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
  96 #include "internal.h"
  97
  98 /*
  99    Assumptions:
 100    - if device has no dev->hard_header routine, it adds and removes ll header
 101      inside itself. In this case ll header is invisible outside of device,
 102      but higher levels still should reserve dev->hard_header_len.
 103      Some devices are enough clever to reallocate skb, when header
 104      will not fit to reserved space (tunnel), another ones are silly
 105      (PPP).
 106    - packet socket receives packets with pulled ll header,
 107      so that SOCK_RAW should push it back.
 108
 109 On receive:
 110 -----------
 111
 112 Incoming, dev->hard_header!=NULL
 113    mac_header -> ll header
 114    data       -> data
 115
 116 Outgoing, dev->hard_header!=NULL
 117    mac_header -> ll header
 118    data       -> ll header
 119
 120 Incoming, dev->hard_header==NULL
 121    mac_header -> UNKNOWN position. It is very likely, that it points to ll
 122                  header.  PPP makes it, that is wrong, because introduce
 123                  assymetry between rx and tx paths.
 124    data       -> data
 125
 126 Outgoing, dev->hard_header==NULL
 127    mac_header -> data. ll header is still not built!
 128    data       -> data
 129
 130 Resume
 131   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 132
 133
 134 On transmit:
 135 ------------
 136
 137 dev->hard_header != NULL
 138    mac_header -> ll header
 139    data       -> ll header
 140
 141 dev->hard_header == NULL (ll header is added by device, we cannot control it)
 142    mac_header -> data
 143    data       -> data
 144
 145    We should set nh.raw on output to correct posistion,
 146    packet classifier depends on it.
 147  */
 148
 149 /* Private packet socket structures. */
 150
 151 /* identical to struct packet_mreq except it has
 152  * a longer address field.
 153  */
 154 struct packet_mreq_max {
 155         int             mr_ifindex;
 156         unsigned short  mr_type;
 157         unsigned short  mr_alen;
 158         unsigned char   mr_address[MAX_ADDR_LEN];
 159 };
 160
 161 union tpacket_uhdr {
 162         struct tpacket_hdr  *h1;
 163         struct tpacket2_hdr *h2;
 164         struct tpacket3_hdr *h3;
 165         void *raw;
 166 };
 167
 168 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 169                 int closing, int tx_ring);
 170
 171 #define V3_ALIGNMENT    (8)
 172
 173 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 174
 175 #define BLK_PLUS_PRIV(sz_of_priv) \
 176         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 177
 178 #define PGV_FROM_VMALLOC 1
 179
 180 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
 181 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
 182 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
 183 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
 184 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
 185 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
 186 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
 187
 188 struct packet_sock;
 189 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
 190 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 191                        struct packet_type *pt, struct net_device *orig_dev);
 192
 193 static void *packet_previous_frame(struct packet_sock *po,
 194                 struct packet_ring_buffer *rb,
 195                 int status);
 196 static void packet_increment_head(struct packet_ring_buffer *buff);
 197 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
 198                         struct tpacket_block_desc *);
 199 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 200                         struct packet_sock *);
 201 static void prb_retire_current_block(struct tpacket_kbdq_core *,
 202                 struct packet_sock *, unsigned int status);
 203 static int prb_queue_frozen(struct tpacket_kbdq_core *);
 204 static void prb_open_block(struct tpacket_kbdq_core *,
 205                 struct tpacket_block_desc *);
 206 static void prb_retire_rx_blk_timer_expired(unsigned long);
 207 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 208 static void prb_init_blk_timer(struct packet_sock *,
 209                 struct tpacket_kbdq_core *,
 210                 void (*func) (unsigned long));
 211 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 212 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 213                 struct tpacket3_hdr *);
 214 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 215                 struct tpacket3_hdr *);
 216 static void packet_flush_mclist(struct sock *sk);
 217
 218 struct packet_skb_cb {
 219         union {
 220                 struct sockaddr_pkt pkt;
 221                 union {
 222                         /* Trick: alias skb original length with
 223                          * ll.sll_family and ll.protocol in order
 224                          * to save room.
 225                          */
 226                         unsigned int origlen;
 227                         struct sockaddr_ll ll;
 228                 };
 229         } sa;
 230 };
 231
 232 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 233
 234 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 235 #define GET_PBLOCK_DESC(x, bid) \
 236         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 237 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 238         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 239 #define GET_NEXT_PRB_BLK_NUM(x) \
 240         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 241         ((x)->kactive_blk_num+1) : 0)
 242
 243 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 244 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 245
 246 static int packet_direct_xmit(struct sk_buff *skb)
 247 {
 248         struct net_device *dev = skb->dev;
 249         netdev_features_t features;
 250         struct netdev_queue *txq;
 251         int ret = NETDEV_TX_BUSY;
 252
 253         if (unlikely(!netif_running(dev) ||
 254                      !netif_carrier_ok(dev)))
 255                 goto drop;
 256
 257         features = netif_skb_features(skb);
 258         if (skb_needs_linearize(skb, features) &&
 259             __skb_linearize(skb))
 260                 goto drop;
 261
 262         txq = skb_get_tx_queue(dev, skb);
 263
 264         local_bh_disable();
 265
 266         HARD_TX_LOCK(dev, txq, smp_processor_id());
 267         if (!netif_xmit_frozen_or_drv_stopped(txq))
 268                 ret = netdev_start_xmit(skb, dev, txq, false);
 269         HARD_TX_UNLOCK(dev, txq);
 270
 271         local_bh_enable();
 272
 273         if (!dev_xmit_complete(ret))
 274                 kfree_skb(skb);
 275
 276         return ret;
 277 drop:
 278         atomic_long_inc(&dev->tx_dropped);
 279         kfree_skb(skb);
 280         return NET_XMIT_DROP;
 281 }
 282
 283 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
 284 {
 285         struct net_device *dev;
 286
 287         rcu_read_lock();
 288         dev = rcu_dereference(po->cached_dev);
 289         if (likely(dev))
 290                 dev_hold(dev);
 291         rcu_read_unlock();
 292
 293         return dev;
 294 }
 295
 296 static void packet_cached_dev_assign(struct packet_sock *po,
 297                                      struct net_device *dev)
 298 {
 299         rcu_assign_pointer(po->cached_dev, dev);
 300 }
 301
 302 static void packet_cached_dev_reset(struct packet_sock *po)
 303 {
 304         RCU_INIT_POINTER(po->cached_dev, NULL);
 305 }
 306
 307 static bool packet_use_direct_xmit(const struct packet_sock *po)
 308 {
 309         return po->xmit == packet_direct_xmit;
 310 }
 311
 312 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
 313 {
 314         return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
 315 }
 316
 317 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
 318 {
 319         const struct net_device_ops *ops = dev->netdev_ops;
 320         u16 queue_index;
 321
 322         if (ops->ndo_select_queue) {
 323                 queue_index = ops->ndo_select_queue(dev, skb, NULL,
 324                                                     __packet_pick_tx_queue);
 325                 queue_index = netdev_cap_txqueue(dev, queue_index);
 326         } else {
 327                 queue_index = __packet_pick_tx_queue(dev, skb);
 328         }
 329
 330         skb_set_queue_mapping(skb, queue_index);
 331 }
 332
 333 /* register_prot_hook must be invoked with the po->bind_lock held,
 334  * or from a context in which asynchronous accesses to the packet
 335  * socket is not possible (packet_create()).
 336  */
 337 static void register_prot_hook(struct sock *sk)
 338 {
 339         struct packet_sock *po = pkt_sk(sk);
 340
 341         if (!po->running) {
 342                 if (po->fanout)
 343                         __fanout_link(sk, po);
 344                 else
 345                         dev_add_pack(&po->prot_hook);
 346
 347                 sock_hold(sk);
 348                 po->running = 1;
 349         }
 350 }
 351
 352 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
 353  * held.   If the sync parameter is true, we will temporarily drop
 354  * the po->bind_lock and do a synchronize_net to make sure no
 355  * asynchronous packet processing paths still refer to the elements
 356  * of po->prot_hook.  If the sync parameter is false, it is the
 357  * callers responsibility to take care of this.
 358  */
 359 static void __unregister_prot_hook(struct sock *sk, bool sync)
 360 {
 361         struct packet_sock *po = pkt_sk(sk);
 362
 363         po->running = 0;
 364
 365         if (po->fanout)
 366                 __fanout_unlink(sk, po);
 367         else
 368                 __dev_remove_pack(&po->prot_hook);
 369
 370         __sock_put(sk);
 371
 372         if (sync) {
 373                 spin_unlock(&po->bind_lock);
 374                 synchronize_net();
 375                 spin_lock(&po->bind_lock);
 376         }
 377 }
 378
 379 static void unregister_prot_hook(struct sock *sk, bool sync)
 380 {
 381         struct packet_sock *po = pkt_sk(sk);
 382
 383         if (po->running)
 384                 __unregister_prot_hook(sk, sync);
 385 }
 386
 387 static inline struct page * __pure pgv_to_page(void *addr)
 388 {
 389         if (is_vmalloc_addr(addr))
 390                 return vmalloc_to_page(addr);
 391         return virt_to_page(addr);
 392 }
 393
 394 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 395 {
 396         union tpacket_uhdr h;
 397
 398         h.raw = frame;
 399         switch (po->tp_version) {
 400         case TPACKET_V1:
 401                 h.h1->tp_status = status;
 402                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 403                 break;
 404         case TPACKET_V2:
 405                 h.h2->tp_status = status;
 406                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 407                 break;
 408         case TPACKET_V3:
 409         default:
 410                 WARN(1, "TPACKET version not supported.\n");
 411                 BUG();
 412         }
 413
 414         smp_wmb();
 415 }
 416
 417 static int __packet_get_status(struct packet_sock *po, void *frame)
 418 {
 419         union tpacket_uhdr h;
 420
 421         smp_rmb();
 422
 423         h.raw = frame;
 424         switch (po->tp_version) {
 425         case TPACKET_V1:
 426                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 427                 return h.h1->tp_status;
 428         case TPACKET_V2:
 429                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 430                 return h.h2->tp_status;
 431         case TPACKET_V3:
 432         default:
 433                 WARN(1, "TPACKET version not supported.\n");
 434                 BUG();
 435                 return 0;
 436         }
 437 }
 438
 439 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
 440                                    unsigned int flags)
 441 {
 442         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 443
 444         if (shhwtstamps &&
 445             (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
 446             ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
 447                 return TP_STATUS_TS_RAW_HARDWARE;
 448
 449         if (ktime_to_timespec_cond(skb->tstamp, ts))
 450                 return TP_STATUS_TS_SOFTWARE;
 451
 452         return 0;
 453 }
 454
 455 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
 456                                     struct sk_buff *skb)
 457 {
 458         union tpacket_uhdr h;
 459         struct timespec ts;
 460         __u32 ts_status;
 461
 462         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
 463                 return 0;
 464
 465         h.raw = frame;
 466         switch (po->tp_version) {
 467         case TPACKET_V1:
 468                 h.h1->tp_sec = ts.tv_sec;
 469                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
 470                 break;
 471         case TPACKET_V2:
 472                 h.h2->tp_sec = ts.tv_sec;
 473                 h.h2->tp_nsec = ts.tv_nsec;
 474                 break;
 475         case TPACKET_V3:
 476         default:
 477                 WARN(1, "TPACKET version not supported.\n");
 478                 BUG();
 479         }
 480
 481         /* one flush is safe, as both fields always lie on the same cacheline */
 482         flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
 483         smp_wmb();
 484
 485         return ts_status;
 486 }
 487
 488 static void *packet_lookup_frame(struct packet_sock *po,
 489                 struct packet_ring_buffer *rb,
 490                 unsigned int position,
 491                 int status)
 492 {
 493         unsigned int pg_vec_pos, frame_offset;
 494         union tpacket_uhdr h;
 495
 496         pg_vec_pos = position / rb->frames_per_block;
 497         frame_offset = position % rb->frames_per_block;
 498
 499         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 500                 (frame_offset * rb->frame_size);
 501
 502         if (status != __packet_get_status(po, h.raw))
 503                 return NULL;
 504
 505         return h.raw;
 506 }
 507
 508 static void *packet_current_frame(struct packet_sock *po,
 509                 struct packet_ring_buffer *rb,
 510                 int status)
 511 {
 512         return packet_lookup_frame(po, rb, rb->head, status);
 513 }
 514
 515 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 516 {
 517         del_timer_sync(&pkc->retire_blk_timer);
 518 }
 519
 520 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 521                 int tx_ring,
 522                 struct sk_buff_head *rb_queue)
 523 {
 524         struct tpacket_kbdq_core *pkc;
 525
 526         pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
 527                         GET_PBDQC_FROM_RB(&po->rx_ring);
 528
 529         spin_lock_bh(&rb_queue->lock);
 530         pkc->delete_blk_timer = 1;
 531         spin_unlock_bh(&rb_queue->lock);
 532
 533         prb_del_retire_blk_timer(pkc);
 534 }
 535
 536 static void prb_init_blk_timer(struct packet_sock *po,
 537                 struct tpacket_kbdq_core *pkc,
 538                 void (*func) (unsigned long))
 539 {
 540         init_timer(&pkc->retire_blk_timer);
 541         pkc->retire_blk_timer.data = (long)po;
 542         pkc->retire_blk_timer.function = func;
 543         pkc->retire_blk_timer.expires = jiffies;
 544 }
 545
 546 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
 547 {
 548         struct tpacket_kbdq_core *pkc;
 549
 550         if (tx_ring)
 551                 BUG();
 552
 553         pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
 554                         GET_PBDQC_FROM_RB(&po->rx_ring);
 555         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
 556 }
 557
 558 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 559                                 int blk_size_in_bytes)
 560 {
 561         struct net_device *dev;
 562         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 563         struct ethtool_cmd ecmd;
 564         int err;
 565         u32 speed;
 566
 567         rtnl_lock();
 568         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 569         if (unlikely(!dev)) {
 570                 rtnl_unlock();
 571                 return DEFAULT_PRB_RETIRE_TOV;
 572         }
 573         err = __ethtool_get_settings(dev, &ecmd);
 574         speed = ethtool_cmd_speed(&ecmd);
 575         rtnl_unlock();
 576         if (!err) {
 577                 /*
 578                  * If the link speed is so slow you don't really
 579                  * need to worry about perf anyways
 580                  */
 581                 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
 582                         return DEFAULT_PRB_RETIRE_TOV;
 583                 } else {
 584                         msec = 1;
 585                         div = speed / 1000;
 586                 }
 587         }
 588
 589         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 590
 591         if (div)
 592                 mbits /= div;
 593
 594         tmo = mbits * msec;
 595
 596         if (div)
 597                 return tmo+1;
 598         return tmo;
 599 }
 600
 601 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 602                         union tpacket_req_u *req_u)
 603 {
 604         p1->feature_req_word = req_u->req3.tp_feature_req_word;
 605 }
 606
 607 static void init_prb_bdqc(struct packet_sock *po,
 608                         struct packet_ring_buffer *rb,
 609                         struct pgv *pg_vec,
 610                         union tpacket_req_u *req_u, int tx_ring)
 611 {
 612         struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
 613         struct tpacket_block_desc *pbd;
 614
 615         memset(p1, 0x0, sizeof(*p1));
 616
 617         p1->knxt_seq_num = 1;
 618         p1->pkbdq = pg_vec;
 619         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 620         p1->pkblk_start = pg_vec[0].buffer;
 621         p1->kblk_size = req_u->req3.tp_block_size;
 622         p1->knum_blocks = req_u->req3.tp_block_nr;
 623         p1->hdrlen = po->tp_hdrlen;
 624         p1->version = po->tp_version;
 625         p1->last_kactive_blk_num = 0;
 626         po->stats.stats3.tp_freeze_q_cnt = 0;
 627         if (req_u->req3.tp_retire_blk_tov)
 628                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 629         else
 630                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 631                                                 req_u->req3.tp_block_size);
 632         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 633         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 634
 635         p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
 636         prb_init_ft_ops(p1, req_u);
 637         prb_setup_retire_blk_timer(po, tx_ring);
 638         prb_open_block(p1, pbd);
 639 }
 640
 641 /*  Do NOT update the last_blk_num first.
