net/packet/af_packet.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              PACKET - implements raw packet sockets.
   7  *
   8  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  11  *
  12  * Fixes:
  13  *              Alan Cox        :       verify_area() now used correctly
  14  *              Alan Cox        :       new skbuff lists, look ma no backlogs!
  15  *              Alan Cox        :       tidied skbuff lists.
  16  *              Alan Cox        :       Now uses generic datagram routines I
  17  *                                      added. Also fixed the peek/read crash
  18  *                                      from all old Linux datagram code.
  19  *              Alan Cox        :       Uses the improved datagram code.
  20  *              Alan Cox        :       Added NULL's for socket options.
  21  *              Alan Cox        :       Re-commented the code.
  22  *              Alan Cox        :       Use new kernel side addressing
  23  *              Rob Janssen     :       Correct MTU usage.
  24  *              Dave Platt      :       Counter leaks caused by incorrect
  25  *                                      interrupt locking and some slightly
  26  *                                      dubious gcc output. Can you read
  27  *                                      compiler: it said _VOLATILE_
  28  *      Richard Kooijman        :       Timestamp fixes.
  29  *              Alan Cox        :       New buffers. Use sk->mac.raw.
  30  *              Alan Cox        :       sendmsg/recvmsg support.
  31  *              Alan Cox        :       Protocol setting support
  32  *      Alexey Kuznetsov        :       Untied from IPv4 stack.
  33  *      Cyrus Durgin            :       Fixed kerneld for kmod.
  34  *      Michal Ostrowski        :       Module initialization cleanup.
  35  *         Ulises Alonso        :       Frame number limit removal and
  36  *                                      packet_set_ring memory leak.
  37  *              Eric Biederman  :       Allow for > 8 byte hardware addresses.
  38  *                                      The convention is that longer addresses
  39  *                                      will simply extend the hardware address
  40  *                                      byte arrays at the end of sockaddr_ll
  41  *                                      and packet_mreq.
  42  *              Johann Baudy    :       Added TX RING.
  43  *              Chetan Loke     :       Implemented TPACKET_V3 block abstraction
  44  *                                      layer.
  45  *                                      Copyright (C) 2011, <lokec@ccs.neu.edu>
  46  *
  47  *
  48  *              This program is free software; you can redistribute it and/or
  49  *              modify it under the terms of the GNU General Public License
  50  *              as published by the Free Software Foundation; either version
  51  *              2 of the License, or (at your option) any later version.
  52  *
  53  */
  54
  55 #include <linux/types.h>
  56 #include <linux/mm.h>
  57 #include <linux/capability.h>
  58 #include <linux/fcntl.h>
  59 #include <linux/socket.h>
  60 #include <linux/in.h>
  61 #include <linux/inet.h>
  62 #include <linux/netdevice.h>
  63 #include <linux/if_packet.h>
  64 #include <linux/wireless.h>
  65 #include <linux/kernel.h>
  66 #include <linux/kmod.h>
  67 #include <linux/slab.h>
  68 #include <linux/vmalloc.h>
  69 #include <net/net_namespace.h>
  70 #include <net/ip.h>
  71 #include <net/protocol.h>
  72 #include <linux/skbuff.h>
  73 #include <net/sock.h>
  74 #include <linux/errno.h>
  75 #include <linux/timer.h>
  76 #include <asm/uaccess.h>
  77 #include <asm/ioctls.h>
  78 #include <asm/page.h>
  79 #include <asm/cacheflush.h>
  80 #include <asm/io.h>
  81 #include <linux/proc_fs.h>
  82 #include <linux/seq_file.h>
  83 #include <linux/poll.h>
  84 #include <linux/module.h>
  85 #include <linux/init.h>
  86 #include <linux/mutex.h>
  87 #include <linux/if_vlan.h>
  88 #include <linux/virtio_net.h>
  89 #include <linux/errqueue.h>
  90 #include <linux/net_tstamp.h>
  91 #include <linux/percpu.h>
  92 #ifdef CONFIG_INET
  93 #include <net/inet_common.h>
  94 #endif
  95 #include <linux/bpf.h>
  96
  97 #include "internal.h"
  98
  99 /*
 100    Assumptions:
 101    - if device has no dev->hard_header routine, it adds and removes ll header
 102      inside itself. In this case ll header is invisible outside of device,
 103      but higher levels still should reserve dev->hard_header_len.
 104      Some devices are enough clever to reallocate skb, when header
 105      will not fit to reserved space (tunnel), another ones are silly
 106      (PPP).
 107    - packet socket receives packets with pulled ll header,
 108      so that SOCK_RAW should push it back.
 109
 110 On receive:
 111 -----------
 112
 113 Incoming, dev->hard_header!=NULL
 114    mac_header -> ll header
 115    data       -> data
 116
 117 Outgoing, dev->hard_header!=NULL
 118    mac_header -> ll header
 119    data       -> ll header
 120
 121 Incoming, dev->hard_header==NULL
 122    mac_header -> UNKNOWN position. It is very likely, that it points to ll
 123                  header.  PPP makes it, that is wrong, because introduce
 124                  assymetry between rx and tx paths.
 125    data       -> data
 126
 127 Outgoing, dev->hard_header==NULL
 128    mac_header -> data. ll header is still not built!
 129    data       -> data
 130
 131 Resume
 132   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 133
 134
 135 On transmit:
 136 ------------
 137
 138 dev->hard_header != NULL
 139    mac_header -> ll header
 140    data       -> ll header
 141
 142 dev->hard_header == NULL (ll header is added by device, we cannot control it)
 143    mac_header -> data
 144    data       -> data
 145
 146    We should set nh.raw on output to correct posistion,
 147    packet classifier depends on it.
 148  */
 149
 150 /* Private packet socket structures. */
 151
 152 /* identical to struct packet_mreq except it has
 153  * a longer address field.
 154  */
 155 struct packet_mreq_max {
 156         int             mr_ifindex;
 157         unsigned short  mr_type;
 158         unsigned short  mr_alen;
 159         unsigned char   mr_address[MAX_ADDR_LEN];
 160 };
 161
 162 union tpacket_uhdr {
 163         struct tpacket_hdr  *h1;
 164         struct tpacket2_hdr *h2;
 165         struct tpacket3_hdr *h3;
 166         void *raw;
 167 };
 168
 169 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 170                 int closing, int tx_ring);
 171
 172 #define V3_ALIGNMENT    (8)
 173
 174 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 175
 176 #define BLK_PLUS_PRIV(sz_of_priv) \
 177         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 178
 179 #define PGV_FROM_VMALLOC 1
 180
 181 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
 182 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
 183 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
 184 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
 185 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
 186 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
 187 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
 188
 189 struct packet_sock;
 190 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
 191 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 192                        struct packet_type *pt, struct net_device *orig_dev);
 193
 194 static void *packet_previous_frame(struct packet_sock *po,
 195                 struct packet_ring_buffer *rb,
 196                 int status);
 197 static void packet_increment_head(struct packet_ring_buffer *buff);
 198 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
 199                         struct tpacket_block_desc *);
 200 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 201                         struct packet_sock *);
 202 static void prb_retire_current_block(struct tpacket_kbdq_core *,
 203                 struct packet_sock *, unsigned int status);
 204 static int prb_queue_frozen(struct tpacket_kbdq_core *);
 205 static void prb_open_block(struct tpacket_kbdq_core *,
 206                 struct tpacket_block_desc *);
 207 static void prb_retire_rx_blk_timer_expired(unsigned long);
 208 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 209 static void prb_init_blk_timer(struct packet_sock *,
 210                 struct tpacket_kbdq_core *,
 211                 void (*func) (unsigned long));
 212 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 213 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 214                 struct tpacket3_hdr *);
 215 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 216                 struct tpacket3_hdr *);
 217 static void packet_flush_mclist(struct sock *sk);
 218
 219 struct packet_skb_cb {
 220         union {
 221                 struct sockaddr_pkt pkt;
 222                 union {
 223                         /* Trick: alias skb original length with
 224                          * ll.sll_family and ll.protocol in order
 225                          * to save room.
 226                          */
 227                         unsigned int origlen;
 228                         struct sockaddr_ll ll;
 229                 };
 230         } sa;
 231 };
 232
 233 #define vio_le() virtio_legacy_is_little_endian()
 234
 235 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 236
 237 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 238 #define GET_PBLOCK_DESC(x, bid) \
 239         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 240 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 241         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 242 #define GET_NEXT_PRB_BLK_NUM(x) \
 243         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 244         ((x)->kactive_blk_num+1) : 0)
 245
 246 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 247 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 248
 249 static int packet_direct_xmit(struct sk_buff *skb)
 250 {
 251         struct net_device *dev = skb->dev;
 252         netdev_features_t features;
 253         struct netdev_queue *txq;
 254         int ret = NETDEV_TX_BUSY;
 255
 256         if (unlikely(!netif_running(dev) ||
 257                      !netif_carrier_ok(dev)))
 258                 goto drop;
 259
 260         features = netif_skb_features(skb);
 261         if (skb_needs_linearize(skb, features) &&
 262             __skb_linearize(skb))
 263                 goto drop;
 264
 265         txq = skb_get_tx_queue(dev, skb);
 266
 267         local_bh_disable();
 268
 269         HARD_TX_LOCK(dev, txq, smp_processor_id());
 270         if (!netif_xmit_frozen_or_drv_stopped(txq))
 271                 ret = netdev_start_xmit(skb, dev, txq, false);
 272         HARD_TX_UNLOCK(dev, txq);
 273
 274         local_bh_enable();
 275
 276         if (!dev_xmit_complete(ret))
 277                 kfree_skb(skb);
 278
 279         return ret;
 280 drop:
 281         atomic_long_inc(&dev->tx_dropped);
 282         kfree_skb(skb);
 283         return NET_XMIT_DROP;
 284 }
 285
 286 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
 287 {
 288         struct net_device *dev;
 289
 290         rcu_read_lock();
 291         dev = rcu_dereference(po->cached_dev);
 292         if (likely(dev))
 293                 dev_hold(dev);
 294         rcu_read_unlock();
 295
 296         return dev;
 297 }
 298
 299 static void packet_cached_dev_assign(struct packet_sock *po,
 300                                      struct net_device *dev)
 301 {
 302         rcu_assign_pointer(po->cached_dev, dev);
 303 }
 304
 305 static void packet_cached_dev_reset(struct packet_sock *po)
 306 {
 307         RCU_INIT_POINTER(po->cached_dev, NULL);
 308 }
 309
 310 static bool packet_use_direct_xmit(const struct packet_sock *po)
 311 {
 312         return po->xmit == packet_direct_xmit;
 313 }
 314
 315 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
 316 {
 317         return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
 318 }
 319
 320 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
 321 {
 322         const struct net_device_ops *ops = dev->netdev_ops;
 323         u16 queue_index;
 324
 325         if (ops->ndo_select_queue) {
 326                 queue_index = ops->ndo_select_queue(dev, skb, NULL,
 327                                                     __packet_pick_tx_queue);
 328                 queue_index = netdev_cap_txqueue(dev, queue_index);
 329         } else {
 330                 queue_index = __packet_pick_tx_queue(dev, skb);
 331         }
 332
 333         skb_set_queue_mapping(skb, queue_index);
 334 }
 335
 336 /* register_prot_hook must be invoked with the po->bind_lock held,
 337  * or from a context in which asynchronous accesses to the packet
 338  * socket is not possible (packet_create()).
 339  */
 340 static void register_prot_hook(struct sock *sk)
 341 {
 342         struct packet_sock *po = pkt_sk(sk);
 343
 344         if (!po->running) {
 345                 if (po->fanout)
 346                         __fanout_link(sk, po);
 347                 else
 348                         dev_add_pack(&po->prot_hook);
 349
 350                 sock_hold(sk);
 351                 po->running = 1;
 352         }
 353 }
 354
 355 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
 356  * held.   If the sync parameter is true, we will temporarily drop
 357  * the po->bind_lock and do a synchronize_net to make sure no
 358  * asynchronous packet processing paths still refer to the elements
 359  * of po->prot_hook.  If the sync parameter is false, it is the
 360  * callers responsibility to take care of this.
 361  */
 362 static void __unregister_prot_hook(struct sock *sk, bool sync)
 363 {
 364         struct packet_sock *po = pkt_sk(sk);
 365
 366         po->running = 0;
 367
 368         if (po->fanout)
 369                 __fanout_unlink(sk, po);
 370         else
 371                 __dev_remove_pack(&po->prot_hook);
 372
 373         __sock_put(sk);
 374
 375         if (sync) {
 376                 spin_unlock(&po->bind_lock);
 377                 synchronize_net();
 378                 spin_lock(&po->bind_lock);
 379         }
 380 }
 381
 382 static void unregister_prot_hook(struct sock *sk, bool sync)
 383 {
 384         struct packet_sock *po = pkt_sk(sk);
 385
 386         if (po->running)
 387                 __unregister_prot_hook(sk, sync);
 388 }
 389
 390 static inline struct page * __pure pgv_to_page(void *addr)
 391 {
 392         if (is_vmalloc_addr(addr))
 393                 return vmalloc_to_page(addr);
 394         return virt_to_page(addr);
 395 }
 396
 397 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 398 {
 399         union tpacket_uhdr h;
 400
 401         h.raw = frame;
 402         switch (po->tp_version) {
 403         case TPACKET_V1:
 404                 h.h1->tp_status = status;
 405                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 406                 break;
 407         case TPACKET_V2:
 408                 h.h2->tp_status = status;
 409                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 410                 break;
 411         case TPACKET_V3:
 412         default:
 413                 WARN(1, "TPACKET version not supported.\n");
 414                 BUG();
 415         }
 416
 417         smp_wmb();
 418 }
 419
 420 static int __packet_get_status(struct packet_sock *po, void *frame)
 421 {
 422         union tpacket_uhdr h;
 423
 424         smp_rmb();
 425
 426         h.raw = frame;
 427         switch (po->tp_version) {
 428         case TPACKET_V1:
 429                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 430                 return h.h1->tp_status;
 431         case TPACKET_V2:
 432                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 433                 return h.h2->tp_status;
 434         case TPACKET_V3:
 435         default:
 436                 WARN(1, "TPACKET version not supported.\n");
 437                 BUG();
 438                 return 0;
 439         }
 440 }
 441
 442 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
 443                                    unsigned int flags)
 444 {
 445         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 446
 447         if (shhwtstamps &&
 448             (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
 449             ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
 450                 return TP_STATUS_TS_RAW_HARDWARE;
 451
 452         if (ktime_to_timespec_cond(skb->tstamp, ts))
 453                 return TP_STATUS_TS_SOFTWARE;
 454
 455         return 0;
 456 }
 457
 458 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
 459                                     struct sk_buff *skb)
 460 {
 461         union tpacket_uhdr h;
 462         struct timespec ts;
 463         __u32 ts_status;
 464
 465         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
 466                 return 0;
 467
 468         h.raw = frame;
 469         switch (po->tp_version) {
 470         case TPACKET_V1:
 471                 h.h1->tp_sec = ts.tv_sec;
 472                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
 473                 break;
 474         case TPACKET_V2:
 475                 h.h2->tp_sec = ts.tv_sec;
 476                 h.h2->tp_nsec = ts.tv_nsec;
 477                 break;
 478         case TPACKET_V3:
 479         default:
 480                 WARN(1, "TPACKET version not supported.\n");
 481                 BUG();
 482         }
 483
 484         /* one flush is safe, as both fields always lie on the same cacheline */
 485         flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
 486         smp_wmb();
 487
 488         return ts_status;
 489 }
 490
 491 static void *packet_lookup_frame(struct packet_sock *po,
 492                 struct packet_ring_buffer *rb,
 493                 unsigned int position,
 494                 int status)
 495 {
 496         unsigned int pg_vec_pos, frame_offset;
 497         union tpacket_uhdr h;
 498
 499         pg_vec_pos = position / rb->frames_per_block;
 500         frame_offset = position % rb->frames_per_block;
 501
 502         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 503                 (frame_offset * rb->frame_size);
 504
 505         if (status != __packet_get_status(po, h.raw))
 506                 return NULL;
 507
 508         return h.raw;
 509 }
 510
 511 static void *packet_current_frame(struct packet_sock *po,
 512                 struct packet_ring_buffer *rb,
 513                 int status)
 514 {
 515         return packet_lookup_frame(po, rb, rb->head, status);
 516 }
 517
 518 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 519 {
 520         del_timer_sync(&pkc->retire_blk_timer);
 521 }
 522
 523 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 524                 struct sk_buff_head *rb_queue)
 525 {
 526         struct tpacket_kbdq_core *pkc;
 527
 528         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 529
 530         spin_lock_bh(&rb_queue->lock);
 531         pkc->delete_blk_timer = 1;
 532         spin_unlock_bh(&rb_queue->lock);
 533
 534         prb_del_retire_blk_timer(pkc);
 535 }
 536
 537 static void prb_init_blk_timer(struct packet_sock *po,
 538                 struct tpacket_kbdq_core *pkc,
 539                 void (*func) (unsigned long))
 540 {
 541         init_timer(&pkc->retire_blk_timer);
 542         pkc->retire_blk_timer.data = (long)po;
 543         pkc->retire_blk_timer.function = func;
 544         pkc->retire_blk_timer.expires = jiffies;
 545 }
 546
 547 static void prb_setup_retire_blk_timer(struct packet_sock *po)
 548 {
 549         struct tpacket_kbdq_core *pkc;
 550
 551         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 552         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
 553 }
 554
 555 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 556                                 int blk_size_in_bytes)
 557 {
 558         struct net_device *dev;
 559         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 560         struct ethtool_cmd ecmd;
 561         int err;
 562         u32 speed;
 563
 564         rtnl_lock();
 565         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 566         if (unlikely(!dev)) {
 567                 rtnl_unlock();
 568                 return DEFAULT_PRB_RETIRE_TOV;
 569         }
 570         err = __ethtool_get_settings(dev, &ecmd);
 571         speed = ethtool_cmd_speed(&ecmd);
 572         rtnl_unlock();
 573         if (!err) {
 574                 /*
 575                  * If the link speed is so slow you don't really
 576                  * need to worry about perf anyways
 577                  */
 578                 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
 579                         return DEFAULT_PRB_RETIRE_TOV;
 580                 } else {
 581                         msec = 1;
 582                         div = speed / 1000;
 583                 }
 584         }
 585
 586         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 587
 588         if (div)
 589                 mbits /= div;
 590
 591         tmo = mbits * msec;
 592
 593         if (div)
 594                 return tmo+1;
 595         return tmo;
 596 }
 597
 598 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 599                         union tpacket_req_u *req_u)
 600 {
 601         p1->feature_req_word = req_u->req3.tp_feature_req_word;
 602 }
 603
 604 static void init_prb_bdqc(struct packet_sock *po,
 605                         struct packet_ring_buffer *rb,
 606                         struct pgv *pg_vec,
 607                         union tpacket_req_u *req_u)
 608 {
 609         struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
 610         struct tpacket_block_desc *pbd;
 611
 612         memset(p1, 0x0, sizeof(*p1));
 613
 614         p1->knxt_seq_num = 1;
 615         p1->pkbdq = pg_vec;
 616         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 617         p1->pkblk_start = pg_vec[0].buffer;
 618         p1->kblk_size = req_u->req3.tp_block_size;
 619         p1->knum_blocks = req_u->req3.tp_block_nr;
 620         p1->hdrlen = po->tp_hdrlen;
 621         p1->version = po->tp_version;
 622         p1->last_kactive_blk_num = 0;
 623         po->stats.stats3.tp_freeze_q_cnt = 0;
 624         if (req_u->req3.tp_retire_blk_tov)
 625                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 626         else
 627                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 628                                                 req_u->req3.tp_block_size);
 629         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 630         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 631
 632         p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
 633         prb_init_ft_ops(p1, req_u);
 634         prb_setup_retire_blk_timer(po);
 635         prb_open_block(p1, pbd);
 636 }
 637
 638 /*  Do NOT update the last_blk_num first.
