&net->packet.sklist_lock);
2896         sk_del_node_init_rcu(sk);
2897         mutex_unlock(&net->packet.sklist_lock);
2898
2899         preempt_disable();
2900         sock_prot_inuse_add(net, sk->sk_prot, -1);
2901         preempt_enable();
2902
2903         spin_lock(&po->bind_lock);
2904         unregister_prot_hook(sk, false);
2905         packet_cached_dev_reset(po);
2906
2907         if (po->prot_hook.dev) {
2908                 dev_put(po->prot_hook.dev);
2909                 po->prot_hook.dev = NULL;
2910         }
2911         spin_unlock(&po->bind_lock);
2912
2913         packet_flush_mclist(sk);
2914
2915         lock_sock(sk);
2916         if (po->rx_ring.pg_vec) {
2917                 memset(&req_u, 0, sizeof(req_u));
2918                 packet_set_ring(sk, &req_u, 1, 0);
2919         }
2920
2921         if (po->tx_ring.pg_vec) {
2922                 memset(&req_u, 0, sizeof(req_u));
2923                 packet_set_ring(sk, &req_u, 1, 1);
2924         }
2925         release_sock(sk);
2926
2927         f = fanout_release(sk);
2928
2929         synchronize_net();
2930
2931         if (f) {
2932                 kfree(po->rollover);
2933                 fanout_release_data(f);
2934                 kfree(f);
2935         }
2936         /*
2937          *      Now the socket is dead. No more input will appear.
2938          */
2939         sock_orphan(sk);
2940         sock->sk = NULL;
2941
2942         /* Purge queues */
2943
2944         skb_queue_purge(&sk->sk_receive_queue);
2945         packet_free_pending(po);
2946         sk_refcnt_debug_release(sk);
2947
2948         sock_put(sk);
2949         return 0;
2950 }
2951
2952 /*
2953  *      Attach a packet hook.
2954  */
2955
2956 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
2957                           __be16 proto)
2958 {
2959         struct packet_sock *po = pkt_sk(sk);
2960         struct net_device *dev_curr;
2961         __be16 proto_curr;
2962         bool need_rehook;
2963         struct net_device *dev = NULL;
2964         int ret = 0;
2965         bool unlisted = false;
2966
2967         lock_sock(sk);
2968         spin_lock(&po->bind_lock);
2969         rcu_read_lock();
2970
2971         if (po->fanout) {
2972                 ret = -EINVAL;
2973                 goto out_unlock;
2974         }
2975
2976         if (name) {
2977                 dev = dev_get_by_name_rcu(sock_net(sk), name);
2978                 if (!dev) {
2979                         ret = -ENODEV;
2980                         goto out_unlock;
2981                 }
2982         } else if (ifindex) {
2983                 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
2984                 if (!dev) {
2985                         ret = -ENODEV;
2986                         goto out_unlock;
2987                 }
2988         }
2989
2990         if (dev)
2991                 dev_hold(dev);
2992
2993         proto_curr = po->prot_hook.type;
2994         dev_curr = po->prot_hook.dev;
2995
2996         need_rehook = proto_curr != proto || dev_curr != dev;
2997
2998         if (need_rehook) {
2999                 if (po->running) {
3000                         rcu_read_unlock();
3001                         /* prevents packet_notifier() from calling
3002                          * register_prot_hook()
3003                          */
3004                         po->num = 0;
3005                         __unregister_prot_hook(sk, true);
3006                         rcu_read_lock();
3007                         dev_curr = po->prot_hook.dev;
3008                         if (dev)
3009                                 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3010                                                                  dev->ifindex);
3011                 }
3012
3013                 BUG_ON(po->running);
3014                 po->num = proto;
3015                 po->prot_hook.type = proto;
3016
3017                 if (unlikely(unlisted)) {
3018                         dev_put(dev);
3019                         po->prot_hook.dev = NULL;
3020                         po->ifindex = -1;
3021                         packet_cached_dev_reset(po);
3022                 } else {
3023                         po->prot_hook.dev = dev;
3024                         po->ifindex = dev ? dev->ifindex : 0;
3025                         packet_cached_dev_assign(po, dev);
3026                 }
3027         }
3028         if (dev_curr)
3029                 dev_put(dev_curr);
3030
3031         if (proto == 0 || !need_rehook)
3032                 goto out_unlock;
3033
3034         if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3035                 register_prot_hook(sk);
3036         } else {
3037                 sk->sk_err = ENETDOWN;
3038                 if (!sock_flag(sk, SOCK_DEAD))
3039                         sk->sk_error_report(sk);
3040         }
3041
3042 out_unlock:
3043         rcu_read_unlock();
3044         spin_unlock(&po->bind_lock);
3045         release_sock(sk);
3046         return ret;
3047 }
3048
3049 /*
3050  *      Bind a packet socket to a device
3051  */
3052
3053 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3054                             int addr_len)
3055 {
3056         struct sock *sk = sock->sk;
3057         char name[sizeof(uaddr->sa_data) + 1];
3058
3059         /*
3060          *      Check legality
3061          */
3062
3063         if (addr_len != sizeof(struct sockaddr))
3064                 return -EINVAL;
3065         /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3066          * zero-terminated.
3067          */
3068         memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3069         name[sizeof(uaddr->sa_data)] = 0;
3070
3071         return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3072 }
3073
3074 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3075 {
3076         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3077         struct sock *sk = sock->sk;
3078
3079         /*
3080          *      Check legality
3081          */
3082
3083         if (addr_len < sizeof(struct sockaddr_ll))
3084                 return -EINVAL;
3085         if (sll->sll_family != AF_PACKET)
3086                 return -EINVAL;
3087
3088         return packet_do_bind(sk, NULL, sll->sll_ifindex,
3089                               sll->sll_protocol ? : pkt_sk(sk)->num);
3090 }
3091
3092 static struct proto packet_proto = {
3093         .name     = "PACKET",
3094         .owner    = THIS_MODULE,
3095         .obj_size = sizeof(struct packet_sock),
3096 };
3097
3098 /*
3099  *      Create a packet of type SOCK_PACKET.