 642  *  Assumes sk_buff_head lock is held.
 643  */
 644 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 645 {
 646         mod_timer(&pkc->retire_blk_timer,
 647                         jiffies + pkc->tov_in_jiffies);
 648         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 649 }
 650
 651 /*
 652  * Timer logic:
 653  * 1) We refresh the timer only when we open a block.
 654  *    By doing this we don't waste cycles refreshing the timer
 655  *        on packet-by-packet basis.
 656  *
 657  * With a 1MB block-size, on a 1Gbps line, it will take
 658  * i) ~8 ms to fill a block + ii) memcpy etc.
 659  * In this cut we are not accounting for the memcpy time.
 660  *
 661  * So, if the user sets the 'tmo' to 10ms then the timer
 662  * will never fire while the block is still getting filled
 663  * (which is what we want). However, the user could choose
 664  * to close a block early and that's fine.
 665  *
 666  * But when the timer does fire, we check whether or not to refresh it.
 667  * Since the tmo granularity is in msecs, it is not too expensive
 668  * to refresh the timer, lets say every '8' msecs.
 669  * Either the user can set the 'tmo' or we can derive it based on
 670  * a) line-speed and b) block-size.
 671  * prb_calc_retire_blk_tmo() calculates the tmo.
 672  *
 673  */
 674 static void prb_retire_rx_blk_timer_expired(unsigned long data)
 675 {
 676         struct packet_sock *po = (struct packet_sock *)data;
 677         struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 678         unsigned int frozen;
 679         struct tpacket_block_desc *pbd;
 680
 681         spin_lock(&po->sk.sk_receive_queue.lock);
 682
 683         frozen = prb_queue_frozen(pkc);
 684         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 685
 686         if (unlikely(pkc->delete_blk_timer))
 687                 goto out;
 688
 689         /* We only need to plug the race when the block is partially filled.
 690          * tpacket_rcv:
 691          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 692          *              copy_bits() is in progress ...
 693          *              timer fires on other cpu:
 694          *              we can't retire the current block because copy_bits
 695          *              is in progress.
 696          *
 697          */
 698         if (BLOCK_NUM_PKTS(pbd)) {
 699                 while (atomic_read(&pkc->blk_fill_in_prog)) {
 700                         /* Waiting for skb_copy_bits to finish... */
 701                         cpu_relax();
 702                 }
 703         }
 704
 705         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 706                 if (!frozen) {
 707                         if (!BLOCK_NUM_PKTS(pbd)) {
 708                                 /* An empty block. Just refresh the timer. */
 709                                 goto refresh_timer;
 710                         }
 711                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 712                         if (!prb_dispatch_next_block(pkc, po))
 713                                 goto refresh_timer;
 714                         else
 715                                 goto out;
 716                 } else {
 717                         /* Case 1. Queue was frozen because user-space was
 718                          *         lagging behind.
 719                          */
 720                         if (prb_curr_blk_in_use(pkc, pbd)) {
 721                                 /*
 722                                  * Ok, user-space is still behind.
 723                                  * So just refresh the timer.
 724                                  */
 725                                 goto refresh_timer;
 726                         } else {
 727                                /* Case 2. queue was frozen,user-space caught up,
 728                                 * now the link went idle && the timer fired.
 729                                 * We don't have a block to close.So we open this
 730                                 * block and restart the timer.
 731                                 * opening a block thaws the queue,restarts timer
 732                                 * Thawing/timer-refresh is a side effect.
 733                                 */
 734                                 prb_open_block(pkc, pbd);
 735                                 goto out;
 736                         }
 737                 }
 738         }
 739
 740 refresh_timer:
 741         _prb_refresh_rx_retire_blk_timer(pkc);
 742
 743 out:
 744         spin_unlock(&po->sk.sk_receive_queue.lock);
 745 }
 746
 747 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 748                 struct tpacket_block_desc *pbd1, __u32 status)
 749 {
 750         /* Flush everything minus the block header */
 751
 752 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 753         u8 *start, *end;
 754
 755         start = (u8 *)pbd1;
 756
 757         /* Skip the block header(we know header WILL fit in 4K) */
 758         start += PAGE_SIZE;
 759
 760         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 761         for (; start < end; start += PAGE_SIZE)
 762                 flush_dcache_page(pgv_to_page(start));
 763
 764         smp_wmb();
 765 #endif
 766
 767         /* Now update the block status. */
 768
 769         BLOCK_STATUS(pbd1) = status;
 770
 771         /* Flush the block header */
 772
 773 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 774         start = (u8 *)pbd1;
 775         flush_dcache_page(pgv_to_page(start));
 776
 777         smp_wmb();
 778 #endif
 779 }
 780
 781 /*
 782  * Side effect:
 783  *
 784  * 1) flush the block
 785  * 2) Increment active_blk_num
 786  *
 787  * Note:We DONT refresh the timer on purpose.
 788  *      Because almost always the next block will be opened.
 789  */
 790 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 791                 struct tpacket_block_desc *pbd1,
 792                 struct packet_sock *po, unsigned int stat)
 793 {
 794         __u32 status = TP_STATUS_USER | stat;
 795
 796         struct tpacket3_hdr *last_pkt;
 797         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 798         struct sock *sk = &po->sk;
 799
 800         if (po->stats.stats3.tp_drops)
 801                 status |= TP_STATUS_LOSING;
 802
 803         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 804         last_pkt->tp_next_offset = 0;
 805
 806         /* Get the ts of the last pkt */
 807         if (BLOCK_NUM_PKTS(pbd1)) {
 808                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 809                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 810         } else {
 811                 /* Ok, we tmo'd - so get the current time.
 812                  *
 813                  * It shouldn't really happen as we don't close empty
 814                  * blocks. See prb_retire_rx_blk_timer_expired().
 815                  */
 816                 struct timespec ts;
 817                 getnstimeofday(&ts);
 818                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 819                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 820         }
 821
 822         smp_wmb();
 823
 824         /* Flush the block */
 825         prb_flush_block(pkc1, pbd1, status);
 826
 827         sk->sk_data_ready(sk);
 828
 829         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 830 }
 831
 832 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 833 {
 834         pkc->reset_pending_on_curr_blk = 0;
 835 }
 836
 837 /*
 838  * Side effect of opening a block:
 839  *
 840  * 1) prb_queue is thawed.
 841  * 2) retire_blk_timer is refreshed.
 842  *
 843  */
 844 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 845         struct tpacket_block_desc *pbd1)
 846 {
 847         struct timespec ts;
 848         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 849
 850         smp_rmb();
 851
 852         /* We could have just memset this but we will lose the
 853          * flexibility of making the priv area sticky
 854          */
 855
 856         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 857         BLOCK_NUM_PKTS(pbd1) = 0;
 858         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 859
 860         getnstimeofday(&ts);
 861
 862         h1->ts_first_pkt.ts_sec = ts.tv_sec;
 863         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 864
 865         pkc1->pkblk_start = (char *)pbd1;
 866         pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 867
 868         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 869         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 870
 871         pbd1->version = pkc1->version;
 872         pkc1->prev = pkc1->nxt_offset;
 873         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 874
 875         prb_thaw_queue(pkc1);
 876         _prb_refresh_rx_retire_blk_timer(pkc1);
 877
 878         smp_wmb();
 879 }
 880
 881 /*
 882  * Queue freeze logic:
 883  * 1) Assume tp_block_nr = 8 blocks.
 884  * 2) At time 't0', user opens Rx ring.
 885  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 886  * 4) user-space is either sleeping or processing block '0'.
 887  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 888  *    it will close block-7,loop around and try to fill block '0'.
 889  *    call-flow:
 890  *    __packet_lookup_frame_in_block
 891  *      prb_retire_current_block()
 892  *      prb_dispatch_next_block()
 893  *        |->(BLOCK_STATUS == USER) evaluates to true
 894  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 895  * 6) Now there are two cases:
 896  *    6.1) Link goes idle right after the queue is frozen.
 897  *         But remember, the last open_block() refreshed the timer.
 898  *         When this timer expires,it will refresh itself so that we can
 899  *         re-open block-0 in near future.
 900  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 901  *         case and __packet_lookup_frame_in_block will check if block-0
 902  *         is free and can now be re-used.
 903  */
 904 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 905                                   struct packet_sock *po)
 906 {
 907         pkc->reset_pending_on_curr_blk = 1;
 908         po->stats.stats3.tp_freeze_q_cnt++;
 909 }
 910
 911 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 912
 913 /*
 914  * If the next block is free then we will dispatch it
 915  * and return a good offset.
 916  * Else, we will freeze the queue.
 917  * So, caller must check the return value.
 918  */
 919 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 920                 struct packet_sock *po)
 921 {
 922         struct tpacket_block_desc *pbd;
 923
 924         smp_rmb();
 925
 926         /* 1. Get current block num */
 927         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 928
 929         /* 2. If this block is currently in_use then freeze the queue */
 930         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 931                 prb_freeze_queue(pkc, po);
 932                 return NULL;
 933         }
 934
 935         /*
 936          * 3.
 937          * open this block and return the offset where the first packet
 938          * needs to get stored.
 939          */
 940         prb_open_block(pkc, pbd);
 941         return (void *)pkc->nxt_offset;
 942 }
 943
 944 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 945                 struct packet_sock *po, unsigned int status)
 946 {
 947         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 948
 949         /* retire/close the current block */
 950         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 951                 /*
 952                  * Plug the case where copy_bits() is in progress on
 953                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 954                  * have space to copy the pkt in the current block and
 955                  * called prb_retire_current_block()
 956                  *
 957                  * We don't need to worry about the TMO case because
 958                  * the timer-handler already handled this case.
 959                  */
 960                 if (!(status & TP_STATUS_BLK_TMO)) {
 961                         while (atomic_read(&pkc->blk_fill_in_prog)) {
 962                                 /* Waiting for skb_copy_bits to finish... */
 963                                 cpu_relax();
 964                         }
 965                 }
 966                 prb_close_block(pkc, pbd, po, status);
 967                 return;
 968         }
 969 }
 970
 971 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
 972                                       struct tpacket_block_desc *pbd)
 973 {
 974         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 975 }
 976
 977 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 978 {
 979         return pkc->reset_pending_on_curr_blk;
 980 }
 981
 982 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 983 {
 984         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 985         atomic_dec(&pkc->blk_fill_in_prog);
 986 }
 987
 988 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 989                         struct tpacket3_hdr *ppd)
 990 {
 991         ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
 992 }
 993
 994 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 995                         struct tpacket3_hdr *ppd)
 996 {
 997         ppd->hv1.tp_rxhash = 0;
 998 }
 999
1000 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1001                         struct tpacket3_hdr *ppd)
1002 {
1003         if (skb_vlan_tag_present(pkc->skb)) {
1004                 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1005                 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1006                 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1007         } else {
1008                 ppd->hv1.tp_vlan_tci = 0;
1009                 ppd->hv1.tp_vlan_tpid = 0;
1010                 ppd->tp_status = TP_STATUS_AVAILABLE;
1011         }
1012 }
1013
1014 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1015                         struct tpacket3_hdr *ppd)
1016 {
1017         ppd->hv1.tp_padding = 0;
1018         prb_fill_vlan_info(pkc, ppd);
1019
1020         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1021                 prb_fill_rxhash(pkc, ppd);
1022         else
1023                 prb_clear_rxhash(pkc, ppd);
1024 }
1025
1026 static void prb_fill_curr_block(char *curr,
1027                                 struct tpacket_kbdq_core *pkc,
1028                                 struct tpacket_block_desc *pbd,
1029                                 unsigned int len)
1030 {
1031         struct tpacket3_hdr *ppd;
1032
1033         ppd  = (struct tpacket3_hdr *)curr;
1034         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1035         pkc->prev = curr;
1036         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1037         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1038         BLOCK_NUM_PKTS(pbd) += 1;
1039         atomic_inc(&pkc->blk_fill_in_prog);
1040         prb_run_all_ft_ops(pkc, ppd);
1041 }
1042
1043 /* Assumes caller has the sk->rx_queue.lock */
1044 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1045                                             struct sk_buff *skb,
1046                                                 int status,
1047                                             unsigned int len
1048                                             )
1049 {
1050         struct tpacket_kbdq_core *pkc;
1051         struct tpacket_block_desc *pbd;
1052         char *curr, *end;
1053
1054         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1055         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1056
1057         /* Queue is frozen when user space is lagging behind */
1058         if (prb_queue_frozen(pkc)) {
1059                 /*
1060                  * Check if that last block which caused the queue to freeze,
1061                  * is still in_use by user-space.
1062                  */
1063                 if (prb_curr_blk_in_use(pkc, pbd)) {
1064                         /* Can't record this packet */
1065                         return NULL;
1066                 } else {
1067                         /*
1068                          * Ok, the block was released by user-space.
1069                          * Now let's open that block.
1070                          * opening a block also thaws the queue.
1071                          * Thawing is a side effect.
1072                          */
1073                         prb_open_block(pkc, pbd);
1074                 }
1075         }
1076
1077         smp_mb();
1078         curr = pkc->nxt_offset;
1079         pkc->skb = skb;
1080         end = (char *)pbd + pkc->kblk_size;
1081
1082         /* first try the current block */
1083         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1084                 prb_fill_curr_block(curr, pkc, pbd, len);
1085                 return (void *)curr;
1086         }
1087
1088         /* Ok, close the current block */
1089         prb_retire_current_block(pkc, po, 0);
1090
1091         /* Now, try to dispatch the next block */
1092         curr = (char *)prb_dispatch_next_block(pkc, po);
1093         if (curr) {
1094                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1095                 prb_fill_curr_block(curr, pkc, pbd, len);
1096                 return (void *)curr;
1097         }
1098
1099         /*
1100          * No free blocks are available.user_space hasn't caught up yet.
1101          * Queue was just frozen and now this packet will get dropped.