 639  *  Assumes sk_buff_head lock is held.
 640  */
 641 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 642 {
 643         mod_timer(&pkc->retire_blk_timer,
 644                         jiffies + pkc->tov_in_jiffies);
 645         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 646 }
 647
 648 /*
 649  * Timer logic:
 650  * 1) We refresh the timer only when we open a block.
 651  *    By doing this we don't waste cycles refreshing the timer
 652  *        on packet-by-packet basis.
 653  *
 654  * With a 1MB block-size, on a 1Gbps line, it will take
 655  * i) ~8 ms to fill a block + ii) memcpy etc.
 656  * In this cut we are not accounting for the memcpy time.
 657  *
 658  * So, if the user sets the 'tmo' to 10ms then the timer
 659  * will never fire while the block is still getting filled
 660  * (which is what we want). However, the user could choose
 661  * to close a block early and that's fine.
 662  *
 663  * But when the timer does fire, we check whether or not to refresh it.
 664  * Since the tmo granularity is in msecs, it is not too expensive
 665  * to refresh the timer, lets say every '8' msecs.
 666  * Either the user can set the 'tmo' or we can derive it based on
 667  * a) line-speed and b) block-size.
 668  * prb_calc_retire_blk_tmo() calculates the tmo.
 669  *
 670  */
 671 static void prb_retire_rx_blk_timer_expired(unsigned long data)
 672 {
 673         struct packet_sock *po = (struct packet_sock *)data;
 674         struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 675         unsigned int frozen;
 676         struct tpacket_block_desc *pbd;
 677
 678         spin_lock(&po->sk.sk_receive_queue.lock);
 679
 680         frozen = prb_queue_frozen(pkc);
 681         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 682
 683         if (unlikely(pkc->delete_blk_timer))
 684                 goto out;
 685
 686         /* We only need to plug the race when the block is partially filled.
 687          * tpacket_rcv:
 688          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 689          *              copy_bits() is in progress ...
 690          *              timer fires on other cpu:
 691          *              we can't retire the current block because copy_bits
 692          *              is in progress.
 693          *
 694          */
 695         if (BLOCK_NUM_PKTS(pbd)) {
 696                 while (atomic_read(&pkc->blk_fill_in_prog)) {
 697                         /* Waiting for skb_copy_bits to finish... */
 698                         cpu_relax();
 699                 }
 700         }
 701
 702         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 703                 if (!frozen) {
 704                         if (!BLOCK_NUM_PKTS(pbd)) {
 705                                 /* An empty block. Just refresh the timer. */
 706                                 goto refresh_timer;
 707                         }
 708                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 709                         if (!prb_dispatch_next_block(pkc, po))
 710                                 goto refresh_timer;
 711                         else
 712                                 goto out;
 713                 } else {
 714                         /* Case 1. Queue was frozen because user-space was
 715                          *         lagging behind.
 716                          */
 717                         if (prb_curr_blk_in_use(pkc, pbd)) {
 718                                 /*
 719                                  * Ok, user-space is still behind.
 720                                  * So just refresh the timer.
 721                                  */
 722                                 goto refresh_timer;
 723                         } else {
 724                                /* Case 2. queue was frozen,user-space caught up,
 725                                 * now the link went idle && the timer fired.
 726                                 * We don't have a block to close.So we open this
 727                                 * block and restart the timer.
 728                                 * opening a block thaws the queue,restarts timer
 729                                 * Thawing/timer-refresh is a side effect.
 730                                 */
 731                                 prb_open_block(pkc, pbd);
 732                                 goto out;
 733                         }
 734                 }
 735         }
 736
 737 refresh_timer:
 738         _prb_refresh_rx_retire_blk_timer(pkc);
 739
 740 out:
 741         spin_unlock(&po->sk.sk_receive_queue.lock);
 742 }
 743
 744 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 745                 struct tpacket_block_desc *pbd1, __u32 status)
 746 {
 747         /* Flush everything minus the block header */
 748
 749 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 750         u8 *start, *end;
 751
 752         start = (u8 *)pbd1;
 753
 754         /* Skip the block header(we know header WILL fit in 4K) */
 755         start += PAGE_SIZE;
 756
 757         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 758         for (; start < end; start += PAGE_SIZE)
 759                 flush_dcache_page(pgv_to_page(start));
 760
 761         smp_wmb();
 762 #endif
 763
 764         /* Now update the block status. */
 765
 766         BLOCK_STATUS(pbd1) = status;
 767
 768         /* Flush the block header */
 769
 770 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 771         start = (u8 *)pbd1;
 772         flush_dcache_page(pgv_to_page(start));
 773
 774         smp_wmb();
 775 #endif
 776 }
 777
 778 /*
 779  * Side effect:
 780  *
 781  * 1) flush the block
 782  * 2) Increment active_blk_num
 783  *
 784  * Note:We DONT refresh the timer on purpose.
 785  *      Because almost always the next block will be opened.
 786  */
 787 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 788                 struct tpacket_block_desc *pbd1,
 789                 struct packet_sock *po, unsigned int stat)
 790 {
 791         __u32 status = TP_STATUS_USER | stat;
 792
 793         struct tpacket3_hdr *last_pkt;
 794         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 795         struct sock *sk = &po->sk;
 796
 797         if (po->stats.stats3.tp_drops)
 798                 status |= TP_STATUS_LOSING;
 799
 800         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 801         last_pkt->tp_next_offset = 0;
 802
 803         /* Get the ts of the last pkt */
 804         if (BLOCK_NUM_PKTS(pbd1)) {
 805                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 806                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 807         } else {
 808                 /* Ok, we tmo'd - so get the current time.
 809                  *
 810                  * It shouldn't really happen as we don't close empty
 811                  * blocks. See prb_retire_rx_blk_timer_expired().
 812                  */
 813                 struct timespec ts;
 814                 getnstimeofday(&ts);
 815                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 816                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 817         }
 818
 819         smp_wmb();
 820
 821         /* Flush the block */
 822         prb_flush_block(pkc1, pbd1, status);
 823
 824         sk->sk_data_ready(sk);
 825
 826         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 827 }
 828
 829 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 830 {
 831         pkc->reset_pending_on_curr_blk = 0;
 832 }
 833
 834 /*
 835  * Side effect of opening a block:
 836  *
 837  * 1) prb_queue is thawed.
 838  * 2) retire_blk_timer is refreshed.
 839  *
 840  */
 841 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 842         struct tpacket_block_desc *pbd1)
 843 {
 844         struct timespec ts;
 845         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 846
 847         smp_rmb();
 848
 849         /* We could have just memset this but we will lose the
 850          * flexibility of making the priv area sticky
 851          */
 852
 853         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 854         BLOCK_NUM_PKTS(pbd1) = 0;
 855         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 856
 857         getnstimeofday(&ts);
 858
 859         h1->ts_first_pkt.ts_sec = ts.tv_sec;
 860         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 861
 862         pkc1->pkblk_start = (char *)pbd1;
 863         pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 864
 865         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 866         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 867
 868         pbd1->version = pkc1->version;
 869         pkc1->prev = pkc1->nxt_offset;
 870         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 871
 872         prb_thaw_queue(pkc1);
 873         _prb_refresh_rx_retire_blk_timer(pkc1);
 874
 875         smp_wmb();
 876 }
 877
 878 /*
 879  * Queue freeze logic:
 880  * 1) Assume tp_block_nr = 8 blocks.
 881  * 2) At time 't0', user opens Rx ring.
 882  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 883  * 4) user-space is either sleeping or processing block '0'.
 884  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 885  *    it will close block-7,loop around and try to fill block '0'.
 886  *    call-flow:
 887  *    __packet_lookup_frame_in_block
 888  *      prb_retire_current_block()
 889  *      prb_dispatch_next_block()
 890  *        |->(BLOCK_STATUS == USER) evaluates to true
 891  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 892  * 6) Now there are two cases:
 893  *    6.1) Link goes idle right after the queue is frozen.
 894  *         But remember, the last open_block() refreshed the timer.
 895  *         When this timer expires,it will refresh itself so that we can
 896  *         re-open block-0 in near future.
 897  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 898  *         case and __packet_lookup_frame_in_block will check if block-0
 899  *         is free and can now be re-used.
 900  */
 901 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 902                                   struct packet_sock *po)
 903 {
 904         pkc->reset_pending_on_curr_blk = 1;
 905         po->stats.stats3.tp_freeze_q_cnt++;
 906 }
 907
 908 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 909
 910 /*
 911  * If the next block is free then we will dispatch it
 912  * and return a good offset.
 913  * Else, we will freeze the queue.
 914  * So, caller must check the return value.
 915  */
 916 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 917                 struct packet_sock *po)
 918 {
 919         struct tpacket_block_desc *pbd;
 920
 921         smp_rmb();
 922
 923         /* 1. Get current block num */
 924         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 925
 926         /* 2. If this block is currently in_use then freeze the queue */
 927         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 928                 prb_freeze_queue(pkc, po);
 929                 return NULL;
 930         }
 931
 932         /*
 933          * 3.
 934          * open this block and return the offset where the first packet
 935          * needs to get stored.
 936          */
 937         prb_open_block(pkc, pbd);
 938         return (void *)pkc->nxt_offset;
 939 }
 940
 941 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 942                 struct packet_sock *po, unsigned int status)
 943 {
 944         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 945
 946         /* retire/close the current block */
 947         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 948                 /*
 949                  * Plug the case where copy_bits() is in progress on
 950                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 951                  * have space to copy the pkt in the current block and
 952                  * called prb_retire_current_block()
 953                  *
 954                  * We don't need to worry about the TMO case because
 955                  * the timer-handler already handled this case.
 956                  */
 957                 if (!(status & TP_STATUS_BLK_TMO)) {
 958                         while (atomic_read(&pkc->blk_fill_in_prog)) {
 959                                 /* Waiting for skb_copy_bits to finish... */
 960                                 cpu_relax();
 961                         }
 962                 }
 963                 prb_close_block(pkc, pbd, po, status);
 964                 return;
 965         }
 966 }
 967
 968 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
 969                                       struct tpacket_block_desc *pbd)
 970 {
 971         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 972 }
 973
 974 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 975 {
 976         return pkc->reset_pending_on_curr_blk;
 977 }
 978
 979 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 980 {
 981         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 982         atomic_dec(&pkc->blk_fill_in_prog);
 983 }
 984
 985 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 986                         struct tpacket3_hdr *ppd)
 987 {
 988         ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
 989 }
 990
 991 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 992                         struct tpacket3_hdr *ppd)
 993 {
 994         ppd->hv1.tp_rxhash = 0;
 995 }
 996
 997 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
 998                         struct tpacket3_hdr *ppd)
 999 {
1000         if (skb_vlan_tag_present(pkc->skb)) {
1001                 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1002                 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1003                 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1004         } else {
1005                 ppd->hv1.tp_vlan_tci = 0;
1006                 ppd->hv1.tp_vlan_tpid = 0;
1007                 ppd->tp_status = TP_STATUS_AVAILABLE;
1008         }
1009 }
1010
1011 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1012                         struct tpacket3_hdr *ppd)
1013 {
1014         ppd->hv1.tp_padding = 0;
1015         prb_fill_vlan_info(pkc, ppd);
1016
1017         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1018                 prb_fill_rxhash(pkc, ppd);
1019         else
1020                 prb_clear_rxhash(pkc, ppd);
1021 }
1022
1023 static void prb_fill_curr_block(char *curr,
1024                                 struct tpacket_kbdq_core *pkc,
1025                                 struct tpacket_block_desc *pbd,
1026                                 unsigned int len)
1027 {
1028         struct tpacket3_hdr *ppd;
1029
1030         ppd  = (struct tpacket3_hdr *)curr;
1031         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1032         pkc->prev = curr;
1033         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1034         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1035         BLOCK_NUM_PKTS(pbd) += 1;
1036         atomic_inc(&pkc->blk_fill_in_prog);
1037         prb_run_all_ft_ops(pkc, ppd);
1038 }
1039
1040 /* Assumes caller has the sk->rx_queue.lock */
1041 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1042                                             struct sk_buff *skb,
1043                                                 int status,
1044                                             unsigned int len
1045                                             )
1046 {
1047         struct tpacket_kbdq_core *pkc;
1048         struct tpacket_block_desc *pbd;
1049         char *curr, *end;
1050
1051         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1052         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1053
1054         /* Queue is frozen when user space is lagging behind */
1055         if (prb_queue_frozen(pkc)) {
1056                 /*
1057                  * Check if that last block which caused the queue to freeze,
1058                  * is still in_use by user-space.
1059                  */
1060                 if (prb_curr_blk_in_use(pkc, pbd)) {
1061                         /* Can't record this packet */
1062                         return NULL;
1063                 } else {
1064                         /*
1065                          * Ok, the block was released by user-space.
1066                          * Now let's open that block.
1067                          * opening a block also thaws the queue.
1068                          * Thawing is a side effect.
1069                          */
1070                         prb_open_block(pkc, pbd);
1071                 }
1072         }
1073
1074         smp_mb();
1075         curr = pkc->nxt_offset;
1076         pkc->skb = skb;
1077         end = (char *)pbd + pkc->kblk_size;
1078
1079         /* first try the current block */
1080         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1081                 prb_fill_curr_block(curr, pkc, pbd, len);
1082                 return (void *)curr;
1083         }
1084
1085         /* Ok, close the current block */
1086         prb_retire_current_block(pkc, po, 0);
1087
1088         /* Now, try to dispatch the next block */
1089         curr = (char *)prb_dispatch_next_block(pkc, po);
1090         if (curr) {
1091                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1092                 prb_fill_curr_block(curr, pkc, pbd, len);
1093                 return (void *)curr;
1094         }
1095
1096         /*
1097          * No free blocks are available.user_space hasn't caught up yet.
1098          * Queue was just frozen and now this packet will get dropped.
