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(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(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 packet_sock *po;
1443         unsigned int idx;
1444
1445         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1446             !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(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 (ptype->af_packet_priv == (void *)((struct packet_sock *)sk)->fanout)
1523                 return true;
1524
1525         return false;
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);
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 const struct proto_ops packet_ops;
1745
1746 static const struct proto_ops packet_ops_spkt;
1747
1748 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1749                            struct packet_type *pt, struct net_device *orig_dev)
1750 {
1751         struct sock *sk;
1752         struct sockaddr_pkt *spkt;
1753
1754         /*
1755          *      When we registered the protocol we saved the socket in the data
1756          *      field for just this event.
1757          */
1758
1759         sk = pt->af_packet_priv;
1760
1761         /*
1762          *      Yank back the headers [hope the device set this
1763          *      right or kerboom...]
1764          *
1765          *      Incoming packets have ll header pulled,
1766          *      push it back.
1767          *
1768          *      For outgoing ones skb->data == skb_mac_header(skb)
1769          *      so that this procedure is noop.
1770          */
1771
1772         if (skb->pkt_type == PACKET_LOOPBACK)
1773                 goto out;
1774
1775         if (!net_eq(dev_net(dev), sock_net(sk)))
1776                 goto out;
1777
1778         skb = skb_share_check(skb, GFP_ATOMIC);
1779         if (skb == NULL)
1780                 goto oom;
1781
1782         /* drop any routing info */
1783         skb_dst_drop(skb);
1784
1785         /* drop conntrack reference */
1786         nf_reset(skb);
1787
1788         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1789
1790         skb_push(skb, skb->data - skb_mac_header(skb));
1791
1792         /*
1793          *      The SOCK_PACKET socket receives _all_ frames.
1794          */
1795
1796         spkt->spkt_family = dev->type;
1797         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1798         spkt->spkt_protocol = skb->protocol;
1799
1800         /*
1801          *      Charge the memory to the socket. This is done specifically
1802          *      to prevent sockets using all the memory up.
1803          */
1804
1805         if (sock_queue_rcv_skb(sk, skb) == 0)
1806                 return 0;
1807
1808 out:
1809         kfree_skb(skb);
1810 oom:
1811         return 0;
1812 }
1813
1814
1815 /*
1816  *      Output a raw packet to a device layer. This bypasses all the other
1817  *      protocol layers and you must therefore supply it with a complete frame
1818  */
1819
1820 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1821                                size_t len)
1822 {
1823         struct sock *sk = sock->sk;
1824         DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1825         struct sk_buff *skb = NULL;
1826         struct net_device *dev;
1827         __be16 proto = 0;
1828         int err;
1829         int extra_len = 0;
1830
1831         /*
1832          *      Get and verify the address.
1833          */
1834
1835         if (saddr) {
1836                 if (msg->msg_namelen < sizeof(struct sockaddr))
1837                         return -EINVAL;
1838                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1839                         proto = saddr->spkt_protocol;
1840         } else
1841                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1842
1843         /*
1844          *      Find the device first to size check it
1845          */
1846
1847         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1848 retry:
1849         rcu_read_lock();
1850         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1851         err = -ENODEV;
1852         if (dev == NULL)
1853                 goto out_unlock;
1854
1855         err = -ENETDOWN;
1856         if (!(dev->flags & IFF_UP))
1857                 goto out_unlock;
1858
1859         /*
1860          * You may not queue a frame bigger than the mtu. This is the lowest level
1861          * raw protocol and you must do your own fragmentation at this level.
1862          */
1863
1864         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1865                 if (!netif_supports_nofcs(dev)) {
1866                         err = -EPROTONOSUPPORT;
1867                         goto out_unlock;
1868                 }
1869                 extra_len = 4; /* We're doing our own CRC */
1870         }
1871
1872         err = -EMSGSIZE;
1873         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1874                 goto out_unlock;
1875
1876         if (!skb) {
1877                 size_t reserved = LL_RESERVED_SPACE(dev);
1878                 int tlen = dev->needed_tailroom;
1879                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1880
1881                 rcu_read_unlock();
1882                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1883                 if (skb == NULL)
1884                         return -ENOBUFS;
1885                 /* FIXME: Save some space for broken drivers that write a hard
1886                  * header at transmission time by themselves. PPP is the notable
1887                  * one here. This should really be fixed at the driver level.
1888                  */
1889                 skb_reserve(skb, reserved);
1890                 skb_reset_network_header(skb);
1891
1892                 /* Try to align data part correctly */
1893                 if (hhlen) {
1894                         skb->data -= hhlen;
1895                         skb->tail -= hhlen;
1896                         if (len < hhlen)
1897                                 skb_reset_network_header(skb);
1898                 }
1899                 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1900                 if (err)
1901                         goto out_free;
1902                 goto retry;
1903         }
1904
1905         if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1906                 /* Earlier code assumed this would be a VLAN pkt,
1907                  * double-check this now that we have the actual
1908                  * packet in hand.
1909                  */
1910                 struct ethhdr *ehdr;
1911                 skb_reset_mac_header(skb);
1912                 ehdr = eth_hdr(skb);
1913                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1914                         err = -EMSGSIZE;
1915                         goto out_unlock;
1916                 }
1917         }
1918
1919         skb->protocol = proto;
1920         skb->dev = dev;
1921         skb->priority = sk->sk_priority;
1922         skb->mark = sk->sk_mark;
1923
1924         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1925
1926         if (unlikely(extra_len == 4))
1927                 skb->no_fcs = 1;
1928
1929         skb_probe_transport_header(skb, 0);
1930
1931         dev_queue_xmit(skb);
1932         rcu_read_unlock();
1933         return len;
1934
1935 out_unlock:
1936         rcu_read_unlock();
1937 out_free:
1938         kfree_skb(skb);
1939         return err;
1940 }
1941
1942 static unsigned int run_filter(const struct sk_buff *skb,
1943                                       const struct sock *sk,
1944                                       unsigned int res)
1945 {
1946         struct sk_filter *filter;
1947
1948         rcu_read_lock();
1949         filter = rcu_dereference(sk->sk_filter);
1950         if (filter != NULL)
1951                 res = SK_RUN_FILTER(filter, skb);
1952         rcu_read_unlock();
1953
1954         return res;
1955 }
1956
1957 /*
1958  * This function makes lazy skb cloning in hope that most of packets
1959  * are discarded by BPF.
1960  *
1961  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1962  * and skb->cb are mangled. It works because (and until) packets
1963  * falling here are owned by current CPU. Output packets are cloned
1964  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1965  * sequencially, so that if we return skb to original state on exit,
1966  * we will not harm anyone.
1967  */
1968
1969 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1970                       struct packet_type *pt, struct net_device *orig_dev)
1971 {
1972         struct sock *sk;
1973         struct sockaddr_ll *sll;
1974         struct packet_sock *po;
1975         u8 *skb_head = skb->data;
1976         int skb_len = skb->len;
1977         unsigned int snaplen, res;
1978
1979         if (skb->pkt_type == PACKET_LOOPBACK)
1980                 goto drop;
1981
1982         sk = pt->af_packet_priv;
1983         po = pkt_sk(sk);
1984
1985         if (!net_eq(dev_net(dev), sock_net(sk)))
1986                 goto drop;
1987
1988         skb->dev = dev;
1989
1990         if (dev->header_ops) {
1991                 /* The device has an explicit notion of ll header,
1992                  * exported to higher levels.
1993                  *
1994                  * Otherwise, the device hides details of its frame
1995                  * structure, so that corresponding packet head is
1996                  * never delivered to user.
1997                  */
1998                 if (sk->sk_type != SOCK_DGRAM)
1999                         skb_push(skb, skb->data - skb_mac_header(skb));
2000                 else if (skb->pkt_type == PACKET_OUTGOING) {
2001                         /* Special case: outgoing packets have ll header at head */
2002                         skb_pull(skb, skb_network_offset(skb));
2003                 }
2004         }
2005
2006         snaplen = skb->len;
2007
2008         res = run_filter(skb, sk, snaplen);
2009         if (!res)
2010                 goto drop_n_restore;
2011         if (snaplen > res)
2012                 snaplen = res;
2013
2014         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2015                 goto drop_n_acct;
2016
2017         if (skb_shared(skb)) {
2018                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2019                 if (nskb == NULL)
2020                         goto drop_n_acct;
2021
2022                 if (skb_head != skb->data) {
2023                         skb->data = skb_head;
2024                         skb->len = skb_len;
2025                 }
2026                 consume_skb(skb);
2027                 skb = nskb;
2028         }
2029
2030         sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2031
2032         sll = &PACKET_SKB_CB(skb)->sa.ll;
2033         sll->sll_hatype = dev->type;
2034         sll->sll_pkttype = skb->pkt_type;
2035         if (unlikely(po->origdev))
2036                 sll->sll_ifindex = orig_dev->ifindex;
2037         else
2038                 sll->sll_ifindex = dev->ifindex;
2039
2040         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2041
2042         /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2043          * Use their space for storing the original skb length.