1102          */
1103         return NULL;
1104 }
1105
1106 static void *packet_current_rx_frame(struct packet_sock *po,
1107                                             struct sk_buff *skb,
1108                                             int status, unsigned int len)
1109 {
1110         char *curr = NULL;
1111         switch (po->tp_version) {
1112         case TPACKET_V1:
1113         case TPACKET_V2:
1114                 curr = packet_lookup_frame(po, &po->rx_ring,
1115                                         po->rx_ring.head, status);
1116                 return curr;
1117         case TPACKET_V3:
1118                 return __packet_lookup_frame_in_block(po, skb, status, len);
1119         default:
1120                 WARN(1, "TPACKET version not supported\n");
1121                 BUG();
1122                 return NULL;
1123         }
1124 }
1125
1126 static void *prb_lookup_block(struct packet_sock *po,
1127                                      struct packet_ring_buffer *rb,
1128                                      unsigned int idx,
1129                                      int status)
1130 {
1131         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1132         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1133
1134         if (status != BLOCK_STATUS(pbd))
1135                 return NULL;
1136         return pbd;
1137 }
1138
1139 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1140 {
1141         unsigned int prev;
1142         if (rb->prb_bdqc.kactive_blk_num)
1143                 prev = rb->prb_bdqc.kactive_blk_num-1;
1144         else
1145                 prev = rb->prb_bdqc.knum_blocks-1;
1146         return prev;
1147 }
1148
1149 /* Assumes caller has held the rx_queue.lock */
1150 static void *__prb_previous_block(struct packet_sock *po,
1151                                          struct packet_ring_buffer *rb,
1152                                          int status)
1153 {
1154         unsigned int previous = prb_previous_blk_num(rb);
1155         return prb_lookup_block(po, rb, previous, status);
1156 }
1157
1158 static void *packet_previous_rx_frame(struct packet_sock *po,
1159                                              struct packet_ring_buffer *rb,
1160                                              int status)
1161 {
1162         if (po->tp_version <= TPACKET_V2)
1163                 return packet_previous_frame(po, rb, status);
1164
1165         return __prb_previous_block(po, rb, status);
1166 }
1167
1168 static void packet_increment_rx_head(struct packet_sock *po,
1169                                             struct packet_ring_buffer *rb)
1170 {
1171         switch (po->tp_version) {
1172         case TPACKET_V1:
1173         case TPACKET_V2:
1174                 return packet_increment_head(rb);
1175         case TPACKET_V3:
1176         default:
1177                 WARN(1, "TPACKET version not supported.\n");
1178                 BUG();
1179                 return;
1180         }
1181 }
1182
1183 static void *packet_previous_frame(struct packet_sock *po,
1184                 struct packet_ring_buffer *rb,
1185                 int status)
1186 {
1187         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1188         return packet_lookup_frame(po, rb, previous, status);
1189 }
1190
1191 static void packet_increment_head(struct packet_ring_buffer *buff)
1192 {
1193         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1194 }
1195
1196 static void packet_inc_pending(struct packet_ring_buffer *rb)
1197 {
1198         this_cpu_inc(*rb->pending_refcnt);
1199 }
1200
1201 static void packet_dec_pending(struct packet_ring_buffer *rb)
1202 {
1203         this_cpu_dec(*rb->pending_refcnt);
1204 }
1205
1206 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1207 {
1208         unsigned int refcnt = 0;
1209         int cpu;
1210
1211         /* We don't use pending refcount in rx_ring. */
1212         if (rb->pending_refcnt == NULL)
1213                 return 0;
1214
1215         for_each_possible_cpu(cpu)
1216                 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1217
1218         return refcnt;
1219 }
1220
1221 static int packet_alloc_pending(struct packet_sock *po)
1222 {
1223         po->rx_ring.pending_refcnt = NULL;
1224
1225         po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1226         if (unlikely(po->tx_ring.pending_refcnt == NULL))
1227                 return -ENOBUFS;
1228
1229         return 0;
1230 }
1231
1232 static void packet_free_pending(struct packet_sock *po)
1233 {
1234         free_percpu(po->tx_ring.pending_refcnt);
1235 }
1236
1237 #define ROOM_POW_OFF    2
1238 #define ROOM_NONE       0x0
1239 #define ROOM_LOW        0x1
1240 #define ROOM_NORMAL     0x2
1241
1242 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1243 {
1244         int idx, len;
1245
1246         len = po->rx_ring.frame_max + 1;
1247         idx = po->rx_ring.head;
1248         if (pow_off)
1249                 idx += len >> pow_off;
1250         if (idx >= len)
1251                 idx -= len;
1252         return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1253 }
1254
1255 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1256 {
1257         int idx, len;
1258
1259         len = po->rx_ring.prb_bdqc.knum_blocks;
1260         idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1261         if (pow_off)
1262                 idx += len >> pow_off;
1263         if (idx >= len)
1264                 idx -= len;
1265         return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1266 }
1267
1268 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1269 {
1270         struct sock *sk = &po->sk;
1271         int ret = ROOM_NONE;
1272
1273         if (po->prot_hook.func != tpacket_rcv) {
1274                 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1275                                           - skb->truesize;
1276                 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1277                         return ROOM_NORMAL;
1278                 else if (avail > 0)
1279                         return ROOM_LOW;
1280                 else
1281                         return ROOM_NONE;
1282         }
1283
1284         spin_lock(&sk->sk_receive_queue.lock);
1285         if (po->tp_version == TPACKET_V3) {
1286                 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1287                         ret = ROOM_NORMAL;
1288                 else if (__tpacket_v3_has_room(po, 0))
1289                         ret = ROOM_LOW;
1290         } else {
1291                 if (__tpacket_has_room(po, ROOM_POW_OFF))
1292                         ret = ROOM_NORMAL;
1293                 else if (__tpacket_has_room(po, 0))
1294                         ret = ROOM_LOW;
1295         }
1296         spin_unlock(&sk->sk_receive_queue.lock);
1297
1298         return ret;
1299 }
1300
1301 static void packet_sock_destruct(struct sock *sk)
1302 {
1303         skb_queue_purge(&sk->sk_error_queue);
1304
1305         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1306         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1307
1308         if (!sock_flag(sk, SOCK_DEAD)) {
1309                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1310                 return;
1311         }
1312
1313         sk_refcnt_debug_dec(sk);
1314 }
1315
1316 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1317 {
1318         int x = atomic_read(&f->rr_cur) + 1;
1319
1320         if (x >= num)
1321                 x = 0;
1322
1323         return x;
1324 }
1325
1326 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1327                                       struct sk_buff *skb,
1328                                       unsigned int num)
1329 {
1330         return reciprocal_scale(skb_get_hash(skb), num);
1331 }
1332
1333 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1334                                     struct sk_buff *skb,
1335                                     unsigned int num)
1336 {
1337         int cur, old;
1338
1339         cur = atomic_read(&f->rr_cur);
1340         while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1341                                      fanout_rr_next(f, num))) != cur)
1342                 cur = old;
1343         return cur;
1344 }
1345
1346 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1347                                      struct sk_buff *skb,
1348                                      unsigned int num)
1349 {
1350         return smp_processor_id() % num;
1351 }
1352
1353 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1354                                      struct sk_buff *skb,
1355                                      unsigned int num)
1356 {
1357         return prandom_u32_max(num);
1358 }
1359
1360 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1361                                           struct sk_buff *skb,
1362                                           unsigned int idx, bool try_self,
1363                                           unsigned int num)
1364 {
1365         struct packet_sock *po;
1366         unsigned int i, j;
1367
1368         po = pkt_sk(f->arr[idx]);
1369         if (try_self && packet_rcv_has_room(po, skb) != ROOM_NONE)
1370                 return idx;
1371
1372         i = j = min_t(int, po->rollover->sock, num - 1);
1373         do {
1374                 if (i != idx &&
1375                     packet_rcv_has_room(pkt_sk(f->arr[i]), skb) == ROOM_NORMAL) {
1376                         if (i != j)
1377                                 po->rollover->sock = i;
1378                         return i;
1379                 }
1380
1381                 if (++i == num)
1382                         i = 0;
1383         } while (i != j);
1384
1385         return idx;
1386 }
1387
1388 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1389                                     struct sk_buff *skb,
1390                                     unsigned int num)
1391 {
1392         return skb_get_queue_mapping(skb) % num;
1393 }
1394
1395 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1396 {
1397         return f->flags & (flag >> 8);
1398 }
1399
1400 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1401                              struct packet_type *pt, struct net_device *orig_dev)
1402 {
1403         struct packet_fanout *f = pt->af_packet_priv;
1404         unsigned int num = f->num_members;
1405         struct packet_sock *po;
1406         unsigned int idx;
1407
1408         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1409             !num) {
1410                 kfree_skb(skb);
1411                 return 0;
1412         }
1413
1414         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1415                 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1416                 if (!skb)
1417                         return 0;
1418         }
1419         switch (f->type) {
1420         case PACKET_FANOUT_HASH:
1421         default:
1422                 idx = fanout_demux_hash(f, skb, num);
1423                 break;
1424         case PACKET_FANOUT_LB:
1425                 idx = fanout_demux_lb(f, skb, num);
1426                 break;
1427         case PACKET_FANOUT_CPU:
1428                 idx = fanout_demux_cpu(f, skb, num);
1429                 break;
1430         case PACKET_FANOUT_RND:
1431                 idx = fanout_demux_rnd(f, skb, num);
1432                 break;
1433         case PACKET_FANOUT_QM:
1434                 idx = fanout_demux_qm(f, skb, num);
1435                 break;
1436         case PACKET_FANOUT_ROLLOVER:
1437                 idx = fanout_demux_rollover(f, skb, 0, false, num);
1438                 break;
1439         }
1440
1441         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1442                 idx = fanout_demux_rollover(f, skb, idx, true, num);
1443
1444         po = pkt_sk(f->arr[idx]);
1445         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1446 }
1447
1448 DEFINE_MUTEX(fanout_mutex);
1449 EXPORT_SYMBOL_GPL(fanout_mutex);
1450 static LIST_HEAD(fanout_list);
1451
1452 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1453 {
1454         struct packet_fanout *f = po->fanout;
1455
1456         spin_lock(&f->lock);
1457         f->arr[f->num_members] = sk;
1458         smp_wmb();
1459         f->num_members++;
1460         spin_unlock(&f->lock);
1461 }
1462
1463 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1464 {
1465         struct packet_fanout *f = po->fanout;
1466         int i;
1467
1468         spin_lock(&f->lock);
1469         for (i = 0; i < f->num_members; i++) {
1470                 if (f->arr[i] == sk)
1471                         break;
1472         }
1473         BUG_ON(i >= f->num_members);
1474         f->arr[i] = f->arr[f->num_members - 1];
1475         f->num_members--;
1476         spin_unlock(&f->lock);
1477 }
1478
1479 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1480 {
1481         if (ptype->af_packet_priv == (void *)((struct packet_sock *)sk)->fanout)
1482                 return true;
1483
1484         return false;
1485 }
1486
1487 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1488 {
1489         struct packet_sock *po = pkt_sk(sk);
1490         struct packet_fanout *f, *match;
1491         u8 type = type_flags & 0xff;
1492         u8 flags = type_flags >> 8;
1493         int err;
1494
1495         switch (type) {
1496         case PACKET_FANOUT_ROLLOVER:
1497                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1498                         return -EINVAL;
1499         case PACKET_FANOUT_HASH:
1500         case PACKET_FANOUT_LB:
1501         case PACKET_FANOUT_CPU:
1502         case PACKET_FANOUT_RND:
1503         case PACKET_FANOUT_QM:
1504                 break;
1505         default:
1506                 return -EINVAL;
1507         }
1508
1509         if (!po->running)
1510                 return -EINVAL;
1511
1512         if (po->fanout)
1513                 return -EALREADY;
1514
1515         if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER) {
1516                 po->rollover = kzalloc(sizeof(*po->rollover), GFP_KERNEL);
1517                 if (!po->rollover)
1518                         return -ENOMEM;
1519         }
1520
1521         mutex_lock(&fanout_mutex);
1522         match = NULL;
1523         list_for_each_entry(f, &fanout_list, list) {
1524                 if (f->id == id &&
1525                     read_pnet(&f->net) == sock_net(sk)) {
1526                         match = f;
1527                         break;
1528                 }
1529         }
1530         err = -EINVAL;
1531         if (match && match->flags != flags)
1532                 goto out;
1533         if (!match) {
1534                 err = -ENOMEM;
1535                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1536                 if (!match)
1537                         goto out;
1538                 write_pnet(&match->net, sock_net(sk));
1539                 match->id = id;
1540                 match->type = type;
1541                 match->flags = flags;
1542                 atomic_set(&match->rr_cur, 0);
1543                 INIT_LIST_HEAD(&match->list);
1544                 spin_lock_init(&match->lock);
1545                 atomic_set(&match->sk_ref, 0);
1546                 match->prot_hook.type = po->prot_hook.type;
1547                 match->prot_hook.dev = po->prot_hook.dev;
1548                 match->prot_hook.func = packet_rcv_fanout;
1549                 match->prot_hook.af_packet_priv = match;
1550                 match->prot_hook.id_match = match_fanout_group;
1551                 dev_add_pack(&match->prot_hook);
1552                 list_add(&match->list, &fanout_list);
1553         }
1554         err = -EINVAL;
1555         if (match->type == type &&
1556             match->prot_hook.type == po->prot_hook.type &&
1557             match->prot_hook.dev == po->prot_hook.dev) {
1558                 err = -ENOSPC;
1559                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1560                         __dev_remove_pack(&po->prot_hook);
1561                         po->fanout = match;
1562                         atomic_inc(&match->sk_ref);
1563                         __fanout_link(sk, po);
1564                         err = 0;
1565                 }
1566         }
1567 out:
1568         mutex_unlock(&fanout_mutex);
1569         if (err) {
1570                 kfree(po->rollover);
1571                 po->rollover = NULL;
1572         }
1573         return err;
1574 }
1575
1576 static void fanout_release(struct sock *sk)
1577 {
1578         struct packet_sock *po = pkt_sk(sk);
1579         struct packet_fanout *f;
1580
1581         f = po->fanout;
1582         if (!f)
1583                 return;
1584
1585         mutex_lock(&fanout_mutex);
1586         po->fanout = NULL;
1587
1588         if (atomic_dec_and_test(&f->sk_ref)) {
1589                 list_del(&f->list);
1590                 dev_remove_pack(&f->prot_hook);
1591                 kfree(f);
1592         }
1593         mutex_unlock(&fanout_mutex);
1594
1595         kfree(po->rollover);
1596 }
1597
1598 static const struct proto_ops packet_ops;
1599
1600 static const struct proto_ops packet_ops_spkt;
1601
1602 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1603                            struct packet_type *pt, struct net_device *orig_dev)
1604 {
1605         struct sock *sk;
1606         struct sockaddr_pkt *spkt;
1607
1608         /*
1609          *      When we registered the protocol we saved the socket in the data
1610          *      field for just this event.
1611          */
1612
1613         sk = pt->af_packet_priv;
1614
1615         /*
1616          *      Yank back the headers [hope the device set this
1617          *      right or kerboom...]
1618          *
1619          *      Incoming packets have ll header pulled,
1620          *      push it back.
1621          *
1622          *      For outgoing ones skb->data == skb_mac_header(skb)
1623          *      so that this procedure is noop.
1624          */
1625
1626         if (skb->pkt_type == PACKET_LOOPBACK)
1627                 goto out;
1628
1629         if (!net_eq(dev_net(dev), sock_net(sk)))
1630                 goto out;
1631
1632         skb = skb_share_check(skb, GFP_ATOMIC);
1633         if (skb == NULL)
1634                 goto oom;
1635
1636         /* drop any routing info */
1637         skb_dst_drop(skb);
1638
1639         /* drop conntrack reference */
1640         nf_reset(skb);
1641
1642         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1643
1644         skb_push(skb, skb->data - skb_mac_header(skb));
1645
1646         /*
1647          *      The SOCK_PACKET socket receives _all_ frames.
1648          */
1649
1650         spkt->spkt_family = dev->type;
1651         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1652         spkt->spkt_protocol = skb->protocol;
1653
1654         /*
1655          *      Charge the memory to the socket. This is done specifically
1656          *      to prevent sockets using all the memory up.
1657          */
1658
1659         if (sock_queue_rcv_skb(sk, skb) == 0)
1660                 return 0;
1661
1662 out:
1663         kfree_skb(skb);
1664 oom:
1665         return 0;
1666 }
1667
1668
1669 /*
1670  *      Output a raw packet to a device layer. This bypasses all the other
1671  *      protocol layers and you must therefore supply it with a complete frame
1672  */
1673
1674 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1675                                size_t len)
1676 {
1677         struct sock *sk = sock->sk;
1678         DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1679         struct sk_buff *skb = NULL;
1680         struct net_device *dev;
1681         __be16 proto = 0;
1682         int err;
1683         int extra_len = 0;
1684
1685         /*
1686          *      Get and verify the address.
1687          */
1688
1689         if (saddr) {
1690                 if (msg->msg_namelen < sizeof(struct sockaddr))
1691                         return -EINVAL;
1692                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1693                         proto = saddr->spkt_protocol;
1694         } else
1695                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1696
1697         /*
1698          *      Find the device first to size check it
1699          */
1700
1701         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1702 retry:
1703         rcu_read_lock();
1704         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1705         err = -ENODEV;
1706         if (dev == NULL)
1707                 goto out_unlock;
1708
1709         err = -ENETDOWN;
1710         if (!(dev->flags & IFF_UP))
1711                 goto out_unlock;
1712
1713         /*
1714          * You may not queue a frame bigger than the mtu. This is the lowest level
1715          * raw protocol and you must do your own fragmentation at this level.