1099          */
1100         return NULL;
1101 }
1102
1103 static void *packet_current_rx_frame(struct packet_sock *po,
1104                                             struct sk_buff *skb,
1105                                             int status, unsigned int len)
1106 {
1107         char *curr = NULL;
1108         switch (po->tp_version) {
1109         case TPACKET_V1:
1110         case TPACKET_V2:
1111                 curr = packet_lookup_frame(po, &po->rx_ring,
1112                                         po->rx_ring.head, status);
1113                 return curr;
1114         case TPACKET_V3:
1115                 return __packet_lookup_frame_in_block(po, skb, status, len);
1116         default:
1117                 WARN(1, "TPACKET version not supported\n");
1118                 BUG();
1119                 return NULL;
1120         }
1121 }
1122
1123 static void *prb_lookup_block(struct packet_sock *po,
1124                                      struct packet_ring_buffer *rb,
1125                                      unsigned int idx,
1126                                      int status)
1127 {
1128         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1129         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1130
1131         if (status != BLOCK_STATUS(pbd))
1132                 return NULL;
1133         return pbd;
1134 }
1135
1136 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1137 {
1138         unsigned int prev;
1139         if (rb->prb_bdqc.kactive_blk_num)
1140                 prev = rb->prb_bdqc.kactive_blk_num-1;
1141         else
1142                 prev = rb->prb_bdqc.knum_blocks-1;
1143         return prev;
1144 }
1145
1146 /* Assumes caller has held the rx_queue.lock */
1147 static void *__prb_previous_block(struct packet_sock *po,
1148                                          struct packet_ring_buffer *rb,
1149                                          int status)
1150 {
1151         unsigned int previous = prb_previous_blk_num(rb);
1152         return prb_lookup_block(po, rb, previous, status);
1153 }
1154
1155 static void *packet_previous_rx_frame(struct packet_sock *po,
1156                                              struct packet_ring_buffer *rb,
1157                                              int status)
1158 {
1159         if (po->tp_version <= TPACKET_V2)
1160                 return packet_previous_frame(po, rb, status);
1161
1162         return __prb_previous_block(po, rb, status);
1163 }
1164
1165 static void packet_increment_rx_head(struct packet_sock *po,
1166                                             struct packet_ring_buffer *rb)
1167 {
1168         switch (po->tp_version) {
1169         case TPACKET_V1:
1170         case TPACKET_V2:
1171                 return packet_increment_head(rb);
1172         case TPACKET_V3:
1173         default:
1174                 WARN(1, "TPACKET version not supported.\n");
1175                 BUG();
1176                 return;
1177         }
1178 }
1179
1180 static void *packet_previous_frame(struct packet_sock *po,
1181                 struct packet_ring_buffer *rb,
1182                 int status)
1183 {
1184         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1185         return packet_lookup_frame(po, rb, previous, status);
1186 }
1187
1188 static void packet_increment_head(struct packet_ring_buffer *buff)
1189 {
1190         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1191 }
1192
1193 static void packet_inc_pending(struct packet_ring_buffer *rb)
1194 {
1195         this_cpu_inc(*rb->pending_refcnt);
1196 }
1197
1198 static void packet_dec_pending(struct packet_ring_buffer *rb)
1199 {
1200         this_cpu_dec(*rb->pending_refcnt);
1201 }
1202
1203 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1204 {
1205         unsigned int refcnt = 0;
1206         int cpu;
1207
1208         /* We don't use pending refcount in rx_ring. */
1209         if (rb->pending_refcnt == NULL)
1210                 return 0;
1211
1212         for_each_possible_cpu(cpu)
1213                 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1214
1215         return refcnt;
1216 }
1217
1218 static int packet_alloc_pending(struct packet_sock *po)
1219 {
1220         po->rx_ring.pending_refcnt = NULL;
1221
1222         po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1223         if (unlikely(po->tx_ring.pending_refcnt == NULL))
1224                 return -ENOBUFS;
1225
1226         return 0;
1227 }
1228
1229 static void packet_free_pending(struct packet_sock *po)
1230 {
1231         free_percpu(po->tx_ring.pending_refcnt);
1232 }
1233
1234 #define ROOM_POW_OFF    2
1235 #define ROOM_NONE       0x0
1236 #define ROOM_LOW        0x1
1237 #define ROOM_NORMAL     0x2
1238
1239 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1240 {
1241         int idx, len;
1242
1243         len = po->rx_ring.frame_max + 1;
1244         idx = po->rx_ring.head;
1245         if (pow_off)
1246                 idx += len >> pow_off;
1247         if (idx >= len)
1248                 idx -= len;
1249         return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1250 }
1251
1252 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1253 {
1254         int idx, len;
1255
1256         len = po->rx_ring.prb_bdqc.knum_blocks;
1257         idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1258         if (pow_off)
1259                 idx += len >> pow_off;
1260         if (idx >= len)
1261                 idx -= len;
1262         return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1263 }
1264
1265 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1266 {
1267         struct sock *sk = &po->sk;
1268         int ret = ROOM_NONE;
1269
1270         if (po->prot_hook.func != tpacket_rcv) {
1271                 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1272                                           - (skb ? skb->truesize : 0);
1273                 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1274                         return ROOM_NORMAL;
1275                 else if (avail > 0)
1276                         return ROOM_LOW;
1277                 else
1278                         return ROOM_NONE;
1279         }
1280
1281         if (po->tp_version == TPACKET_V3) {
1282                 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1283                         ret = ROOM_NORMAL;
1284                 else if (__tpacket_v3_has_room(po, 0))
1285                         ret = ROOM_LOW;
1286         } else {
1287                 if (__tpacket_has_room(po, ROOM_POW_OFF))
1288                         ret = ROOM_NORMAL;
1289                 else if (__tpacket_has_room(po, 0))
1290                         ret = ROOM_LOW;
1291         }
1292
1293         return ret;
1294 }
1295
1296 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1297 {
1298         int ret;
1299         bool has_room;
1300
1301         spin_lock_bh(&po->sk.sk_receive_queue.lock);
1302         ret = __packet_rcv_has_room(po, skb);
1303         has_room = ret == ROOM_NORMAL;
1304         if (po->pressure == has_room)
1305                 po->pressure = !has_room;
1306         spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1307
1308         return ret;
1309 }
1310
1311 static void packet_sock_destruct(struct sock *sk)
1312 {
1313         skb_queue_purge(&sk->sk_error_queue);
1314
1315         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1316         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1317
1318         if (!sock_flag(sk, SOCK_DEAD)) {
1319                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1320                 return;
1321         }
1322
1323         sk_refcnt_debug_dec(sk);
1324 }
1325
1326 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1327 {
1328         u32 rxhash;
1329         int i, count = 0;
1330
1331         rxhash = skb_get_hash(skb);
1332         for (i = 0; i < ROLLOVER_HLEN; i++)
1333                 if (po->rollover->history[i] == rxhash)
1334                         count++;
1335
1336         po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1337         return count > (ROLLOVER_HLEN >> 1);
1338 }
1339
1340 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1341                                       struct sk_buff *skb,
1342                                       unsigned int num)
1343 {
1344         return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1345 }
1346
1347 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1348                                     struct sk_buff *skb,
1349                                     unsigned int num)
1350 {
1351         unsigned int val = atomic_inc_return(&f->rr_cur);
1352
1353         return val % num;
1354 }
1355
1356 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1357                                      struct sk_buff *skb,
1358                                      unsigned int num)
1359 {
1360         return smp_processor_id() % num;
1361 }
1362
1363 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1364                                      struct sk_buff *skb,
1365                                      unsigned int num)
1366 {
1367         return prandom_u32_max(num);
1368 }
1369
1370 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1371                                           struct sk_buff *skb,
1372                                           unsigned int idx, bool try_self,
1373                                           unsigned int num)
1374 {
1375         struct packet_sock *po, *po_next, *po_skip = NULL;
1376         unsigned int i, j, room = ROOM_NONE;
1377
1378         po = pkt_sk(f->arr[idx]);
1379
1380         if (try_self) {
1381                 room = packet_rcv_has_room(po, skb);
1382                 if (room == ROOM_NORMAL ||
1383                     (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1384                         return idx;
1385                 po_skip = po;
1386         }
1387
1388         i = j = min_t(int, po->rollover->sock, num - 1);
1389         do {
1390                 po_next = pkt_sk(f->arr[i]);
1391                 if (po_next != po_skip && !po_next->pressure &&
1392                     packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1393                         if (i != j)
1394                                 po->rollover->sock = i;
1395                         atomic_long_inc(&po->rollover->num);
1396                         if (room == ROOM_LOW)
1397                                 atomic_long_inc(&po->rollover->num_huge);
1398                         return i;
1399                 }
1400
1401                 if (++i == num)
1402                         i = 0;
1403         } while (i != j);
1404
1405         atomic_long_inc(&po->rollover->num_failed);
1406         return idx;
1407 }
1408
1409 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1410                                     struct sk_buff *skb,
1411                                     unsigned int num)
1412 {
1413         return skb_get_queue_mapping(skb) % num;
1414 }
1415
1416 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1417                                      struct sk_buff *skb,
1418                                      unsigned int num)
1419 {
1420         struct bpf_prog *prog;
1421         unsigned int ret = 0;
1422
1423         rcu_read_lock();
1424         prog = rcu_dereference(f->bpf_prog);
1425         if (prog)
1426                 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1427         rcu_read_unlock();
1428
1429         return ret;
1430 }
1431
1432 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1433 {
1434         return f->flags & (flag >> 8);
1435 }
1436
1437 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1438                              struct packet_type *pt, struct net_device *orig_dev)
1439 {
1440         struct packet_fanout *f = pt->af_packet_priv;
1441         unsigned int num = READ_ONCE(f->num_members);
1442         struct net *net = read_pnet(&f->net);
1443         struct packet_sock *po;
1444         unsigned int idx;
1445
1446         if (!net_eq(dev_net(dev), net) || !num) {
1447                 kfree_skb(skb);
1448                 return 0;
1449         }
1450
1451         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1452                 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1453                 if (!skb)
1454                         return 0;
1455         }
1456         switch (f->type) {
1457         case PACKET_FANOUT_HASH:
1458         default:
1459                 idx = fanout_demux_hash(f, skb, num);
1460                 break;
1461         case PACKET_FANOUT_LB:
1462                 idx = fanout_demux_lb(f, skb, num);
1463                 break;
1464         case PACKET_FANOUT_CPU:
1465                 idx = fanout_demux_cpu(f, skb, num);
1466                 break;
1467         case PACKET_FANOUT_RND:
1468                 idx = fanout_demux_rnd(f, skb, num);
1469                 break;
1470         case PACKET_FANOUT_QM:
1471                 idx = fanout_demux_qm(f, skb, num);
1472                 break;
1473         case PACKET_FANOUT_ROLLOVER:
1474                 idx = fanout_demux_rollover(f, skb, 0, false, num);
1475                 break;
1476         case PACKET_FANOUT_CBPF:
1477         case PACKET_FANOUT_EBPF:
1478                 idx = fanout_demux_bpf(f, skb, num);
1479                 break;
1480         }
1481
1482         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1483                 idx = fanout_demux_rollover(f, skb, idx, true, num);
1484
1485         po = pkt_sk(f->arr[idx]);
1486         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1487 }
1488
1489 DEFINE_MUTEX(fanout_mutex);
1490 EXPORT_SYMBOL_GPL(fanout_mutex);
1491 static LIST_HEAD(fanout_list);
1492
1493 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1494 {
1495         struct packet_fanout *f = po->fanout;
1496
1497         spin_lock(&f->lock);
1498         f->arr[f->num_members] = sk;
1499         smp_wmb();
1500         f->num_members++;
1501         spin_unlock(&f->lock);
1502 }
1503
1504 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1505 {
1506         struct packet_fanout *f = po->fanout;
1507         int i;
1508
1509         spin_lock(&f->lock);
1510         for (i = 0; i < f->num_members; i++) {
1511                 if (f->arr[i] == sk)
1512                         break;
1513         }
1514         BUG_ON(i >= f->num_members);
1515         f->arr[i] = f->arr[f->num_members - 1];
1516         f->num_members--;
1517         spin_unlock(&f->lock);
1518 }
1519
1520 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1521 {
1522         if (sk->sk_family != PF_PACKET)
1523                 return false;
1524
1525         return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1526 }
1527
1528 static void fanout_init_data(struct packet_fanout *f)
1529 {
1530         switch (f->type) {
1531         case PACKET_FANOUT_LB:
1532                 atomic_set(&f->rr_cur, 0);
1533                 break;
1534         case PACKET_FANOUT_CBPF:
1535         case PACKET_FANOUT_EBPF:
1536                 RCU_INIT_POINTER(f->bpf_prog, NULL);
1537                 break;
1538         }
1539 }
1540
1541 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1542 {
1543         struct bpf_prog *old;
1544
1545         spin_lock(&f->lock);
1546         old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1547         rcu_assign_pointer(f->bpf_prog, new);
1548         spin_unlock(&f->lock);
1549
1550         if (old) {
1551                 synchronize_net();
1552                 bpf_prog_destroy(old);
1553         }
1554 }
1555
1556 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1557                                 unsigned int len)
1558 {
1559         struct bpf_prog *new;
1560         struct sock_fprog fprog;
1561         int ret;
1562
1563         if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1564                 return -EPERM;
1565         if (len != sizeof(fprog))
1566                 return -EINVAL;
1567         if (copy_from_user(&fprog, data, len))
1568                 return -EFAULT;
1569
1570         ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1571         if (ret)
1572                 return ret;
1573
1574         __fanout_set_data_bpf(po->fanout, new);
1575         return 0;
1576 }
1577
1578 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1579                                 unsigned int len)
1580 {
1581         struct bpf_prog *new;
1582         u32 fd;
1583
1584         if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1585                 return -EPERM;
1586         if (len != sizeof(fd))
1587                 return -EINVAL;
1588         if (copy_from_user(&fd, data, len))
1589                 return -EFAULT;
1590
1591         new = bpf_prog_get(fd);
1592         if (IS_ERR(new))
1593                 return PTR_ERR(new);
1594         if (new->type != BPF_PROG_TYPE_SOCKET_FILTER) {
1595                 bpf_prog_put(new);
1596                 return -EINVAL;
1597         }
1598
1599         __fanout_set_data_bpf(po->fanout, new);
1600         return 0;
1601 }
1602
1603 static int fanout_set_data(struct packet_sock *po, char __user *data,
1604                            unsigned int len)
1605 {
1606         switch (po->fanout->type) {
1607         case PACKET_FANOUT_CBPF:
1608                 return fanout_set_data_cbpf(po, data, len);
1609         case PACKET_FANOUT_EBPF:
1610                 return fanout_set_data_ebpf(po, data, len);
1611         default:
1612                 return -EINVAL;
1613         };
1614 }
1615
1616 static void fanout_release_data(struct packet_fanout *f)
1617 {
1618         switch (f->type) {
1619         case PACKET_FANOUT_CBPF:
1620         case PACKET_FANOUT_EBPF:
1621                 __fanout_set_data_bpf(f, NULL);
1622         };
1623 }
1624
1625 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1626 {
1627         struct packet_sock *po = pkt_sk(sk);
1628         struct packet_fanout *f, *match;
1629         u8 type = type_flags & 0xff;
1630         u8 flags = type_flags >> 8;
1631         int err;
1632
1633         switch (type) {
1634         case PACKET_FANOUT_ROLLOVER:
1635                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1636                         return -EINVAL;
1637         case PACKET_FANOUT_HASH:
1638         case PACKET_FANOUT_LB:
1639         case PACKET_FANOUT_CPU:
1640         case PACKET_FANOUT_RND:
1641         case PACKET_FANOUT_QM:
1642         case PACKET_FANOUT_CBPF:
1643         case PACKET_FANOUT_EBPF:
1644                 break;
1645         default:
1646                 return -EINVAL;
1647         }
1648
1649         if (!po->running)
1650                 return -EINVAL;
1651
1652         if (po->fanout)
1653                 return -EALREADY;
1654
1655         if (type == PACKET_FANOUT_ROLLOVER ||
1656             (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1657                 po->rollover = kzalloc(sizeof(*po->rollover), GFP_KERNEL);
1658                 if (!po->rollover)
1659                         return -ENOMEM;
1660                 atomic_long_set(&po->rollover->num, 0);
1661                 atomic_long_set(&po->rollover->num_huge, 0);
1662                 atomic_long_set(&po->rollover->num_failed, 0);
1663         }
1664
1665         mutex_lock(&fanout_mutex);
1666         match = NULL;
1667         list_for_each_entry(f, &fanout_list, list) {
1668                 if (f->id == id &&
1669                     read_pnet(&f->net) == sock_net(sk)) {
1670                         match = f;
1671                         break;
1672                 }
1673         }
1674         err = -EINVAL;
1675         if (match && match->flags != flags)
1676                 goto out;
1677         if (!match) {
1678                 err = -ENOMEM;
1679                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1680                 if (!match)
1681                         goto out;
1682                 write_pnet(&match->net, sock_net(sk));
1683                 match->id = id;
1684                 match->type = type;
1685                 match->flags = flags;
1686                 INIT_LIST_HEAD(&match->list);
1687                 spin_lock_init(&match->lock);
1688                 atomic_set(&match->sk_ref, 0);
1689                 fanout_init_data(match);
1690                 match->prot_hook.type = po->prot_hook.type;
1691                 match->prot_hook.dev = po->prot_hook.dev;
1692                 match->prot_hook.func = packet_rcv_fanout;
1693                 match->prot_hook.af_packet_priv = match;
1694                 match->prot_hook.id_match = match_fanout_group;
1695                 dev_add_pack(&match->prot_hook);
1696                 list_add(&match->list, &fanout_list);
1697         }
1698         err = -EINVAL;
1699         if (match->type == type &&
1700             match->prot_hook.type == po->prot_hook.type &&
1701             match->prot_hook.dev == po->prot_hook.dev) {
1702                 err = -ENOSPC;
1703                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1704                         __dev_remove_pack(&po->prot_hook);
1705                         po->fanout = match;
1706                         atomic_inc(&match->sk_ref);
1707                         __fanout_link(sk, po);
1708                         err = 0;
1709                 }
1710         }
1711 out:
1712         mutex_unlock(&fanout_mutex);
1713         if (err) {
1714                 kfree(po->rollover);
1715                 po->rollover = NULL;
1716         }
1717         return err;
1718 }
1719
1720 static void fanout_release(struct sock *sk)
1721 {
1722         struct packet_sock *po = pkt_sk(sk);
1723         struct packet_fanout *f;
1724
1725         f = po->fanout;
1726         if (!f)
1727                 return;
1728
1729         mutex_lock(&fanout_mutex);
1730         po->fanout = NULL;
1731
1732         if (atomic_dec_and_test(&f->sk_ref)) {
1733                 list_del(&f->list);
1734                 dev_remove_pack(&f->prot_hook);
1735                 fanout_release_data(f);
1736                 kfree(f);
1737         }
1738         mutex_unlock(&fanout_mutex);
1739
1740         if (po->rollover)
1741                 kfree_rcu(po->rollover, rcu);
1742 }
1743
1744 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1745                                           struct sk_buff *skb)
1746 {
1747         /* Earlier code assumed this would be a VLAN pkt, double-check
1748          * this now that we have the actual packet in hand. We can only
1749          * do this check on Ethernet devices.