3100  */
3101
3102 static int packet_create(struct net *net, struct socket *sock, int protocol,
3103                          int kern)
3104 {
3105         struct sock *sk;
3106         struct packet_sock *po;
3107         __be16 proto = (__force __be16)protocol; /* weird, but documented */
3108         int err;
3109
3110         if (!ns_capable(net->user_ns, CAP_NET_RAW))
3111                 return -EPERM;
3112         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3113             sock->type != SOCK_PACKET)
3114                 return -ESOCKTNOSUPPORT;
3115
3116         sock->state = SS_UNCONNECTED;
3117
3118         err = -ENOBUFS;
3119         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3120         if (sk == NULL)
3121                 goto out;
3122
3123         sock->ops = &packet_ops;
3124         if (sock->type == SOCK_PACKET)
3125                 sock->ops = &packet_ops_spkt;
3126
3127         sock_init_data(sock, sk);
3128
3129         po = pkt_sk(sk);
3130         sk->sk_family = PF_PACKET;
3131         po->num = proto;
3132         po->xmit = dev_queue_xmit;
3133
3134         err = packet_alloc_pending(po);
3135         if (err)
3136                 goto out2;
3137
3138         packet_cached_dev_reset(po);
3139
3140         sk->sk_destruct = packet_sock_destruct;
3141         sk_refcnt_debug_inc(sk);
3142
3143         /*
3144          *      Attach a protocol block
3145          */
3146
3147         spin_lock_init(&po->bind_lock);
3148         mutex_init(&po->pg_vec_lock);
3149         po->rollover = NULL;
3150         po->prot_hook.func = packet_rcv;
3151
3152         if (sock->type == SOCK_PACKET)
3153                 po->prot_hook.func = packet_rcv_spkt;
3154
3155         po->prot_hook.af_packet_priv = sk;
3156
3157         if (proto) {
3158                 po->prot_hook.type = proto;
3159                 __register_prot_hook(sk);
3160         }
3161
3162         mutex_lock(&net->packet.sklist_lock);
3163         sk_add_node_tail_rcu(sk, &net->packet.sklist);
3164         mutex_unlock(&net->packet.sklist_lock);
3165
3166         preempt_disable();
3167         sock_prot_inuse_add(net, &packet_proto, 1);
3168         preempt_enable();
3169
3170         return 0;
3171 out2:
3172         sk_free(sk);
3173 out:
3174         return err;
3175 }
3176
3177 /*
3178  *      Pull a packet from our receive queue and hand it to the user.
3179  *      If necessary we block.
3180  */
3181
3182 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3183                           int flags)
3184 {
3185         struct sock *sk = sock->sk;
3186         struct sk_buff *skb;
3187         int copied, err;
3188         int vnet_hdr_len = 0;
3189         unsigned int origlen = 0;
3190
3191         err = -EINVAL;
3192         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3193                 goto out;
3194
3195 #if 0
3196         /* What error should we return now? EUNATTACH? */
3197         if (pkt_sk(sk)->ifindex < 0)
3198                 return -ENODEV;
3199 #endif
3200
3201         if (flags & MSG_ERRQUEUE) {
3202                 err = sock_recv_errqueue(sk, msg, len,
3203                                          SOL_PACKET, PACKET_TX_TIMESTAMP);
3204                 goto out;
3205         }
3206
3207         /*
3208          *      Call the generic datagram receiver. This handles all sorts
3209          *      of horrible races and re-entrancy so we can forget about it
3210          *      in the protocol layers.
3211          *
3212          *      Now it will return ENETDOWN, if device have just gone down,
3213          *      but then it will block.
3214          */
3215
3216         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3217
3218         /*
3219          *      An error occurred so return it. Because skb_recv_datagram()
3220          *      handles the blocking we don't see and worry about blocking
3221          *      retries.
3222          */
3223
3224         if (skb == NULL)
3225                 goto out;
3226
3227         if (pkt_sk(sk)->pressure)
3228                 packet_rcv_has_room(pkt_sk(sk), NULL);
3229
3230         if (pkt_sk(sk)->has_vnet_hdr) {
3231                 struct virtio_net_hdr vnet_hdr = { 0 };
3232
3233                 err = -EINVAL;
3234                 vnet_hdr_len = sizeof(vnet_hdr);
3235                 if (len < vnet_hdr_len)
3236                         goto out_free;
3237
3238                 len -= vnet_hdr_len;
3239
3240                 if (skb_is_gso(skb)) {
3241                         struct skb_shared_info *sinfo = skb_shinfo(skb);
3242
3243                         /* This is a hint as to how much should be linear. */
3244                         vnet_hdr.hdr_len =
3245                                 __cpu_to_virtio16(vio_le(), skb_headlen(skb));
3246                         vnet_hdr.gso_size =
3247                                 __cpu_to_virtio16(vio_le(), sinfo->gso_size);
3248                         if (sinfo->gso_type & SKB_GSO_TCPV4)
3249                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
3250                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
3251                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
3252                         else if (sinfo->gso_type & SKB_GSO_UDP)
3253                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
3254                         else if (sinfo->gso_type & SKB_GSO_FCOE)
3255                                 goto out_free;
3256                         else
3257                                 BUG();
3258                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
3259                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
3260                 } else
3261                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
3262
3263                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3264                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
3265                         vnet_hdr.csum_start = __cpu_to_virtio16(vio_le(),
3266                                           skb_checksum_start_offset(skb));
3267                         vnet_hdr.csum_offset = __cpu_to_virtio16(vio_le(),
3268                                                          skb->csum_offset);
3269                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3270                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
3271                 } /* else everything is zero */
3272
3273                 err = memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_len);
3274                 if (err < 0)
3275                         goto out_free;
3276         }
3277
3278         /* You lose any data beyond the buffer you gave. If it worries
3279          * a user program they can ask the device for its MTU
3280          * anyway.
3281          */
3282         copied = skb->len;
3283         if (copied > len) {
3284                 copied = len;
3285                 msg->msg_flags |= MSG_TRUNC;
3286         }
3287
3288         err = skb_copy_datagram_msg(skb, 0, msg, copied);
3289         if (err)
3290                 goto out_free;
3291
3292         if (sock->type != SOCK_PACKET) {
3293                 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3294
3295                 /* Original length was stored in sockaddr_ll fields */
3296                 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3297                 sll->sll_family = AF_PACKET;
3298                 sll->sll_protocol = skb->protocol;
3299         }
3300
3301         sock_recv_ts_and_drops(msg, sk, skb);
3302
3303         if (msg->msg_name) {
3304                 /* If the address length field is there to be filled
3305                  * in, we fill it in now.