1716          */
1717
1718         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1719                 if (!netif_supports_nofcs(dev)) {
1720                         err = -EPROTONOSUPPORT;
1721                         goto out_unlock;
1722                 }
1723                 extra_len = 4; /* We're doing our own CRC */
1724         }
1725
1726         err = -EMSGSIZE;
1727         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1728                 goto out_unlock;
1729
1730         if (!skb) {
1731                 size_t reserved = LL_RESERVED_SPACE(dev);
1732                 int tlen = dev->needed_tailroom;
1733                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1734
1735                 rcu_read_unlock();
1736                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1737                 if (skb == NULL)
1738                         return -ENOBUFS;
1739                 /* FIXME: Save some space for broken drivers that write a hard
1740                  * header at transmission time by themselves. PPP is the notable
1741                  * one here. This should really be fixed at the driver level.
1742                  */
1743                 skb_reserve(skb, reserved);
1744                 skb_reset_network_header(skb);
1745
1746                 /* Try to align data part correctly */
1747                 if (hhlen) {
1748                         skb->data -= hhlen;
1749                         skb->tail -= hhlen;
1750                         if (len < hhlen)
1751                                 skb_reset_network_header(skb);
1752                 }
1753                 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1754                 if (err)
1755                         goto out_free;
1756                 goto retry;
1757         }
1758
1759         if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1760                 /* Earlier code assumed this would be a VLAN pkt,
1761                  * double-check this now that we have the actual
1762                  * packet in hand.
1763                  */
1764                 struct ethhdr *ehdr;
1765                 skb_reset_mac_header(skb);
1766                 ehdr = eth_hdr(skb);
1767                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1768                         err = -EMSGSIZE;
1769                         goto out_unlock;
1770                 }
1771         }
1772
1773         skb->protocol = proto;
1774         skb->dev = dev;
1775         skb->priority = sk->sk_priority;
1776         skb->mark = sk->sk_mark;
1777
1778         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1779
1780         if (unlikely(extra_len == 4))
1781                 skb->no_fcs = 1;
1782
1783         skb_probe_transport_header(skb, 0);
1784
1785         dev_queue_xmit(skb);
1786         rcu_read_unlock();
1787         return len;
1788
1789 out_unlock:
1790         rcu_read_unlock();
1791 out_free:
1792         kfree_skb(skb);
1793         return err;
1794 }
1795
1796 static unsigned int run_filter(const struct sk_buff *skb,
1797                                       const struct sock *sk,
1798                                       unsigned int res)
1799 {
1800         struct sk_filter *filter;
1801
1802         rcu_read_lock();
1803         filter = rcu_dereference(sk->sk_filter);
1804         if (filter != NULL)
1805                 res = SK_RUN_FILTER(filter, skb);
1806         rcu_read_unlock();
1807
1808         return res;
1809 }
1810
1811 /*
1812  * This function makes lazy skb cloning in hope that most of packets
1813  * are discarded by BPF.
1814  *
1815  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1816  * and skb->cb are mangled. It works because (and until) packets
1817  * falling here are owned by current CPU. Output packets are cloned
1818  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1819  * sequencially, so that if we return skb to original state on exit,
1820  * we will not harm anyone.
1821  */
1822
1823 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1824                       struct packet_type *pt, struct net_device *orig_dev)
1825 {
1826         struct sock *sk;
1827         struct sockaddr_ll *sll;
1828         struct packet_sock *po;
1829         u8 *skb_head = skb->data;
1830         int skb_len = skb->len;
1831         unsigned int snaplen, res;
1832
1833         if (skb->pkt_type == PACKET_LOOPBACK)
1834                 goto drop;
1835
1836         sk = pt->af_packet_priv;
1837         po = pkt_sk(sk);
1838
1839         if (!net_eq(dev_net(dev), sock_net(sk)))
1840                 goto drop;
1841
1842         skb->dev = dev;
1843
1844         if (dev->header_ops) {
1845                 /* The device has an explicit notion of ll header,
1846                  * exported to higher levels.
1847                  *
1848                  * Otherwise, the device hides details of its frame
1849                  * structure, so that corresponding packet head is
1850                  * never delivered to user.
1851                  */
1852                 if (sk->sk_type != SOCK_DGRAM)
1853                         skb_push(skb, skb->data - skb_mac_header(skb));
1854                 else if (skb->pkt_type == PACKET_OUTGOING) {
1855                         /* Special case: outgoing packets have ll header at head */
1856                         skb_pull(skb, skb_network_offset(skb));
1857                 }
1858         }
1859
1860         snaplen = skb->len;
1861
1862         res = run_filter(skb, sk, snaplen);
1863         if (!res)
1864                 goto drop_n_restore;
1865         if (snaplen > res)
1866                 snaplen = res;
1867
1868         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1869                 goto drop_n_acct;
1870
1871         if (skb_shared(skb)) {
1872                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1873                 if (nskb == NULL)
1874                         goto drop_n_acct;
1875
1876                 if (skb_head != skb->data) {
1877                         skb->data = skb_head;
1878                         skb->len = skb_len;
1879                 }
1880                 consume_skb(skb);
1881                 skb = nskb;
1882         }
1883
1884         sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
1885
1886         sll = &PACKET_SKB_CB(skb)->sa.ll;
1887         sll->sll_hatype = dev->type;
1888         sll->sll_pkttype = skb->pkt_type;
1889         if (unlikely(po->origdev))
1890                 sll->sll_ifindex = orig_dev->ifindex;
1891         else
1892                 sll->sll_ifindex = dev->ifindex;
1893
1894         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1895
1896         /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
1897          * Use their space for storing the original skb length.
1898          */
1899         PACKET_SKB_CB(skb)->sa.origlen = skb->len;
1900
1901         if (pskb_trim(skb, snaplen))
1902                 goto drop_n_acct;
1903
1904         skb_set_owner_r(skb, sk);
1905         skb->dev = NULL;
1906         skb_dst_drop(skb);
1907
1908         /* drop conntrack reference */
1909         nf_reset(skb);
1910
1911         spin_lock(&sk->sk_receive_queue.lock);
1912         po->stats.stats1.tp_packets++;
1913         sock_skb_set_dropcount(sk, skb);
1914         __skb_queue_tail(&sk->sk_receive_queue, skb);
1915         spin_unlock(&sk->sk_receive_queue.lock);
1916         sk->sk_data_ready(sk);
1917         return 0;
1918
1919 drop_n_acct:
1920         spin_lock(&sk->sk_receive_queue.lock);
1921         po->stats.stats1.tp_drops++;
1922         atomic_inc(&sk->sk_drops);
1923         spin_unlock(&sk->sk_receive_queue.lock);
1924
1925 drop_n_restore:
1926         if (skb_head != skb->data && skb_shared(skb)) {
1927                 skb->data = skb_head;
1928                 skb->len = skb_len;
1929         }
1930 drop:
1931         consume_skb(skb);
1932         return 0;
1933 }
1934
1935 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1936                        struct packet_type *pt, struct net_device *orig_dev)
1937 {
1938         struct sock *sk;
1939         struct packet_sock *po;
1940         struct sockaddr_ll *sll;
1941         union tpacket_uhdr h;
1942         u8 *skb_head = skb->data;
1943         int skb_len = skb->len;
1944         unsigned int snaplen, res;
1945         unsigned long status = TP_STATUS_USER;
1946         unsigned short macoff, netoff, hdrlen;
1947         struct sk_buff *copy_skb = NULL;
1948         struct timespec ts;
1949         __u32 ts_status;
1950
1951         /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
1952          * We may add members to them until current aligned size without forcing
1953          * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
1954          */
1955         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
1956         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
1957
1958         if (skb->pkt_type == PACKET_LOOPBACK)
1959                 goto drop;
1960
1961         sk = pt->af_packet_priv;
1962         po = pkt_sk(sk);
1963
1964         if (!net_eq(dev_net(dev), sock_net(sk)))
1965                 goto drop;
1966
1967         if (dev->header_ops) {
1968                 if (sk->sk_type != SOCK_DGRAM)
1969                         skb_push(skb, skb->data - skb_mac_header(skb));
1970                 else if (skb->pkt_type == PACKET_OUTGOING) {
1971                         /* Special case: outgoing packets have ll header at head */
1972                         skb_pull(skb, skb_network_offset(skb));
1973                 }
1974         }
1975
1976         snaplen = skb->len;
1977
1978         res = run_filter(skb, sk, snaplen);
1979         if (!res)
1980                 goto drop_n_restore;
1981
1982         if (skb->ip_summed == CHECKSUM_PARTIAL)
1983                 status |= TP_STATUS_CSUMNOTREADY;
1984         else if (skb->pkt_type != PACKET_OUTGOING &&
1985                  (skb->ip_summed == CHECKSUM_COMPLETE ||
1986                   skb_csum_unnecessary(skb)))
1987                 status |= TP_STATUS_CSUM_VALID;
1988
1989         if (snaplen > res)
1990                 snaplen = res;
1991
1992         if (sk->sk_type == SOCK_DGRAM) {
1993                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1994                                   po->tp_reserve;
1995         } else {
1996                 unsigned int maclen = skb_network_offset(skb);
1997                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1998                                        (maclen < 16 ? 16 : maclen)) +
1999                         po->tp_reserve;
2000                 macoff = netoff - maclen;
2001         }
2002         if (po->tp_version <= TPACKET_V2) {
2003                 if (macoff + snaplen > po->rx_ring.frame_size) {
2004                         if (po->copy_thresh &&
2005                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2006                                 if (skb_shared(skb)) {
2007                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
2008                                 } else {
2009                                         copy_skb = skb_get(skb);
2010                                         skb_head = skb->data;
2011                                 }
2012                                 if (copy_skb)
2013                                         skb_set_owner_r(copy_skb, sk);
2014                         }
2015                         snaplen = po->rx_ring.frame_size - macoff;
2016                         if ((int)snaplen < 0)
2017                                 snaplen = 0;
2018                 }
2019         } else if (unlikely(macoff + snaplen >
2020                             GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2021                 u32 nval;
2022
2023                 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2024                 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2025                             snaplen, nval, macoff);
2026                 snaplen = nval;
2027                 if (unlikely((int)snaplen < 0)) {
2028                         snaplen = 0;
2029                         macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2030                 }
2031         }
2032         spin_lock(&sk->sk_receive_queue.lock);
2033         h.raw = packet_current_rx_frame(po, skb,
2034                                         TP_STATUS_KERNEL, (macoff+snaplen));
2035         if (!h.raw)
2036                 goto ring_is_full;
2037         if (po->tp_version <= TPACKET_V2) {
2038                 packet_increment_rx_head(po, &po->rx_ring);
2039         /*
2040          * LOSING will be reported till you read the stats,
2041          * because it's COR - Clear On Read.
2042          * Anyways, moving it for V1/V2 only as V3 doesn't need this
2043          * at packet level.
2044          */
2045                 if (po->stats.stats1.tp_drops)
2046                         status |= TP_STATUS_LOSING;
2047         }
2048         po->stats.stats1.tp_packets++;
2049         if (copy_skb) {
2050                 status |= TP_STATUS_COPY;
2051                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2052         }
2053         spin_unlock(&sk->sk_receive_queue.lock);
2054
2055         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2056
2057         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2058                 getnstimeofday(&ts);
2059
2060         status |= ts_status;
2061
2062         switch (po->tp_version) {
2063         case TPACKET_V1:
2064                 h.h1->tp_len = skb->len;
2065                 h.h1->tp_snaplen = snaplen;
2066                 h.h1->tp_mac = macoff;
2067                 h.h1->tp_net = netoff;
2068                 h.h1->tp_sec = ts.tv_sec;
2069                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2070                 hdrlen = sizeof(*h.h1);
2071                 break;
2072         case TPACKET_V2:
2073                 h.h2->tp_len = skb->len;
2074                 h.h2->tp_snaplen = snaplen;
2075                 h.h2->tp_mac = macoff;
2076                 h.h2->tp_net = netoff;
2077                 h.h2->tp_sec = ts.tv_sec;
2078                 h.h2->tp_nsec = ts.tv_nsec;
2079                 if (skb_vlan_tag_present(skb)) {
2080                         h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2081                         h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2082                         status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2083                 } else {
2084                         h.h2->tp_vlan_tci = 0;
2085                         h.h2->tp_vlan_tpid = 0;
2086                 }
2087                 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2088                 hdrlen = sizeof(*h.h2);
2089                 break;
2090         case TPACKET_V3:
2091                 /* tp_nxt_offset,vlan are already populated above.