1750          */
1751         if (unlikely(dev->type != ARPHRD_ETHER))
1752                 return false;
1753
1754         skb_reset_mac_header(skb);
1755         return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1756 }
1757
1758 static const struct proto_ops packet_ops;
1759
1760 static const struct proto_ops packet_ops_spkt;
1761
1762 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1763                            struct packet_type *pt, struct net_device *orig_dev)
1764 {
1765         struct sock *sk;
1766         struct sockaddr_pkt *spkt;
1767
1768         /*
1769          *      When we registered the protocol we saved the socket in the data
1770          *      field for just this event.
1771          */
1772
1773         sk = pt->af_packet_priv;
1774
1775         /*
1776          *      Yank back the headers [hope the device set this
1777          *      right or kerboom...]
1778          *
1779          *      Incoming packets have ll header pulled,
1780          *      push it back.
1781          *
1782          *      For outgoing ones skb->data == skb_mac_header(skb)
1783          *      so that this procedure is noop.
1784          */
1785
1786         if (skb->pkt_type == PACKET_LOOPBACK)
1787                 goto out;
1788
1789         if (!net_eq(dev_net(dev), sock_net(sk)))
1790                 goto out;
1791
1792         skb = skb_share_check(skb, GFP_ATOMIC);
1793         if (skb == NULL)
1794                 goto oom;
1795
1796         /* drop any routing info */
1797         skb_dst_drop(skb);
1798
1799         /* drop conntrack reference */
1800         nf_reset(skb);
1801
1802         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1803
1804         skb_push(skb, skb->data - skb_mac_header(skb));
1805
1806         /*
1807          *      The SOCK_PACKET socket receives _all_ frames.
1808          */
1809
1810         spkt->spkt_family = dev->type;
1811         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1812         spkt->spkt_protocol = skb->protocol;
1813
1814         /*
1815          *      Charge the memory to the socket. This is done specifically
1816          *      to prevent sockets using all the memory up.
1817          */
1818
1819         if (sock_queue_rcv_skb(sk, skb) == 0)
1820                 return 0;
1821
1822 out:
1823         kfree_skb(skb);
1824 oom:
1825         return 0;
1826 }
1827
1828
1829 /*
1830  *      Output a raw packet to a device layer. This bypasses all the other
1831  *      protocol layers and you must therefore supply it with a complete frame
1832  */
1833
1834 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1835                                size_t len)
1836 {
1837         struct sock *sk = sock->sk;
1838         DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1839         struct sk_buff *skb = NULL;
1840         struct net_device *dev;
1841         __be16 proto = 0;
1842         int err;
1843         int extra_len = 0;
1844
1845         /*
1846          *      Get and verify the address.
1847          */
1848
1849         if (saddr) {
1850                 if (msg->msg_namelen < sizeof(struct sockaddr))
1851                         return -EINVAL;
1852                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1853                         proto = saddr->spkt_protocol;
1854         } else
1855                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1856
1857         /*
1858          *      Find the device first to size check it
1859          */
1860
1861         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1862 retry:
1863         rcu_read_lock();
1864         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1865         err = -ENODEV;
1866         if (dev == NULL)
1867                 goto out_unlock;
1868
1869         err = -ENETDOWN;
1870         if (!(dev->flags & IFF_UP))
1871                 goto out_unlock;
1872
1873         /*
1874          * You may not queue a frame bigger than the mtu. This is the lowest level
1875          * raw protocol and you must do your own fragmentation at this level.
1876          */
1877
1878         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1879                 if (!netif_supports_nofcs(dev)) {
1880                         err = -EPROTONOSUPPORT;
1881                         goto out_unlock;
1882                 }
1883                 extra_len = 4; /* We're doing our own CRC */
1884         }
1885
1886         err = -EMSGSIZE;
1887         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1888                 goto out_unlock;
1889
1890         if (!skb) {
1891                 size_t reserved = LL_RESERVED_SPACE(dev);
1892                 int tlen = dev->needed_tailroom;
1893                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1894
1895                 rcu_read_unlock();
1896                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1897                 if (skb == NULL)
1898                         return -ENOBUFS;
1899                 /* FIXME: Save some space for broken drivers that write a hard
1900                  * header at transmission time by themselves. PPP is the notable
1901                  * one here. This should really be fixed at the driver level.
1902                  */
1903                 skb_reserve(skb, reserved);
1904                 skb_reset_network_header(skb);
1905
1906                 /* Try to align data part correctly */
1907                 if (hhlen) {
1908                         skb->data -= hhlen;
1909                         skb->tail -= hhlen;
1910                         if (len < hhlen)
1911                                 skb_reset_network_header(skb);
1912                 }
1913                 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1914                 if (err)
1915                         goto out_free;
1916                 goto retry;
1917         }
1918
1919         if (!dev_validate_header(dev, skb->data, len)) {
1920                 err = -EINVAL;
1921                 goto out_unlock;
1922         }
1923         if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1924             !packet_extra_vlan_len_allowed(dev, skb)) {
1925                 err = -EMSGSIZE;
1926                 goto out_unlock;
1927         }
1928
1929         skb->protocol = proto;
1930         skb->dev = dev;
1931         skb->priority = sk->sk_priority;
1932         skb->mark = sk->sk_mark;
1933
1934         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1935
1936         if (unlikely(extra_len == 4))
1937                 skb->no_fcs = 1;
1938
1939         skb_probe_transport_header(skb, 0);
1940
1941         dev_queue_xmit(skb);
1942         rcu_read_unlock();
1943         return len;
1944
1945 out_unlock:
1946         rcu_read_unlock();
1947 out_free:
1948         kfree_skb(skb);
1949         return err;
1950 }
1951
1952 static unsigned int run_filter(struct sk_buff *skb,
1953                                const struct sock *sk,
1954                                unsigned int res)
1955 {
1956         struct sk_filter *filter;
1957
1958         rcu_read_lock();
1959         filter = rcu_dereference(sk->sk_filter);
1960         if (filter != NULL)
1961                 res = bpf_prog_run_clear_cb(filter->prog, skb);
1962         rcu_read_unlock();
1963
1964         return res;
1965 }
1966
1967 /*
1968  * This function makes lazy skb cloning in hope that most of packets
1969  * are discarded by BPF.
1970  *
1971  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1972  * and skb->cb are mangled. It works because (and until) packets
1973  * falling here are owned by current CPU. Output packets are cloned
1974  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1975  * sequencially, so that if we return skb to original state on exit,
1976  * we will not harm anyone.
1977  */
1978
1979 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1980                       struct packet_type *pt, struct net_device *orig_dev)
1981 {
1982         struct sock *sk;
1983         struct sockaddr_ll *sll;
1984         struct packet_sock *po;
1985         u8 *skb_head = skb->data;
1986         int skb_len = skb->len;
1987         unsigned int snaplen, res;
1988
1989         if (skb->pkt_type == PACKET_LOOPBACK)
1990                 goto drop;
1991
1992         sk = pt->af_packet_priv;
1993         po = pkt_sk(sk);
1994
1995         if (!net_eq(dev_net(dev), sock_net(sk)))
1996                 goto drop;
1997
1998         skb->dev = dev;
1999
2000         if (dev->header_ops) {
2001                 /* The device has an explicit notion of ll header,
2002                  * exported to higher levels.
2003                  *
2004                  * Otherwise, the device hides details of its frame
2005                  * structure, so that corresponding packet head is
2006                  * never delivered to user.
2007                  */
2008                 if (sk->sk_type != SOCK_DGRAM)
2009                         skb_push(skb, skb->data - skb_mac_header(skb));
2010                 else if (skb->pkt_type == PACKET_OUTGOING) {
2011                         /* Special case: outgoing packets have ll header at head */
2012                         skb_pull(skb, skb_network_offset(skb));
2013                 }
2014         }
2015
2016         snaplen = skb->len;
2017
2018         res = run_filter(skb, sk, snaplen);
2019         if (!res)
2020                 goto drop_n_restore;
2021         if (snaplen > res)
2022                 snaplen = res;
2023
2024         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2025                 goto drop_n_acct;
2026
2027         if (skb_shared(skb)) {
2028                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2029                 if (nskb == NULL)
2030                         goto drop_n_acct;
2031
2032                 if (skb_head != skb->data) {
2033                         skb->data = skb_head;
2034                         skb->len = skb_len;
2035                 }
2036                 consume_skb(skb);
2037                 skb = nskb;
2038         }
2039
2040         sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2041
2042         sll = &PACKET_SKB_CB(skb)->sa.ll;
2043         sll->sll_hatype = dev->type;
2044         sll->sll_pkttype = skb->pkt_type;
2045         if (unlikely(po->origdev))
2046                 sll->sll_ifindex = orig_dev->ifindex;
2047         else
2048                 sll->sll_ifindex = dev->ifindex;
2049
2050         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2051
2052         /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2053          * Use their space for storing the original skb length.
2054          */
2055         PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2056
2057         if (pskb_trim(skb, snaplen))
2058                 goto drop_n_acct;
2059
2060         skb_set_owner_r(skb, sk);
2061         skb->dev = NULL;
2062         skb_dst_drop(skb);
2063
2064         /* drop conntrack reference */
2065         nf_reset(skb);
2066
2067         spin_lock(&sk->sk_receive_queue.lock);
2068         po->stats.stats1.tp_packets++;
2069         sock_skb_set_dropcount(sk, skb);
2070         __skb_queue_tail(&sk->sk_receive_queue, skb);
2071         spin_unlock(&sk->sk_receive_queue.lock);
2072         sk->sk_data_ready(sk);
2073         return 0;
2074
2075 drop_n_acct:
2076         spin_lock(&sk->sk_receive_queue.lock);
2077         po->stats.stats1.tp_drops++;
2078         atomic_inc(&sk->sk_drops);
2079         spin_unlock(&sk->sk_receive_queue.lock);
2080
2081 drop_n_restore:
2082         if (skb_head != skb->data && skb_shared(skb)) {
2083                 skb->data = skb_head;
2084                 skb->len = skb_len;
2085         }
2086 drop:
2087         consume_skb(skb);
2088         return 0;
2089 }
2090
2091 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2092                        struct packet_type *pt, struct net_device *orig_dev)
2093 {
2094         struct sock *sk;
2095         struct packet_sock *po;
2096         struct sockaddr_ll *sll;
2097         union tpacket_uhdr h;
2098         u8 *skb_head = skb->data;
2099         int skb_len = skb->len;
2100         unsigned int snaplen, res;
2101         unsigned long status = TP_STATUS_USER;
2102         unsigned short macoff, netoff, hdrlen;
2103         struct sk_buff *copy_skb = NULL;
2104         struct timespec ts;
2105         __u32 ts_status;
2106
2107         /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2108          * We may add members to them until current aligned size without forcing
2109          * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2110          */
2111         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2112         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2113
2114         if (skb->pkt_type == PACKET_LOOPBACK)
2115                 goto drop;
2116
2117         sk = pt->af_packet_priv;
2118         po = pkt_sk(sk);
2119
2120         if (!net_eq(dev_net(dev), sock_net(sk)))
2121                 goto drop;
2122
2123         if (dev->header_ops) {
2124                 if (sk->sk_type != SOCK_DGRAM)
2125                         skb_push(skb, skb->data - skb_mac_header(skb));
2126                 else if (skb->pkt_type == PACKET_OUTGOING) {
2127                         /* Special case: outgoing packets have ll header at head */
2128                         skb_pull(skb, skb_network_offset(skb));
2129                 }
2130         }
2131
2132         snaplen = skb->len;
2133
2134         res = run_filter(skb, sk, snaplen);
2135         if (!res)
2136                 goto drop_n_restore;
2137
2138         if (skb->ip_summed == CHECKSUM_PARTIAL)
2139                 status |= TP_STATUS_CSUMNOTREADY;
2140         else if (skb->pkt_type != PACKET_OUTGOING &&
2141                  (skb->ip_summed == CHECKSUM_COMPLETE ||
2142                   skb_csum_unnecessary(skb)))
2143                 status |= TP_STATUS_CSUM_VALID;
2144
2145         if (snaplen > res)
2146                 snaplen = res;
2147
2148         if (sk->sk_type == SOCK_DGRAM) {
2149                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2150                                   po->tp_reserve;
2151         } else {
2152                 unsigned int maclen = skb_network_offset(skb);
2153                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2154                                        (maclen < 16 ? 16 : maclen)) +
2155                         po->tp_reserve;
2156                 macoff = netoff - maclen;
2157         }
2158         if (po->tp_version <= TPACKET_V2) {
2159                 if (macoff + snaplen > po->rx_ring.frame_size) {
2160                         if (po->copy_thresh &&
2161                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2162                                 if (skb_shared(skb)) {
2163                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
2164                                 } else {
2165                                         copy_skb = skb_get(skb);
2166                                         skb_head = skb->data;
2167                                 }
2168                                 if (copy_skb)
2169                                         skb_set_owner_r(copy_skb, sk);
2170                         }
2171                         snaplen = po->rx_ring.frame_size - macoff;
2172                         if ((int)snaplen < 0)
2173                                 snaplen = 0;
2174                 }
2175         } else if (unlikely(macoff + snaplen >
2176                             GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2177                 u32 nval;
2178
2179                 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2180                 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2181                             snaplen, nval, macoff);
2182                 snaplen = nval;
2183                 if (unlikely((int)snaplen < 0)) {
2184                         snaplen = 0;
2185                         macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2186                 }
2187         }
2188         spin_lock(&sk->sk_receive_queue.lock);
2189         h.raw = packet_current_rx_frame(po, skb,
2190                                         TP_STATUS_KERNEL, (macoff+snaplen));
2191         if (!h.raw)
2192                 goto ring_is_full;
2193         if (po->tp_version <= TPACKET_V2) {
2194                 packet_increment_rx_head(po, &po->rx_ring);
2195         /*
2196          * LOSING will be reported till you read the stats,
2197          * because it's COR - Clear On Read.
2198          * Anyways, moving it for V1/V2 only as V3 doesn't need this
2199          * at packet level.
2200          */
2201                 if (po->stats.stats1.tp_drops)
2202                         status |= TP_STATUS_LOSING;
2203         }
2204         po->stats.stats1.tp_packets++;
2205         if (copy_skb) {
2206                 status |= TP_STATUS_COPY;
2207                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2208         }
2209         spin_unlock(&sk->sk_receive_queue.lock);
2210
2211         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2212
2213         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2214                 getnstimeofday(&ts);
2215
2216         status |= ts_status;
2217
2218         switch (po->tp_version) {
2219         case TPACKET_V1:
2220                 h.h1->tp_len = skb->len;
2221                 h.h1->tp_snaplen = snaplen;
2222                 h.h1->tp_mac = macoff;
2223                 h.h1->tp_net = netoff;
2224                 h.h1->tp_sec = ts.tv_sec;
2225                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2226                 hdrlen = sizeof(*h.h1);
2227                 break;
2228         case TPACKET_V2:
2229                 h.h2->tp_len = skb->len;
2230                 h.h2->tp_snaplen = snaplen;
2231                 h.h2->tp_mac = macoff;
2232                 h.h2->tp_net = netoff;
2233                 h.h2->tp_sec = ts.tv_sec;
2234                 h.h2->tp_nsec = ts.tv_nsec;
2235                 if (skb_vlan_tag_present(skb)) {
2236                         h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2237                         h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2238                         status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2239                 } else {
2240                         h.h2->tp_vlan_tci = 0;
2241                         h.h2->tp_vlan_tpid = 0;
2242                 }
2243                 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2244                 hdrlen = sizeof(*h.h2);
2245                 break;
2246         case TPACKET_V3:
2247                 /* tp_nxt_offset,vlan are already populated above.