2044          */
2045         PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2046
2047         if (pskb_trim(skb, snaplen))
2048                 goto drop_n_acct;
2049
2050         skb_set_owner_r(skb, sk);
2051         skb->dev = NULL;
2052         skb_dst_drop(skb);
2053
2054         /* drop conntrack reference */
2055         nf_reset(skb);
2056
2057         spin_lock(&sk->sk_receive_queue.lock);
2058         po->stats.stats1.tp_packets++;
2059         sock_skb_set_dropcount(sk, skb);
2060         __skb_queue_tail(&sk->sk_receive_queue, skb);
2061         spin_unlock(&sk->sk_receive_queue.lock);
2062         sk->sk_data_ready(sk);
2063         return 0;
2064
2065 drop_n_acct:
2066         spin_lock(&sk->sk_receive_queue.lock);
2067         po->stats.stats1.tp_drops++;
2068         atomic_inc(&sk->sk_drops);
2069         spin_unlock(&sk->sk_receive_queue.lock);
2070
2071 drop_n_restore:
2072         if (skb_head != skb->data && skb_shared(skb)) {
2073                 skb->data = skb_head;
2074                 skb->len = skb_len;
2075         }
2076 drop:
2077         consume_skb(skb);
2078         return 0;
2079 }
2080
2081 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2082                        struct packet_type *pt, struct net_device *orig_dev)
2083 {
2084         struct sock *sk;
2085         struct packet_sock *po;
2086         struct sockaddr_ll *sll;
2087         union tpacket_uhdr h;
2088         u8 *skb_head = skb->data;
2089         int skb_len = skb->len;
2090         unsigned int snaplen, res;
2091         unsigned long status = TP_STATUS_USER;
2092         unsigned short macoff, netoff, hdrlen;
2093         struct sk_buff *copy_skb = NULL;
2094         struct timespec ts;
2095         __u32 ts_status;
2096
2097         /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2098          * We may add members to them until current aligned size without forcing
2099          * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2100          */
2101         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2102         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2103
2104         if (skb->pkt_type == PACKET_LOOPBACK)
2105                 goto drop;
2106
2107         sk = pt->af_packet_priv;
2108         po = pkt_sk(sk);
2109
2110         if (!net_eq(dev_net(dev), sock_net(sk)))
2111                 goto drop;
2112
2113         if (dev->header_ops) {
2114                 if (sk->sk_type != SOCK_DGRAM)
2115                         skb_push(skb, skb->data - skb_mac_header(skb));
2116                 else if (skb->pkt_type == PACKET_OUTGOING) {
2117                         /* Special case: outgoing packets have ll header at head */
2118                         skb_pull(skb, skb_network_offset(skb));
2119                 }
2120         }
2121
2122         snaplen = skb->len;
2123
2124         res = run_filter(skb, sk, snaplen);
2125         if (!res)
2126                 goto drop_n_restore;
2127
2128         if (skb->ip_summed == CHECKSUM_PARTIAL)
2129                 status |= TP_STATUS_CSUMNOTREADY;
2130         else if (skb->pkt_type != PACKET_OUTGOING &&
2131                  (skb->ip_summed == CHECKSUM_COMPLETE ||
2132                   skb_csum_unnecessary(skb)))
2133                 status |= TP_STATUS_CSUM_VALID;
2134
2135         if (snaplen > res)
2136                 snaplen = res;
2137
2138         if (sk->sk_type == SOCK_DGRAM) {
2139                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2140                                   po->tp_reserve;
2141         } else {
2142                 unsigned int maclen = skb_network_offset(skb);
2143                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2144                                        (maclen < 16 ? 16 : maclen)) +
2145                         po->tp_reserve;
2146                 macoff = netoff - maclen;
2147         }
2148         if (po->tp_version <= TPACKET_V2) {
2149                 if (macoff + snaplen > po->rx_ring.frame_size) {
2150                         if (po->copy_thresh &&
2151                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2152                                 if (skb_shared(skb)) {
2153                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
2154                                 } else {
2155                                         copy_skb = skb_get(skb);
2156                                         skb_head = skb->data;
2157                                 }
2158                                 if (copy_skb)
2159                                         skb_set_owner_r(copy_skb, sk);
2160                         }
2161                         snaplen = po->rx_ring.frame_size - macoff;
2162                         if ((int)snaplen < 0)
2163                                 snaplen = 0;
2164                 }
2165         } else if (unlikely(macoff + snaplen >
2166                             GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2167                 u32 nval;
2168
2169                 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2170                 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2171                             snaplen, nval, macoff);
2172                 snaplen = nval;
2173                 if (unlikely((int)snaplen < 0)) {
2174                         snaplen = 0;
2175                         macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2176                 }
2177         }
2178         spin_lock(&sk->sk_receive_queue.lock);
2179         h.raw = packet_current_rx_frame(po, skb,
2180                                         TP_STATUS_KERNEL, (macoff+snaplen));
2181         if (!h.raw)
2182                 goto ring_is_full;
2183         if (po->tp_version <= TPACKET_V2) {
2184                 packet_increment_rx_head(po, &po->rx_ring);
2185         /*
2186          * LOSING will be reported till you read the stats,
2187          * because it's COR - Clear On Read.
2188          * Anyways, moving it for V1/V2 only as V3 doesn't need this
2189          * at packet level.
2190          */
2191                 if (po->stats.stats1.tp_drops)
2192                         status |= TP_STATUS_LOSING;
2193         }
2194         po->stats.stats1.tp_packets++;
2195         if (copy_skb) {
2196                 status |= TP_STATUS_COPY;
2197                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2198         }
2199         spin_unlock(&sk->sk_receive_queue.lock);
2200
2201         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2202
2203         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2204                 getnstimeofday(&ts);
2205
2206         status |= ts_status;
2207
2208         switch (po->tp_version) {
2209         case TPACKET_V1:
2210                 h.h1->tp_len = skb->len;
2211                 h.h1->tp_snaplen = snaplen;
2212                 h.h1->tp_mac = macoff;
2213                 h.h1->tp_net = netoff;
2214                 h.h1->tp_sec = ts.tv_sec;
2215                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2216                 hdrlen = sizeof(*h.h1);
2217                 break;
2218         case TPACKET_V2:
2219                 h.h2->tp_len = skb->len;
2220                 h.h2->tp_snaplen = snaplen;
2221                 h.h2->tp_mac = macoff;
2222                 h.h2->tp_net = netoff;
2223                 h.h2->tp_sec = ts.tv_sec;
2224                 h.h2->tp_nsec = ts.tv_nsec;
2225                 if (skb_vlan_tag_present(skb)) {
2226                         h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2227                         h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2228                         status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2229                 } else {
2230                         h.h2->tp_vlan_tci = 0;
2231                         h.h2->tp_vlan_tpid = 0;
2232                 }
2233                 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2234                 hdrlen = sizeof(*h.h2);
2235                 break;
2236         case TPACKET_V3:
2237                 /* tp_nxt_offset,vlan are already populated above.