3306                  */
3307                 if (sock->type == SOCK_PACKET) {
3308                         __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3309                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
3310                 } else {
3311                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3312
3313                         msg->msg_namelen = sll->sll_halen +
3314                                 offsetof(struct sockaddr_ll, sll_addr);
3315                 }
3316                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3317                        msg->msg_namelen);
3318         }
3319
3320         if (pkt_sk(sk)->auxdata) {
3321                 struct tpacket_auxdata aux;
3322
3323                 aux.tp_status = TP_STATUS_USER;
3324                 if (skb->ip_summed == CHECKSUM_PARTIAL)
3325                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3326                 else if (skb->pkt_type != PACKET_OUTGOING &&
3327                          (skb->ip_summed == CHECKSUM_COMPLETE ||
3328                           skb_csum_unnecessary(skb)))
3329                         aux.tp_status |= TP_STATUS_CSUM_VALID;
3330
3331                 aux.tp_len = origlen;
3332                 aux.tp_snaplen = skb->len;
3333                 aux.tp_mac = 0;
3334                 aux.tp_net = skb_network_offset(skb);
3335                 if (skb_vlan_tag_present(skb)) {
3336                         aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3337                         aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3338                         aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3339                 } else {
3340                         aux.tp_vlan_tci = 0;
3341                         aux.tp_vlan_tpid = 0;
3342                 }
3343                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3344         }
3345
3346         /*
3347          *      Free or return the buffer as appropriate. Again this
3348          *      hides all the races and re-entrancy issues from us.
3349          */
3350         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3351
3352 out_free:
3353         skb_free_datagram(sk, skb);
3354 out:
3355         return err;
3356 }
3357
3358 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3359                                int *uaddr_len, int peer)
3360 {
3361         struct net_device *dev;
3362         struct sock *sk = sock->sk;
3363
3364         if (peer)
3365                 return -EOPNOTSUPP;
3366
3367         uaddr->sa_family = AF_PACKET;
3368         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3369         rcu_read_lock();
3370         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3371         if (dev)
3372                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3373         rcu_read_unlock();
3374         *uaddr_len = sizeof(*uaddr);
3375
3376         return 0;
3377 }
3378
3379 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3380                           int *uaddr_len, int peer)
3381 {
3382         struct net_device *dev;
3383         struct sock *sk = sock->sk;
3384         struct packet_sock *po = pkt_sk(sk);
3385         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3386
3387         if (peer)
3388                 return -EOPNOTSUPP;
3389
3390         sll->sll_family = AF_PACKET;
3391         sll->sll_ifindex = po->ifindex;
3392         sll->sll_protocol = po->num;
3393         sll->sll_pkttype = 0;
3394         rcu_read_lock();
3395         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3396         if (dev) {
3397                 sll->sll_hatype = dev->type;
3398                 sll->sll_halen = dev->addr_len;
3399                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3400         } else {
3401                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
3402                 sll->sll_halen = 0;
3403         }
3404         rcu_read_unlock();
3405         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3406
3407         return 0;
3408 }
3409
3410 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3411                          int what)
3412 {
3413         switch (i->type) {
3414         case PACKET_MR_MULTICAST:
3415                 if (i->alen != dev->addr_len)
3416                         return -EINVAL;
3417                 if (what > 0)
3418                         return dev_mc_add(dev, i->addr);
3419                 else
3420                         return dev_mc_del(dev, i->addr);
3421                 break;
3422         case PACKET_MR_PROMISC:
3423                 return dev_set_promiscuity(dev, what);
3424         case PACKET_MR_ALLMULTI:
3425                 return dev_set_allmulti(dev, what);
3426         case PACKET_MR_UNICAST:
3427                 if (i->alen != dev->addr_len)
3428                         return -EINVAL;
3429                 if (what > 0)
3430                         return dev_uc_add(dev, i->addr);
3431                 else
3432                         return dev_uc_del(dev, i->addr);
3433                 break;
3434         default:
3435                 break;
3436         }
3437         return 0;
3438 }
3439
3440 static void packet_dev_mclist_delete(struct net_device *dev,
3441                                      struct packet_mclist **mlp)
3442 {
3443         struct packet_mclist *ml;
3444
3445         while ((ml = *mlp) != NULL) {
3446                 if (ml->ifindex == dev->ifindex) {
3447                         packet_dev_mc(dev, ml, -1);
3448                         *mlp = ml->next;
3449                         kfree(ml);
3450                 } else
3451                         mlp = &ml->next;
3452         }
3453 }
3454
3455 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3456 {
3457         struct packet_sock *po = pkt_sk(sk);
3458         struct packet_mclist *ml, *i;
3459         struct net_device *dev;
3460         int err;
3461
3462         rtnl_lock();
3463
3464         err = -ENODEV;
3465         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3466         if (!dev)
3467                 goto done;
3468
3469         err = -EINVAL;
3470         if (mreq->mr_alen > dev->addr_len)
3471                 goto done;
3472
3473         err = -ENOBUFS;
3474         i = kmalloc(sizeof(*i), GFP_KERNEL);
3475         if (i == NULL)
3476                 goto done;
3477
3478         err = 0;
3479         for (ml = po->mclist; ml; ml = ml->next) {
3480                 if (ml->ifindex == mreq->mr_ifindex &&
3481                     ml->type == mreq->mr_type &&
3482                     ml->alen == mreq->mr_alen &&
3483                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3484                         ml->count++;
3485                         /* Free the new element ... */
3486                         kfree(i);
3487                         goto done;
3488                 }
3489         }
3490
3491         i->type = mreq->mr_type;
3492         i->ifindex = mreq->mr_ifindex;
3493         i->alen = mreq->mr_alen;
3494         memcpy(i->addr, mreq->mr_address, i->alen);
3495         memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3496         i->count = 1;
3497         i->next = po->mclist;
3498         po->mclist = i;
3499         err = packet_dev_mc(dev, i, 1);
3500         if (err) {
3501                 po->mclist = i->next;
3502                 kfree(i);
3503         }
3504
3505 done:
3506         rtnl_unlock();
3507         return err;
3508 }
3509
3510 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3511 {
3512         struct packet_mclist *ml, **mlp;