2248                  * So DONT clear those fields here
2249                  */
2250                 h.h3->tp_status |= status;
2251                 h.h3->tp_len = skb->len;
2252                 h.h3->tp_snaplen = snaplen;
2253                 h.h3->tp_mac = macoff;
2254                 h.h3->tp_net = netoff;
2255                 h.h3->tp_sec  = ts.tv_sec;
2256                 h.h3->tp_nsec = ts.tv_nsec;
2257                 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2258                 hdrlen = sizeof(*h.h3);
2259                 break;
2260         default:
2261                 BUG();
2262         }
2263
2264         sll = h.raw + TPACKET_ALIGN(hdrlen);
2265         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2266         sll->sll_family = AF_PACKET;
2267         sll->sll_hatype = dev->type;
2268         sll->sll_protocol = skb->protocol;
2269         sll->sll_pkttype = skb->pkt_type;
2270         if (unlikely(po->origdev))
2271                 sll->sll_ifindex = orig_dev->ifindex;
2272         else
2273                 sll->sll_ifindex = dev->ifindex;
2274
2275         smp_mb();
2276
2277 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2278         if (po->tp_version <= TPACKET_V2) {
2279                 u8 *start, *end;
2280
2281                 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2282                                         macoff + snaplen);
2283
2284                 for (start = h.raw; start < end; start += PAGE_SIZE)
2285                         flush_dcache_page(pgv_to_page(start));
2286         }
2287         smp_wmb();
2288 #endif
2289
2290         if (po->tp_version <= TPACKET_V2) {
2291                 __packet_set_status(po, h.raw, status);
2292                 sk->sk_data_ready(sk);
2293         } else {
2294                 prb_clear_blk_fill_status(&po->rx_ring);
2295         }
2296
2297 drop_n_restore:
2298         if (skb_head != skb->data && skb_shared(skb)) {
2299                 skb->data = skb_head;
2300                 skb->len = skb_len;
2301         }
2302 drop:
2303         kfree_skb(skb);
2304         return 0;
2305
2306 ring_is_full:
2307         po->stats.stats1.tp_drops++;
2308         spin_unlock(&sk->sk_receive_queue.lock);
2309
2310         sk->sk_data_ready(sk);
2311         kfree_skb(copy_skb);
2312         goto drop_n_restore;
2313 }
2314
2315 static void tpacket_destruct_skb(struct sk_buff *skb)
2316 {
2317         struct packet_sock *po = pkt_sk(skb->sk);
2318
2319         if (likely(po->tx_ring.pg_vec)) {
2320                 void *ph;
2321                 __u32 ts;
2322
2323                 ph = skb_shinfo(skb)->destructor_arg;
2324                 packet_dec_pending(&po->tx_ring);
2325
2326                 ts = __packet_set_timestamp(po, ph, skb);
2327                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2328         }
2329
2330         sock_wfree(skb);
2331 }
2332
2333 static void tpacket_set_protocol(const struct net_device *dev,
2334                                  struct sk_buff *skb)
2335 {
2336         if (dev->type == ARPHRD_ETHER) {
2337                 skb_reset_mac_header(skb);
2338                 skb->protocol = eth_hdr(skb)->h_proto;
2339         }
2340 }
2341
2342 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2343                 void *frame, struct net_device *dev, int size_max,
2344                 __be16 proto, unsigned char *addr, int hlen)
2345 {
2346         union tpacket_uhdr ph;
2347         int to_write, offset, len, tp_len, nr_frags, len_max;
2348         struct socket *sock = po->sk.sk_socket;
2349         struct page *page;
2350         void *data;
2351         int err;
2352
2353         ph.raw = frame;
2354
2355         skb->protocol = proto;
2356         skb->dev = dev;
2357         skb->priority = po->sk.sk_priority;
2358         skb->mark = po->sk.sk_mark;
2359         sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2360         skb_shinfo(skb)->destructor_arg = ph.raw;
2361
2362         switch (po->tp_version) {
2363         case TPACKET_V2:
2364                 tp_len = ph.h2->tp_len;
2365                 break;
2366         default:
2367                 tp_len = ph.h1->tp_len;
2368                 break;
2369         }
2370         if (unlikely(tp_len > size_max)) {
2371                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2372                 return -EMSGSIZE;
2373         }
2374
2375         skb_reserve(skb, hlen);
2376         skb_reset_network_header(skb);
2377
2378         if (unlikely(po->tp_tx_has_off)) {
2379                 int off_min, off_max, off;
2380                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2381                 off_max = po->tx_ring.frame_size - tp_len;
2382                 if (sock->type == SOCK_DGRAM) {
2383                         switch (po->tp_version) {
2384                         case TPACKET_V2:
2385                                 off = ph.h2->tp_net;
2386                                 break;
2387                         default:
2388                                 off = ph.h1->tp_net;
2389                                 break;
2390                         }
2391                 } else {
2392                         switch (po->tp_version) {
2393                         case TPACKET_V2:
2394                                 off = ph.h2->tp_mac;
2395                                 break;
2396                         default:
2397                                 off = ph.h1->tp_mac;
2398                                 break;
2399                         }
2400                 }
2401                 if (unlikely((off < off_min) || (off_max < off)))
2402                         return -EINVAL;
2403                 data = ph.raw + off;
2404         } else {
2405                 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2406         }
2407         to_write = tp_len;
2408
2409         if (sock->type == SOCK_DGRAM) {
2410                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2411                                 NULL, tp_len);
2412                 if (unlikely(err < 0))
2413                         return -EINVAL;
2414         } else if (dev->hard_header_len) {
2415                 int hdrlen = min_t(int, dev->hard_header_len, tp_len);
2416
2417                 skb_push(skb, dev->hard_header_len);
2418                 err = skb_store_bits(skb, 0, data, hdrlen);
2419                 if (unlikely(err))
2420                         return err;
2421                 if (!dev_validate_header(dev, skb->data, hdrlen))
2422                         return -EINVAL;
2423                 if (!skb->protocol)
2424                         tpacket_set_protocol(dev, skb);
2425
2426                 data += hdrlen;
2427                 to_write -= hdrlen;
2428         }
2429
2430         offset = offset_in_page(data);
2431         len_max = PAGE_SIZE - offset;
2432         len = ((to_write > len_max) ? len_max : to_write);
2433
2434         skb->data_len = to_write;
2435         skb->len += to_write;
2436         skb->truesize += to_write;
2437         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2438
2439         while (likely(to_write)) {
2440                 nr_frags = skb_shinfo(skb)->nr_frags;
2441
2442                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2443                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2444                                MAX_SKB_FRAGS);
2445                         return -EFAULT;
2446                 }
2447
2448                 page = pgv_to_page(data);
2449                 data += len;
2450                 flush_dcache_page(page);
2451                 get_page(page);
2452                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2453                 to_write -= len;
2454                 offset = 0;
2455                 len_max = PAGE_SIZE;
2456                 len = ((to_write > len_max) ? len_max : to_write);
2457         }
2458
2459         skb_probe_transport_header(skb, 0);
2460
2461         return tp_len;
2462 }
2463
2464 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2465 {
2466         struct sk_buff *skb;
2467         struct net_device *dev;
2468         __be16 proto;
2469         int err, reserve = 0;
2470         void *ph;
2471         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2472         bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2473         int tp_len, size_max;
2474         unsigned char *addr;
2475         int len_sum = 0;
2476         int status = TP_STATUS_AVAILABLE;
2477         int hlen, tlen;
2478
2479         mutex_lock(&po->pg_vec_lock);
2480
2481         if (likely(saddr == NULL)) {
2482                 dev     = packet_cached_dev_get(po);
2483                 proto   = po->num;
2484                 addr    = NULL;
2485         } else {
2486                 err = -EINVAL;
2487                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2488                         goto out;
2489                 if (msg->msg_namelen < (saddr->sll_halen
2490                                         + offsetof(struct sockaddr_ll,
2491                                                 sll_addr)))
2492                         goto out;
2493                 proto   = saddr->sll_protocol;
2494                 addr    = saddr->sll_addr;
2495                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2496         }
2497
2498         err = -ENXIO;
2499         if (unlikely(dev == NULL))
2500                 goto out;
2501         err = -ENETDOWN;
2502         if (unlikely(!(dev->flags & IFF_UP)))
2503                 goto out_put;
2504
2505         if (po->sk.sk_socket->type == SOCK_RAW)
2506                 reserve = dev->hard_header_len;
2507         size_max = po->tx_ring.frame_size
2508                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2509
2510         if (size_max > dev->mtu + reserve + VLAN_HLEN)
2511                 size_max = dev->mtu + reserve + VLAN_HLEN;
2512
2513         do {
2514                 ph = packet_current_frame(po, &po->tx_ring,
2515                                           TP_STATUS_SEND_REQUEST);
2516                 if (unlikely(ph == NULL)) {
2517                         if (need_wait && need_resched())
2518                                 schedule();
2519                         continue;
2520                 }
2521
2522                 status = TP_STATUS_SEND_REQUEST;
2523                 hlen = LL_RESERVED_SPACE(dev);
2524                 tlen = dev->needed_tailroom;
2525                 skb = sock_alloc_send_skb(&po->sk,
2526                                 hlen + tlen + sizeof(struct sockaddr_ll),
2527                                 !need_wait, &err);
2528
2529                 if (unlikely(skb == NULL)) {
2530                         /* we assume the socket was initially writeable ... */
2531                         if (likely(len_sum > 0))
2532                                 err = len_sum;
2533                         goto out_status;
2534                 }
2535                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2536                                           addr, hlen);
2537                 if (likely(tp_len >= 0) &&
2538                     tp_len > dev->mtu + reserve &&
2539                     !packet_extra_vlan_len_allowed(dev, skb))
2540                         tp_len = -EMSGSIZE;
2541
2542                 if (unlikely(tp_len < 0)) {
2543                         if (po->tp_loss) {
2544                                 __packet_set_status(po, ph,
2545                                                 TP_STATUS_AVAILABLE);
2546                                 packet_increment_head(&po->tx_ring);
2547                                 kfree_skb(skb);
2548                                 continue;
2549                         } else {
2550                                 status = TP_STATUS_WRONG_FORMAT;
2551                                 err = tp_len;
2552                                 goto out_status;
2553                         }
2554                 }
2555
2556                 packet_pick_tx_queue(dev, skb);
2557
2558                 skb->destructor = tpacket_destruct_skb;
2559                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2560                 packet_inc_pending(&po->tx_ring);
2561
2562                 status = TP_STATUS_SEND_REQUEST;
2563                 err = po->xmit(skb);
2564                 if (unlikely(err > 0)) {
2565                         err = net_xmit_errno(err);
2566                         if (err && __packet_get_status(po, ph) ==
2567                                    TP_STATUS_AVAILABLE) {
2568                                 /* skb was destructed already */
2569                                 skb = NULL;
2570                                 goto out_status;
2571                         }
2572                         /*
2573                          * skb was dropped but not destructed yet;
2574                          * let's treat it like congestion or err < 0
2575                          */
2576                         err = 0;
2577                 }
2578                 packet_increment_head(&po->tx_ring);
2579                 len_sum += tp_len;
2580         } while (likely((ph != NULL) ||
2581                 /* Note: packet_read_pending() might be slow if we have
2582                  * to call it as it's per_cpu variable, but in fast-path
2583                  * we already short-circuit the loop with the first
2584                  * condition, and luckily don't have to go that path
2585                  * anyway.
2586                  */
2587                  (need_wait && packet_read_pending(&po->tx_ring))));
2588
2589         err = len_sum;
2590         goto out_put;
2591
2592 out_status:
2593         __packet_set_status(po, ph, status);
2594         kfree_skb(skb);
2595 out_put:
2596         dev_put(dev);
2597 out:
2598         mutex_unlock(&po->pg_vec_lock);
2599         return err;
2600 }
2601
2602 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2603                                         size_t reserve, size_t len,
2604                                         size_t linear, int noblock,
2605                                         int *err)
2606 {
2607         struct sk_buff *skb;
2608
2609         /* Under a page?  Don't bother with paged skb. */
2610         if (prepad + len < PAGE_SIZE || !linear)
2611                 linear = len;
2612
2613         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2614                                    err, 0);
2615         if (!skb)
2616                 return NULL;
2617
2618         skb_reserve(skb, reserve);
2619         skb_put(skb, linear);
2620         skb->data_len = len - linear;
2621         skb->len += len - linear;
2622
2623         return skb;
2624 }
2625
2626 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2627 {
2628         struct sock *sk = sock->sk;
2629         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2630         struct sk_buff *skb;
2631         struct net_device *dev;
2632         __be16 proto;
2633         unsigned char *addr;
2634         int err, reserve = 0;
2635         struct sockcm_cookie sockc;
2636         struct virtio_net_hdr vnet_hdr = { 0 };
2637         int offset = 0;
2638         int vnet_hdr_len;
2639         struct packet_sock *po = pkt_sk(sk);
2640         unsigned short gso_type = 0;
2641         int hlen, tlen;
2642         int extra_len = 0;
2643         ssize_t n;
2644
2645         /*
2646          *      Get and verify the address.
2647          */
2648
2649         if (likely(saddr == NULL)) {
2650                 dev     = packet_cached_dev_get(po);
2651                 proto   = po->num;
2652                 addr    = NULL;
2653         } else {
2654                 err = -EINVAL;
2655                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2656                         goto out;
2657                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2658                         goto out;
2659                 proto   = saddr->sll_protocol;
2660                 addr    = saddr->sll_addr;
2661                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2662         }
2663
2664         err = -ENXIO;
2665         if (unlikely(dev == NULL))
2666                 goto out_unlock;
2667         err = -ENETDOWN;
2668         if (unlikely(!(dev->flags & IFF_UP)))
2669                 goto out_unlock;
2670
2671         sockc.mark = sk->sk_mark;
2672         if (msg->msg_controllen) {
2673                 err = sock_cmsg_send(sk, msg, &sockc);
2674                 if (unlikely(err))
2675                         goto out_unlock;
2676         }
2677
2678         if (sock->type == SOCK_RAW)
2679                 reserve = dev->hard_header_len;
2680         if (po->has_vnet_hdr) {
2681                 vnet_hdr_len = sizeof(vnet_hdr);
2682
2683                 err = -EINVAL;
2684                 if (len < vnet_hdr_len)
2685                         goto out_unlock;
2686
2687                 len -= vnet_hdr_len;
2688
2689                 err = -EFAULT;
2690                 n = copy_from_iter(&vnet_hdr, vnet_hdr_len, &msg->msg_iter);
2691                 if (n != vnet_hdr_len)
2692                         goto out_unlock;
2693
2694                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2695                     (__virtio16_to_cpu(vio_le(), vnet_hdr.csum_start) +
2696                      __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset) + 2 >
2697                       __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len)))
2698                         vnet_hdr.hdr_len = __cpu_to_virtio16(vio_le(),
2699                                  __virtio16_to_cpu(vio_le(), vnet_hdr.csum_start) +
2700                                 __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset) + 2);
2701
2702                 err = -EINVAL;
2703                 if (__virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len) > len)
2704                         goto out_unlock;
2705
2706                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2707                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2708                         case VIRTIO_NET_HDR_GSO_TCPV4:
2709                                 gso_type = SKB_GSO_TCPV4;
2710                                 break;
2711                         case VIRTIO_NET_HDR_GSO_TCPV6:
2712                                 gso_type = SKB_GSO_TCPV6;
2713                                 break;
2714                         case VIRTIO_NET_HDR_GSO_UDP:
2715                                 gso_type = SKB_GSO_UDP;
2716                                 break;
2717                         default:
2718                                 goto out_unlock;
2719                         }
2720
2721                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2722                                 gso_type |= SKB_GSO_TCP_ECN;
2723
2724                         if (vnet_hdr.gso_size == 0)
2725                                 goto out_unlock;
2726
2727                 }
2728         }
2729
2730         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2731                 if (!netif_supports_nofcs(dev)) {
2732                         err = -EPROTONOSUPPORT;
2733                         goto out_unlock;
2734                 }
2735                 extra_len = 4; /* We're doing our own CRC */
2736         }
2737
2738         err = -EMSGSIZE;
2739         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2740                 goto out_unlock;
2741
2742         err = -ENOBUFS;
2743         hlen = LL_RESERVED_SPACE(dev);
2744         tlen = dev->needed_tailroom;
2745         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
2746                                __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len),
2747                                msg->msg_flags & MSG_DONTWAIT, &err);
2748         if (skb == NULL)
2749                 goto out_unlock;
2750
2751         skb_set_network_header(skb, reserve);
2752
2753         err = -EINVAL;
2754         if (sock->type == SOCK_DGRAM) {
2755                 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2756                 if (unlikely(offset < 0))
2757                         goto out_free;
2758         }
2759
2760         /* Returns -EFAULT on error */
2761         err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2762         if (err)
2763                 goto out_free;
2764
2765         if (sock->type == SOCK_RAW &&
2766             !dev_validate_header(dev, skb->data, len)) {
2767                 err = -EINVAL;
2768                 goto out_free;
2769         }
2770
2771         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2772
2773         if (!gso_type && (len > dev->mtu + reserve + extra_len) &&
2774             !packet_extra_vlan_len_allowed(dev, skb)) {
2775                 err = -EMSGSIZE;
2776                 goto out_free;
2777         }
2778
2779         skb->protocol = proto;
2780         skb->dev = dev;
2781         skb->priority = sk->sk_priority;
2782         skb->mark = sockc.mark;
2783
2784         packet_pick_tx_queue(dev, skb);
2785
2786         if (po->has_vnet_hdr) {
2787                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2788                         u16 s = __virtio16_to_cpu(vio_le(), vnet_hdr.csum_start);
2789                         u16 o = __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset);
2790                         if (!skb_partial_csum_set(skb, s, o)) {
2791                                 err = -EINVAL;
2792                                 goto out_free;
2793                         }
2794                 }
2795
2796                 skb_shinfo(skb)->gso_size =
2797                         __virtio16_to_cpu(vio_le(), vnet_hdr.gso_size);
2798                 skb_shinfo(skb)->gso_type = gso_type;
2799
2800                 /* Header must be checked, and gso_segs computed. */
2801                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2802                 skb_shinfo(skb)->gso_segs = 0;
2803
2804                 len += vnet_hdr_len;
2805         }
2806
2807         skb_probe_transport_header(skb, reserve);
2808
2809         if (unlikely(extra_len == 4))
2810                 skb->no_fcs = 1;
2811
2812         err = po->xmit(skb);
2813         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2814                 goto out_unlock;
2815
2816         dev_put(dev);
2817
2818         return len;
2819
2820 out_free:
2821         kfree_skb(skb);
2822 out_unlock:
2823         if (dev)
2824                 dev_put(dev);
2825 out:
2826         return err;
2827 }
2828
2829 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2830 {
2831         struct sock *sk = sock->sk;
2832         struct packet_sock *po = pkt_sk(sk);
2833
2834         if (po->tx_ring.pg_vec)
2835                 return tpacket_snd(po, msg);
2836         else
2837                 return packet_snd(sock, msg, len);
2838 }
2839
2840 /*
2841  *      Close a PACKET socket. This is fairly simple. We immediately go
2842  *      to 'closed' state and remove our protocol entry in the device list.