2092                  * So DONT clear those fields here
2093                  */
2094                 h.h3->tp_status |= status;
2095                 h.h3->tp_len = skb->len;
2096                 h.h3->tp_snaplen = snaplen;
2097                 h.h3->tp_mac = macoff;
2098                 h.h3->tp_net = netoff;
2099                 h.h3->tp_sec  = ts.tv_sec;
2100                 h.h3->tp_nsec = ts.tv_nsec;
2101                 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2102                 hdrlen = sizeof(*h.h3);
2103                 break;
2104         default:
2105                 BUG();
2106         }
2107
2108         sll = h.raw + TPACKET_ALIGN(hdrlen);
2109         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2110         sll->sll_family = AF_PACKET;
2111         sll->sll_hatype = dev->type;
2112         sll->sll_protocol = skb->protocol;
2113         sll->sll_pkttype = skb->pkt_type;
2114         if (unlikely(po->origdev))
2115                 sll->sll_ifindex = orig_dev->ifindex;
2116         else
2117                 sll->sll_ifindex = dev->ifindex;
2118
2119         smp_mb();
2120
2121 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2122         if (po->tp_version <= TPACKET_V2) {
2123                 u8 *start, *end;
2124
2125                 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2126                                         macoff + snaplen);
2127
2128                 for (start = h.raw; start < end; start += PAGE_SIZE)
2129                         flush_dcache_page(pgv_to_page(start));
2130         }
2131         smp_wmb();
2132 #endif
2133
2134         if (po->tp_version <= TPACKET_V2) {
2135                 __packet_set_status(po, h.raw, status);
2136                 sk->sk_data_ready(sk);
2137         } else {
2138                 prb_clear_blk_fill_status(&po->rx_ring);
2139         }
2140
2141 drop_n_restore:
2142         if (skb_head != skb->data && skb_shared(skb)) {
2143                 skb->data = skb_head;
2144                 skb->len = skb_len;
2145         }
2146 drop:
2147         kfree_skb(skb);
2148         return 0;
2149
2150 ring_is_full:
2151         po->stats.stats1.tp_drops++;
2152         spin_unlock(&sk->sk_receive_queue.lock);
2153
2154         sk->sk_data_ready(sk);
2155         kfree_skb(copy_skb);
2156         goto drop_n_restore;
2157 }
2158
2159 static void tpacket_destruct_skb(struct sk_buff *skb)
2160 {
2161         struct packet_sock *po = pkt_sk(skb->sk);
2162
2163         if (likely(po->tx_ring.pg_vec)) {
2164                 void *ph;
2165                 __u32 ts;
2166
2167                 ph = skb_shinfo(skb)->destructor_arg;
2168                 packet_dec_pending(&po->tx_ring);
2169
2170                 ts = __packet_set_timestamp(po, ph, skb);
2171                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2172         }
2173
2174         sock_wfree(skb);
2175 }
2176
2177 static bool ll_header_truncated(const struct net_device *dev, int len)
2178 {
2179         /* net device doesn't like empty head */
2180         if (unlikely(len <= dev->hard_header_len)) {
2181                 net_warn_ratelimited("%s: packet size is too short (%d <= %d)\n",
2182                                      current->comm, len, dev->hard_header_len);
2183                 return true;
2184         }
2185
2186         return false;
2187 }
2188
2189 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2190                 void *frame, struct net_device *dev, int size_max,
2191                 __be16 proto, unsigned char *addr, int hlen)
2192 {
2193         union tpacket_uhdr ph;
2194         int to_write, offset, len, tp_len, nr_frags, len_max;
2195         struct socket *sock = po->sk.sk_socket;
2196         struct page *page;
2197         void *data;
2198         int err;
2199
2200         ph.raw = frame;
2201
2202         skb->protocol = proto;
2203         skb->dev = dev;
2204         skb->priority = po->sk.sk_priority;
2205         skb->mark = po->sk.sk_mark;
2206         sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2207         skb_shinfo(skb)->destructor_arg = ph.raw;
2208
2209         switch (po->tp_version) {
2210         case TPACKET_V2:
2211                 tp_len = ph.h2->tp_len;
2212                 break;
2213         default:
2214                 tp_len = ph.h1->tp_len;
2215                 break;
2216         }
2217         if (unlikely(tp_len > size_max)) {
2218                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2219                 return -EMSGSIZE;
2220         }
2221
2222         skb_reserve(skb, hlen);
2223         skb_reset_network_header(skb);
2224
2225         if (!packet_use_direct_xmit(po))
2226                 skb_probe_transport_header(skb, 0);
2227         if (unlikely(po->tp_tx_has_off)) {
2228                 int off_min, off_max, off;
2229                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2230                 off_max = po->tx_ring.frame_size - tp_len;
2231                 if (sock->type == SOCK_DGRAM) {
2232                         switch (po->tp_version) {
2233                         case TPACKET_V2:
2234                                 off = ph.h2->tp_net;
2235                                 break;
2236                         default:
2237                                 off = ph.h1->tp_net;
2238                                 break;
2239                         }
2240                 } else {
2241                         switch (po->tp_version) {
2242                         case TPACKET_V2:
2243                                 off = ph.h2->tp_mac;
2244                                 break;
2245                         default:
2246                                 off = ph.h1->tp_mac;
2247                                 break;
2248                         }
2249                 }
2250                 if (unlikely((off < off_min) || (off_max < off)))
2251                         return -EINVAL;
2252                 data = ph.raw + off;
2253         } else {
2254                 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2255         }
2256         to_write = tp_len;
2257
2258         if (sock->type == SOCK_DGRAM) {
2259                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2260                                 NULL, tp_len);
2261                 if (unlikely(err < 0))
2262                         return -EINVAL;
2263         } else if (dev->hard_header_len) {
2264                 if (ll_header_truncated(dev, tp_len))
2265                         return -EINVAL;
2266
2267                 skb_push(skb, dev->hard_header_len);
2268                 err = skb_store_bits(skb, 0, data,
2269                                 dev->hard_header_len);
2270                 if (unlikely(err))
2271                         return err;
2272
2273                 data += dev->hard_header_len;
2274                 to_write -= dev->hard_header_len;
2275         }
2276
2277         offset = offset_in_page(data);
2278         len_max = PAGE_SIZE - offset;
2279         len = ((to_write > len_max) ? len_max : to_write);
2280
2281         skb->data_len = to_write;
2282         skb->len += to_write;
2283         skb->truesize += to_write;
2284         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2285
2286         while (likely(to_write)) {
2287                 nr_frags = skb_shinfo(skb)->nr_frags;
2288
2289                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2290                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2291                                MAX_SKB_FRAGS);
2292                         return -EFAULT;
2293                 }
2294
2295                 page = pgv_to_page(data);
2296                 data += len;
2297                 flush_dcache_page(page);
2298                 get_page(page);
2299                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2300                 to_write -= len;
2301                 offset = 0;
2302                 len_max = PAGE_SIZE;
2303                 len = ((to_write > len_max) ? len_max : to_write);
2304         }
2305
2306         return tp_len;
2307 }
2308
2309 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2310 {
2311         struct sk_buff *skb;
2312         struct net_device *dev;
2313         __be16 proto;
2314         int err, reserve = 0;
2315         void *ph;
2316         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2317         bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2318         int tp_len, size_max;
2319         unsigned char *addr;
2320         int len_sum = 0;
2321         int status = TP_STATUS_AVAILABLE;
2322         int hlen, tlen;
2323
2324         mutex_lock(&po->pg_vec_lock);
2325
2326         if (likely(saddr == NULL)) {
2327                 dev     = packet_cached_dev_get(po);
2328                 proto   = po->num;
2329                 addr    = NULL;
2330         } else {
2331                 err = -EINVAL;
2332                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2333                         goto out;
2334                 if (msg->msg_namelen < (saddr->sll_halen
2335                                         + offsetof(struct sockaddr_ll,
2336                                                 sll_addr)))
2337                         goto out;
2338                 proto   = saddr->sll_protocol;
2339                 addr    = saddr->sll_addr;
2340                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2341         }
2342
2343         err = -ENXIO;
2344         if (unlikely(dev == NULL))
2345                 goto out;
2346         err = -ENETDOWN;
2347         if (unlikely(!(dev->flags & IFF_UP)))
2348                 goto out_put;
2349
2350         reserve = dev->hard_header_len + VLAN_HLEN;
2351         size_max = po->tx_ring.frame_size
2352                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2353
2354         if (size_max > dev->mtu + reserve)
2355                 size_max = dev->mtu + reserve;
2356
2357         do {
2358                 ph = packet_current_frame(po, &po->tx_ring,
2359                                           TP_STATUS_SEND_REQUEST);
2360                 if (unlikely(ph == NULL)) {
2361                         if (need_wait && need_resched())
2362                                 schedule();
2363                         continue;
2364                 }
2365
2366                 status = TP_STATUS_SEND_REQUEST;
2367                 hlen = LL_RESERVED_SPACE(dev);
2368                 tlen = dev->needed_tailroom;
2369                 skb = sock_alloc_send_skb(&po->sk,
2370                                 hlen + tlen + sizeof(struct sockaddr_ll),
2371                                 !need_wait, &err);
2372
2373                 if (unlikely(skb == NULL)) {
2374                         /* we assume the socket was initially writeable ... */
2375                         if (likely(len_sum > 0))
2376                                 err = len_sum;
2377                         goto out_status;
2378                 }
2379                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2380                                           addr, hlen);
2381                 if (tp_len > dev->mtu + dev->hard_header_len) {
2382                         struct ethhdr *ehdr;
2383                         /* Earlier code assumed this would be a VLAN pkt,
2384                          * double-check this now that we have the actual
2385                          * packet in hand.
2386                          */
2387
2388                         skb_reset_mac_header(skb);
2389                         ehdr = eth_hdr(skb);
2390                         if (ehdr->h_proto != htons(ETH_P_8021Q))
2391                                 tp_len = -EMSGSIZE;
2392                 }
2393                 if (unlikely(tp_len < 0)) {
2394                         if (po->tp_loss) {
2395                                 __packet_set_status(po, ph,
2396                                                 TP_STATUS_AVAILABLE);
2397                                 packet_increment_head(&po->tx_ring);
2398                                 kfree_skb(skb);
2399                                 continue;
2400                         } else {
2401                                 status = TP_STATUS_WRONG_FORMAT;
2402                                 err = tp_len;
2403                                 goto out_status;
2404                         }
2405                 }
2406
2407                 packet_pick_tx_queue(dev, skb);
2408
2409                 skb->destructor = tpacket_destruct_skb;
2410                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2411                 packet_inc_pending(&po->tx_ring);
2412
2413                 status = TP_STATUS_SEND_REQUEST;
2414                 err = po->xmit(skb);
2415                 if (unlikely(err > 0)) {
2416                         err = net_xmit_errno(err);
2417                         if (err && __packet_get_status(po, ph) ==
2418                                    TP_STATUS_AVAILABLE) {
2419                                 /* skb was destructed already */
2420                                 skb = NULL;
2421                                 goto out_status;
2422                         }
2423                         /*
2424                          * skb was dropped but not destructed yet;
2425                          * let's treat it like congestion or err < 0
2426                          */
2427                         err = 0;
2428                 }
2429                 packet_increment_head(&po->tx_ring);
2430                 len_sum += tp_len;
2431         } while (likely((ph != NULL) ||
2432                 /* Note: packet_read_pending() might be slow if we have
2433                  * to call it as it's per_cpu variable, but in fast-path
2434                  * we already short-circuit the loop with the first
2435                  * condition, and luckily don't have to go that path
2436                  * anyway.
2437                  */
2438                  (need_wait && packet_read_pending(&po->tx_ring))));
2439
2440         err = len_sum;
2441         goto out_put;
2442
2443 out_status:
2444         __packet_set_status(po, ph, status);
2445         kfree_skb(skb);
2446 out_put:
2447         dev_put(dev);
2448 out:
2449         mutex_unlock(&po->pg_vec_lock);
2450         return err;
2451 }
2452
2453 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2454                                         size_t reserve, size_t len,
2455                                         size_t linear, int noblock,
2456                                         int *err)
2457 {
2458         struct sk_buff *skb;
2459
2460         /* Under a page?  Don't bother with paged skb. */
2461         if (prepad + len < PAGE_SIZE || !linear)
2462                 linear = len;
2463
2464         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2465                                    err, 0);
2466         if (!skb)
2467                 return NULL;
2468
2469         skb_reserve(skb, reserve);
2470         skb_put(skb, linear);
2471         skb->data_len = len - linear;
2472         skb->len += len - linear;
2473
2474         return skb;
2475 }
2476
2477 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2478 {
2479         struct sock *sk = sock->sk;
2480         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2481         struct sk_buff *skb;
2482         struct net_device *dev;
2483         __be16 proto;
2484         unsigned char *addr;
2485         int err, reserve = 0;
2486         struct virtio_net_hdr vnet_hdr = { 0 };
2487         int offset = 0;
2488         int vnet_hdr_len;
2489         struct packet_sock *po = pkt_sk(sk);
2490         unsigned short gso_type = 0;
2491         int hlen, tlen;
2492         int extra_len = 0;
2493         ssize_t n;
2494
2495         /*
2496          *      Get and verify the address.
2497          */
2498
2499         if (likely(saddr == NULL)) {
2500                 dev     = packet_cached_dev_get(po);
2501                 proto   = po->num;
2502                 addr    = NULL;
2503         } else {
2504                 err = -EINVAL;
2505                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2506                         goto out;
2507                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2508                         goto out;
2509                 proto   = saddr->sll_protocol;
2510                 addr    = saddr->sll_addr;
2511                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2512         }
2513
2514         err = -ENXIO;
2515         if (unlikely(dev == NULL))
2516                 goto out_unlock;
2517         err = -ENETDOWN;
2518         if (unlikely(!(dev->flags & IFF_UP)))
2519                 goto out_unlock;
2520
2521         if (sock->type == SOCK_RAW)
2522                 reserve = dev->hard_header_len;
2523         if (po->has_vnet_hdr) {
2524                 vnet_hdr_len = sizeof(vnet_hdr);
2525
2526                 err = -EINVAL;
2527                 if (len < vnet_hdr_len)
2528                         goto out_unlock;
2529
2530                 len -= vnet_hdr_len;
2531
2532                 err = -EFAULT;
2533                 n = copy_from_iter(&vnet_hdr, vnet_hdr_len, &msg->msg_iter);
2534                 if (n != vnet_hdr_len)
2535                         goto out_unlock;
2536
2537                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2538                     (__virtio16_to_cpu(false, vnet_hdr.csum_start) +
2539                      __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2 >
2540                       __virtio16_to_cpu(false, vnet_hdr.hdr_len)))
2541                         vnet_hdr.hdr_len = __cpu_to_virtio16(false,
2542                                  __virtio16_to_cpu(false, vnet_hdr.csum_start) +
2543                                 __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2);
2544
2545                 err = -EINVAL;
2546                 if (__virtio16_to_cpu(false, vnet_hdr.hdr_len) > len)
2547                         goto out_unlock;
2548
2549                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2550                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2551                         case VIRTIO_NET_HDR_GSO_TCPV4:
2552                                 gso_type = SKB_GSO_TCPV4;
2553                                 break;
2554                         case VIRTIO_NET_HDR_GSO_TCPV6:
2555                                 gso_type = SKB_GSO_TCPV6;
2556                                 break;
2557                         case VIRTIO_NET_HDR_GSO_UDP:
2558                                 gso_type = SKB_GSO_UDP;
2559                                 break;
2560                         default:
2561                                 goto out_unlock;
2562                         }
2563
2564                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2565                                 gso_type |= SKB_GSO_TCP_ECN;
2566
2567                         if (vnet_hdr.gso_size == 0)
2568                                 goto out_unlock;
2569
2570                 }
2571         }
2572
2573         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2574                 if (!netif_supports_nofcs(dev)) {
2575                         err = -EPROTONOSUPPORT;
2576                         goto out_unlock;
2577                 }
2578                 extra_len = 4; /* We're doing our own CRC */
2579         }
2580
2581         err = -EMSGSIZE;
2582         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2583                 goto out_unlock;
2584
2585         err = -ENOBUFS;
2586         hlen = LL_RESERVED_SPACE(dev);
2587         tlen = dev->needed_tailroom;
2588         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
2589                                __virtio16_to_cpu(false, vnet_hdr.hdr_len),
2590                                msg->msg_flags & MSG_DONTWAIT, &err);
2591         if (skb == NULL)
2592                 goto out_unlock;
2593
2594         skb_set_network_header(skb, reserve);
2595
2596         err = -EINVAL;
2597         if (sock->type == SOCK_DGRAM) {
2598                 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2599                 if (unlikely(offset < 0))
2600                         goto out_free;
2601         } else {
2602                 if (ll_header_truncated(dev, len))
2603                         goto out_free;
2604         }
2605
2606         /* Returns -EFAULT on error */
2607         err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2608         if (err)
2609                 goto out_free;
2610
2611         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2612
2613         if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2614                 /* Earlier code assumed this would be a VLAN pkt,
2615                  * double-check this now that we have the actual
2616                  * packet in hand.
2617                  */
2618                 struct ethhdr *ehdr;
2619                 skb_reset_mac_header(skb);
2620                 ehdr = eth_hdr(skb);
2621                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2622                         err = -EMSGSIZE;
2623                         goto out_free;
2624                 }
2625         }
2626
2627         skb->protocol = proto;
2628         skb->dev = dev;
2629         skb->priority = sk->sk_priority;
2630         skb->mark = sk->sk_mark;
2631
2632         packet_pick_tx_queue(dev, skb);
2633
2634         if (po->has_vnet_hdr) {
2635                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2636                         u16 s = __virtio16_to_cpu(false, vnet_hdr.csum_start);
2637                         u16 o = __virtio16_to_cpu(false, vnet_hdr.csum_offset);
2638                         if (!skb_partial_csum_set(skb, s, o)) {
2639                                 err = -EINVAL;
2640                                 goto out_free;
2641                         }
2642                 }
2643
2644                 skb_shinfo(skb)->gso_size =
2645                         __virtio16_to_cpu(false, vnet_hdr.gso_size);
2646                 skb_shinfo(skb)->gso_type = gso_type;
2647
2648                 /* Header must be checked, and gso_segs computed. */
2649                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2650                 skb_shinfo(skb)->gso_segs = 0;
2651
2652                 len += vnet_hdr_len;
2653         }
2654
2655         if (!packet_use_direct_xmit(po))
2656                 skb_probe_transport_header(skb, reserve);
2657         if (unlikely(extra_len == 4))
2658                 skb->no_fcs = 1;
2659
2660         err = po->xmit(skb);
2661         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2662                 goto out_unlock;
2663
2664         dev_put(dev);
2665
2666         return len;
2667
2668 out_free:
2669         kfree_skb(skb);
2670 out_unlock:
2671         if (dev)
2672                 dev_put(dev);
2673 out:
2674         return err;
2675 }
2676
2677 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2678 {
2679         struct sock *sk = sock->sk;
2680         struct packet_sock *po = pkt_sk(sk);
2681
2682         if (po->tx_ring.pg_vec)
2683                 return tpacket_snd(po, msg);
2684         else
2685                 return packet_snd(sock, msg, len);
2686 }
2687
2688 /*
2689  *      Close a PACKET socket. This is fairly simple. We immediately go
2690  *      to 'closed' state and remove our protocol entry in the device list.