3513
3514         rtnl_lock();
3515
3516         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3517                 if (ml->ifindex == mreq->mr_ifindex &&
3518                     ml->type == mreq->mr_type &&
3519                     ml->alen == mreq->mr_alen &&
3520                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3521                         if (--ml->count == 0) {
3522                                 struct net_device *dev;
3523                                 *mlp = ml->next;
3524                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3525                                 if (dev)
3526                                         packet_dev_mc(dev, ml, -1);
3527                                 kfree(ml);
3528                         }
3529                         break;
3530                 }
3531         }
3532         rtnl_unlock();
3533         return 0;
3534 }
3535
3536 static void packet_flush_mclist(struct sock *sk)
3537 {
3538         struct packet_sock *po = pkt_sk(sk);
3539         struct packet_mclist *ml;
3540
3541         if (!po->mclist)
3542                 return;
3543
3544         rtnl_lock();
3545         while ((ml = po->mclist) != NULL) {
3546                 struct net_device *dev;
3547
3548                 po->mclist = ml->next;
3549                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3550                 if (dev != NULL)
3551                         packet_dev_mc(dev, ml, -1);
3552                 kfree(ml);
3553         }
3554         rtnl_unlock();
3555 }
3556
3557 static int
3558 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3559 {
3560         struct sock *sk = sock->sk;
3561         struct packet_sock *po = pkt_sk(sk);
3562         int ret;
3563
3564         if (level != SOL_PACKET)
3565                 return -ENOPROTOOPT;
3566
3567         switch (optname) {
3568         case PACKET_ADD_MEMBERSHIP:
3569         case PACKET_DROP_MEMBERSHIP:
3570         {
3571                 struct packet_mreq_max mreq;
3572                 int len = optlen;
3573                 memset(&mreq, 0, sizeof(mreq));
3574                 if (len < sizeof(struct packet_mreq))
3575                         return -EINVAL;
3576                 if (len > sizeof(mreq))
3577                         len = sizeof(mreq);
3578                 if (copy_from_user(&mreq, optval, len))
3579                         return -EFAULT;
3580                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3581                         return -EINVAL;
3582                 if (optname == PACKET_ADD_MEMBERSHIP)
3583                         ret = packet_mc_add(sk, &mreq);
3584                 else
3585                         ret = packet_mc_drop(sk, &mreq);
3586                 return ret;
3587         }
3588
3589         case PACKET_RX_RING:
3590         case PACKET_TX_RING:
3591         {
3592                 union tpacket_req_u req_u;
3593                 int len;
3594
3595                 lock_sock(sk);
3596                 switch (po->tp_version) {
3597                 case TPACKET_V1:
3598                 case TPACKET_V2:
3599                         len = sizeof(req_u.req);
3600                         break;
3601                 case TPACKET_V3:
3602                 default:
3603                         len = sizeof(req_u.req3);
3604                         break;
3605                 }
3606                 if (optlen < len) {
3607                         ret = -EINVAL;
3608                 } else {
3609                         if (pkt_sk(sk)->has_vnet_hdr) {
3610                                 ret = -EINVAL;
3611                         } else {
3612                                 if (copy_from_user(&req_u.req, optval, len))
3613                                         ret = -EFAULT;
3614                                 else
3615                                         ret = packet_set_ring(sk, &req_u, 0,
3616                                                               optname == PACKET_TX_RING);
3617                         }
3618                 }
3619                 release_sock(sk);
3620                 return ret;
3621         }
3622         case PACKET_COPY_THRESH:
3623         {
3624                 int val;
3625
3626                 if (optlen != sizeof(val))
3627                         return -EINVAL;
3628                 if (copy_from_user(&val, optval, sizeof(val)))
3629                         return -EFAULT;
3630
3631                 pkt_sk(sk)->copy_thresh = val;
3632                 return 0;
3633         }
3634         case PACKET_VERSION:
3635         {
3636                 int val;
3637
3638                 if (optlen != sizeof(val))
3639                         return -EINVAL;
3640                 if (copy_from_user(&val, optval, sizeof(val)))
3641                         return -EFAULT;
3642                 switch (val) {
3643                 case TPACKET_V1:
3644                 case TPACKET_V2:
3645                 case TPACKET_V3:
3646                         break;
3647                 default:
3648                         return -EINVAL;
3649                 }
3650                 lock_sock(sk);
3651                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3652                         ret = -EBUSY;
3653                 } else {
3654                         po->tp_version = val;
3655                         ret = 0;
3656                 }
3657                 release_sock(sk);
3658                 return ret;
3659         }
3660         case PACKET_RESERVE:
3661         {
3662                 unsigned int val;
3663
3664                 if (optlen != sizeof(val))
3665                         return -EINVAL;
3666                 if (copy_from_user(&val, optval, sizeof(val)))
3667                         return -EFAULT;
3668                 if (val > INT_MAX)
3669                         return -EINVAL;
3670                 lock_sock(sk);
3671                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3672                         ret = -EBUSY;
3673                 } else {
3674                         po->tp_reserve = val;
3675                         ret = 0;
3676                 }
3677                 release_sock(sk);
3678                 return ret;
3679         }
3680         case PACKET_LOSS:
3681         {
3682                 unsigned int val;
3683
3684                 if (optlen != sizeof(val))
3685                         return -EINVAL;
3686                 if (copy_from_user(&val, optval, sizeof(val)))
3687                         return -EFAULT;
3688
3689                 lock_sock(sk);
3690                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3691                         ret = -EBUSY;
3692                 } else {
3693                         po->tp_loss = !!val;
3694                         ret = 0;
3695                 }
3696                 release_sock(sk);
3697                 return ret;
3698         }
3699         case PACKET_AUXDATA:
3700         {
3701                 int val;
3702
3703                 if (optlen < sizeof(val))
3704                         return -EINVAL;
3705                 if (copy_from_user(&val, optval, sizeof(val)))
3706                         return -EFAULT;
3707
3708                 lock_sock(sk);
3709                 po->auxdata = !!val;
3710                 release_sock(sk);
3711                 return 0;
3712         }
3713         case PACKET_ORIGDEV:
3714         {
3715                 int val;
3716
3717                 if (optlen < sizeof(val))
3718                         return -EINVAL;
3719                 if (copy_from_user(&val, optval, sizeof(val)))