2843  */
2844
2845 static int packet_release(struct socket *sock)
2846 {
2847         struct sock *sk = sock->sk;
2848         struct packet_sock *po;
2849         struct net *net;
2850         union tpacket_req_u req_u;
2851
2852         if (!sk)
2853                 return 0;
2854
2855         net = sock_net(sk);
2856         po = pkt_sk(sk);
2857
2858         mutex_lock(&net->packet.sklist_lock);
2859         sk_del_node_init_rcu(sk);
2860         mutex_unlock(&net->packet.sklist_lock);
2861
2862         preempt_disable();
2863         sock_prot_inuse_add(net, sk->sk_prot, -1);
2864         preempt_enable();
2865
2866         spin_lock(&po->bind_lock);
2867         unregister_prot_hook(sk, false);
2868         packet_cached_dev_reset(po);
2869
2870         if (po->prot_hook.dev) {
2871                 dev_put(po->prot_hook.dev);
2872                 po->prot_hook.dev = NULL;
2873         }
2874         spin_unlock(&po->bind_lock);
2875
2876         packet_flush_mclist(sk);
2877
2878         if (po->rx_ring.pg_vec) {
2879                 memset(&req_u, 0, sizeof(req_u));
2880                 packet_set_ring(sk, &req_u, 1, 0);
2881         }
2882
2883         if (po->tx_ring.pg_vec) {
2884                 memset(&req_u, 0, sizeof(req_u));
2885                 packet_set_ring(sk, &req_u, 1, 1);
2886         }
2887
2888         fanout_release(sk);
2889
2890         synchronize_net();
2891         /*
2892          *      Now the socket is dead. No more input will appear.
2893          */
2894         sock_orphan(sk);
2895         sock->sk = NULL;
2896
2897         /* Purge queues */
2898
2899         skb_queue_purge(&sk->sk_receive_queue);
2900         packet_free_pending(po);
2901         sk_refcnt_debug_release(sk);
2902
2903         sock_put(sk);
2904         return 0;
2905 }
2906
2907 /*
2908  *      Attach a packet hook.
2909  */
2910
2911 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
2912                           __be16 proto)
2913 {
2914         struct packet_sock *po = pkt_sk(sk);
2915         struct net_device *dev_curr;
2916         __be16 proto_curr;
2917         bool need_rehook;
2918         struct net_device *dev = NULL;
2919         int ret = 0;
2920         bool unlisted = false;
2921
2922         if (po->fanout)
2923                 return -EINVAL;
2924
2925         lock_sock(sk);
2926         spin_lock(&po->bind_lock);
2927         rcu_read_lock();
2928
2929         if (name) {
2930                 dev = dev_get_by_name_rcu(sock_net(sk), name);
2931                 if (!dev) {
2932                         ret = -ENODEV;
2933                         goto out_unlock;
2934                 }
2935         } else if (ifindex) {
2936                 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
2937                 if (!dev) {
2938                         ret = -ENODEV;
2939                         goto out_unlock;
2940                 }
2941         }
2942
2943         if (dev)
2944                 dev_hold(dev);
2945
2946         proto_curr = po->prot_hook.type;
2947         dev_curr = po->prot_hook.dev;
2948
2949         need_rehook = proto_curr != proto || dev_curr != dev;
2950
2951         if (need_rehook) {
2952                 if (po->running) {
2953                         rcu_read_unlock();
2954                         __unregister_prot_hook(sk, true);
2955                         rcu_read_lock();
2956                         dev_curr = po->prot_hook.dev;
2957                         if (dev)
2958                                 unlisted = !dev_get_by_index_rcu(sock_net(sk),
2959                                                                  dev->ifindex);
2960                 }
2961
2962                 po->num = proto;
2963                 po->prot_hook.type = proto;
2964
2965                 if (unlikely(unlisted)) {
2966                         dev_put(dev);
2967                         po->prot_hook.dev = NULL;
2968                         po->ifindex = -1;
2969                         packet_cached_dev_reset(po);
2970                 } else {
2971                         po->prot_hook.dev = dev;
2972                         po->ifindex = dev ? dev->ifindex : 0;
2973                         packet_cached_dev_assign(po, dev);
2974                 }
2975         }
2976         if (dev_curr)
2977                 dev_put(dev_curr);
2978
2979         if (proto == 0 || !need_rehook)
2980                 goto out_unlock;
2981
2982         if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
2983                 register_prot_hook(sk);
2984         } else {
2985                 sk->sk_err = ENETDOWN;
2986                 if (!sock_flag(sk, SOCK_DEAD))
2987                         sk->sk_error_report(sk);
2988         }
2989
2990 out_unlock:
2991         rcu_read_unlock();
2992         spin_unlock(&po->bind_lock);
2993         release_sock(sk);
2994         return ret;
2995 }
2996
2997 /*
2998  *      Bind a packet socket to a device
2999  */
3000
3001 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3002                             int addr_len)
3003 {
3004         struct sock *sk = sock->sk;
3005         char name[15];
3006
3007         /*
3008          *      Check legality
3009          */
3010
3011         if (addr_len != sizeof(struct sockaddr))
3012                 return -EINVAL;
3013         strlcpy(name, uaddr->sa_data, sizeof(name));
3014
3015         return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3016 }
3017
3018 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3019 {
3020         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3021         struct sock *sk = sock->sk;
3022
3023         /*
3024          *      Check legality
3025          */
3026
3027         if (addr_len < sizeof(struct sockaddr_ll))
3028                 return -EINVAL;
3029         if (sll->sll_family != AF_PACKET)
3030                 return -EINVAL;
3031
3032         return packet_do_bind(sk, NULL, sll->sll_ifindex,
3033                               sll->sll_protocol ? : pkt_sk(sk)->num);
3034 }
3035
3036 static struct proto packet_proto = {
3037         .name     = "PACKET",
3038         .owner    = THIS_MODULE,
3039         .obj_size = sizeof(struct packet_sock),
3040 };
3041
3042 /*
3043  *      Create a packet of type SOCK_PACKET.
3044  */
3045
3046 static int packet_create(struct net *net, struct socket *sock, int protocol,
3047                          int kern)
3048 {
3049         struct sock *sk;
3050         struct packet_sock *po;
3051         __be16 proto = (__force __be16)protocol; /* weird, but documented */
3052         int err;
3053
3054         if (!ns_capable(net->user_ns, CAP_NET_RAW))
3055                 return -EPERM;
3056         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3057             sock->type != SOCK_PACKET)
3058                 return -ESOCKTNOSUPPORT;
3059
3060         sock->state = SS_UNCONNECTED;
3061
3062         err = -ENOBUFS;
3063         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3064         if (sk == NULL)
3065                 goto out;
3066
3067         sock->ops = &packet_ops;
3068         if (sock->type == SOCK_PACKET)
3069                 sock->ops = &packet_ops_spkt;
3070
3071         sock_init_data(sock, sk);
3072
3073         po = pkt_sk(sk);
3074         sk->sk_family = PF_PACKET;
3075         po->num = proto;
3076         po->xmit = dev_queue_xmit;
3077
3078         err = packet_alloc_pending(po);
3079         if (err)
3080                 goto out2;
3081
3082         packet_cached_dev_reset(po);
3083
3084         sk->sk_destruct = packet_sock_destruct;
3085         sk_refcnt_debug_inc(sk);
3086
3087         /*
3088          *      Attach a protocol block
3089          */
3090
3091         spin_lock_init(&po->bind_lock);
3092         mutex_init(&po->pg_vec_lock);
3093         po->rollover = NULL;
3094         po->prot_hook.func = packet_rcv;
3095
3096         if (sock->type == SOCK_PACKET)
3097                 po->prot_hook.func = packet_rcv_spkt;
3098
3099         po->prot_hook.af_packet_priv = sk;
3100
3101         if (proto) {
3102                 po->prot_hook.type = proto;
3103                 register_prot_hook(sk);
3104         }
3105
3106         mutex_lock(&net->packet.sklist_lock);
3107         sk_add_node_rcu(sk, &net->packet.sklist);
3108         mutex_unlock(&net->packet.sklist_lock);
3109
3110         preempt_disable();
3111         sock_prot_inuse_add(net, &packet_proto, 1);
3112         preempt_enable();
3113
3114         return 0;
3115 out2:
3116         sk_free(sk);
3117 out:
3118         return err;
3119 }
3120
3121 /*
3122  *      Pull a packet from our receive queue and hand it to the user.
3123  *      If necessary we block.
3124  */
3125
3126 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3127                           int flags)
3128 {
3129         struct sock *sk = sock->sk;
3130         struct sk_buff *skb;
3131         int copied, err;
3132         int vnet_hdr_len = 0;
3133         unsigned int origlen = 0;
3134
3135         err = -EINVAL;
3136         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3137                 goto out;
3138
3139 #if 0
3140         /* What error should we return now? EUNATTACH? */
3141         if (pkt_sk(sk)->ifindex < 0)
3142                 return -ENODEV;
3143 #endif
3144
3145         if (flags & MSG_ERRQUEUE) {
3146                 err = sock_recv_errqueue(sk, msg, len,
3147                                          SOL_PACKET, PACKET_TX_TIMESTAMP);
3148                 goto out;
3149         }
3150
3151         /*
3152          *      Call the generic datagram receiver. This handles all sorts
3153          *      of horrible races and re-entrancy so we can forget about it
3154          *      in the protocol layers.
3155          *
3156          *      Now it will return ENETDOWN, if device have just gone down,
3157          *      but then it will block.
3158          */
3159
3160         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3161
3162         /*
3163          *      An error occurred so return it. Because skb_recv_datagram()
3164          *      handles the blocking we don't see and worry about blocking
3165          *      retries.
3166          */
3167
3168         if (skb == NULL)
3169                 goto out;
3170
3171         if (pkt_sk(sk)->pressure)
3172                 packet_rcv_has_room(pkt_sk(sk), NULL);
3173
3174         if (pkt_sk(sk)->has_vnet_hdr) {
3175                 struct virtio_net_hdr vnet_hdr = { 0 };
3176
3177                 err = -EINVAL;
3178                 vnet_hdr_len = sizeof(vnet_hdr);
3179                 if (len < vnet_hdr_len)
3180                         goto out_free;
3181
3182                 len -= vnet_hdr_len;
3183
3184                 if (skb_is_gso(skb)) {
3185                         struct skb_shared_info *sinfo = skb_shinfo(skb);
3186
3187                         /* This is a hint as to how much should be linear. */
3188                         vnet_hdr.hdr_len =
3189                                 __cpu_to_virtio16(vio_le(), skb_headlen(skb));
3190                         vnet_hdr.gso_size =
3191                                 __cpu_to_virtio16(vio_le(), sinfo->gso_size);
3192                         if (sinfo->gso_type & SKB_GSO_TCPV4)
3193                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
3194                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
3195                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
3196                         else if (sinfo->gso_type & SKB_GSO_UDP)
3197                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
3198                         else if (sinfo->gso_type & SKB_GSO_FCOE)
3199                                 goto out_free;
3200                         else
3201                                 BUG();
3202                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
3203                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
3204                 } else
3205                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
3206
3207                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3208                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
3209                         vnet_hdr.csum_start = __cpu_to_virtio16(vio_le(),
3210                                           skb_checksum_start_offset(skb));
3211                         vnet_hdr.csum_offset = __cpu_to_virtio16(vio_le(),
3212                                                          skb->csum_offset);
3213                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3214                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
3215                 } /* else everything is zero */
3216
3217                 err = memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_len);
3218                 if (err < 0)
3219                         goto out_free;
3220         }
3221
3222         /* You lose any data beyond the buffer you gave. If it worries
3223          * a user program they can ask the device for its MTU
3224          * anyway.
3225          */
3226         copied = skb->len;
3227         if (copied > len) {
3228                 copied = len;
3229                 msg->msg_flags |= MSG_TRUNC;
3230         }
3231
3232         err = skb_copy_datagram_msg(skb, 0, msg, copied);
3233         if (err)
3234                 goto out_free;
3235
3236         if (sock->type != SOCK_PACKET) {
3237                 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3238
3239                 /* Original length was stored in sockaddr_ll fields */
3240                 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3241                 sll->sll_family = AF_PACKET;
3242                 sll->sll_protocol = skb->protocol;
3243         }
3244
3245         sock_recv_ts_and_drops(msg, sk, skb);
3246
3247         if (msg->msg_name) {
3248                 /* If the address length field is there to be filled
3249                  * in, we fill it in now.
3250                  */
3251                 if (sock->type == SOCK_PACKET) {
3252                         __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3253                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
3254                 } else {
3255                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3256
3257                         msg->msg_namelen = sll->sll_halen +
3258                                 offsetof(struct sockaddr_ll, sll_addr);
3259                 }
3260                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3261                        msg->msg_namelen);
3262         }
3263
3264         if (pkt_sk(sk)->auxdata) {
3265                 struct tpacket_auxdata aux;
3266
3267                 aux.tp_status = TP_STATUS_USER;
3268                 if (skb->ip_summed == CHECKSUM_PARTIAL)
3269                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3270                 else if (skb->pkt_type != PACKET_OUTGOING &&
3271                          (skb->ip_summed == CHECKSUM_COMPLETE ||
3272                           skb_csum_unnecessary(skb)))
3273                         aux.tp_status |= TP_STATUS_CSUM_VALID;
3274
3275                 aux.tp_len = origlen;
3276                 aux.tp_snaplen = skb->len;
3277                 aux.tp_mac = 0;
3278                 aux.tp_net = skb_network_offset(skb);
3279                 if (skb_vlan_tag_present(skb)) {
3280                         aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3281                         aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3282                         aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3283                 } else {
3284                         aux.tp_vlan_tci = 0;
3285                         aux.tp_vlan_tpid = 0;
3286                 }
3287                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3288         }
3289
3290         /*
3291          *      Free or return the buffer as appropriate. Again this
3292          *      hides all the races and re-entrancy issues from us.