2691  */
2692
2693 static int packet_release(struct socket *sock)
2694 {
2695         struct sock *sk = sock->sk;
2696         struct packet_sock *po;
2697         struct net *net;
2698         union tpacket_req_u req_u;
2699
2700         if (!sk)
2701                 return 0;
2702
2703         net = sock_net(sk);
2704         po = pkt_sk(sk);
2705
2706         mutex_lock(&net->packet.sklist_lock);
2707         sk_del_node_init_rcu(sk);
2708         mutex_unlock(&net->packet.sklist_lock);
2709
2710         preempt_disable();
2711         sock_prot_inuse_add(net, sk->sk_prot, -1);
2712         preempt_enable();
2713
2714         spin_lock(&po->bind_lock);
2715         unregister_prot_hook(sk, false);
2716         packet_cached_dev_reset(po);
2717
2718         if (po->prot_hook.dev) {
2719                 dev_put(po->prot_hook.dev);
2720                 po->prot_hook.dev = NULL;
2721         }
2722         spin_unlock(&po->bind_lock);
2723
2724         packet_flush_mclist(sk);
2725
2726         if (po->rx_ring.pg_vec) {
2727                 memset(&req_u, 0, sizeof(req_u));
2728                 packet_set_ring(sk, &req_u, 1, 0);
2729         }
2730
2731         if (po->tx_ring.pg_vec) {
2732                 memset(&req_u, 0, sizeof(req_u));
2733                 packet_set_ring(sk, &req_u, 1, 1);
2734         }
2735
2736         fanout_release(sk);
2737
2738         synchronize_net();
2739         /*
2740          *      Now the socket is dead. No more input will appear.
2741          */
2742         sock_orphan(sk);
2743         sock->sk = NULL;
2744
2745         /* Purge queues */
2746
2747         skb_queue_purge(&sk->sk_receive_queue);
2748         packet_free_pending(po);
2749         sk_refcnt_debug_release(sk);
2750
2751         sock_put(sk);
2752         return 0;
2753 }
2754
2755 /*
2756  *      Attach a packet hook.
2757  */
2758
2759 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 proto)
2760 {
2761         struct packet_sock *po = pkt_sk(sk);
2762         const struct net_device *dev_curr;
2763         __be16 proto_curr;
2764         bool need_rehook;
2765
2766         if (po->fanout) {
2767                 if (dev)
2768                         dev_put(dev);
2769
2770                 return -EINVAL;
2771         }
2772
2773         lock_sock(sk);
2774         spin_lock(&po->bind_lock);
2775
2776         proto_curr = po->prot_hook.type;
2777         dev_curr = po->prot_hook.dev;
2778
2779         need_rehook = proto_curr != proto || dev_curr != dev;
2780
2781         if (need_rehook) {
2782                 unregister_prot_hook(sk, true);
2783
2784                 po->num = proto;
2785                 po->prot_hook.type = proto;
2786
2787                 if (po->prot_hook.dev)
2788                         dev_put(po->prot_hook.dev);
2789
2790                 po->prot_hook.dev = dev;
2791
2792                 po->ifindex = dev ? dev->ifindex : 0;
2793                 packet_cached_dev_assign(po, dev);
2794         }
2795
2796         if (proto == 0 || !need_rehook)
2797                 goto out_unlock;
2798
2799         if (!dev || (dev->flags & IFF_UP)) {
2800                 register_prot_hook(sk);
2801         } else {
2802                 sk->sk_err = ENETDOWN;
2803                 if (!sock_flag(sk, SOCK_DEAD))
2804                         sk->sk_error_report(sk);
2805         }
2806
2807 out_unlock:
2808         spin_unlock(&po->bind_lock);
2809         release_sock(sk);
2810         return 0;
2811 }
2812
2813 /*
2814  *      Bind a packet socket to a device
2815  */
2816
2817 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2818                             int addr_len)
2819 {
2820         struct sock *sk = sock->sk;
2821         char name[15];
2822         struct net_device *dev;
2823         int err = -ENODEV;
2824
2825         /*
2826          *      Check legality
2827          */
2828
2829         if (addr_len != sizeof(struct sockaddr))
2830                 return -EINVAL;
2831         strlcpy(name, uaddr->sa_data, sizeof(name));
2832
2833         dev = dev_get_by_name(sock_net(sk), name);
2834         if (dev)
2835                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2836         return err;
2837 }
2838
2839 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2840 {
2841         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2842         struct sock *sk = sock->sk;
2843         struct net_device *dev = NULL;
2844         int err;
2845
2846
2847         /*
2848          *      Check legality
2849          */
2850
2851         if (addr_len < sizeof(struct sockaddr_ll))
2852                 return -EINVAL;
2853         if (sll->sll_family != AF_PACKET)
2854                 return -EINVAL;
2855
2856         if (sll->sll_ifindex) {
2857                 err = -ENODEV;
2858                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2859                 if (dev == NULL)
2860                         goto out;
2861         }
2862         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2863
2864 out:
2865         return err;
2866 }
2867
2868 static struct proto packet_proto = {
2869         .name     = "PACKET",
2870         .owner    = THIS_MODULE,
2871         .obj_size = sizeof(struct packet_sock),
2872 };
2873
2874 /*
2875  *      Create a packet of type SOCK_PACKET.
2876  */
2877
2878 static int packet_create(struct net *net, struct socket *sock, int protocol,
2879                          int kern)
2880 {
2881         struct sock *sk;
2882         struct packet_sock *po;
2883         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2884         int err;
2885
2886         if (!ns_capable(net->user_ns, CAP_NET_RAW))
2887                 return -EPERM;
2888         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2889             sock->type != SOCK_PACKET)
2890                 return -ESOCKTNOSUPPORT;
2891
2892         sock->state = SS_UNCONNECTED;
2893
2894         err = -ENOBUFS;
2895         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
2896         if (sk == NULL)
2897                 goto out;
2898
2899         sock->ops = &packet_ops;
2900         if (sock->type == SOCK_PACKET)
2901                 sock->ops = &packet_ops_spkt;
2902
2903         sock_init_data(sock, sk);
2904
2905         po = pkt_sk(sk);
2906         sk->sk_family = PF_PACKET;
2907         po->num = proto;
2908         po->xmit = dev_queue_xmit;
2909
2910         err = packet_alloc_pending(po);
2911         if (err)
2912                 goto out2;
2913
2914         packet_cached_dev_reset(po);
2915
2916         sk->sk_destruct = packet_sock_destruct;
2917         sk_refcnt_debug_inc(sk);
2918
2919         /*
2920          *      Attach a protocol block
2921          */
2922
2923         spin_lock_init(&po->bind_lock);
2924         mutex_init(&po->pg_vec_lock);
2925         po->rollover = NULL;
2926         po->prot_hook.func = packet_rcv;
2927
2928         if (sock->type == SOCK_PACKET)
2929                 po->prot_hook.func = packet_rcv_spkt;
2930
2931         po->prot_hook.af_packet_priv = sk;
2932
2933         if (proto) {
2934                 po->prot_hook.type = proto;
2935                 register_prot_hook(sk);
2936         }
2937
2938         mutex_lock(&net->packet.sklist_lock);
2939         sk_add_node_rcu(sk, &net->packet.sklist);
2940         mutex_unlock(&net->packet.sklist_lock);
2941
2942         preempt_disable();
2943         sock_prot_inuse_add(net, &packet_proto, 1);
2944         preempt_enable();
2945
2946         return 0;
2947 out2:
2948         sk_free(sk);
2949 out:
2950         return err;
2951 }
2952
2953 /*
2954  *      Pull a packet from our receive queue and hand it to the user.
2955  *      If necessary we block.
2956  */
2957
2958 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2959                           int flags)
2960 {
2961         struct sock *sk = sock->sk;
2962         struct sk_buff *skb;
2963         int copied, err;
2964         int vnet_hdr_len = 0;
2965         unsigned int origlen = 0;
2966
2967         err = -EINVAL;
2968         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2969                 goto out;
2970
2971 #if 0
2972         /* What error should we return now? EUNATTACH? */
2973         if (pkt_sk(sk)->ifindex < 0)
2974                 return -ENODEV;
2975 #endif
2976
2977         if (flags & MSG_ERRQUEUE) {
2978                 err = sock_recv_errqueue(sk, msg, len,
2979                                          SOL_PACKET, PACKET_TX_TIMESTAMP);
2980                 goto out;
2981         }
2982
2983         /*
2984          *      Call the generic datagram receiver. This handles all sorts
2985          *      of horrible races and re-entrancy so we can forget about it
2986          *      in the protocol layers.
2987          *
2988          *      Now it will return ENETDOWN, if device have just gone down,
2989          *      but then it will block.
2990          */
2991
2992         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2993
2994         /*
2995          *      An error occurred so return it. Because skb_recv_datagram()
2996          *      handles the blocking we don't see and worry about blocking
2997          *      retries.
2998          */
2999
3000         if (skb == NULL)
3001                 goto out;
3002
3003         if (pkt_sk(sk)->has_vnet_hdr) {
3004                 struct virtio_net_hdr vnet_hdr = { 0 };
3005
3006                 err = -EINVAL;
3007                 vnet_hdr_len = sizeof(vnet_hdr);
3008                 if (len < vnet_hdr_len)
3009                         goto out_free;
3010
3011                 len -= vnet_hdr_len;
3012
3013                 if (skb_is_gso(skb)) {
3014                         struct skb_shared_info *sinfo = skb_shinfo(skb);
3015
3016                         /* This is a hint as to how much should be linear. */
3017                         vnet_hdr.hdr_len =
3018                                 __cpu_to_virtio16(false, skb_headlen(skb));
3019                         vnet_hdr.gso_size =
3020                                 __cpu_to_virtio16(false, sinfo->gso_size);
3021                         if (sinfo->gso_type & SKB_GSO_TCPV4)
3022                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
3023                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
3024                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
3025                         else if (sinfo->gso_type & SKB_GSO_UDP)
3026                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
3027                         else if (sinfo->gso_type & SKB_GSO_FCOE)
3028                                 goto out_free;
3029                         else
3030                                 BUG();
3031                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
3032                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
3033                 } else
3034                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
3035
3036                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3037                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
3038                         vnet_hdr.csum_start = __cpu_to_virtio16(false,
3039                                           skb_checksum_start_offset(skb));
3040                         vnet_hdr.csum_offset = __cpu_to_virtio16(false,
3041                                                          skb->csum_offset);
3042                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3043                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
3044                 } /* else everything is zero */
3045
3046                 err = memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_len);
3047                 if (err < 0)
3048                         goto out_free;
3049         }
3050
3051         /* You lose any data beyond the buffer you gave. If it worries
3052          * a user program they can ask the device for its MTU
3053          * anyway.
3054          */
3055         copied = skb->len;
3056         if (copied > len) {
3057                 copied = len;
3058                 msg->msg_flags |= MSG_TRUNC;
3059         }
3060
3061         err = skb_copy_datagram_msg(skb, 0, msg, copied);
3062         if (err)
3063                 goto out_free;
3064
3065         if (sock->type != SOCK_PACKET) {
3066                 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3067
3068                 /* Original length was stored in sockaddr_ll fields */
3069                 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3070                 sll->sll_family = AF_PACKET;
3071                 sll->sll_protocol = skb->protocol;
3072         }
3073
3074         sock_recv_ts_and_drops(msg, sk, skb);
3075
3076         if (msg->msg_name) {
3077                 /* If the address length field is there to be filled
3078                  * in, we fill it in now.
3079                  */
3080                 if (sock->type == SOCK_PACKET) {
3081                         __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3082                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
3083                 } else {
3084                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3085
3086                         msg->msg_namelen = sll->sll_halen +
3087                                 offsetof(struct sockaddr_ll, sll_addr);
3088                 }
3089                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3090                        msg->msg_namelen);
3091         }
3092
3093         if (pkt_sk(sk)->auxdata) {
3094                 struct tpacket_auxdata aux;
3095
3096                 aux.tp_status = TP_STATUS_USER;
3097                 if (skb->ip_summed == CHECKSUM_PARTIAL)
3098                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3099                 else if (skb->pkt_type != PACKET_OUTGOING &&
3100                          (skb->ip_summed == CHECKSUM_COMPLETE ||
3101                           skb_csum_unnecessary(skb)))
3102                         aux.tp_status |= TP_STATUS_CSUM_VALID;
3103
3104                 aux.tp_len = origlen;
3105                 aux.tp_snaplen = skb->len;
3106                 aux.tp_mac = 0;
3107                 aux.tp_net = skb_network_offset(skb);
3108                 if (skb_vlan_tag_present(skb)) {
3109                         aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3110                         aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3111                         aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3112                 } else {
3113                         aux.tp_vlan_tci = 0;
3114                         aux.tp_vlan_tpid = 0;
3115                 }
3116                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3117         }
3118
3119         /*
3120          *      Free or return the buffer as appropriate. Again this
3121          *      hides all the races and re-entrancy issues from us.