3720                         return -EFAULT;
3721
3722                 lock_sock(sk);
3723                 po->origdev = !!val;
3724                 release_sock(sk);
3725                 return 0;
3726         }
3727         case PACKET_VNET_HDR:
3728         {
3729                 int val;
3730
3731                 if (sock->type != SOCK_RAW)
3732                         return -EINVAL;
3733                 if (optlen < sizeof(val))
3734                         return -EINVAL;
3735                 if (copy_from_user(&val, optval, sizeof(val)))
3736                         return -EFAULT;
3737
3738                 lock_sock(sk);
3739                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3740                         ret = -EBUSY;
3741                 } else {
3742                         po->has_vnet_hdr = !!val;
3743                         ret = 0;
3744                 }
3745                 release_sock(sk);
3746                 return ret;
3747         }
3748         case PACKET_TIMESTAMP:
3749         {
3750                 int val;
3751
3752                 if (optlen != sizeof(val))
3753                         return -EINVAL;
3754                 if (copy_from_user(&val, optval, sizeof(val)))
3755                         return -EFAULT;
3756
3757                 po->tp_tstamp = val;
3758                 return 0;
3759         }
3760         case PACKET_FANOUT:
3761         {
3762                 int val;
3763
3764                 if (optlen != sizeof(val))
3765                         return -EINVAL;
3766                 if (copy_from_user(&val, optval, sizeof(val)))
3767                         return -EFAULT;
3768
3769                 return fanout_add(sk, val & 0xffff, val >> 16);
3770         }
3771         case PACKET_FANOUT_DATA:
3772         {
3773                 if (!po->fanout)
3774                         return -EINVAL;
3775
3776                 return fanout_set_data(po, optval, optlen);
3777         }
3778         case PACKET_TX_HAS_OFF:
3779         {
3780                 unsigned int val;
3781
3782                 if (optlen != sizeof(val))
3783                         return -EINVAL;
3784                 if (copy_from_user(&val, optval, sizeof(val)))
3785                         return -EFAULT;
3786
3787                 lock_sock(sk);
3788                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3789                         ret = -EBUSY;
3790                 } else {
3791                         po->tp_tx_has_off = !!val;
3792                         ret = 0;
3793                 }
3794                 release_sock(sk);
3795                 return 0;
3796         }
3797         case PACKET_QDISC_BYPASS:
3798         {
3799                 int val;
3800
3801                 if (optlen != sizeof(val))
3802                         return -EINVAL;
3803                 if (copy_from_user(&val, optval, sizeof(val)))
3804                         return -EFAULT;
3805
3806                 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3807                 return 0;
3808         }
3809         default:
3810                 return -ENOPROTOOPT;
3811         }
3812 }
3813
3814 static int packet_getsockopt(struct socket *sock, int level, int optname,
3815                              char __user *optval, int __user *optlen)
3816 {
3817         int len;
3818         int val, lv = sizeof(val);
3819         struct sock *sk = sock->sk;
3820         struct packet_sock *po = pkt_sk(sk);
3821         void *data = &val;
3822         union tpacket_stats_u st;
3823         struct tpacket_rollover_stats rstats;
3824
3825         if (level != SOL_PACKET)
3826                 return -ENOPROTOOPT;
3827
3828         if (get_user(len, optlen))
3829                 return -EFAULT;
3830
3831         if (len < 0)
3832                 return -EINVAL;
3833
3834         switch (optname) {
3835         case PACKET_STATISTICS:
3836                 spin_lock_bh(&sk->sk_receive_queue.lock);
3837                 memcpy(&st, &po->stats, sizeof(st));
3838                 memset(&po->stats, 0, sizeof(po->stats));
3839                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3840
3841                 if (po->tp_version == TPACKET_V3) {
3842                         lv = sizeof(struct tpacket_stats_v3);
3843                         st.stats3.tp_packets += st.stats3.tp_drops;
3844                         data = &st.stats3;
3845                 } else {
3846                         lv = sizeof(struct tpacket_stats);
3847                         st.stats1.tp_packets += st.stats1.tp_drops;
3848                         data = &st.stats1;
3849                 }
3850
3851                 break;
3852         case PACKET_AUXDATA:
3853                 val = po->auxdata;
3854                 break;
3855         case PACKET_ORIGDEV:
3856                 val = po->origdev;
3857                 break;
3858         case PACKET_VNET_HDR:
3859                 val = po->has_vnet_hdr;
3860                 break;
3861         case PACKET_VERSION:
3862                 val = po->tp_version;
3863                 break;
3864         case PACKET_HDRLEN:
3865                 if (len > sizeof(int))
3866                         len = sizeof(int);
3867                 if (len < sizeof(int))
3868                         return -EINVAL;
3869                 if (copy_from_user(&val, optval, len))
3870                         return -EFAULT;
3871                 switch (val) {
3872                 case TPACKET_V1:
3873                         val = sizeof(struct tpacket_hdr);
3874                         break;
3875                 case TPACKET_V2:
3876                         val = sizeof(struct tpacket2_hdr);
3877                         break;
3878                 case TPACKET_V3:
3879                         val = sizeof(struct tpacket3_hdr);
3880                         break;
3881                 default:
3882                         return -EINVAL;
3883                 }
3884                 break;
3885         case PACKET_RESERVE:
3886                 val = po->tp_reserve;
3887                 break;
3888         case PACKET_LOSS:
3889                 val = po->tp_loss;
3890                 break;
3891         case PACKET_TIMESTAMP:
3892                 val = po->tp_tstamp;
3893                 break;
3894         case PACKET_FANOUT:
3895                 val = (po->fanout ?
3896                        ((u32)po->fanout->id |
3897                         ((u32)po->fanout->type << 16) |
3898                         ((u32)po->fanout->flags << 24)) :
3899                        0);
3900                 break;
3901         case PACKET_ROLLOVER_STATS:
3902                 if (!po->rollover)
3903                         return -EINVAL;
3904                 rstats.tp_all = atomic_long_read(&po->rollover->num);
3905                 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3906                 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3907                 data = &rstats;
3908                 lv = sizeof(rstats);
3909                 break;
3910         case PACKET_TX_HAS_OFF:
3911                 val = po->tp_tx_has_off;
3912                 break;
3913         case PACKET_QDISC_BYPASS:
3914                 val = packet_use_direct_xmit(po);
3915                 break;
3916         default:
3917                 return -ENOPROTOOPT;
3918         }
3919
3920         if (len > lv)
3921                 len = lv;
3922         if (put_user(len, optlen))
3923                 return -EFAULT;
3924         if (copy_to_user(optval, data, len))
3925                 return -EFAULT;
3926         return 0;
3927 }
3928