3293          */
3294         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3295
3296 out_free:
3297         skb_free_datagram(sk, skb);
3298 out:
3299         return err;
3300 }
3301
3302 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3303                                int *uaddr_len, int peer)
3304 {
3305         struct net_device *dev;
3306         struct sock *sk = sock->sk;
3307
3308         if (peer)
3309                 return -EOPNOTSUPP;
3310
3311         uaddr->sa_family = AF_PACKET;
3312         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3313         rcu_read_lock();
3314         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3315         if (dev)
3316                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3317         rcu_read_unlock();
3318         *uaddr_len = sizeof(*uaddr);
3319
3320         return 0;
3321 }
3322
3323 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3324                           int *uaddr_len, int peer)
3325 {
3326         struct net_device *dev;
3327         struct sock *sk = sock->sk;
3328         struct packet_sock *po = pkt_sk(sk);
3329         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3330
3331         if (peer)
3332                 return -EOPNOTSUPP;
3333
3334         sll->sll_family = AF_PACKET;
3335         sll->sll_ifindex = po->ifindex;
3336         sll->sll_protocol = po->num;
3337         sll->sll_pkttype = 0;
3338         rcu_read_lock();
3339         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3340         if (dev) {
3341                 sll->sll_hatype = dev->type;
3342                 sll->sll_halen = dev->addr_len;
3343                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3344         } else {
3345                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
3346                 sll->sll_halen = 0;
3347         }
3348         rcu_read_unlock();
3349         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3350
3351         return 0;
3352 }
3353
3354 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3355                          int what)
3356 {
3357         switch (i->type) {
3358         case PACKET_MR_MULTICAST:
3359                 if (i->alen != dev->addr_len)
3360                         return -EINVAL;
3361                 if (what > 0)
3362                         return dev_mc_add(dev, i->addr);
3363                 else
3364                         return dev_mc_del(dev, i->addr);
3365                 break;
3366         case PACKET_MR_PROMISC:
3367                 return dev_set_promiscuity(dev, what);
3368         case PACKET_MR_ALLMULTI:
3369                 return dev_set_allmulti(dev, what);
3370         case PACKET_MR_UNICAST:
3371                 if (i->alen != dev->addr_len)
3372                         return -EINVAL;
3373                 if (what > 0)
3374                         return dev_uc_add(dev, i->addr);
3375                 else
3376                         return dev_uc_del(dev, i->addr);
3377                 break;
3378         default:
3379                 break;
3380         }
3381         return 0;
3382 }
3383
3384 static void packet_dev_mclist_delete(struct net_device *dev,
3385                                      struct packet_mclist **mlp)
3386 {
3387         struct packet_mclist *ml;
3388
3389         while ((ml = *mlp) != NULL) {
3390                 if (ml->ifindex == dev->ifindex) {
3391                         packet_dev_mc(dev, ml, -1);
3392                         *mlp = ml->next;
3393                         kfree(ml);
3394                 } else
3395                         mlp = &ml->next;
3396         }
3397 }
3398
3399 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3400 {
3401         struct packet_sock *po = pkt_sk(sk);
3402         struct packet_mclist *ml, *i;
3403         struct net_device *dev;
3404         int err;
3405
3406         rtnl_lock();
3407
3408         err = -ENODEV;
3409         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3410         if (!dev)
3411                 goto done;
3412
3413         err = -EINVAL;
3414         if (mreq->mr_alen > dev->addr_len)
3415                 goto done;
3416
3417         err = -ENOBUFS;
3418         i = kmalloc(sizeof(*i), GFP_KERNEL);
3419         if (i == NULL)
3420                 goto done;
3421
3422         err = 0;
3423         for (ml = po->mclist; ml; ml = ml->next) {
3424                 if (ml->ifindex == mreq->mr_ifindex &&
3425                     ml->type == mreq->mr_type &&
3426                     ml->alen == mreq->mr_alen &&
3427                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3428                         ml->count++;
3429                         /* Free the new element ... */
3430                         kfree(i);
3431                         goto done;
3432                 }
3433         }
3434
3435         i->type = mreq->mr_type;
3436         i->ifindex = mreq->mr_ifindex;
3437         i->alen = mreq->mr_alen;
3438         memcpy(i->addr, mreq->mr_address, i->alen);
3439         memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3440         i->count = 1;
3441         i->next = po->mclist;
3442         po->mclist = i;
3443         err = packet_dev_mc(dev, i, 1);
3444         if (err) {
3445                 po->mclist = i->next;
3446                 kfree(i);
3447         }
3448
3449 done:
3450         rtnl_unlock();
3451         return err;
3452 }
3453
3454 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3455 {
3456         struct packet_mclist *ml, **mlp;
3457
3458         rtnl_lock();
3459
3460         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3461                 if (ml->ifindex == mreq->mr_ifindex &&
3462                     ml->type == mreq->mr_type &&
3463                     ml->alen == mreq->mr_alen &&
3464                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3465                         if (--ml->count == 0) {
3466                                 struct net_device *dev;
3467                                 *mlp = ml->next;
3468                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3469                                 if (dev)
3470                                         packet_dev_mc(dev, ml, -1);
3471                                 kfree(ml);
3472                         }
3473                         break;
3474                 }
3475         }
3476         rtnl_unlock();
3477         return 0;
3478 }
3479
3480 static void packet_flush_mclist(struct sock *sk)
3481 {
3482         struct packet_sock *po = pkt_sk(sk);
3483         struct packet_mclist *ml;
3484
3485         if (!po->mclist)
3486                 return;
3487
3488         rtnl_lock();
3489         while ((ml = po->mclist) != NULL) {
3490                 struct net_device *dev;
3491
3492                 po->mclist = ml->next;
3493                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3494                 if (dev != NULL)
3495                         packet_dev_mc(dev, ml, -1);
3496                 kfree(ml);
3497         }
3498         rtnl_unlock();
3499 }
3500
3501 static int
3502 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3503 {
3504         struct sock *sk = sock->sk;
3505         struct packet_sock *po = pkt_sk(sk);
3506         int ret;
3507
3508         if (level != SOL_PACKET)
3509                 return -ENOPROTOOPT;
3510
3511         switch (optname) {
3512         case PACKET_ADD_MEMBERSHIP:
3513         case PACKET_DROP_MEMBERSHIP:
3514         {
3515                 struct packet_mreq_max mreq;
3516                 int len = optlen;
3517                 memset(&mreq, 0, sizeof(mreq));
3518                 if (len < sizeof(struct packet_mreq))
3519                         return -EINVAL;
3520                 if (len > sizeof(mreq))
3521                         len = sizeof(mreq);
3522                 if (copy_from_user(&mreq, optval, len))
3523                         return -EFAULT;
3524                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3525                         return -EINVAL;
3526                 if (optname == PACKET_ADD_MEMBERSHIP)
3527                         ret = packet_mc_add(sk, &mreq);
3528                 else
3529                         ret = packet_mc_drop(sk, &mreq);
3530                 return ret;
3531         }
3532
3533         case PACKET_RX_RING:
3534         case PACKET_TX_RING:
3535         {
3536                 union tpacket_req_u req_u;
3537                 int len;
3538
3539                 switch (po->tp_version) {
3540                 case TPACKET_V1:
3541                 case TPACKET_V2:
3542                         len = sizeof(req_u.req);
3543                         break;
3544                 case TPACKET_V3:
3545                 default:
3546                         len = sizeof(req_u.req3);
3547                         break;
3548                 }
3549                 if (optlen < len)
3550                         return -EINVAL;
3551                 if (pkt_sk(sk)->has_vnet_hdr)
3552                         return -EINVAL;
3553                 if (copy_from_user(&req_u.req, optval, len))
3554                         return -EFAULT;
3555                 return packet_set_ring(sk, &req_u, 0,
3556                         optname == PACKET_TX_RING);
3557         }
3558         case PACKET_COPY_THRESH:
3559         {
3560                 int val;
3561
3562                 if (optlen != sizeof(val))
3563                         return -EINVAL;
3564                 if (copy_from_user(&val, optval, sizeof(val)))
3565                         return -EFAULT;
3566
3567                 pkt_sk(sk)->copy_thresh = val;
3568                 return 0;
3569         }
3570         case PACKET_VERSION:
3571         {
3572                 int val;
3573
3574                 if (optlen != sizeof(val))
3575                         return -EINVAL;
3576                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3577                         return -EBUSY;
3578                 if (copy_from_user(&val, optval, sizeof(val)))
3579                         return -EFAULT;
3580                 switch (val) {
3581                 case TPACKET_V1:
3582                 case TPACKET_V2:
3583                 case TPACKET_V3:
3584                         po->tp_version = val;
3585                         return 0;
3586                 default:
3587                         return -EINVAL;
3588                 }
3589         }
3590         case PACKET_RESERVE:
3591         {
3592                 unsigned int val;
3593
3594                 if (optlen != sizeof(val))
3595                         return -EINVAL;
3596                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3597                         return -EBUSY;
3598                 if (copy_from_user(&val, optval, sizeof(val)))
3599                         return -EFAULT;
3600                 po->tp_reserve = val;
3601                 return 0;
3602         }
3603         case PACKET_LOSS:
3604         {
3605                 unsigned int val;
3606
3607                 if (optlen != sizeof(val))
3608                         return -EINVAL;
3609                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3610                         return -EBUSY;
3611                 if (copy_from_user(&val, optval, sizeof(val)))
3612                         return -EFAULT;
3613                 po->tp_loss = !!val;
3614                 return 0;
3615         }
3616         case PACKET_AUXDATA:
3617         {
3618                 int val;
3619
3620                 if (optlen < sizeof(val))
3621                         return -EINVAL;
3622                 if (copy_from_user(&val, optval, sizeof(val)))
3623                         return -EFAULT;
3624
3625                 po->auxdata = !!val;
3626                 return 0;
3627         }
3628         case PACKET_ORIGDEV:
3629         {
3630                 int val;
3631
3632                 if (optlen < sizeof(val))
3633                         return -EINVAL;
3634                 if (copy_from_user(&val, optval, sizeof(val)))
3635                         return -EFAULT;
3636
3637                 po->origdev = !!val;
3638                 return 0;
3639         }
3640         case PACKET_VNET_HDR:
3641         {
3642                 int val;
3643
3644                 if (sock->type != SOCK_RAW)
3645                         return -EINVAL;
3646                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3647                         return -EBUSY;
3648                 if (optlen < sizeof(val))
3649                         return -EINVAL;
3650                 if (copy_from_user(&val, optval, sizeof(val)))
3651                         return -EFAULT;
3652
3653                 po->has_vnet_hdr = !!val;
3654                 return 0;
3655         }
3656         case PACKET_TIMESTAMP:
3657         {
3658                 int val;
3659
3660                 if (optlen != sizeof(val))
3661                         return -EINVAL;
3662                 if (copy_from_user(&val, optval, sizeof(val)))
3663                         return -EFAULT;
3664
3665                 po->tp_tstamp = val;
3666                 return 0;
3667         }
3668         case PACKET_FANOUT:
3669         {
3670                 int val;
3671
3672                 if (optlen != sizeof(val))
3673                         return -EINVAL;
3674                 if (copy_from_user(&val, optval, sizeof(val)))
3675                         return -EFAULT;
3676
3677                 return fanout_add(sk, val & 0xffff, val >> 16);
3678         }
3679         case PACKET_FANOUT_DATA:
3680         {
3681                 if (!po->fanout)
3682                         return -EINVAL;
3683
3684                 return fanout_set_data(po, optval, optlen);
3685         }
3686         case PACKET_TX_HAS_OFF:
3687         {
3688                 unsigned int val;
3689
3690                 if (optlen != sizeof(val))
3691                         return -EINVAL;
3692                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3693                         return -EBUSY;
3694                 if (copy_from_user(&val, optval, sizeof(val)))
3695                         return -EFAULT;
3696                 po->tp_tx_has_off = !!val;
3697                 return 0;
3698         }
3699         case PACKET_QDISC_BYPASS:
3700         {
3701                 int val;
3702
3703                 if (optlen != sizeof(val))
3704                         return -EINVAL;
3705                 if (copy_from_user(&val, optval, sizeof(val)))
3706                         return -EFAULT;
3707
3708                 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3709                 return 0;
3710         }
3711         default:
3712                 return -ENOPROTOOPT;
3713         }
3714 }
3715
3716 static int packet_getsockopt(struct socket *sock, int level, int optname,
3717                              char __user *optval, int __user *optlen)
3718 {
3719         int len;
3720         int val, lv = sizeof(val);
3721         struct sock *sk = sock->sk;
3722         struct packet_sock *po = pkt_sk(sk);
3723         void *data = &val;
3724         union tpacket_stats_u st;
3725         struct tpacket_rollover_stats rstats;
3726
3727         if (level != SOL_PACKET)
3728                 return -ENOPROTOOPT;
3729
3730         if (get_user(len, optlen))
3731                 return -EFAULT;
3732
3733         if (len < 0)
3734                 return -EINVAL;
3735
3736         switch (optname) {
3737         case PACKET_STATISTICS:
3738                 spin_lock_bh(&sk->sk_receive_queue.lock);
3739                 memcpy(&st, &po->stats, sizeof(st));
3740                 memset(&po->stats, 0, sizeof(po->stats));
3741                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3742
3743                 if (po->tp_version == TPACKET_V3) {
3744                         lv = sizeof(struct tpacket_stats_v3);
3745                         st.stats3.tp_packets += st.stats3.tp_drops;
3746                         data = &st.stats3;
3747                 } else {
3748                         lv = sizeof(struct tpacket_stats);
3749                         st.stats1.tp_packets += st.stats1.tp_drops;
3750                         data = &st.stats1;
3751                 }
3752
3753                 break;
3754         case PACKET_AUXDATA:
3755                 val = po->auxdata;
3756                 break;
3757         case PACKET_ORIGDEV:
3758                 val = po->origdev;
3759                 break;
3760         case PACKET_VNET_HDR:
3761                 val = po->has_vnet_hdr;
3762                 break;
3763         case PACKET_VERSION:
3764                 val = po->tp_version;
3765                 break;
3766         case PACKET_HDRLEN:
3767                 if (len > sizeof(int))
3768                         len = sizeof(int);
3769                 if (copy_from_user(&val, optval, len))
3770                         return -EFAULT;
3771                 switch (val) {
3772                 case TPACKET_V1:
3773                         val = sizeof(struct tpacket_hdr);
3774                         break;
3775                 case TPACKET_V2:
3776                         val = sizeof(struct tpacket2_hdr);
3777                         break;
3778                 case TPACKET_V3:
3779                         val = sizeof(struct tpacket3_hdr);
3780                         break;
3781                 default:
3782                         return -EINVAL;
3783                 }
3784                 break;
3785         case PACKET_RESERVE:
3786                 val = po->tp_reserve;
3787                 break;
3788         case PACKET_LOSS:
3789                 val = po->tp_loss;
3790                 break;
3791         case PACKET_TIMESTAMP:
3792                 val = po->tp_tstamp;
3793                 break;
3794         case PACKET_FANOUT:
3795                 val = (po->fanout ?
3796                        ((u32)po->fanout->id |
3797                         ((u32)po->fanout->type << 16) |
3798                         ((u32)po->fanout->flags << 24)) :
3799                        0);
3800                 break;
3801         case PACKET_ROLLOVER_STATS:
3802                 if (!po->rollover)
3803                         return -EINVAL;
3804                 rstats.tp_all = atomic_long_read(&po->rollover->num);
3805                 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3806                 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3807                 data = &rstats;
3808                 lv = sizeof(rstats);
3809                 break;
3810         case PACKET_TX_HAS_OFF:
3811                 val = po->tp_tx_has_off;
3812                 break;
3813         case PACKET_QDISC_BYPASS:
3814                 val = packet_use_direct_xmit(po);
3815                 break;
3816         default:
3817                 return -ENOPROTOOPT;
3818         }
3819
3820         if (len > lv)
3821                 len = lv;
3822         if (put_user(len, optlen))
3823                 return -EFAULT;
3824         if (copy_to_user(optval, data, len))
3825                 return -EFAULT;
3826         return 0;
3827 }
3828
3829
3830 static int packet_notifier(struct notifier_block *this,
3831                            unsigned long msg, void *ptr)
3832 {
3833         struct sock *sk;
3834         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3835         struct net *net = dev_net(dev);
3836
3837         rcu_read_lock();
3838         sk_for_each_rcu(sk, &net->packet.sklist) {
3839                 struct packet_sock *po = pkt_sk(sk);
3840
3841                 switch (msg) {
3842                 case NETDEV_UNREGISTER:
3843                         if (po->mclist)
3844                                 packet_dev_mclist_delete(dev, &po->mclist);
3845                         /* fallthrough */
3846
3847                 case NETDEV_DOWN:
3848                         if (dev->ifindex == po->ifindex) {
3849                                 spin_lock(&po->bind_lock);
3850                                 if (po->running) {
3851                                         __unregister_prot_hook(sk, false);
3852                                         sk->sk_err = ENETDOWN;
3853                                         if (!sock_flag(sk, SOCK_DEAD))
3854                                                 sk->sk_error_report(sk);
3855                                 }
3856                                 if (msg == NETDEV_UNREGISTER) {
3857                                         packet_cached_dev_reset(po);
3858                                         po->ifindex = -1;
3859                                         if (po->prot_hook.dev)
3860                                                 dev_put(po->prot_hook.dev);
3861                                         po->prot_hook.dev = NULL;
3862                                 }
3863                                 spin_unlock(&po->bind_lock);
3864                         }
3865                         break;
3866                 case NETDEV_UP:
3867                         if (dev->ifindex == po->ifindex) {
3868                                 spin_lock(&po->bind_lock);
3869                                 if (po->num)
3870                                         register_prot_hook(sk);
3871                                 spin_unlock(&po->bind_lock);
3872                         }
3873                         break;
3874                 }
3875         }
3876         rcu_read_unlock();
3877         return NOTIFY_DONE;
3878 }
3879
3880
3881 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3882                         unsigned long arg)
3883 {
3884         struct sock *sk = sock->sk;
3885
3886         switch (cmd) {
3887         case SIOCOUTQ:
3888         {
3889                 int amount = sk_wmem_alloc_get(sk);
3890
3891                 return put_user(amount, (int __user *)arg);
3892         }
3893         case SIOCINQ:
3894         {
3895                 struct sk_buff *skb;
3896                 int amount = 0;
3897
3898                 spin_lock_bh(&sk->sk_receive_queue.lock);
3899                 skb = skb_peek(&sk->sk_receive_queue);
3900                 if (skb)
3901                         amount = skb->len;
3902                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3903                 return put_user(amount, (int __user *)arg);
3904         }
3905         case SIOCGSTAMP:
3906                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3907         case SIOCGSTAMPNS:
3908                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3909
3910 #ifdef CONFIG_INET
3911         case SIOCADDRT:
3912         case SIOCDELRT:
3913         case SIOCDARP:
3914         case SIOCGARP:
3915         case SIOCSARP:
3916         case SIOCGIFADDR:
3917         case SIOCSIFADDR:
3918         case SIOCGIFBRDADDR:
3919         case SIOCSIFBRDADDR:
3920         case SIOCGIFNETMASK:
3921         case SIOCSIFNETMASK:
3922         case SIOCGIFDSTADDR:
3923         case SIOCSIFDSTADDR:
3924         case SIOCSIFFLAGS:
3925                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3926 #endif
3927
3928         default:
3929                 return -ENOIOCTLCMD;
3930         }
3931         return 0;
3932 }
3933
3934 static unsigned int packet_poll(struct file *file, struct socket *sock,
3935                                 poll_table *wait)
3936 {
3937         struct sock *sk = sock->sk;
3938         struct packet_sock *po = pkt_sk(sk);
3939         unsigned int mask = datagram_poll(file, sock, wait);
3940
3941         spin_lock_bh(&sk->sk_receive_queue.lock);
3942         if (po->rx_ring.pg_vec) {
3943                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3944                         TP_STATUS_KERNEL))
3945                         mask |= POLLIN | POLLRDNORM;
3946         }
3947         if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
3948                 po->pressure = 0;
3949         spin_unlock_bh(&sk->sk_receive_queue.lock);
3950         spin_lock_bh(&sk->sk_write_queue.lock);
3951         if (po->tx_ring.pg_vec) {
3952                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3953                         mask |= POLLOUT | POLLWRNORM;
3954         }
3955         spin_unlock_bh(&sk->sk_write_queue.lock);
3956         return mask;
3957 }
3958
3959
3960 /* Dirty? Well, I still did not learn better way to account
3961  * for user mmaps.