2238                  * So DONT clear those fields here
2239                  */
2240                 h.h3->tp_status |= status;
2241                 h.h3->tp_len = skb->len;
2242                 h.h3->tp_snaplen = snaplen;
2243                 h.h3->tp_mac = macoff;
2244                 h.h3->tp_net = netoff;
2245                 h.h3->tp_sec  = ts.tv_sec;
2246                 h.h3->tp_nsec = ts.tv_nsec;
2247                 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2248                 hdrlen = sizeof(*h.h3);
2249                 break;
2250         default:
2251                 BUG();
2252         }
2253
2254         sll = h.raw + TPACKET_ALIGN(hdrlen);
2255         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2256         sll->sll_family = AF_PACKET;
2257         sll->sll_hatype = dev->type;
2258         sll->sll_protocol = skb->protocol;
2259         sll->sll_pkttype = skb->pkt_type;
2260         if (unlikely(po->origdev))
2261                 sll->sll_ifindex = orig_dev->ifindex;
2262         else
2263                 sll->sll_ifindex = dev->ifindex;
2264
2265         smp_mb();
2266
2267 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2268         if (po->tp_version <= TPACKET_V2) {
2269                 u8 *start, *end;
2270
2271                 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2272                                         macoff + snaplen);
2273
2274                 for (start = h.raw; start < end; start += PAGE_SIZE)
2275                         flush_dcache_page(pgv_to_page(start));
2276         }
2277         smp_wmb();
2278 #endif
2279
2280         if (po->tp_version <= TPACKET_V2) {
2281                 __packet_set_status(po, h.raw, status);
2282                 sk->sk_data_ready(sk);
2283         } else {
2284                 prb_clear_blk_fill_status(&po->rx_ring);
2285         }
2286
2287 drop_n_restore:
2288         if (skb_head != skb->data && skb_shared(skb)) {
2289                 skb->data = skb_head;
2290                 skb->len = skb_len;
2291         }
2292 drop:
2293         kfree_skb(skb);
2294         return 0;
2295
2296 ring_is_full:
2297         po->stats.stats1.tp_drops++;
2298         spin_unlock(&sk->sk_receive_queue.lock);
2299
2300         sk->sk_data_ready(sk);
2301         kfree_skb(copy_skb);
2302         goto drop_n_restore;
2303 }
2304
2305 static void tpacket_destruct_skb(struct sk_buff *skb)
2306 {
2307         struct packet_sock *po = pkt_sk(skb->sk);
2308
2309         if (likely(po->tx_ring.pg_vec)) {
2310                 void *ph;
2311                 __u32 ts;
2312
2313                 ph = skb_shinfo(skb)->destructor_arg;
2314                 packet_dec_pending(&po->tx_ring);
2315
2316                 ts = __packet_set_timestamp(po, ph, skb);
2317                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2318         }
2319
2320         sock_wfree(skb);
2321 }
2322
2323 static bool ll_header_truncated(const struct net_device *dev, int len)
2324 {
2325         /* net device doesn't like empty head */
2326         if (unlikely(len <= dev->hard_header_len)) {
2327                 net_warn_ratelimited("%s: packet size is too short (%d <= %d)\n",
2328                                      current->comm, len, dev->hard_header_len);
2329                 return true;
2330         }
2331
2332         return false;
2333 }
2334
2335 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2336                 void *frame, struct net_device *dev, int size_max,
2337                 __be16 proto, unsigned char *addr, int hlen)
2338 {
2339         union tpacket_uhdr ph;
2340         int to_write, offset, len, tp_len, nr_frags, len_max;
2341         struct socket *sock = po->sk.sk_socket;
2342         struct page *page;
2343         void *data;
2344         int err;
2345
2346         ph.raw = frame;
2347
2348         skb->protocol = proto;
2349         skb->dev = dev;
2350         skb->priority = po->sk.sk_priority;
2351         skb->mark = po->sk.sk_mark;
2352         sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2353         skb_shinfo(skb)->destructor_arg = ph.raw;
2354
2355         switch (po->tp_version) {
2356         case TPACKET_V2:
2357                 tp_len = ph.h2->tp_len;
2358                 break;
2359         default:
2360                 tp_len = ph.h1->tp_len;
2361                 break;
2362         }
2363         if (unlikely(tp_len > size_max)) {
2364                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2365                 return -EMSGSIZE;
2366         }
2367
2368         skb_reserve(skb, hlen);
2369         skb_reset_network_header(skb);
2370
2371         if (!packet_use_direct_xmit(po))
2372                 skb_probe_transport_header(skb, 0);
2373         if (unlikely(po->tp_tx_has_off)) {
2374                 int off_min, off_max, off;
2375                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2376                 off_max = po->tx_ring.frame_size - tp_len;
2377                 if (sock->type == SOCK_DGRAM) {
2378                         switch (po->tp_version) {
2379                         case TPACKET_V2:
2380                                 off = ph.h2->tp_net;
2381                                 break;
2382                         default:
2383                                 off = ph.h1->tp_net;
2384                                 break;
2385                         }
2386                 } else {
2387                         switch (po->tp_version) {
2388                         case TPACKET_V2:
2389                                 off = ph.h2->tp_mac;
2390                                 break;
2391                         default:
2392                                 off = ph.h1->tp_mac;
2393                                 break;
2394                         }
2395                 }
2396                 if (unlikely((off < off_min) || (off_max < off)))
2397                         return -EINVAL;
2398                 data = ph.raw + off;
2399         } else {
2400                 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2401         }
2402         to_write = tp_len;
2403
2404         if (sock->type == SOCK_DGRAM) {
2405                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2406                                 NULL, tp_len);
2407                 if (unlikely(err < 0))
2408                         return -EINVAL;
2409         } else if (dev->hard_header_len) {
2410                 if (ll_header_truncated(dev, tp_len))
2411                         return -EINVAL;
2412
2413                 skb_push(skb, dev->hard_header_len);
2414                 err = skb_store_bits(skb, 0, data,
2415                                 dev->hard_header_len);
2416                 if (unlikely(err))
2417                         return err;
2418
2419                 data += dev->hard_header_len;
2420                 to_write -= dev->hard_header_len;
2421         }
2422
2423         offset = offset_in_page(data);
2424         len_max = PAGE_SIZE - offset;
2425         len = ((to_write > len_max) ? len_max : to_write);
2426
2427         skb->data_len = to_write;
2428         skb->len += to_write;
2429         skb->truesize += to_write;
2430         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2431
2432         while (likely(to_write)) {
2433                 nr_frags = skb_shinfo(skb)->nr_frags;
2434
2435                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2436                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2437                                MAX_SKB_FRAGS);
2438                         return -EFAULT;
2439                 }
2440
2441                 page = pgv_to_page(data);
2442                 data += len;
2443                 flush_dcache_page(page);
2444                 get_page(page);
2445                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2446                 to_write -= len;
2447                 offset = 0;
2448                 len_max = PAGE_SIZE;
2449                 len = ((to_write > len_max) ? len_max : to_write);
2450         }
2451
2452         return tp_len;
2453 }
2454
2455 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2456 {
2457         struct sk_buff *skb;
2458         struct net_device *dev;
2459         __be16 proto;
2460         int err, reserve = 0;
2461         void *ph;
2462         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2463         bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2464         int tp_len, size_max;
2465         unsigned char *addr;
2466         int len_sum = 0;
2467         int status = TP_STATUS_AVAILABLE;
2468         int hlen, tlen;
2469
2470         mutex_lock(&po->pg_vec_lock);
2471
2472         if (likely(saddr == NULL)) {
2473                 dev     = packet_cached_dev_get(po);
2474                 proto   = po->num;
2475                 addr    = NULL;
2476         } else {
2477                 err = -EINVAL;
2478                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2479                         goto out;
2480                 if (msg->msg_namelen < (saddr->sll_halen
2481                                         + offsetof(struct sockaddr_ll,
2482                                                 sll_addr)))
2483                         goto out;
2484                 proto   = saddr->sll_protocol;
2485                 addr    = saddr->sll_addr;
2486                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2487         }
2488
2489         err = -ENXIO;
2490         if (unlikely(dev == NULL))
2491                 goto out;
2492         err = -ENETDOWN;
2493         if (unlikely(!(dev->flags & IFF_UP)))
2494                 goto out_put;
2495
2496         reserve = dev->hard_header_len + VLAN_HLEN;
2497         size_max = po->tx_ring.frame_size
2498                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2499
2500         if (size_max > dev->mtu + reserve)
2501                 size_max = dev->mtu + reserve;
2502
2503         do {
2504                 ph = packet_current_frame(po, &po->tx_ring,
2505                                           TP_STATUS_SEND_REQUEST);
2506                 if (unlikely(ph == NULL)) {
2507                         if (need_wait && need_resched())
2508                                 schedule();
2509                         continue;
2510                 }
2511
2512                 status = TP_STATUS_SEND_REQUEST;
2513                 hlen = LL_RESERVED_SPACE(dev);
2514                 tlen = dev->needed_tailroom;
2515                 skb = sock_alloc_send_skb(&po->sk,
2516                                 hlen + tlen + sizeof(struct sockaddr_ll),
2517                                 !need_wait, &err);
2518
2519                 if (unlikely(skb == NULL)) {
2520                         /* we assume the socket was initially writeable ... */
2521                         if (likely(len_sum > 0))
2522                                 err = len_sum;
2523                         goto out_status;
2524                 }
2525                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2526                                           addr, hlen);
2527                 if (likely(tp_len >= 0) &&
2528                     tp_len > dev->mtu + dev->hard_header_len) {
2529                         struct ethhdr *ehdr;
2530                         /* Earlier code assumed this would be a VLAN pkt,
2531                          * double-check this now that we have the actual
2532                          * packet in hand.
2533                          */
2534
2535                         skb_reset_mac_header(skb);
2536                         ehdr = eth_hdr(skb);
2537                         if (ehdr->h_proto != htons(ETH_P_8021Q))
2538                                 tp_len = -EMSGSIZE;
2539                 }
2540                 if (unlikely(tp_len < 0)) {
2541                         if (po->tp_loss) {
2542                                 __packet_set_status(po, ph,
2543                                                 TP_STATUS_AVAILABLE);
2544                                 packet_increment_head(&po->tx_ring);
2545                                 kfree_skb(skb);
2546                                 continue;
2547                         } else {
2548                                 status = TP_STATUS_WRONG_FORMAT;
2549                                 err = tp_len;
2550                                 goto out_status;
2551                         }
2552                 }
2553
2554                 packet_pick_tx_queue(dev, skb);
2555
2556                 skb->destructor = tpacket_destruct_skb;
2557                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2558                 packet_inc_pending(&po->tx_ring);
2559
2560                 status = TP_STATUS_SEND_REQUEST;
2561                 err = po->xmit(skb);
2562                 if (unlikely(err > 0)) {
2563                         err = net_xmit_errno(err);
2564                         if (err && __packet_get_status(po, ph) ==
2565                                    TP_STATUS_AVAILABLE) {
2566                                 /* skb was destructed already */
2567                                 skb = NULL;
2568                                 goto out_status;
2569                         }
2570                         /*
2571                          * skb was dropped but not destructed yet;
2572                          * let's treat it like congestion or err < 0
2573                          */
2574                         err = 0;
2575                 }
2576                 packet_increment_head(&po->tx_ring);
2577                 len_sum += tp_len;
2578         } while (likely((ph != NULL) ||
2579                 /* Note: packet_read_pending() might be slow if we have
2580                  * to call it as it's per_cpu variable, but in fast-path
2581                  * we already short-circuit the loop with the first
2582                  * condition, and luckily don't have to go that path
2583                  * anyway.