3122          */
3123         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3124
3125 out_free:
3126         skb_free_datagram(sk, skb);
3127 out:
3128         return err;
3129 }
3130
3131 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3132                                int *uaddr_len, int peer)
3133 {
3134         struct net_device *dev;
3135         struct sock *sk = sock->sk;
3136
3137         if (peer)
3138                 return -EOPNOTSUPP;
3139
3140         uaddr->sa_family = AF_PACKET;
3141         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3142         rcu_read_lock();
3143         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3144         if (dev)
3145                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3146         rcu_read_unlock();
3147         *uaddr_len = sizeof(*uaddr);
3148
3149         return 0;
3150 }
3151
3152 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3153                           int *uaddr_len, int peer)
3154 {
3155         struct net_device *dev;
3156         struct sock *sk = sock->sk;
3157         struct packet_sock *po = pkt_sk(sk);
3158         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3159
3160         if (peer)
3161                 return -EOPNOTSUPP;
3162
3163         sll->sll_family = AF_PACKET;
3164         sll->sll_ifindex = po->ifindex;
3165         sll->sll_protocol = po->num;
3166         sll->sll_pkttype = 0;
3167         rcu_read_lock();
3168         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3169         if (dev) {
3170                 sll->sll_hatype = dev->type;
3171                 sll->sll_halen = dev->addr_len;
3172                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3173         } else {
3174                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
3175                 sll->sll_halen = 0;
3176         }
3177         rcu_read_unlock();
3178         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3179
3180         return 0;
3181 }
3182
3183 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3184                          int what)
3185 {
3186         switch (i->type) {
3187         case PACKET_MR_MULTICAST:
3188                 if (i->alen != dev->addr_len)
3189                         return -EINVAL;
3190                 if (what > 0)
3191                         return dev_mc_add(dev, i->addr);
3192                 else
3193                         return dev_mc_del(dev, i->addr);
3194                 break;
3195         case PACKET_MR_PROMISC:
3196                 return dev_set_promiscuity(dev, what);
3197         case PACKET_MR_ALLMULTI:
3198                 return dev_set_allmulti(dev, what);
3199         case PACKET_MR_UNICAST:
3200                 if (i->alen != dev->addr_len)
3201                         return -EINVAL;
3202                 if (what > 0)
3203                         return dev_uc_add(dev, i->addr);
3204                 else
3205                         return dev_uc_del(dev, i->addr);
3206                 break;
3207         default:
3208                 break;
3209         }
3210         return 0;
3211 }
3212
3213 static void packet_dev_mclist_delete(struct net_device *dev,
3214                                      struct packet_mclist **mlp)
3215 {
3216         struct packet_mclist *ml;
3217
3218         while ((ml = *mlp) != NULL) {
3219                 if (ml->ifindex == dev->ifindex) {
3220                         packet_dev_mc(dev, ml, -1);
3221                         *mlp = ml->next;
3222                         kfree(ml);
3223                 } else
3224                         mlp = &ml->next;
3225         }
3226 }
3227
3228 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3229 {
3230         struct packet_sock *po = pkt_sk(sk);
3231         struct packet_mclist *ml, *i;
3232         struct net_device *dev;
3233         int err;
3234
3235         rtnl_lock();
3236
3237         err = -ENODEV;
3238         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3239         if (!dev)
3240                 goto done;
3241
3242         err = -EINVAL;
3243         if (mreq->mr_alen > dev->addr_len)
3244                 goto done;
3245
3246         err = -ENOBUFS;
3247         i = kmalloc(sizeof(*i), GFP_KERNEL);
3248         if (i == NULL)
3249                 goto done;
3250
3251         err = 0;
3252         for (ml = po->mclist; ml; ml = ml->next) {
3253                 if (ml->ifindex == mreq->mr_ifindex &&
3254                     ml->type == mreq->mr_type &&
3255                     ml->alen == mreq->mr_alen &&
3256                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3257                         ml->count++;
3258                         /* Free the new element ... */
3259                         kfree(i);
3260                         goto done;
3261                 }
3262         }
3263
3264         i->type = mreq->mr_type;
3265         i->ifindex = mreq->mr_ifindex;
3266         i->alen = mreq->mr_alen;
3267         memcpy(i->addr, mreq->mr_address, i->alen);
3268         i->count = 1;
3269         i->next = po->mclist;
3270         po->mclist = i;
3271         err = packet_dev_mc(dev, i, 1);
3272         if (err) {
3273                 po->mclist = i->next;
3274                 kfree(i);
3275         }
3276
3277 done:
3278         rtnl_unlock();
3279         return err;
3280 }
3281
3282 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3283 {
3284         struct packet_mclist *ml, **mlp;
3285
3286         rtnl_lock();
3287
3288         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3289                 if (ml->ifindex == mreq->mr_ifindex &&
3290                     ml->type == mreq->mr_type &&
3291                     ml->alen == mreq->mr_alen &&
3292                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3293                         if (--ml->count == 0) {
3294                                 struct net_device *dev;
3295                                 *mlp = ml->next;
3296                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3297                                 if (dev)
3298                                         packet_dev_mc(dev, ml, -1);
3299                                 kfree(ml);
3300                         }
3301                         break;
3302                 }
3303         }
3304         rtnl_unlock();
3305         return 0;
3306 }
3307
3308 static void packet_flush_mclist(struct sock *sk)
3309 {
3310         struct packet_sock *po = pkt_sk(sk);
3311         struct packet_mclist *ml;
3312
3313         if (!po->mclist)
3314                 return;
3315
3316         rtnl_lock();
3317         while ((ml = po->mclist) != NULL) {
3318                 struct net_device *dev;
3319
3320                 po->mclist = ml->next;
3321                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3322                 if (dev != NULL)
3323                         packet_dev_mc(dev, ml, -1);
3324                 kfree(ml);
3325         }
3326         rtnl_unlock();
3327 }
3328
3329 static int
3330 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3331 {
3332         struct sock *sk = sock->sk;
3333         struct packet_sock *po = pkt_sk(sk);
3334         int ret;
3335
3336         if (level != SOL_PACKET)
3337                 return -ENOPROTOOPT;
3338
3339         switch (optname) {
3340         case PACKET_ADD_MEMBERSHIP:
3341         case PACKET_DROP_MEMBERSHIP:
3342         {
3343                 struct packet_mreq_max mreq;
3344                 int len = optlen;
3345                 memset(&mreq, 0, sizeof(mreq));
3346                 if (len < sizeof(struct packet_mreq))
3347                         return -EINVAL;
3348                 if (len > sizeof(mreq))
3349                         len = sizeof(mreq);
3350                 if (copy_from_user(&mreq, optval, len))
3351                         return -EFAULT;
3352                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3353                         return -EINVAL;
3354                 if (optname == PACKET_ADD_MEMBERSHIP)
3355                         ret = packet_mc_add(sk, &mreq);
3356                 else
3357                         ret = packet_mc_drop(sk, &mreq);
3358                 return ret;
3359         }
3360
3361         case PACKET_RX_RING:
3362         case PACKET_TX_RING:
3363         {
3364                 union tpacket_req_u req_u;
3365                 int len;
3366
3367                 switch (po->tp_version) {
3368                 case TPACKET_V1:
3369                 case TPACKET_V2:
3370                         len = sizeof(req_u.req);
3371                         break;
3372                 case TPACKET_V3:
3373                 default:
3374                         len = sizeof(req_u.req3);
3375                         break;
3376                 }
3377                 if (optlen < len)
3378                         return -EINVAL;
3379                 if (pkt_sk(sk)->has_vnet_hdr)
3380                         return -EINVAL;
3381                 if (copy_from_user(&req_u.req, optval, len))
3382                         return -EFAULT;
3383                 return packet_set_ring(sk, &req_u, 0,
3384                         optname == PACKET_TX_RING);
3385         }
3386         case PACKET_COPY_THRESH:
3387         {
3388                 int val;
3389
3390                 if (optlen != sizeof(val))
3391                         return -EINVAL;
3392                 if (copy_from_user(&val, optval, sizeof(val)))
3393                         return -EFAULT;
3394
3395                 pkt_sk(sk)->copy_thresh = val;
3396                 return 0;
3397         }
3398         case PACKET_VERSION:
3399         {
3400                 int val;
3401
3402                 if (optlen != sizeof(val))
3403                         return -EINVAL;
3404                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3405                         return -EBUSY;
3406                 if (copy_from_user(&val, optval, sizeof(val)))
3407                         return -EFAULT;
3408                 switch (val) {
3409                 case TPACKET_V1:
3410                 case TPACKET_V2:
3411                 case TPACKET_V3:
3412                         po->tp_version = val;
3413                         return 0;
3414                 default:
3415                         return -EINVAL;
3416                 }
3417         }
3418         case PACKET_RESERVE:
3419         {
3420                 unsigned int val;
3421
3422                 if (optlen != sizeof(val))
3423                         return -EINVAL;
3424                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3425                         return -EBUSY;
3426                 if (copy_from_user(&val, optval, sizeof(val)))
3427                         return -EFAULT;
3428                 po->tp_reserve = val;
3429                 return 0;
3430         }
3431         case PACKET_LOSS:
3432         {
3433                 unsigned int val;
3434
3435                 if (optlen != sizeof(val))
3436                         return -EINVAL;
3437                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3438                         return -EBUSY;
3439                 if (copy_from_user(&val, optval, sizeof(val)))
3440                         return -EFAULT;
3441                 po->tp_loss = !!val;
3442                 return 0;
3443         }
3444         case PACKET_AUXDATA:
3445         {
3446                 int val;
3447
3448                 if (optlen < sizeof(val))
3449                         return -EINVAL;
3450                 if (copy_from_user(&val, optval, sizeof(val)))
3451                         return -EFAULT;
3452
3453                 po->auxdata = !!val;
3454                 return 0;
3455         }
3456         case PACKET_ORIGDEV:
3457         {
3458                 int val;
3459
3460                 if (optlen < sizeof(val))
3461                         return -EINVAL;
3462                 if (copy_from_user(&val, optval, sizeof(val)))
3463                         return -EFAULT;
3464
3465                 po->origdev = !!val;
3466                 return 0;
3467         }
3468         case PACKET_VNET_HDR:
3469         {
3470                 int val;
3471
3472                 if (sock->type != SOCK_RAW)
3473                         return -EINVAL;
3474                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3475                         return -EBUSY;
3476                 if (optlen < sizeof(val))
3477                         return -EINVAL;
3478                 if (copy_from_user(&val, optval, sizeof(val)))
3479                         return -EFAULT;
3480
3481                 po->has_vnet_hdr = !!val;
3482                 return 0;
3483         }
3484         case PACKET_TIMESTAMP:
3485         {
3486                 int val;
3487
3488                 if (optlen != sizeof(val))
3489                         return -EINVAL;
3490                 if (copy_from_user(&val, optval, sizeof(val)))
3491                         return -EFAULT;
3492
3493                 po->tp_tstamp = val;
3494                 return 0;
3495         }
3496         case PACKET_FANOUT:
3497         {
3498                 int val;
3499
3500                 if (optlen != sizeof(val))
3501                         return -EINVAL;
3502                 if (copy_from_user(&val, optval, sizeof(val)))
3503                         return -EFAULT;
3504
3505                 return fanout_add(sk, val & 0xffff, val >> 16);
3506         }
3507         case PACKET_TX_HAS_OFF:
3508         {
3509                 unsigned int val;
3510
3511                 if (optlen != sizeof(val))
3512                         return -EINVAL;
3513                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3514                         return -EBUSY;
3515                 if (copy_from_user(&val, optval, sizeof(val)))
3516                         return -EFAULT;
3517                 po->tp_tx_has_off = !!val;
3518                 return 0;
3519         }
3520         case PACKET_QDISC_BYPASS:
3521         {
3522                 int val;
3523
3524                 if (optlen != sizeof(val))
3525                         return -EINVAL;
3526                 if (copy_from_user(&val, optval, sizeof(val)))
3527                         return -EFAULT;
3528
3529                 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3530                 return 0;
3531         }
3532         default:
3533                 return -ENOPROTOOPT;
3534         }
3535 }
3536
3537 static int packet_getsockopt(struct socket *sock, int level, int optname,
3538                              char __user *optval, int __user *optlen)
3539 {
3540         int len;
3541         int val, lv = sizeof(val);
3542         struct sock *sk = sock->sk;
3543         struct packet_sock *po = pkt_sk(sk);
3544         void *data = &val;
3545         union tpacket_stats_u st;
3546
3547         if (level != SOL_PACKET)
3548                 return -ENOPROTOOPT;
3549
3550         if (get_user(len, optlen))
3551                 return -EFAULT;
3552
3553         if (len < 0)
3554                 return -EINVAL;
3555
3556         switch (optname) {
3557         case PACKET_STATISTICS:
3558                 spin_lock_bh(&sk->sk_receive_queue.lock);
3559                 memcpy(&st, &po->stats, sizeof(st));
3560                 memset(&po->stats, 0, sizeof(po->stats));
3561                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3562
3563                 if (po->tp_version == TPACKET_V3) {
3564                         lv = sizeof(struct tpacket_stats_v3);
3565                         st.stats3.tp_packets += st.stats3.tp_drops;
3566                         data = &st.stats3;
3567                 } else {
3568                         lv = sizeof(struct tpacket_stats);
3569                         st.stats1.tp_packets += st.stats1.tp_drops;
3570                         data = &st.stats1;
3571                 }
3572
3573                 break;
3574         case PACKET_AUXDATA:
3575                 val = po->auxdata;
3576                 break;
3577         case PACKET_ORIGDEV:
3578                 val = po->origdev;
3579                 break;
3580         case PACKET_VNET_HDR:
3581                 val = po->has_vnet_hdr;
3582                 break;
3583         case PACKET_VERSION:
3584                 val = po->tp_version;
3585                 break;
3586         case PACKET_HDRLEN:
3587                 if (len > sizeof(int))
3588                         len = sizeof(int);
3589                 if (copy_from_user(&val, optval, len))
3590                         return -EFAULT;
3591                 switch (val) {
3592                 case TPACKET_V1:
3593                         val = sizeof(struct tpacket_hdr);
3594                         break;
3595                 case TPACKET_V2:
3596                         val = sizeof(struct tpacket2_hdr);
3597                         break;
3598                 case TPACKET_V3:
3599                         val = sizeof(struct tpacket3_hdr);
3600                         break;
3601                 default:
3602                         return -EINVAL;
3603                 }
3604                 break;
3605         case PACKET_RESERVE:
3606                 val = po->tp_reserve;
3607                 break;
3608         case PACKET_LOSS:
3609                 val = po->tp_loss;
3610                 break;
3611         case PACKET_TIMESTAMP:
3612                 val = po->tp_tstamp;
3613                 break;
3614         case PACKET_FANOUT:
3615                 val = (po->fanout ?
3616                        ((u32)po->fanout->id |
3617                         ((u32)po->fanout->type << 16) |
3618                         ((u32)po->fanout->flags << 24)) :
3619                        0);
3620                 break;
3621         case PACKET_TX_HAS_OFF:
3622                 val = po->tp_tx_has_off;
3623                 break;
3624         case PACKET_QDISC_BYPASS:
3625                 val = packet_use_direct_xmit(po);
3626                 break;
3627         default:
3628                 return -ENOPROTOOPT;
3629         }
3630
3631         if (len > lv)
3632                 len = lv;
3633         if (put_user(len, optlen))
3634                 return -EFAULT;
3635         if (copy_to_user(optval, data, len))
3636                 return -EFAULT;
3637         return 0;
3638 }
3639
3640
3641 static int packet_notifier(struct notifier_block *this,
3642                            unsigned long msg, void *ptr)
3643 {
3644         struct sock *sk;
3645         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3646         struct net *net = dev_net(dev);
3647
3648         rcu_read_lock();
3649         sk_for_each_rcu(sk, &net->packet.sklist) {
3650                 struct packet_sock *po = pkt_sk(sk);
3651
3652                 switch (msg) {
3653                 case NETDEV_UNREGISTER:
3654                         if (po->mclist)
3655                                 packet_dev_mclist_delete(dev, &po->mclist);
3656                         /* fallthrough */
3657
3658                 case NETDEV_DOWN:
3659                         if (dev->ifindex == po->ifindex) {
3660                                 spin_lock(&po->bind_lock);
3661                                 if (po->running) {
3662                                         __unregister_prot_hook(sk, false);
3663                                         sk->sk_err = ENETDOWN;
3664                                         if (!sock_flag(sk, SOCK_DEAD))
3665                                                 sk->sk_error_report(sk);
3666                                 }
3667                                 if (msg == NETDEV_UNREGISTER) {
3668                                         packet_cached_dev_reset(po);
3669                                         po->ifindex = -1;
3670                                         if (po->prot_hook.dev)
3671                                                 dev_put(po->prot_hook.dev);
3672                                         po->prot_hook.dev = NULL;
3673                                 }
3674                                 spin_unlock(&po->bind_lock);
3675                         }
3676                         break;
3677                 case NETDEV_UP:
3678                         if (dev->ifindex == po->ifindex) {
3679                                 spin_lock(&po->bind_lock);
3680                                 if (po->num)
3681                                         register_prot_hook(sk);
3682                                 spin_unlock(&po->bind_lock);
3683                         }
3684                         break;
3685                 }
3686         }
3687         rcu_read_unlock();
3688         return NOTIFY_DONE;
3689 }
3690
3691
3692 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3693                         unsigned long arg)
3694 {
3695         struct sock *sk = sock->sk;
3696
3697         switch (cmd) {
3698         case SIOCOUTQ:
3699         {
3700                 int amount = sk_wmem_alloc_get(sk);
3701
3702                 return put_user(amount, (int __user *)arg);
3703         }
3704         case SIOCINQ:
3705         {
3706                 struct sk_buff *skb;
3707                 int amount = 0;
3708
3709                 spin_lock_bh(&sk->sk_receive_queue.lock);
3710                 skb = skb_peek(&sk->sk_receive_queue);
3711                 if (skb)
3712                         amount = skb->len;
3713                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3714                 return put_user(amount, (int __user *)arg);
3715         }
3716         case SIOCGSTAMP:
3717                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3718         case SIOCGSTAMPNS:
3719                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3720
3721 #ifdef CONFIG_INET
3722         case SIOCADDRT:
3723         case SIOCDELRT:
3724         case SIOCDARP:
3725         case SIOCGARP:
3726         case SIOCSARP:
3727         case SIOCGIFADDR:
3728         case SIOCSIFADDR:
3729         case SIOCGIFBRDADDR:
3730         case SIOCSIFBRDADDR:
3731         case SIOCGIFNETMASK:
3732         case SIOCSIFNETMASK:
3733         case SIOCGIFDSTADDR:
3734         case SIOCSIFDSTADDR:
3735         case SIOCSIFFLAGS:
3736                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3737 #endif
3738
3739         default:
3740                 return -ENOIOCTLCMD;
3741         }
3742         return 0;
3743 }
3744
3745 static unsigned int packet_poll(struct file *file, struct socket *sock,
3746                                 poll_table *wait)
3747 {
3748         struct sock *sk = sock->sk;
3749         struct packet_sock *po = pkt_sk(sk);
3750         unsigned int mask = datagram_poll(file, sock, wait);
3751
3752         spin_lock_bh(&sk->sk_receive_queue.lock);
3753         if (po->rx_ring.pg_vec) {
3754                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3755                         TP_STATUS_KERNEL))
3756                         mask |= POLLIN | POLLRDNORM;
3757         }
3758         spin_unlock_bh(&sk->sk_receive_queue.lock);
3759         spin_lock_bh(&sk->sk_write_queue.lock);
3760         if (po->tx_ring.pg_vec) {
3761                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3762                         mask |= POLLOUT | POLLWRNORM;
3763         }
3764         spin_unlock_bh(&sk->sk_write_queue.lock);
3765         return mask;
3766 }
3767
3768
3769 /* Dirty? Well, I still did not learn better way to account
3770  * for user mmaps.