3929
3930 static int packet_notifier(struct notifier_block *this,
3931                            unsigned long msg, void *ptr)
3932 {
3933         struct sock *sk;
3934         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3935         struct net *net = dev_net(dev);
3936
3937         rcu_read_lock();
3938         sk_for_each_rcu(sk, &net->packet.sklist) {
3939                 struct packet_sock *po = pkt_sk(sk);
3940
3941                 switch (msg) {
3942                 case NETDEV_UNREGISTER:
3943                         if (po->mclist)
3944                                 packet_dev_mclist_delete(dev, &po->mclist);
3945                         /* fallthrough */
3946
3947                 case NETDEV_DOWN:
3948                         if (dev->ifindex == po->ifindex) {
3949                                 spin_lock(&po->bind_lock);
3950                                 if (po->running) {
3951                                         __unregister_prot_hook(sk, false);
3952                                         sk->sk_err = ENETDOWN;
3953                                         if (!sock_flag(sk, SOCK_DEAD))
3954                                                 sk->sk_error_report(sk);
3955                                 }
3956                                 if (msg == NETDEV_UNREGISTER) {
3957                                         packet_cached_dev_reset(po);
3958                                         po->ifindex = -1;
3959                                         if (po->prot_hook.dev)
3960                                                 dev_put(po->prot_hook.dev);
3961                                         po->prot_hook.dev = NULL;
3962                                 }
3963                                 spin_unlock(&po->bind_lock);
3964                         }
3965                         break;
3966                 case NETDEV_UP:
3967                         if (dev->ifindex == po->ifindex) {
3968                                 spin_lock(&po->bind_lock);
3969                                 if (po->num)
3970                                         register_prot_hook(sk);
3971                                 spin_unlock(&po->bind_lock);
3972                         }
3973                         break;
3974                 }
3975         }
3976         rcu_read_unlock();
3977         return NOTIFY_DONE;
3978 }
3979
3980
3981 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3982                         unsigned long arg)
3983 {
3984         struct sock *sk = sock->sk;
3985
3986         switch (cmd) {
3987         case SIOCOUTQ:
3988         {
3989                 int amount = sk_wmem_alloc_get(sk);
3990
3991                 return put_user(amount, (int __user *)arg);
3992         }
3993         case SIOCINQ:
3994         {
3995                 struct sk_buff *skb;
3996                 int amount = 0;
3997
3998                 spin_lock_bh(&sk->sk_receive_queue.lock);
3999                 skb = skb_peek(&sk->sk_receive_queue);
4000                 if (skb)
4001                         amount = skb->len;
4002                 spin_unlock_bh(&sk->sk_receive_queue.lock);
4003                 return put_user(amount, (int __user *)arg);
4004         }
4005         case SIOCGSTAMP:
4006                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
4007         case SIOCGSTAMPNS:
4008                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
4009
4010 #ifdef CONFIG_INET
4011         case SIOCADDRT:
4012         case SIOCDELRT:
4013         case SIOCDARP:
4014         case SIOCGARP:
4015         case SIOCSARP:
4016         case SIOCGIFADDR:
4017         case SIOCSIFADDR:
4018         case SIOCGIFBRDADDR:
4019         case SIOCSIFBRDADDR:
4020         case SIOCGIFNETMASK:
4021         case SIOCSIFNETMASK:
4022         case SIOCGIFDSTADDR:
4023         case SIOCSIFDSTADDR:
4024         case SIOCSIFFLAGS:
4025                 return inet_dgram_ops.ioctl(sock, cmd, arg);
4026 #endif
4027
4028         default:
4029                 return -ENOIOCTLCMD;
4030         }
4031         return 0;
4032 }
4033
4034 static unsigned int packet_poll(struct file *file, struct socket *sock,
4035                                 poll_table *wait)
4036 {
4037         struct sock *sk = sock->sk;
4038         struct packet_sock *po = pkt_sk(sk);
4039         unsigned int mask = datagram_poll(file, sock, wait);
4040
4041         spin_lock_bh(&sk->sk_receive_queue.lock);
4042         if (po->rx_ring.pg_vec) {
4043                 if (!packet_previous_rx_frame(po, &po->rx_ring,
4044                         TP_STATUS_KERNEL))
4045                         mask |= POLLIN | POLLRDNORM;
4046         }
4047         if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
4048                 po->pressure = 0;
4049         spin_unlock_bh(&sk->sk_receive_queue.lock);
4050         spin_lock_bh(&sk->sk_write_queue.lock);
4051         if (po->tx_ring.pg_vec) {
4052                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4053                         mask |= POLLOUT | POLLWRNORM;
4054         }
4055         spin_unlock_bh(&sk->sk_write_queue.lock);
4056         return mask;
4057 }
4058
4059
4060 /* Dirty? Well, I still did not learn better way to account
4061  * for user mmaps.
4062  */
4063
4064 static void packet_mm_open(struct vm_area_struct *vma)
4065 {
4066         struct file *file = vma->vm_file;
4067         struct socket *sock = file->private_data;
4068         struct sock *sk = sock->sk;
4069
4070         if (sk)
4071                 atomic_inc(&pkt_sk(sk)->mapped);
4072 }
4073
4074 static void packet_mm_close(struct vm_area_struct *vma)
4075 {
4076         struct file *file = vma->vm_file;
4077         struct socket *sock = file->private_data;
4078         struct sock *sk = sock->sk;
4079
4080         if (sk)
4081                 atomic_dec(&pkt_sk(sk)->mapped);
4082 }
4083
4084 static const struct vm_operations_struct packet_mmap_ops = {
4085         .open   =       packet_mm_open,
4086         .close  =       packet_mm_close,
4087 };
4088
4089 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4090                         unsigned int len)
4091 {
4092         int i;
4093
4094         for (i = 0; i < len; i++) {
4095                 if (likely(pg_vec[i].buffer)) {
4096                         if (is_vmalloc_addr(pg_vec[i].buffer))
4097                                 vfree(pg_vec[i].buffer);
4098                         else
4099                                 free_pages((unsigned long)pg_vec[i].buffer,
4100                                            order);
4101                         pg_vec[i].buffer = NULL;
4102                 }
4103         }
4104         kfree(pg_vec);
4105 }
4106
4107 static char *alloc_one_pg_vec_page(unsigned long order)
4108 {
4109         char *buffer;
4110         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4111                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4112
4113         buffer = (char *) __get_free_pages(gfp_flags, order);
4114         if (buffer)
4115                 return buffer;
4116
4117         /* __get_free_pages failed, fall back to vmalloc */
4118         buffer = vzalloc((1 << order) * PAGE_SIZE);
4119         if (buffer)
4120                 return buffer;
4121
4122         /* vmalloc failed, lets dig into swap here */
4123         gfp_flags &= ~__GFP_NORETRY;
4124         buffer = (char *) __get_free_pages(gfp_flags, order);
4125         if (buffer)
4126                 return buffer;
4127
4128         /* complete and utter failure */
4129         return NULL;
4130 }
4131
4132 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4133 {