2584                  */
2585                  (need_wait && packet_read_pending(&po->tx_ring))));
2586
2587         err = len_sum;
2588         goto out_put;
2589
2590 out_status:
2591         __packet_set_status(po, ph, status);
2592         kfree_skb(skb);
2593 out_put:
2594         dev_put(dev);
2595 out:
2596         mutex_unlock(&po->pg_vec_lock);
2597         return err;
2598 }
2599
2600 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2601                                         size_t reserve, size_t len,
2602                                         size_t linear, int noblock,
2603                                         int *err)
2604 {
2605         struct sk_buff *skb;
2606
2607         /* Under a page?  Don't bother with paged skb. */
2608         if (prepad + len < PAGE_SIZE || !linear)
2609                 linear = len;
2610
2611         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2612                                    err, 0);
2613         if (!skb)
2614                 return NULL;
2615
2616         skb_reserve(skb, reserve);
2617         skb_put(skb, linear);
2618         skb->data_len = len - linear;
2619         skb->len += len - linear;
2620
2621         return skb;
2622 }
2623
2624 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2625 {
2626         struct sock *sk = sock->sk;
2627         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2628         struct sk_buff *skb;
2629         struct net_device *dev;
2630         __be16 proto;
2631         unsigned char *addr;
2632         int err, reserve = 0;
2633         struct virtio_net_hdr vnet_hdr = { 0 };
2634         int offset = 0;
2635         int vnet_hdr_len;
2636         struct packet_sock *po = pkt_sk(sk);
2637         unsigned short gso_type = 0;
2638         int hlen, tlen;
2639         int extra_len = 0;
2640         ssize_t n;
2641
2642         /*
2643          *      Get and verify the address.
2644          */
2645
2646         if (likely(saddr == NULL)) {
2647                 dev     = packet_cached_dev_get(po);
2648                 proto   = po->num;
2649                 addr    = NULL;
2650         } else {
2651                 err = -EINVAL;
2652                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2653                         goto out;
2654                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2655                         goto out;
2656                 proto   = saddr->sll_protocol;
2657                 addr    = saddr->sll_addr;
2658                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2659         }
2660
2661         err = -ENXIO;
2662         if (unlikely(dev == NULL))
2663                 goto out_unlock;
2664         err = -ENETDOWN;
2665         if (unlikely(!(dev->flags & IFF_UP)))
2666                 goto out_unlock;
2667
2668         if (sock->type == SOCK_RAW)
2669                 reserve = dev->hard_header_len;
2670         if (po->has_vnet_hdr) {
2671                 vnet_hdr_len = sizeof(vnet_hdr);
2672
2673                 err = -EINVAL;
2674                 if (len < vnet_hdr_len)
2675                         goto out_unlock;
2676
2677                 len -= vnet_hdr_len;
2678
2679                 err = -EFAULT;
2680                 n = copy_from_iter(&vnet_hdr, vnet_hdr_len, &msg->msg_iter);
2681                 if (n != vnet_hdr_len)
2682                         goto out_unlock;
2683
2684                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2685                     (__virtio16_to_cpu(vio_le(), vnet_hdr.csum_start) +
2686                      __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset) + 2 >
2687                       __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len)))
2688                         vnet_hdr.hdr_len = __cpu_to_virtio16(vio_le(),
2689                                  __virtio16_to_cpu(vio_le(), vnet_hdr.csum_start) +
2690                                 __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset) + 2);
2691
2692                 err = -EINVAL;
2693                 if (__virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len) > len)
2694                         goto out_unlock;
2695
2696                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2697                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2698                         case VIRTIO_NET_HDR_GSO_TCPV4:
2699                                 gso_type = SKB_GSO_TCPV4;
2700                                 break;
2701                         case VIRTIO_NET_HDR_GSO_TCPV6:
2702                                 gso_type = SKB_GSO_TCPV6;
2703                                 break;
2704                         case VIRTIO_NET_HDR_GSO_UDP:
2705                                 gso_type = SKB_GSO_UDP;
2706                                 break;
2707                         default:
2708                                 goto out_unlock;
2709                         }
2710
2711                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2712                                 gso_type |= SKB_GSO_TCP_ECN;
2713
2714                         if (vnet_hdr.gso_size == 0)
2715                                 goto out_unlock;
2716
2717                 }
2718         }
2719
2720         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2721                 if (!netif_supports_nofcs(dev)) {
2722                         err = -EPROTONOSUPPORT;
2723                         goto out_unlock;
2724                 }
2725                 extra_len = 4; /* We're doing our own CRC */
2726         }
2727
2728         err = -EMSGSIZE;
2729         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2730                 goto out_unlock;
2731
2732         err = -ENOBUFS;
2733         hlen = LL_RESERVED_SPACE(dev);
2734         tlen = dev->needed_tailroom;
2735         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
2736                                __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len),
2737                                msg->msg_flags & MSG_DONTWAIT, &err);
2738         if (skb == NULL)
2739                 goto out_unlock;
2740
2741         skb_set_network_header(skb, reserve);
2742
2743         err = -EINVAL;
2744         if (sock->type == SOCK_DGRAM) {
2745                 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2746                 if (unlikely(offset < 0))
2747                         goto out_free;
2748         } else {
2749                 if (ll_header_truncated(dev, len))
2750                         goto out_free;
2751         }
2752
2753         /* Returns -EFAULT on error */
2754         err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2755         if (err)
2756                 goto out_free;
2757
2758         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2759
2760         if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2761                 /* Earlier code assumed this would be a VLAN pkt,
2762                  * double-check this now that we have the actual
2763                  * packet in hand.
2764                  */
2765                 struct ethhdr *ehdr;
2766                 skb_reset_mac_header(skb);
2767                 ehdr = eth_hdr(skb);
2768                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2769                         err = -EMSGSIZE;
2770                         goto out_free;
2771                 }
2772         }
2773
2774         skb->protocol = proto;
2775         skb->dev = dev;
2776         skb->priority = sk->sk_priority;
2777         skb->mark = sk->sk_mark;
2778
2779         packet_pick_tx_queue(dev, skb);
2780
2781         if (po->has_vnet_hdr) {
2782                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2783                         u16 s = __virtio16_to_cpu(vio_le(), vnet_hdr.csum_start);
2784                         u16 o = __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset);
2785                         if (!skb_partial_csum_set(skb, s, o)) {
2786                                 err = -EINVAL;
2787                                 goto out_free;
2788                         }
2789                 }
2790
2791                 skb_shinfo(skb)->gso_size =
2792                         __virtio16_to_cpu(vio_le(), vnet_hdr.gso_size);
2793                 skb_shinfo(skb)->gso_type = gso_type;
2794
2795                 /* Header must be checked, and gso_segs computed. */
2796                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2797                 skb_shinfo(skb)->gso_segs = 0;
2798
2799                 len += vnet_hdr_len;
2800         }
2801
2802         if (!packet_use_direct_xmit(po))
2803                 skb_probe_transport_header(skb, reserve);
2804         if (unlikely(extra_len == 4))
2805                 skb->no_fcs = 1;
2806
2807         err = po->xmit(skb);
2808         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2809                 goto out_unlock;
2810
2811         dev_put(dev);
2812
2813         return len;
2814
2815 out_free:
2816         kfree_skb(skb);
2817 out_unlock:
2818         if (dev)
2819                 dev_put(dev);
2820 out:
2821         return err;
2822 }
2823
2824 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2825 {
2826         struct sock *sk = sock->sk;
2827         struct packet_sock *po = pkt_sk(sk);
2828
2829         if (po->tx_ring.pg_vec)
2830                 return tpacket_snd(po, msg);
2831         else
2832                 return packet_snd(sock, msg, len);
2833 }
2834
2835 /*
2836  *      Close a PACKET socket. This is fairly simple. We immediately go
2837  *      to 'closed' state and remove our protocol entry in the device list.
2838  */
2839
2840 static int packet_release(struct socket *sock)
2841 {
2842         struct sock *sk = sock->sk;
2843         struct packet_sock *po;
2844         struct net *net;
2845         union tpacket_req_u req_u;
2846
2847         if (!sk)
2848                 return 0;
2849
2850         net = sock_net(sk);
2851         po = pkt_sk(sk);
2852
2853         mutex_lock(&net->packet.sklist_lock);
2854         sk_del_node_init_rcu(sk);
2855         mutex_unlock(&net->packet.sklist_lock);
2856
2857         preempt_disable();
2858         sock_prot_inuse_add(net, sk->sk_prot, -1);
2859         preempt_enable();
2860
2861         spin_lock(&po->bind_lock);
2862         unregister_prot_hook(sk, false);
2863         packet_cached_dev_reset(po);
2864
2865         if (po->prot_hook.dev) {
2866                 dev_put(po->prot_hook.dev);
2867                 po->prot_hook.dev = NULL;
2868         }
2869         spin_unlock(&po->bind_lock);
2870
2871         packet_flush_mclist(sk);
2872
2873         if (po->rx_ring.pg_vec) {
2874                 memset(&req_u, 0, sizeof(req_u));
2875                 packet_set_ring(sk, &req_u, 1, 0);
2876         }
2877
2878         if (po->tx_ring.pg_vec) {
2879                 memset(&req_u, 0, sizeof(req_u));
2880                 packet_set_ring(sk, &req_u, 1, 1);
2881         }
2882
2883         fanout_release(sk);
2884
2885         synchronize_net();
2886         /*
2887          *      Now the socket is dead. No more input will appear.