3771  */
3772
3773 static void packet_mm_open(struct vm_area_struct *vma)
3774 {
3775         struct file *file = vma->vm_file;
3776         struct socket *sock = file->private_data;
3777         struct sock *sk = sock->sk;
3778
3779         if (sk)
3780                 atomic_inc(&pkt_sk(sk)->mapped);
3781 }
3782
3783 static void packet_mm_close(struct vm_area_struct *vma)
3784 {
3785         struct file *file = vma->vm_file;
3786         struct socket *sock = file->private_data;
3787         struct sock *sk = sock->sk;
3788
3789         if (sk)
3790                 atomic_dec(&pkt_sk(sk)->mapped);
3791 }
3792
3793 static const struct vm_operations_struct packet_mmap_ops = {
3794         .open   =       packet_mm_open,
3795         .close  =       packet_mm_close,
3796 };
3797
3798 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3799                         unsigned int len)
3800 {
3801         int i;
3802
3803         for (i = 0; i < len; i++) {
3804                 if (likely(pg_vec[i].buffer)) {
3805                         if (is_vmalloc_addr(pg_vec[i].buffer))
3806                                 vfree(pg_vec[i].buffer);
3807                         else
3808                                 free_pages((unsigned long)pg_vec[i].buffer,
3809                                            order);
3810                         pg_vec[i].buffer = NULL;
3811                 }
3812         }
3813         kfree(pg_vec);
3814 }
3815
3816 static char *alloc_one_pg_vec_page(unsigned long order)
3817 {
3818         char *buffer;
3819         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3820                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3821
3822         buffer = (char *) __get_free_pages(gfp_flags, order);
3823         if (buffer)
3824                 return buffer;
3825
3826         /* __get_free_pages failed, fall back to vmalloc */
3827         buffer = vzalloc((1 << order) * PAGE_SIZE);
3828         if (buffer)
3829                 return buffer;
3830
3831         /* vmalloc failed, lets dig into swap here */
3832         gfp_flags &= ~__GFP_NORETRY;
3833         buffer = (char *) __get_free_pages(gfp_flags, order);
3834         if (buffer)
3835                 return buffer;
3836
3837         /* complete and utter failure */
3838         return NULL;
3839 }
3840
3841 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3842 {
3843         unsigned int block_nr = req->tp_block_nr;
3844         struct pgv *pg_vec;
3845         int i;
3846
3847         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3848         if (unlikely(!pg_vec))
3849                 goto out;
3850
3851         for (i = 0; i < block_nr; i++) {
3852                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3853                 if (unlikely(!pg_vec[i].buffer))
3854                         goto out_free_pgvec;
3855         }
3856
3857 out:
3858         return pg_vec;
3859
3860 out_free_pgvec:
3861         free_pg_vec(pg_vec, order, block_nr);
3862         pg_vec = NULL;
3863         goto out;
3864 }
3865
3866 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3867                 int closing, int tx_ring)
3868 {
3869         struct pgv *pg_vec = NULL;
3870         struct packet_sock *po = pkt_sk(sk);
3871         int was_running, order = 0;
3872         struct packet_ring_buffer *rb;
3873         struct sk_buff_head *rb_queue;
3874         __be16 num;
3875         int err = -EINVAL;
3876         /* Added to avoid minimal code churn */
3877         struct tpacket_req *req = &req_u->req;
3878
3879         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3880         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3881                 WARN(1, "Tx-ring is not supported.\n");
3882                 goto out;
3883         }
3884
3885         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3886         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3887
3888         err = -EBUSY;
3889         if (!closing) {
3890                 if (atomic_read(&po->mapped))
3891                         goto out;
3892                 if (packet_read_pending(rb))
3893                         goto out;
3894         }
3895
3896         if (req->tp_block_nr) {
3897                 /* Sanity tests and some calculations */
3898                 err = -EBUSY;
3899                 if (unlikely(rb->pg_vec))
3900                         goto out;
3901
3902                 switch (po->tp_version) {
3903                 case TPACKET_V1:
3904                         po->tp_hdrlen = TPACKET_HDRLEN;
3905                         break;
3906                 case TPACKET_V2:
3907                         po->tp_hdrlen = TPACKET2_HDRLEN;
3908                         break;
3909                 case TPACKET_V3:
3910                         po->tp_hdrlen = TPACKET3_HDRLEN;
3911                         break;
3912                 }
3913
3914                 err = -EINVAL;
3915                 if (unlikely((int)req->tp_block_size <= 0))
3916                         goto out;
3917                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3918                         goto out;
3919                 if (po->tp_version >= TPACKET_V3 &&
3920                     (int)(req->tp_block_size -
3921                           BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
3922                         goto out;
3923                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3924                                         po->tp_reserve))
3925                         goto out;
3926                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3927                         goto out;
3928
3929                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3930                 if (unlikely(rb->frames_per_block <= 0))
3931                         goto out;
3932                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3933                                         req->tp_frame_nr))
3934                         goto out;
3935
3936                 err = -ENOMEM;
3937                 order = get_order(req->tp_block_size);
3938                 pg_vec = alloc_pg_vec(req, order);
3939                 if (unlikely(!pg_vec))
3940                         goto out;
3941                 switch (po->tp_version) {
3942                 case TPACKET_V3:
3943                 /* Transmit path is not supported. We checked
3944                  * it above but just being paranoid
3945                  */
3946                         if (!tx_ring)
3947                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3948                         break;
3949                 default:
3950                         break;
3951                 }
3952         }
3953         /* Done */
3954         else {
3955                 err = -EINVAL;
3956                 if (unlikely(req->tp_frame_nr))
3957                         goto out;
3958         }
3959
3960         lock_sock(sk);
3961
3962         /* Detach socket from network */
3963         spin_lock(&po->bind_lock);
3964         was_running = po->running;
3965         num = po->num;
3966         if (was_running) {
3967                 po->num = 0;
3968                 __unregister_prot_hook(sk, false);
3969         }
3970         spin_unlock(&po->bind_lock);
3971
3972         synchronize_net();
3973
3974         err = -EBUSY;
3975         mutex_lock(&po->pg_vec_lock);
3976         if (closing || atomic_read(&po->mapped) == 0) {
3977                 err = 0;
3978                 spin_lock_bh(&rb_queue->lock);
3979                 swap(rb->pg_vec, pg_vec);
3980                 rb->frame_max = (req->tp_frame_nr - 1);
3981                 rb->head = 0;
3982                 rb->frame_size = req->tp_frame_size;
3983                 spin_unlock_bh(&rb_queue->lock);
3984
3985                 swap(rb->pg_vec_order, order);
3986                 swap(rb->pg_vec_len, req->tp_block_nr);
3987
3988                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3989                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3990                                                 tpacket_rcv : packet_rcv;
3991                 skb_queue_purge(rb_queue);
3992                 if (atomic_read(&po->mapped))
3993                         pr_err("packet_mmap: vma is busy: %d\n",
3994                                atomic_read(&po->mapped));
3995         }
3996         mutex_unlock(&po->pg_vec_lock);
3997
3998         spin_lock(&po->bind_lock);
3999         if (was_running) {
4000                 po->num = num;
4001                 register_prot_hook(sk);
4002         }
4003         spin_unlock(&po->bind_lock);
4004         if (closing && (po->tp_version > TPACKET_V2)) {
4005                 /* Because we don't support block-based V3 on tx-ring */
4006                 if (!tx_ring)
4007                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
4008         }
4009         release_sock(sk);
4010
4011         if (pg_vec)
4012                 free_pg_vec(pg_vec, order, req->tp_block_nr);
4013 out:
4014         return err;
4015 }
4016
4017 static int packet_mmap(struct file *file, struct socket *sock,
4018                 struct vm_area_struct *vma)
4019 {
4020         struct sock *sk = sock->sk;
4021         struct packet_sock *po = pkt_sk(sk);
4022         unsigned long size, expected_size;
4023         struct packet_ring_buffer *rb;
4024         unsigned long start;
4025         int err = -EINVAL;
4026         int i;
4027
4028         if (vma->vm_pgoff)
4029                 return -EINVAL;
4030
4031         mutex_lock(&po->pg_vec_lock);
4032
4033         expected_size = 0;
4034         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4035                 if (rb->pg_vec) {
4036                         expected_size += rb->pg_vec_len
4037                                                 * rb->pg_vec_pages
4038                                                 * PAGE_SIZE;
4039                 }
4040         }
4041
4042         if (expected_size == 0)
4043                 goto out;
4044
4045         size = vma->vm_end - vma->vm_start;
4046         if (size != expected_size)
4047                 goto out;
4048
4049         start = vma->vm_start;
4050         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4051                 if (rb->pg_vec == NULL)
4052                         continue;
4053
4054                 for (i = 0; i < rb->pg_vec_len; i++) {
4055                         struct page *page;
4056                         void *kaddr = rb->pg_vec[i].buffer;
4057                         int pg_num;
4058
4059                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4060                                 page = pgv_to_page(kaddr);
4061                                 err = vm_insert_page(vma, start, page);
4062                                 if (unlikely(err))
4063                                         goto out;
4064                                 start += PAGE_SIZE;
4065                                 kaddr += PAGE_SIZE;
4066                         }
4067                 }
4068         }
4069
4070         atomic_inc(&po->mapped);
4071         vma->vm_ops = &packet_mmap_ops;
4072         err = 0;
4073
4074 out:
4075         mutex_unlock(&po->pg_vec_lock);
4076         return err;
4077 }
4078
4079 static const struct proto_ops packet_ops_spkt = {
4080         .family =       PF_PACKET,
4081         .owner =        THIS_MODULE,
4082         .release =      packet_release,
4083         .bind =         packet_bind_spkt,
4084         .connect =      sock_no_connect,
4085         .socketpair =   sock_no_socketpair,
4086         .accept =       sock_no_accept,
4087         .getname =      packet_getname_spkt,
4088         .poll =         datagram_poll,
4089         .ioctl =        packet_ioctl,
4090         .listen =       sock_no_listen,
4091         .shutdown =     sock_no_shutdown,
4092         .setsockopt =   sock_no_setsockopt,
4093         .getsockopt =   sock_no_getsockopt,
4094         .sendmsg =      packet_sendmsg_spkt,
4095         .recvmsg =      packet_recvmsg,
4096         .mmap =         sock_no_mmap,
4097         .sendpage =     sock_no_sendpage,
4098 };
4099
4100 static const struct proto_ops packet_ops = {
4101         .family =       PF_PACKET,
4102         .owner =        THIS_MODULE,
4103         .release =      packet_release,
4104         .bind =         packet_bind,
4105         .connect =      sock_no_connect,
4106         .socketpair =   sock_no_socketpair,
4107         .accept =       sock_no_accept,
4108         .getname =      packet_getname,
4109         .poll =         packet_poll,
4110         .ioctl =        packet_ioctl,
4111         .listen =       sock_no_listen,
4112         .shutdown =     sock_no_shutdown,
4113         .setsockopt =   packet_setsockopt,
4114         .getsockopt =   packet_getsockopt,
4115         .sendmsg =      packet_sendmsg,
4116         .recvmsg =      packet_recvmsg,
4117         .mmap =         packet_mmap,
4118         .sendpage =     sock_no_sendpage,
4119 };
4120
4121 static const struct net_proto_family packet_family_ops = {
4122         .family =       PF_PACKET,
4123         .create =       packet_create,
4124         .owner  =       THIS_MODULE,
4125 };
4126
4127 static struct notifier_block packet_netdev_notifier = {
4128         .notifier_call =        packet_notifier,
4129 };
4130
4131 #ifdef CONFIG_PROC_FS
4132
4133 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4134         __acquires(RCU)
4135 {
4136         struct net *net = seq_file_net(seq);
4137
4138         rcu_read_lock();
4139         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4140 }
4141
4142 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4143 {
4144         struct net *net = seq_file_net(seq);
4145         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4146 }
4147
4148 static void packet_seq_stop(struct seq_file *seq, void *v)
4149         __releases(RCU)
4150 {
4151         rcu_read_unlock();
4152 }
4153
4154 static int packet_seq_show(struct seq_file *seq, void *v)
4155 {
4156         if (v == SEQ_START_TOKEN)
4157                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
4158         else {
4159                 struct sock *s = sk_entry(v);
4160                 const struct packet_sock *po = pkt_sk(s);
4161
4162                 seq_printf(seq,
4163                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4164                            s,
4165                            atomic_read(&s->sk_refcnt),
4166                            s->sk_type,
4167                            ntohs(po->num),
4168                            po->ifindex,
4169                            po->running,
4170                            atomic_read(&s->sk_rmem_alloc),
4171                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4172                            sock_i_ino(s));
4173         }
4174
4175         return 0;
4176 }
4177
4178 static const struct seq_operations packet_seq_ops = {
4179         .start  = packet_seq_start,
4180         .next   = packet_seq_next,
4181         .stop   = packet_seq_stop,
4182         .show   = packet_seq_show,
4183 };
4184
4185 static int packet_seq_open(struct inode *inode, struct file *file)
4186 {
4187         return seq_open_net(inode, file, &packet_seq_ops,
4188                             sizeof(struct seq_net_private));
4189 }
4190
4191 static const struct file_operations packet_seq_fops = {
4192         .owner          = THIS_MODULE,
4193         .open           = packet_seq_open,
4194         .read           = seq_read,
4195         .llseek         = seq_lseek,
4196         .release        = seq_release_net,
4197 };
4198
4199 #endif
4200
4201 static int __net_init packet_net_init(struct net *net)
4202 {
4203         mutex_init(&net->packet.sklist_lock);
4204         INIT_HLIST_HEAD(&net->packet.sklist);
4205
4206         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4207                 return -ENOMEM;
4208
4209         return 0;
4210 }
4211
4212 static void __net_exit packet_net_exit(struct net *net)
4213 {
4214         remove_proc_entry("packet", net->proc_net);
4215 }
4216
4217 static struct pernet_operations packet_net_ops = {
4218         .init = packet_net_init,
4219         .exit = packet_net_exit,
4220 };
4221
4222
4223 static void __exit packet_exit(void)
4224 {
4225         unregister_netdevice_notifier(&packet_netdev_notifier);
4226         unregister_pernet_subsys(&packet_net_ops);
4227         sock_unregister(PF_PACKET);
4228         proto_unregister(&packet_proto);
4229 }
4230
4231 static int __init packet_init(void)
4232 {
4233         int rc = proto_register(&packet_proto, 0);
4234
4235         if (rc != 0)
4236                 goto out;
4237
4238         sock_register(&packet_family_ops);
4239         register_pernet_subsys(&packet_net_ops);
4240         register_netdevice_notifier(&packet_netdev_notifier);
4241 out:
4242         return rc;
4243 }
4244
4245 module_init(packet_init);
4246 module_exit(packet_exit);
4247 MODULE_LICENSE("GPL");
4248 MODULE_ALIAS_NETPROTO(PF_PACKET);