4134         unsigned int block_nr = req->tp_block_nr;
4135         struct pgv *pg_vec;
4136         int i;
4137
4138         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4139         if (unlikely(!pg_vec))
4140                 goto out;
4141
4142         for (i = 0; i < block_nr; i++) {
4143                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4144                 if (unlikely(!pg_vec[i].buffer))
4145                         goto out_free_pgvec;
4146         }
4147
4148 out:
4149         return pg_vec;
4150
4151 out_free_pgvec:
4152         free_pg_vec(pg_vec, order, block_nr);
4153         pg_vec = NULL;
4154         goto out;
4155 }
4156
4157 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4158                 int closing, int tx_ring)
4159 {
4160         struct pgv *pg_vec = NULL;
4161         struct packet_sock *po = pkt_sk(sk);
4162         int was_running, order = 0;
4163         struct packet_ring_buffer *rb;
4164         struct sk_buff_head *rb_queue;
4165         __be16 num;
4166         int err = -EINVAL;
4167         /* Added to avoid minimal code churn */
4168         struct tpacket_req *req = &req_u->req;
4169
4170         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
4171         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
4172                 WARN(1, "Tx-ring is not supported.\n");
4173                 goto out;
4174         }
4175
4176         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4177         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4178
4179         err = -EBUSY;
4180         if (!closing) {
4181                 if (atomic_read(&po->mapped))
4182                         goto out;
4183                 if (packet_read_pending(rb))
4184                         goto out;
4185         }
4186
4187         if (req->tp_block_nr) {
4188                 unsigned int min_frame_size;
4189
4190                 /* Sanity tests and some calculations */
4191                 err = -EBUSY;
4192                 if (unlikely(rb->pg_vec))
4193                         goto out;
4194
4195                 switch (po->tp_version) {
4196                 case TPACKET_V1:
4197                         po->tp_hdrlen = TPACKET_HDRLEN;
4198                         break;
4199                 case TPACKET_V2:
4200                         po->tp_hdrlen = TPACKET2_HDRLEN;
4201                         break;
4202                 case TPACKET_V3:
4203                         po->tp_hdrlen = TPACKET3_HDRLEN;
4204                         break;
4205                 }
4206
4207                 err = -EINVAL;
4208                 if (unlikely((int)req->tp_block_size <= 0))
4209                         goto out;
4210                 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4211                         goto out;
4212                 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4213                 if (po->tp_version >= TPACKET_V3 &&
4214                     req->tp_block_size <
4215                     BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4216                         goto out;
4217                 if (unlikely(req->tp_frame_size < min_frame_size))
4218                         goto out;
4219                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4220                         goto out;
4221
4222                 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4223                 if (unlikely(rb->frames_per_block == 0))
4224                         goto out;
4225                 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4226                         goto out;
4227                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4228                                         req->tp_frame_nr))
4229                         goto out;
4230
4231                 err = -ENOMEM;
4232                 order = get_order(req->tp_block_size);
4233                 pg_vec = alloc_pg_vec(req, order);
4234                 if (unlikely(!pg_vec))
4235                         goto out;
4236                 switch (po->tp_version) {
4237                 case TPACKET_V3:
4238                 /* Transmit path is not supported. We checked
4239                  * it above but just being paranoid
4240                  */
4241                         if (!tx_ring)
4242                                 init_prb_bdqc(po, rb, pg_vec, req_u);
4243                         break;
4244                 default:
4245                         break;
4246                 }
4247         }
4248         /* Done */
4249         else {
4250                 err = -EINVAL;
4251                 if (unlikely(req->tp_frame_nr))
4252                         goto out;
4253         }
4254
4255
4256         /* Detach socket from network */
4257         spin_lock(&po->bind_lock);
4258         was_running = po->running;
4259         num = po->num;
4260         if (was_running) {
4261                 po->num = 0;
4262                 __unregister_prot_hook(sk, false);
4263         }
4264         spin_unlock(&po->bind_lock);
4265
4266         synchronize_net();
4267
4268         err = -EBUSY;
4269         mutex_lock(&po->pg_vec_lock);
4270         if (closing || atomic_read(&po->mapped) == 0) {
4271                 err = 0;
4272                 spin_lock_bh(&rb_queue->lock);
4273                 swap(rb->pg_vec, pg_vec);
4274                 rb->frame_max = (req->tp_frame_nr - 1);
4275                 rb->head = 0;
4276                 rb->frame_size = req->tp_frame_size;
4277                 spin_unlock_bh(&rb_queue->lock);
4278
4279                 swap(rb->pg_vec_order, order);
4280                 swap(rb->pg_vec_len, req->tp_block_nr);
4281
4282                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4283                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4284                                                 tpacket_rcv : packet_rcv;
4285                 skb_queue_purge(rb_queue);
4286                 if (atomic_read(&po->mapped))
4287                         pr_err("packet_mmap: vma is busy: %d\n",
4288                                atomic_read(&po->mapped));
4289         }
4290         mutex_unlock(&po->pg_vec_lock);
4291
4292         spin_lock(&po->bind_lock);
4293         if (was_running) {
4294                 po->num = num;
4295                 register_prot_hook(sk);
4296         }
4297         spin_unlock(&po->bind_lock);
4298         if (pg_vec && (po->tp_version > TPACKET_V2)) {
4299                 /* Because we don't support block-based V3 on tx-ring */
4300                 if (!tx_ring)
4301                         prb_shutdown_retire_blk_timer(po, rb_queue);
4302         }
4303
4304         if (pg_vec)
4305                 free_pg_vec(pg_vec, order, req->tp_block_nr);
4306 out:
4307         return err;
4308 }
4309
4310 static int packet_mmap(struct file *file, struct socket *sock,
4311                 struct vm_area_struct *vma)
4312 {
4313         struct sock *sk = sock->sk;
4314         struct packet_sock *po = pkt_sk(sk);
4315         unsigned long size, expected_size;
4316         struct packet_ring_buffer *rb;
4317         unsigned long start;
4318         int err = -EINVAL;
4319         int i;
4320
4321         if (vma->vm_pgoff)
4322                 return -EINVAL;
4323
4324         mutex_lock(&po->pg_vec_lock);
4325
4326         expected_size = 0;
4327         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4328                 if (rb->pg_vec) {
4329                         expected_size += rb->pg_vec_len
4330                                                 * rb->pg_vec_pages