2888          */
2889         sock_orphan(sk);
2890         sock->sk = NULL;
2891
2892         /* Purge queues */
2893
2894         skb_queue_purge(&sk->sk_receive_queue);
2895         packet_free_pending(po);
2896         sk_refcnt_debug_release(sk);
2897
2898         sock_put(sk);
2899         return 0;
2900 }
2901
2902 /*
2903  *      Attach a packet hook.
2904  */
2905
2906 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 proto)
2907 {
2908         struct packet_sock *po = pkt_sk(sk);
2909         struct net_device *dev_curr;
2910         __be16 proto_curr;
2911         bool need_rehook;
2912
2913         if (po->fanout) {
2914                 if (dev)
2915                         dev_put(dev);
2916
2917                 return -EINVAL;
2918         }
2919
2920         lock_sock(sk);
2921         spin_lock(&po->bind_lock);
2922
2923         proto_curr = po->prot_hook.type;
2924         dev_curr = po->prot_hook.dev;
2925
2926         need_rehook = proto_curr != proto || dev_curr != dev;
2927
2928         if (need_rehook) {
2929                 unregister_prot_hook(sk, true);
2930
2931                 po->num = proto;
2932                 po->prot_hook.type = proto;
2933                 po->prot_hook.dev = dev;
2934
2935                 po->ifindex = dev ? dev->ifindex : 0;
2936                 packet_cached_dev_assign(po, dev);
2937         }
2938         if (dev_curr)
2939                 dev_put(dev_curr);
2940
2941         if (proto == 0 || !need_rehook)
2942                 goto out_unlock;
2943
2944         if (!dev || (dev->flags & IFF_UP)) {
2945                 register_prot_hook(sk);
2946         } else {
2947                 sk->sk_err = ENETDOWN;
2948                 if (!sock_flag(sk, SOCK_DEAD))
2949                         sk->sk_error_report(sk);
2950         }
2951
2952 out_unlock:
2953         spin_unlock(&po->bind_lock);
2954         release_sock(sk);
2955         return 0;
2956 }
2957
2958 /*
2959  *      Bind a packet socket to a device
2960  */
2961
2962 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2963                             int addr_len)
2964 {
2965         struct sock *sk = sock->sk;
2966         char name[15];
2967         struct net_device *dev;
2968         int err = -ENODEV;
2969
2970         /*
2971          *      Check legality
2972          */
2973
2974         if (addr_len != sizeof(struct sockaddr))
2975                 return -EINVAL;
2976         strlcpy(name, uaddr->sa_data, sizeof(name));
2977
2978         dev = dev_get_by_name(sock_net(sk), name);
2979         if (dev)
2980                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2981         return err;
2982 }
2983
2984 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2985 {
2986         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2987         struct sock *sk = sock->sk;
2988         struct net_device *dev = NULL;
2989         int err;
2990
2991
2992         /*
2993          *      Check legality
2994          */
2995
2996         if (addr_len < sizeof(struct sockaddr_ll))
2997                 return -EINVAL;
2998         if (sll->sll_family != AF_PACKET)
2999                 return -EINVAL;
3000
3001         if (sll->sll_ifindex) {
3002                 err = -ENODEV;
3003                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
3004                 if (dev == NULL)
3005                         goto out;
3006         }
3007         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
3008
3009 out:
3010         return err;
3011 }
3012
3013 static struct proto packet_proto = {
3014         .name     = "PACKET",
3015         .owner    = THIS_MODULE,
3016         .obj_size = sizeof(struct packet_sock),
3017 };
3018
3019 /*
3020  *      Create a packet of type SOCK_PACKET.
3021  */
3022
3023 static int packet_create(struct net *net, struct socket *sock, int protocol,
3024                          int kern)
3025 {
3026         struct sock *sk;
3027         struct packet_sock *po;
3028         __be16 proto = (__force __be16)protocol; /* weird, but documented */
3029         int err;
3030
3031         if (!ns_capable(net->user_ns, CAP_NET_RAW))
3032                 return -EPERM;
3033         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3034             sock->type != SOCK_PACKET)
3035                 return -ESOCKTNOSUPPORT;
3036
3037         sock->state = SS_UNCONNECTED;
3038
3039         err = -ENOBUFS;
3040         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3041         if (sk == NULL)
3042                 goto out;
3043
3044         sock->ops = &packet_ops;
3045         if (sock->type == SOCK_PACKET)
3046                 sock->ops = &packet_ops_spkt;
3047
3048         sock_init_data(sock, sk);
3049
3050         po = pkt_sk(sk);
3051         sk->sk_family = PF_PACKET;
3052         po->num = proto;
3053         po->xmit = dev_queue_xmit;
3054
3055         err = packet_alloc_pending(po);
3056         if (err)
3057                 goto out2;
3058
3059         packet_cached_dev_reset(po);
3060
3061         sk->sk_destruct = packet_sock_destruct;
3062         sk_refcnt_debug_inc(sk);
3063
3064         /*
3065          *      Attach a protocol block
3066          */
3067
3068         spin_lock_init(&po->bind_lock);
3069         mutex_init(&po->pg_vec_lock);
3070         po->rollover = NULL;
3071         po->prot_hook.func = packet_rcv;
3072
3073         if (sock->type == SOCK_PACKET)
3074                 po->prot_hook.func = packet_rcv_spkt;
3075
3076         po->prot_hook.af_packet_priv = sk;
3077
3078         if (proto) {
3079                 po->prot_hook.type = proto;
3080                 register_prot_hook(sk);
3081         }
3082
3083         mutex_lock(&net->packet.sklist_lock);
3084         sk_add_node_rcu(sk, &net->packet.sklist);
3085         mutex_unlock(&net->packet.sklist_lock);
3086
3087         preempt_disable();
3088         sock_prot_inuse_add(net, &packet_proto, 1);
3089         preempt_enable();
3090
3091         return 0;
3092 out2:
3093         sk_free(sk);
3094 out:
3095         return err;
3096 }
3097
3098 /*
3099  *      Pull a packet from our receive queue and hand it to the user.
3100  *      If necessary we block.
3101  */
3102
3103 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3104                           int flags)
3105 {
3106         struct sock *sk = sock->sk;
3107         struct sk_buff *skb;
3108         int copied, err;
3109         int vnet_hdr_len = 0;
3110         unsigned int origlen = 0;
3111
3112         err = -EINVAL;
3113         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3114                 goto out;
3115
3116 #if 0
3117         /* What error should we return now? EUNATTACH? */
3118         if (pkt_sk(sk)->ifindex < 0)
3119                 return -ENODEV;
3120 #endif
3121
3122         if (flags & MSG_ERRQUEUE) {
3123                 err = sock_recv_errqueue(sk, msg, len,
3124                                          SOL_PACKET, PACKET_TX_TIMESTAMP);
3125                 goto out;
3126         }
3127
3128         /*
3129          *      Call the generic datagram receiver. This handles all sorts
3130          *      of horrible races and re-entrancy so we can forget about it
3131          *      in the protocol layers.
3132          *
3133          *      Now it will return ENETDOWN, if device have just gone down,
3134          *      but then it will block.
3135          */
3136
3137         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3138
3139         /*
3140          *      An error occurred so return it. Because skb_recv_datagram()
3141          *      handles the blocking we don't see and worry about blocking
3142          *      retries.
3143          */
3144
3145         if (skb == NULL)
3146                 goto out;
3147
3148         if (pkt_sk(sk)->pressure)
3149                 packet_rcv_has_room(pkt_sk(sk), NULL);
3150
3151         if (pkt_sk(sk)->has_vnet_hdr) {
3152                 struct virtio_net_hdr vnet_hdr = { 0 };
3153
3154                 err = -EINVAL;
3155                 vnet_hdr_len = sizeof(vnet_hdr);
3156                 if (len < vnet_hdr_len)
3157                         goto out_free;
3158
3159                 len -= vnet_hdr_len;
3160
3161                 if (skb_is_gso(skb)) {
3162                         struct skb_shared_info *sinfo = skb_shinfo(skb);
3163
3164                         /* This is a hint as to how much should be linear. */
3165                         vnet_hdr.hdr_len =
3166                                 __cpu_to_virtio16(vio_le(), skb_headlen(skb));
3167                         vnet_hdr.gso_size =
3168                                 __cpu_to_virtio16(vio_le(), sinfo->gso_size);
3169                         if (sinfo->gso_type & SKB_GSO_TCPV4)
3170                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
3171                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
3172                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
3173                         else if (sinfo->gso_type & SKB_GSO_UDP)
3174                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
3175                         else if (sinfo->gso_type & SKB_GSO_FCOE)
3176                                 goto out_free;
3177                         else
3178                                 BUG();
3179                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
3180                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
3181                 } else
3182                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
3183
3184                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3185                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
3186                         vnet_hdr.csum_start = __cpu_to_virtio16(vio_le(),
3187                                           skb_checksum_start_offset(skb));
3188                         vnet_hdr.csum_offset = __cpu_to_virtio16(vio_le(),
3189                                                          skb->csum_offset);
3190                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3191                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
3192                 } /* else everything is zero */
3193
3194                 err = memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_len);
3195                 if (err < 0)
3196                         goto out_free;
3197         }
3198
3199         /* You lose any data beyond the buffer you gave. If it worries
3200          * a user program they can ask the device for its MTU
3201          * anyway.