3962  */
3963
3964 static void packet_mm_open(struct vm_area_struct *vma)
3965 {
3966         struct file *file = vma->vm_file;
3967         struct socket *sock = file->private_data;
3968         struct sock *sk = sock->sk;
3969
3970         if (sk)
3971                 atomic_inc(&pkt_sk(sk)->mapped);
3972 }
3973
3974 static void packet_mm_close(struct vm_area_struct *vma)
3975 {
3976         struct file *file = vma->vm_file;
3977         struct socket *sock = file->private_data;
3978         struct sock *sk = sock->sk;
3979
3980         if (sk)
3981                 atomic_dec(&pkt_sk(sk)->mapped);
3982 }
3983
3984 static const struct vm_operations_struct packet_mmap_ops = {
3985         .open   =       packet_mm_open,
3986         .close  =       packet_mm_close,
3987 };
3988
3989 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3990                         unsigned int len)
3991 {
3992         int i;
3993
3994         for (i = 0; i < len; i++) {
3995                 if (likely(pg_vec[i].buffer)) {
3996                         if (is_vmalloc_addr(pg_vec[i].buffer))
3997                                 vfree(pg_vec[i].buffer);
3998                         else
3999                                 free_pages((unsigned long)pg_vec[i].buffer,
4000                                            order);
4001                         pg_vec[i].buffer = NULL;
4002                 }
4003         }
4004         kfree(pg_vec);
4005 }
4006
4007 static char *alloc_one_pg_vec_page(unsigned long order)
4008 {
4009         char *buffer;
4010         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4011                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4012
4013         buffer = (char *) __get_free_pages(gfp_flags, order);
4014         if (buffer)
4015                 return buffer;
4016
4017         /* __get_free_pages failed, fall back to vmalloc */
4018         buffer = vzalloc((1 << order) * PAGE_SIZE);
4019         if (buffer)
4020                 return buffer;
4021
4022         /* vmalloc failed, lets dig into swap here */
4023         gfp_flags &= ~__GFP_NORETRY;
4024         buffer = (char *) __get_free_pages(gfp_flags, order);
4025         if (buffer)
4026                 return buffer;
4027
4028         /* complete and utter failure */
4029         return NULL;
4030 }
4031
4032 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4033 {
4034         unsigned int block_nr = req->tp_block_nr;
4035         struct pgv *pg_vec;
4036         int i;
4037
4038         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4039         if (unlikely(!pg_vec))
4040                 goto out;
4041
4042         for (i = 0; i < block_nr; i++) {
4043                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4044                 if (unlikely(!pg_vec[i].buffer))
4045                         goto out_free_pgvec;
4046         }
4047
4048 out:
4049         return pg_vec;
4050
4051 out_free_pgvec:
4052         free_pg_vec(pg_vec, order, block_nr);
4053         pg_vec = NULL;
4054         goto out;
4055 }
4056
4057 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4058                 int closing, int tx_ring)
4059 {
4060         struct pgv *pg_vec = NULL;
4061         struct packet_sock *po = pkt_sk(sk);
4062         int was_running, order = 0;
4063         struct packet_ring_buffer *rb;
4064         struct sk_buff_head *rb_queue;
4065         __be16 num;
4066         int err = -EINVAL;
4067         /* Added to avoid minimal code churn */
4068         struct tpacket_req *req = &req_u->req;
4069
4070         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
4071         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
4072                 WARN(1, "Tx-ring is not supported.\n");
4073                 goto out;
4074         }
4075
4076         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4077         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4078
4079         err = -EBUSY;
4080         if (!closing) {
4081                 if (atomic_read(&po->mapped))
4082                         goto out;
4083                 if (packet_read_pending(rb))
4084                         goto out;
4085         }
4086
4087         if (req->tp_block_nr) {
4088                 /* Sanity tests and some calculations */
4089                 err = -EBUSY;
4090                 if (unlikely(rb->pg_vec))
4091                         goto out;
4092
4093                 switch (po->tp_version) {
4094                 case TPACKET_V1:
4095                         po->tp_hdrlen = TPACKET_HDRLEN;
4096                         break;
4097                 case TPACKET_V2:
4098                         po->tp_hdrlen = TPACKET2_HDRLEN;
4099                         break;
4100                 case TPACKET_V3:
4101                         po->tp_hdrlen = TPACKET3_HDRLEN;
4102                         break;
4103                 }
4104
4105                 err = -EINVAL;
4106                 if (unlikely((int)req->tp_block_size <= 0))
4107                         goto out;
4108                 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4109                         goto out;
4110                 if (po->tp_version >= TPACKET_V3 &&
4111                     (int)(req->tp_block_size -
4112                           BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
4113                         goto out;
4114                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
4115                                         po->tp_reserve))
4116                         goto out;
4117                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4118                         goto out;
4119
4120                 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4121                 if (unlikely(rb->frames_per_block == 0))
4122                         goto out;
4123                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4124                                         req->tp_frame_nr))
4125                         goto out;
4126
4127                 err = -ENOMEM;
4128                 order = get_order(req->tp_block_size);
4129                 pg_vec = alloc_pg_vec(req, order);
4130                 if (unlikely(!pg_vec))
4131                         goto out;
4132                 switch (po->tp_version) {
4133                 case TPACKET_V3:
4134                 /* Transmit path is not supported. We checked
4135                  * it above but just being paranoid
4136                  */
4137                         if (!tx_ring)
4138                                 init_prb_bdqc(po, rb, pg_vec, req_u);
4139                         break;
4140                 default:
4141                         break;
4142                 }
4143         }
4144         /* Done */
4145         else {
4146                 err = -EINVAL;
4147                 if (unlikely(req->tp_frame_nr))
4148                         goto out;
4149         }
4150
4151         lock_sock(sk);
4152
4153         /* Detach socket from network */
4154         spin_lock(&po->bind_lock);
4155         was_running = po->running;
4156         num = po->num;
4157         if (was_running) {
4158                 po->num = 0;
4159                 __unregister_prot_hook(sk, false);
4160         }
4161         spin_unlock(&po->bind_lock);
4162
4163         synchronize_net();
4164
4165         err = -EBUSY;
4166         mutex_lock(&po->pg_vec_lock);
4167         if (closing || atomic_read(&po->mapped) == 0) {
4168                 err = 0;
4169                 spin_lock_bh(&rb_queue->lock);
4170                 swap(rb->pg_vec, pg_vec);
4171                 rb->frame_max = (req->tp_frame_nr - 1);
4172                 rb->head = 0;
4173                 rb->frame_size = req->tp_frame_size;
4174                 spin_unlock_bh(&rb_queue->lock);
4175
4176                 swap(rb->pg_vec_order, order);
4177                 swap(rb->pg_vec_len, req->tp_block_nr);
4178
4179                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4180                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4181                                                 tpacket_rcv : packet_rcv;
4182                 skb_queue_purge(rb_queue);
4183                 if (atomic_read(&po->mapped))
4184                         pr_err("packet_mmap: vma is busy: %d\n",
4185                                atomic_read(&po->mapped));
4186         }
4187         mutex_unlock(&po->pg_vec_lock);
4188
4189         spin_lock(&po->bind_lock);
4190         if (was_running) {
4191                 po->num = num;
4192                 register_prot_hook(sk);
4193         }
4194         spin_unlock(&po->bind_lock);
4195         if (closing && (po->tp_version > TPACKET_V2)) {
4196                 /* Because we don't support block-based V3 on tx-ring */
4197                 if (!tx_ring)
4198                         prb_shutdown_retire_blk_timer(po, rb_queue);
4199         }
4200         release_sock(sk);
4201
4202         if (pg_vec)
4203                 free_pg_vec(pg_vec, order, req->tp_block_nr);
4204 out:
4205         return err;
4206 }
4207
4208 static int packet_mmap(struct file *file, struct socket *sock,
4209                 struct vm_area_struct *vma)
4210 {
4211         struct sock *sk = sock->sk;
4212         struct packet_sock *po = pkt_sk(sk);
4213         unsigned long size, expected_size;
4214         struct packet_ring_buffer *rb;
4215         unsigned long start;
4216         int err = -EINVAL;
4217         int i;
4218
4219         if (vma->vm_pgoff)
4220                 return -EINVAL;
4221
4222         mutex_lock(&po->pg_vec_lock);
4223
4224         expected_size = 0;
4225         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4226                 if (rb->pg_vec) {
4227                         expected_size += rb->pg_vec_len
4228                                                 * rb->pg_vec_pages
4229                                                 * PAGE_SIZE;
4230                 }
4231         }
4232
4233         if (expected_size == 0)
4234                 goto out;
4235
4236         size = vma->vm_end - vma->vm_start;
4237         if (size != expected_size)
4238                 goto out;
4239
4240         start = vma->vm_start;
4241         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4242                 if (rb->pg_vec == NULL)
4243                         continue;
4244
4245                 for (i = 0; i < rb->pg_vec_len; i++) {
4246                         struct page *page;
4247                         void *kaddr = rb->pg_vec[i].buffer;
4248                         int pg_num;
4249
4250                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4251                                 page = pgv_to_page(kaddr);
4252                                 err = vm_insert_page(vma, start, page);
4253                                 if (unlikely(err))
4254                                         goto out;
4255                                 start += PAGE_SIZE;
4256                                 kaddr += PAGE_SIZE;
4257                         }
4258                 }
4259         }
4260
4261         atomic_inc(&po->mapped);
4262         vma->vm_ops = &packet_mmap_ops;
4263         err = 0;
4264
4265 out:
4266         mutex_unlock(&po->pg_vec_lock);
4267         return err;
4268 }
4269
4270 static const struct proto_ops packet_ops_spkt = {
4271         .family =       PF_PACKET,
4272         .owner =        THIS_MODULE,
4273         .release =      packet_release,
4274         .bind =         packet_bind_spkt,
4275         .connect =      sock_no_connect,
4276         .socketpair =   sock_no_socketpair,
4277         .accept =       sock_no_accept,
4278         .getname =      packet_getname_spkt,
4279         .poll =         datagram_poll,
4280         .ioctl =        packet_ioctl,
4281         .listen =       sock_no_listen,
4282         .shutdown =     sock_no_shutdown,
4283         .setsockopt =   sock_no_setsockopt,
4284         .getsockopt =   sock_no_getsockopt,
4285         .sendmsg =      packet_sendmsg_spkt,
4286         .recvmsg =      packet_recvmsg,
4287         .mmap =         sock_no_mmap,
4288         .sendpage =     sock_no_sendpage,
4289 };
4290
4291 static const struct proto_ops packet_ops = {
4292         .family =       PF_PACKET,
4293         .owner =        THIS_MODULE,
4294         .release =      packet_release,
4295         .bind =         packet_bind,
4296         .connect =      sock_no_connect,
4297         .socketpair =   sock_no_socketpair,
4298         .accept =       sock_no_accept,
4299         .getname =      packet_getname,
4300         .poll =         packet_poll,
4301         .ioctl =        packet_ioctl,
4302         .listen =       sock_no_listen,
4303         .shutdown =     sock_no_shutdown,
4304         .setsockopt =   packet_setsockopt,
4305         .getsockopt =   packet_getsockopt,
4306         .sendmsg =      packet_sendmsg,
4307         .recvmsg =      packet_recvmsg,
4308         .mmap =         packet_mmap,
4309         .sendpage =     sock_no_sendpage,
4310 };
4311
4312 static const struct net_proto_family packet_family_ops = {
4313         .family =       PF_PACKET,
4314         .create =       packet_create,
4315         .owner  =       THIS_MODULE,
4316 };
4317
4318 static struct notifier_block packet_netdev_notifier = {
4319         .notifier_call =        packet_notifier,
4320 };
4321
4322 #ifdef CONFIG_PROC_FS
4323
4324 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4325         __acquires(RCU)
4326 {
4327         struct net *net = seq_file_net(seq);
4328
4329         rcu_read_lock();
4330         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4331 }
4332
4333 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4334 {
4335         struct net *net = seq_file_net(seq);
4336         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4337 }
4338
4339 static void packet_seq_stop(struct seq_file *seq, void *v)
4340         __releases(RCU)
4341 {
4342         rcu_read_unlock();
4343 }
4344
4345 static int packet_seq_show(struct seq_file *seq, void *v)
4346 {
4347         if (v == SEQ_START_TOKEN)
4348                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
4349         else {
4350                 struct sock *s = sk_entry(v);
4351                 const struct packet_sock *po = pkt_sk(s);
4352
4353                 seq_printf(seq,
4354                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4355                            s,
4356                            atomic_read(&s->sk_refcnt),
4357                            s->sk_type,
4358                            ntohs(po->num),
4359                            po->ifindex,
4360                            po->running,
4361                            atomic_read(&s->sk_rmem_alloc),
4362                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4363                            sock_i_ino(s));
4364         }
4365
4366         return 0;
4367 }
4368
4369 static const struct seq_operations packet_seq_ops = {
4370         .start  = packet_seq_start,
4371         .next   = packet_seq_next,
4372         .stop   = packet_seq_stop,
4373         .show   = packet_seq_show,
4374 };
4375
4376 static int packet_seq_open(struct inode *inode, struct file *file)
4377 {
4378         return seq_open_net(inode, file, &packet_seq_ops,
4379                             sizeof(struct seq_net_private));
4380 }
4381
4382 static const struct file_operations packet_seq_fops = {
4383         .owner          = THIS_MODULE,
4384         .open           = packet_seq_open,
4385         .read           = seq_read,
4386         .llseek         = seq_lseek,
4387         .release        = seq_release_net,
4388 };
4389
4390 #endif
4391
4392 static int __net_init packet_net_init(struct net *net)
4393 {
4394         mutex_init(&net->packet.sklist_lock);
4395         INIT_HLIST_HEAD(&net->packet.sklist);
4396
4397         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4398                 return -ENOMEM;
4399
4400         return 0;
4401 }
4402
4403 static void __net_exit packet_net_exit(struct net *net)
4404 {
4405         remove_proc_entry("packet", net->proc_net);
4406 }
4407
4408 static struct pernet_operations packet_net_ops = {
4409         .init = packet_net_init,
4410         .exit = packet_net_exit,
4411 };
4412
4413
4414 static void __exit packet_exit(void)
4415 {
4416         unregister_netdevice_notifier(&packet_netdev_notifier);
4417         unregister_pernet_subsys(&packet_net_ops);
4418         sock_unregister(PF_PACKET);
4419         proto_unregister(&packet_proto);
4420 }
4421
4422 static int __init packet_init(void)
4423 {
4424         int rc = proto_register(&packet_proto, 0);
4425
4426         if (rc != 0)
4427                 goto out;
4428
4429         sock_register(&packet_family_ops);
4430         register_pernet_subsys(&packet_net_ops);
4431         register_netdevice_notifier(&packet_netdev_notifier);
4432 out:
4433         return rc;
4434 }
4435
4436 module_init(packet_init);
4437 module_exit(packet_exit);
4438 MODULE_LICENSE("GPL");
4439 MODULE_ALIAS_NETPROTO(PF_PACKET);