4331                                                 * PAGE_SIZE;
4332                 }
4333         }
4334
4335         if (expected_size == 0)
4336                 goto out;
4337
4338         size = vma->vm_end - vma->vm_start;
4339         if (size != expected_size)
4340                 goto out;
4341
4342         start = vma->vm_start;
4343         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4344                 if (rb->pg_vec == NULL)
4345                         continue;
4346
4347                 for (i = 0; i < rb->pg_vec_len; i++) {
4348                         struct page *page;
4349                         void *kaddr = rb->pg_vec[i].buffer;
4350                         int pg_num;
4351
4352                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4353                                 page = pgv_to_page(kaddr);
4354                                 err = vm_insert_page(vma, start, page);
4355                                 if (unlikely(err))
4356                                         goto out;
4357                                 start += PAGE_SIZE;
4358                                 kaddr += PAGE_SIZE;
4359                         }
4360                 }
4361         }
4362
4363         atomic_inc(&po->mapped);
4364         vma->vm_ops = &packet_mmap_ops;
4365         err = 0;
4366
4367 out:
4368         mutex_unlock(&po->pg_vec_lock);
4369         return err;
4370 }
4371
4372 static const struct proto_ops packet_ops_spkt = {
4373         .family =       PF_PACKET,
4374         .owner =        THIS_MODULE,
4375         .release =      packet_release,
4376         .bind =         packet_bind_spkt,
4377         .connect =      sock_no_connect,
4378         .socketpair =   sock_no_socketpair,
4379         .accept =       sock_no_accept,
4380         .getname =      packet_getname_spkt,
4381         .poll =         datagram_poll,
4382         .ioctl =        packet_ioctl,
4383         .listen =       sock_no_listen,
4384         .shutdown =     sock_no_shutdown,
4385         .setsockopt =   sock_no_setsockopt,
4386         .getsockopt =   sock_no_getsockopt,
4387         .sendmsg =      packet_sendmsg_spkt,
4388         .recvmsg =      packet_recvmsg,
4389         .mmap =         sock_no_mmap,
4390         .sendpage =     sock_no_sendpage,
4391 };
4392
4393 static const struct proto_ops packet_ops = {
4394         .family =       PF_PACKET,
4395         .owner =        THIS_MODULE,
4396         .release =      packet_release,
4397         .bind =         packet_bind,
4398         .connect =      sock_no_connect,
4399         .socketpair =   sock_no_socketpair,
4400         .accept =       sock_no_accept,
4401         .getname =      packet_getname,
4402         .poll =         packet_poll,
4403         .ioctl =        packet_ioctl,
4404         .listen =       sock_no_listen,
4405         .shutdown =     sock_no_shutdown,
4406         .setsockopt =   packet_setsockopt,
4407         .getsockopt =   packet_getsockopt,
4408         .sendmsg =      packet_sendmsg,
4409         .recvmsg =      packet_recvmsg,
4410         .mmap =         packet_mmap,
4411         .sendpage =     sock_no_sendpage,
4412 };
4413
4414 static const struct net_proto_family packet_family_ops = {
4415         .family =       PF_PACKET,
4416         .create =       packet_create,
4417         .owner  =       THIS_MODULE,
4418 };
4419
4420 static struct notifier_block packet_netdev_notifier = {
4421         .notifier_call =        packet_notifier,
4422 };
4423
4424 #ifdef CONFIG_PROC_FS
4425
4426 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4427         __acquires(RCU)
4428 {
4429         struct net *net = seq_file_net(seq);
4430
4431         rcu_read_lock();
4432         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4433 }
4434
4435 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4436 {
4437         struct net *net = seq_file_net(seq);
4438         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4439 }
4440
4441 static void packet_seq_stop(struct seq_file *seq, void *v)
4442         __releases(RCU)
4443 {
4444         rcu_read_unlock();
4445 }
4446
4447 static int packet_seq_show(struct seq_file *seq, void *v)
4448 {
4449         if (v == SEQ_START_TOKEN)
4450                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
4451         else {
4452                 struct sock *s = sk_entry(v);
4453                 const struct packet_sock *po = pkt_sk(s);
4454
4455                 seq_printf(seq,
4456                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4457                            s,
4458                            atomic_read(&s->sk_refcnt),
4459                            s->sk_type,
4460                            ntohs(po->num),
4461                            po->ifindex,
4462                            po->running,
4463                            atomic_read(&s->sk_rmem_alloc),
4464                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4465                            sock_i_ino(s));
4466         }
4467
4468         return 0;
4469 }
4470
4471 static const struct seq_operations packet_seq_ops = {
4472         .start  = packet_seq_start,
4473         .next   = packet_seq_next,
4474         .stop   = packet_seq_stop,
4475         .show   = packet_seq_show,
4476 };
4477
4478 static int packet_seq_open(struct inode *inode, struct file *file)
4479 {
4480         return seq_open_net(inode, file, &packet_seq_ops,
4481                             sizeof(struct seq_net_private));
4482 }
4483
4484 static const struct file_operations packet_seq_fops = {
4485         .owner          = THIS_MODULE,
4486         .open           = packet_seq_open,
4487         .read           = seq_read,
4488         .llseek         = seq_lseek,
4489         .release        = seq_release_net,
4490 };
4491
4492 #endif
4493
4494 static int __net_init packet_net_init(struct net *net)
4495 {
4496         mutex_init(&net->packet.sklist_lock);
4497         INIT_HLIST_HEAD(&net->packet.sklist);
4498
4499         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4500                 return -ENOMEM;
4501
4502         return 0;
4503 }
4504
4505 static void __net_exit packet_net_exit(struct net *net)
4506 {
4507         remove_proc_entry("packet", net->proc_net);
4508 }
4509
4510 static struct pernet_operations packet_net_ops = {
4511         .init = packet_net_init,
4512         .exit = packet_net_exit,
4513 };
4514
4515
4516 static void __exit packet_exit(void)
4517 {
4518         unregister_netdevice_notifier(&packet_netdev_notifier);
4519         unregister_pernet_subsys(&packet_net_ops);
4520         sock_unregister(PF_PACKET);
4521         proto_unregister(&packet_proto);
4522 }
4523
4524 static int __init packet_init(void)
4525 {
4526         int rc;
4527
4528         rc = proto_register(&packet_proto, 0);
4529         if (rc)
4530                 goto out;
4531         rc = sock_register(&packet_family_ops);
4532         if (rc)
4533                 goto out_proto;
4534         rc = register_pernet_subsys(&packet_net_ops);
4535         if (rc)
4536                 goto out_sock;
4537         rc = register_netdevice_notifier(&packet_netdev_notifier);
4538         if (rc)
4539                 goto out_pernet;
4540
4541         return 0;
4542
4543 out_pernet:
4544         unregister_pernet_subsys(&packet_net_ops);
4545 out_sock:
4546         sock_unregister(PF_PACKET);
4547 out_proto:
4548         proto_unregister(&packet_proto);
4549 out:
4550         return rc;
4551 }
4552
4553 module_init(packet_init);
4554 module_exit(packet_exit);
4555 MODULE_LICENSE("GPL");
4556 MODULE_ALIAS_NETPROTO(PF_PACKET);