3202          */
3203         copied = skb->len;
3204         if (copied > len) {
3205                 copied = len;
3206                 msg->msg_flags |= MSG_TRUNC;
3207         }
3208
3209         err = skb_copy_datagram_msg(skb, 0, msg, copied);
3210         if (err)
3211                 goto out_free;
3212
3213         if (sock->type != SOCK_PACKET) {
3214                 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3215
3216                 /* Original length was stored in sockaddr_ll fields */
3217                 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3218                 sll->sll_family = AF_PACKET;
3219                 sll->sll_protocol = skb->protocol;
3220         }
3221
3222         sock_recv_ts_and_drops(msg, sk, skb);
3223
3224         if (msg->msg_name) {
3225                 /* If the address length field is there to be filled
3226                  * in, we fill it in now.
3227                  */
3228                 if (sock->type == SOCK_PACKET) {
3229                         __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3230                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
3231                 } else {
3232                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3233
3234                         msg->msg_namelen = sll->sll_halen +
3235                                 offsetof(struct sockaddr_ll, sll_addr);
3236                 }
3237                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3238                        msg->msg_namelen);
3239         }
3240
3241         if (pkt_sk(sk)->auxdata) {
3242                 struct tpacket_auxdata aux;
3243
3244                 aux.tp_status = TP_STATUS_USER;
3245                 if (skb->ip_summed == CHECKSUM_PARTIAL)
3246                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3247                 else if (skb->pkt_type != PACKET_OUTGOING &&
3248                          (skb->ip_summed == CHECKSUM_COMPLETE ||
3249                           skb_csum_unnecessary(skb)))
3250                         aux.tp_status |= TP_STATUS_CSUM_VALID;
3251
3252                 aux.tp_len = origlen;
3253                 aux.tp_snaplen = skb->len;
3254                 aux.tp_mac = 0;
3255                 aux.tp_net = skb_network_offset(skb);
3256                 if (skb_vlan_tag_present(skb)) {
3257                         aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3258                         aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3259                         aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3260                 } else {
3261                         aux.tp_vlan_tci = 0;
3262                         aux.tp_vlan_tpid = 0;
3263                 }
3264                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3265         }
3266
3267         /*
3268          *      Free or return the buffer as appropriate. Again this
3269          *      hides all the races and re-entrancy issues from us.
3270          */
3271         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3272
3273 out_free:
3274         skb_free_datagram(sk, skb);
3275 out:
3276         return err;
3277 }
3278
3279 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3280                                int *uaddr_len, int peer)
3281 {
3282         struct net_device *dev;
3283         struct sock *sk = sock->sk;
3284
3285         if (peer)
3286                 return -EOPNOTSUPP;
3287
3288         uaddr->sa_family = AF_PACKET;
3289         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3290         rcu_read_lock();
3291         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3292         if (dev)
3293                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3294         rcu_read_unlock();
3295         *uaddr_len = sizeof(*uaddr);
3296
3297         return 0;
3298 }
3299
3300 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3301                           int *uaddr_len, int peer)
3302 {
3303         struct net_device *dev;
3304         struct sock *sk = sock->sk;
3305         struct packet_sock *po = pkt_sk(sk);
3306         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3307
3308         if (peer)
3309                 return -EOPNOTSUPP;
3310
3311         sll->sll_family = AF_PACKET;
3312         sll->sll_ifindex = po->ifindex;
3313         sll->sll_protocol = po->num;
3314         sll->sll_pkttype = 0;
3315         rcu_read_lock();
3316         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3317         if (dev) {
3318                 sll->sll_hatype = dev->type;
3319                 sll->sll_halen = dev->addr_len;
3320                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3321         } else {
3322                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
3323                 sll->sll_halen = 0;
3324         }
3325         rcu_read_unlock();
3326         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3327
3328         return 0;
3329 }
3330
3331 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3332                          int what)
3333 {
3334         switch (i->type) {
3335         case PACKET_MR_MULTICAST:
3336                 if (i->alen != dev->addr_len)
3337                         return -EINVAL;
3338                 if (what > 0)
3339                         return dev_mc_add(dev, i->addr);
3340                 else
3341                         return dev_mc_del(dev, i->addr);
3342                 break;
3343         case PACKET_MR_PROMISC:
3344                 return dev_set_promiscuity(dev, what);
3345         case PACKET_MR_ALLMULTI:
3346                 return dev_set_allmulti(dev, what);
3347         case PACKET_MR_UNICAST:
3348                 if (i->alen != dev->addr_len)
3349                         return -EINVAL;
3350                 if (what > 0)
3351                         return dev_uc_add(dev, i->addr);
3352                 else
3353                         return dev_uc_del(dev, i->addr);
3354                 break;
3355         default:
3356                 break;
3357         }
3358         return 0;
3359 }
3360
3361 static void packet_dev_mclist_delete(struct net_device *dev,
3362                                      struct packet_mclist **mlp)
3363 {
3364         struct packet_mclist *ml;
3365
3366         while ((ml = *mlp) != NULL) {
3367                 if (ml->ifindex == dev->ifindex) {
3368                         packet_dev_mc(dev, ml, -1);
3369                         *mlp = ml->next;
3370                         kfree(ml);
3371                 } else
3372                         mlp = &ml->next;
3373         }
3374 }
3375
3376 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3377 {
3378         struct packet_sock *po = pkt_sk(sk);
3379         struct packet_mclist *ml, *i;
3380         struct net_device *dev;
3381         int err;
3382
3383         rtnl_lock();
3384
3385         err = -ENODEV;
3386         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3387         if (!dev)
3388                 goto done;
3389
3390         err = -EINVAL;
3391         if (mreq->mr_alen > dev->addr_len)
3392                 goto done;
3393
3394         err = -ENOBUFS;
3395         i = kmalloc(sizeof(*i), GFP_KERNEL);
3396         if (i == NULL)
3397                 goto done;
3398
3399         err = 0;
3400         for (ml = po->mclist; ml; ml = ml->next) {
3401                 if (ml->ifindex == mreq->mr_ifindex &&
3402                     ml->type == mreq->mr_type &&
3403                     ml->alen == mreq->mr_alen &&
3404                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3405                         ml->count++;
3406                         /* Free the new element ... */
3407                         kfree(i);
3408                         goto done;
3409                 }
3410         }
3411
3412         i->type = mreq->mr_type;
3413         i->ifindex = mreq->mr_ifindex;
3414         i->alen = mreq->mr_alen;
3415         memcpy(i->addr, mreq->mr_address, i->alen);
3416         i->count = 1;
3417         i->next = po->mclist;
3418         po->mclist = i;
3419         err = packet_dev_mc(dev, i, 1);
3420         if (err) {
3421                 po->mclist = i->next;
3422                 kfree(i);
3423         }
3424
3425 done:
3426         rtnl_unlock();
3427         return err;
3428 }
3429
3430 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3431 {
3432         struct packet_mclist *ml, **mlp;
3433
3434         rtnl_lock();
3435
3436         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3437                 if (ml->ifindex == mreq->mr_ifindex &&
3438                     ml->type == mreq->mr_type &&
3439                     ml->alen == mreq->mr_alen &&
3440                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3441                         if (--ml->count == 0) {
3442                                 struct net_device *dev;
3443                                 *mlp = ml->next;
3444                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3445                                 if (dev)
3446                                         packet_dev_mc(dev, ml, -1);
3447                                 kfree(ml);
3448                         }
3449                         break;
3450                 }
3451         }
3452         rtnl_unlock();
3453         return 0;
3454 }
3455
3456 static void packet_flush_mclist(struct sock *sk)
3457 {
3458         struct packet_sock *po = pkt_sk(sk);
3459         struct packet_mclist *ml;
3460
3461         if (!po->mclist)
3462                 return;
3463
3464         rtnl_lock();
3465         while ((ml = po->mclist) != NULL) {
3466                 struct net_device *dev;
3467
3468                 po->mclist = ml->next;
3469                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3470                 if (dev != NULL)
3471                         packet_dev_mc(dev, ml, -1);
3472                 kfree(ml);
3473         }
3474         rtnl_unlock();
3475 }
3476
3477 static int
3478 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3479 {
3480         struct sock *sk = sock->sk;
3481         struct packet_sock *po = pkt_sk(sk);
3482         int ret;
3483
3484         if (level != SOL_PACKET)
3485                 return -ENOPROTOOPT;
3486
3487         switch (optname) {
3488         case PACKET_ADD_MEMBERSHIP:
3489         case PACKET_DROP_MEMBERSHIP:
3490         {
3491                 struct packet_mreq_max mreq;
3492                 int len = optlen;
3493                 memset(&mreq, 0, sizeof(mreq));
3494                 if (len < sizeof(struct packet_mreq))
3495                         return -EINVAL;
3496                 if (len > sizeof(mreq))
3497                         len = sizeof(mreq);
3498                 if (copy_from_user(&mreq, optval, len))
3499                         return -EFAULT;
3500                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3501                         return -EINVAL;
3502                 if (optname == PACKET_ADD_MEMBERSHIP)
3503                         ret = packet_mc_add(sk, &mreq);
3504                 else
3505                         ret = packet_mc_drop(sk, &mreq);
3506                 return ret;
3507         }
3508
3509         case PACKET_RX_RING:
3510         case PACKET_TX_RING:
3511         {
3512                 union tpacket_req_u req_u;
3513                 int len;
3514
3515                 switch (po->tp_version) {
3516                 case TPACKET_V1:
3517                 case TPACKET_V2:
3518                         len = sizeof(req_u.req);
3519                         break;
3520                 case TPACKET_V3:
3521                 default:
3522                         len = sizeof(req_u.req3);
3523                         break;
3524                 }
3525                 if (optlen < len)
3526                         return -EINVAL;
3527                 if (pkt_sk(sk)->has_vnet_hdr)
3528                         return -EINVAL;
3529                 if (copy_from_user(&req_u.req, optval, len))
3530                         return -EFAULT;
3531                 return packet_set_ring(sk, &req_u, 0,
3532                         optname == PACKET_TX_RING);
3533         }
3534         case PACKET_COPY_THRESH:
3535         {
3536                 int val;
3537
3538                 if (optlen != sizeof(val))
3539                         return -EINVAL;
3540                 if (copy_from_user(&val, optval, sizeof(val)))
3541                         return -EFAULT;
3542
3543                 pkt_sk(sk)->copy_thresh = val;
3544                 return 0;
3545         }
3546         case PACKET_VERSION:
3547         {
3548                 int val;
3549
3550                 if (optlen != sizeof(val))
3551                         return -EINVAL;
3552                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3553                         return -EBUSY;
3554                 if (copy_from_user(&val, optval, sizeof(val)))
3555                         return -EFAULT;
3556                 switch (val) {
3557                 case TPACKET_V1:
3558                 case TPACKET_V2:
3559                 case TPACKET_V3:
3560                         po->tp_version = val;
3561                         return 0;
3562                 default:
3563                         return -EINVAL;
3564                 }
3565         }
3566         case PACKET_RESERVE:
3567         {
3568                 unsigned int val;
3569
3570                 if (optlen != sizeof(val))
3571                         return -EINVAL;
3572                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3573                         return -EBUSY;
3574                 if (copy_from_user(&val, optval, sizeof(val)))
3575                         return -EFAULT;
3576                 po->tp_reserve = val;
3577                 return 0;
3578         }
3579         case PACKET_LOSS:
3580         {
3581                 unsigned int val;
3582
3583                 if (optlen != sizeof(val))
3584                         return -EINVAL;
3585                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3586                         return -EBUSY;
3587                 if (copy_from_user(&val, optval, sizeof(val)))
3588                         return -EFAULT;
3589                 po->tp_loss = !!val;
3590                 return 0;
3591         }
3592         case PACKET_AUXDATA:
3593         {
3594                 int val;
3595
3596                 if (optlen < sizeof(val))
3597                         return -EINVAL;
3598                 if (copy_from_user(&val, optval, sizeof(val)))
3599                         return -EFAULT;
3600
3601                 po->auxdata = !!val;
3602                 return 0;
3603         }
3604         case PACKET_ORIGDEV:
3605         {
3606                 int val;
3607
3608                 if (optlen < sizeof(val))
3609                         return -EINVAL;
3610                 if (copy_from_user(&val, optval, sizeof(val)))
3611                         return -EFAULT;
3612
3613                 po->origdev = !!val;
3614                 return 0;
3615         }
3616         case PACKET_VNET_HDR:
3617         {
3618                 int val;
3619
3620                 if (sock->type != SOCK_RAW)
3621                         return -EINVAL;
3622                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3623                         return -EBUSY;
3624                 if (optlen < sizeof(val))
3625                         return -EINVAL;
3626                 if (copy_from_user(&val, optval, sizeof(val)))
3627                         return -EFAULT;
3628
3629                 po->has_vnet_hdr = !!val;
3630                 return 0;
3631         }
3632         case PACKET_TIMESTAMP:
3633         {
3634                 int val;
3635
3636                 if (optlen != sizeof(val))
3637                         return -EINVAL;
3638                 if (copy_from_user(&val, optval, sizeof(val)))
3639                         return -EFAULT;
3640
3641                 po->tp_tstamp = val;
3642                 return 0;
3643         }
3644         case PACKET_FANOUT:
3645         {
3646                 int val;
3647
3648                 if (optlen != sizeof(val))
3649                         return -EINVAL;
3650                 if (copy_from_user(&val, optval, sizeof(val)))
3651                         return -EFAULT;
3652
3653                 return fanout_add(sk, val & 0xffff, val >> 16);
3654         }
3655         case PACKET_FANOUT_DATA:
3656         {
3657                 if (!po->fanout)
3658                         return -EINVAL;
3659
3660                 return fanout_set_data(po, optval, optlen);
3661         }
3662         case PACKET_TX_HAS_OFF:
3663         {
3664                 unsigned int val;
3665
3666                 if (optlen != sizeof(val))
3667                         return -EINVAL;
3668                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3669                         return -EBUSY;
3670                 if (copy_from_user(&val, optval, sizeof(val)))
3671                         return -EFAULT;
3672                 po->tp_tx_has_off = !!val;
3673                 return 0;
3674         }
3675         case PACKET_QDISC_BYPASS:
3676         {
3677                 int val;
3678
3679                 if (optlen != sizeof(val))
3680                         return -EINVAL;
3681                 if (copy_from_user(&val, optval, sizeof(val)))
3682                         return -EFAULT;
3683
3684                 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3685                 return 0;
3686         }
3687         default:
3688                 return -ENOPROTOOPT;
3689         }
3690 }
3691
3692 static int packet_getsockopt(struct socket *sock, int level, int optname,
3693                              char __user *optval, int __user *optlen)
3694 {
3695         int len;
3696         int val, lv = sizeof(val);
3697         struct sock *sk = sock->sk;
3698         struct packet_sock *po = pkt_sk(sk);
3699         void *data = &val;
3700         union tpacket_stats_u st;
3701         struct tpacket_rollover_stats rstats;
3702
3703         if (level != SOL_PACKET)
3704                 return -ENOPROTOOPT;
3705
3706         if (get_user(len, optlen))
3707                 return -EFAULT;
3708
3709         if (len < 0)
3710                 return -EINVAL;
3711
3712         switch (optname) {
3713         case PACKET_STATISTICS:
3714                 spin_lock_bh(&sk->sk_receive_queue.lock);
3715                 memcpy(&st, &po->stats, sizeof(st));
3716                 memset(&po->stats, 0, sizeof(po->stats));
3717                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3718
3719                 if (po->tp_version == TPACKET_V3) {
3720                         lv = sizeof(struct tpacket_stats_v3);
3721                         st.stats3.tp_packets += st.stats3.tp_drops;
3722                         data = &st.stats3;
3723                 } else {
3724                         lv = sizeof(struct tpacket_stats);
3725                         st.stats1.tp_packets += st.stats1.tp_drops;
3726                         data = &st.stats1;
3727                 }
3728
3729                 break;
3730         case PACKET_AUXDATA:
3731                 val = po->auxdata;
3732                 break;
3733         case PACKET_ORIGDEV:
3734                 val = po->origdev;
3735                 break;
3736         case PACKET_VNET_HDR:
3737                 val = po->has_vnet_hdr;
3738                 break;
3739         case PACKET_VERSION:
3740                 val = po->tp_version;
3741                 break;
3742         case PACKET_HDRLEN:
3743                 if (len > sizeof(int))
3744                         len = sizeof(int);
3745                 if (copy_from_user(&val, optval, len))
3746                         return -EFAULT;
3747                 switch (val) {
3748                 case TPACKET_V1:
3749                         val = sizeof(struct tpacket_hdr);
3750                         break;
3751                 case TPACKET_V2:
3752                         val = sizeof(struct tpacket2_hdr);
3753                         break;
3754                 case TPACKET_V3:
3755                         val = sizeof(struct tpacket3_hdr);
3756                         break;
3757                 default:
3758                         return -EINVAL;
3759                 }
3760                 break;
3761         case PACKET_RESERVE:
3762                 val = po->tp_reserve;
3763                 break;
3764         case PACKET_LOSS:
3765                 val = po->tp_loss;
3766                 break;
3767         case PACKET_TIMESTAMP:
3768                 val = po->tp_tstamp;
3769                 break;
3770         case PACKET_FANOUT:
3771                 val = (po->fanout ?
3772                        ((u32)po->fanout->id |
3773                         ((u32)po->fanout->type << 16) |
3774                         ((u32)po->fanout->flags << 24)) :
3775                        0);
3776                 break;
3777         case PACKET_ROLLOVER_STATS:
3778                 if (!po->rollover)
3779                         return -EINVAL;
3780                 rstats.tp_all = atomic_long_read(&po->rollover->num);
3781                 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3782                 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3783                 data = &rstats;
3784                 lv = sizeof(rstats);
3785                 break;
3786         case PACKET_TX_HAS_OFF:
3787                 val = po->tp_tx_has_off;
3788                 break;
3789         case PACKET_QDISC_BYPASS:
3790                 val = packet_use_direct_xmit(po);
3791                 break;
3792         default:
3793                 return -ENOPROTOOPT;
3794         }
3795
3796         if (len > lv)
3797                 len = lv;
3798         if (put_user(len, optlen))
3799                 return -EFAULT;
3800         if (copy_to_user(optval, data, len))
3801                 return -EFAULT;
3802         return 0;
3803 }
3804
3805
3806 static int packet_notifier(struct notifier_block *this,
3807                            unsigned long msg, void *ptr)
3808 {
3809         struct sock *sk;
3810         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3811         struct net *net = dev_net(dev);
3812
3813         rcu_read_lock();
3814         sk_for_each_rcu(sk, &net->packet.sklist) {
3815                 struct packet_sock *po = pkt_sk(sk);
3816
3817                 switch (msg) {
3818                 case NETDEV_UNREGISTER:
3819                         if (po->mclist)
3820                                 packet_dev_mclist_delete(dev, &po->mclist);
3821                         /* fallthrough */
3822
3823                 case NETDEV_DOWN:
3824                         if (dev->ifindex == po->ifindex) {
3825                                 spin_lock(&po->bind_lock);
3826                                 if (po->running) {
3827                                         __unregister_prot_hook(sk, false);
3828                                         sk->sk_err = ENETDOWN;
3829                                         if (!sock_flag(sk, SOCK_DEAD))
3830                                                 sk->sk_error_report(sk);
3831                                 }
3832                                 if (msg == NETDEV_UNREGISTER) {
3833                                         packet_cached_dev_reset(po);
3834                                         po->ifindex = -1;
3835                                         if (po->prot_hook.dev)
3836                                                 dev_put(po->prot_hook.dev);
3837                                         po->prot_hook.dev = NULL;
3838                                 }
3839                                 spin_unlock(&po->bind_lock);
3840                         }
3841                         break;
3842                 case NETDEV_UP:
3843                         if (dev->ifindex == po->ifindex) {
3844                                 spin_lock(&po->bind_lock);
3845                                 if (po->num)
3846                                         register_prot_hook(sk);
3847                                 spin_unlock(&po->bind_lock);
3848                         }
3849                         break;
3850                 }
3851         }
3852         rcu_read_unlock();
3853         return NOTIFY_DONE;
3854 }
3855
3856
3857 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3858                         unsigned long arg)
3859 {
3860         struct sock *sk = sock->sk;
3861
3862         switch (cmd) {
3863         case SIOCOUTQ:
3864         {
3865                 int amount = sk_wmem_alloc_get(sk);
3866
3867                 return put_user(amount, (int __user *)arg);
3868         }
3869         case SIOCINQ:
3870         {
3871                 struct sk_buff *skb;
3872                 int amount = 0;
3873
3874                 spin_lock_bh(&sk->sk_receive_queue.lock);
3875                 skb = skb_peek(&sk->sk_receive_queue);
3876                 if (skb)
3877                         amount = skb->len;
3878                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3879                 return put_user(amount, (int __user *)arg);
3880         }
3881         case SIOCGSTAMP:
3882                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3883         case SIOCGSTAMPNS:
3884                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3885
3886 #ifdef CONFIG_INET
3887         case SIOCADDRT:
3888         case SIOCDELRT:
3889         case SIOCDARP:
3890         case SIOCGARP:
3891         case SIOCSARP:
3892         case SIOCGIFADDR:
3893         case SIOCSIFADDR:
3894         case SIOCGIFBRDADDR:
3895         case SIOCSIFBRDADDR:
3896         case SIOCGIFNETMASK:
3897         case SIOCSIFNETMASK:
3898         case SIOCGIFDSTADDR:
3899         case SIOCSIFDSTADDR:
3900         case SIOCSIFFLAGS:
3901                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3902 #endif
3903
3904         default:
3905                 return -ENOIOCTLCMD;
3906         }
3907         return 0;
3908 }
3909
3910 static unsigned int packet_poll(struct file *file, struct socket *sock,
3911                                 poll_table *wait)
3912 {
3913         struct sock *sk = sock->sk;
3914         struct packet_sock *po = pkt_sk(sk);
3915         unsigned int mask = datagram_poll(file, sock, wait);
3916
3917         spin_lock_bh(&sk->sk_receive_queue.lock);
3918         if (po->rx_ring.pg_vec) {
3919                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3920                         TP_STATUS_KERNEL))
3921                         mask |= POLLIN | POLLRDNORM;
3922         }
3923         if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
3924                 po->pressure = 0;
3925         spin_unlock_bh(&sk->sk_receive_queue.lock);
3926         spin_lock_bh(&sk->sk_write_queue.lock);
3927         if (po->tx_ring.pg_vec) {
3928                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3929                         mask |= POLLOUT | POLLWRNORM;
3930         }
3931         spin_unlock_bh(&sk->sk_write_queue.lock);
3932         return mask;
3933 }
3934
3935
3936 /* Dirty? Well, I still did not learn better way to account
3937  * for user mmaps.
3938  */
3939
3940 static void packet_mm_open(struct vm_area_struct *vma)
3941 {
3942         struct file *file = vma->vm_file;
3943         struct socket *sock = file->private_data;
3944         struct sock *sk = sock->sk;
3945
3946         if (sk)
3947                 atomic_inc(&pkt_sk(sk)->mapped);
3948 }
3949
3950 static void packet_mm_close(struct vm_area_struct *vma)
3951 {
3952         struct file *file = vma->vm_file;
3953         struct socket *sock = file->private_data;
3954         struct sock *sk = sock->sk;
3955
3956         if (sk)
3957                 atomic_dec(&pkt_sk(sk)->mapped);
3958 }
3959
3960 static const struct vm_operations_struct packet_mmap_ops = {
3961         .open   =       packet_mm_open,
3962         .close  =       packet_mm_close,
3963 };
3964
3965 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3966                         unsigned int len)
3967 {
3968         int i;
3969
3970         for (i = 0; i < len; i++) {
3971                 if (likely(pg_vec[i].buffer)) {
3972                         if (is_vmalloc_addr(pg_vec[i].buffer))
3973                                 vfree(pg_vec[i].buffer);
3974                         else
3975                                 free_pages((unsigned long)pg_vec[i].buffer,
3976                                            order);
3977                         pg_vec[i].buffer = NULL;
3978                 }
3979         }
3980         kfree(pg_vec);
3981 }
3982
3983 static char *alloc_one_pg_vec_page(unsigned long order)
3984 {
3985         char *buffer;
3986         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3987                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3988
3989         buffer = (char *) __get_free_pages(gfp_flags, order);
3990         if (buffer)
3991                 return buffer;
3992
3993         /* __get_free_pages failed, fall back to vmalloc */
3994         buffer = vzalloc((1 << order) * PAGE_SIZE);
3995         if (buffer)
3996                 return buffer;
3997
3998         /* vmalloc failed, lets dig into swap here */
3999         gfp_flags &= ~__GFP_NORETRY;
4000         buffer = (char *) __get_free_pages(gfp_flags, order);
4001         if (buffer)
4002                 return buffer;
4003
4004         /* complete and utter failure */
4005         return NULL;
4006 }
4007
4008 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4009 {
4010         unsigned int block_nr = req->tp_block_nr;
4011         struct pgv *pg_vec;
4012         int i;
4013
4014         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4015         if (unlikely(!pg_vec))
4016                 goto out;
4017
4018         for (i = 0; i < block_nr; i++) {
4019                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4020                 if (unlikely(!pg_vec[i].buffer))
4021                         goto out_free_pgvec;
4022         }
4023
4024 out:
4025         return pg_vec;
4026
4027 out_free_pgvec:
4028         free_pg_vec(pg_vec, order, block_nr);
4029         pg_vec = NULL;
4030         goto out;
4031 }
4032
4033 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4034                 int closing, int tx_ring)
4035 {
4036         struct pgv *pg_vec = NULL;
4037         struct packet_sock *po = pkt_sk(sk);
4038         int was_running, order = 0;
4039         struct packet_ring_buffer *rb;
4040         struct sk_buff_head *rb_queue;
4041         __be16 num;
4042         int err = -EINVAL;
4043         /* Added to avoid minimal code churn */
4044         struct tpacket_req *req = &req_u->req;
4045
4046         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
4047         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
4048                 WARN(1, "Tx-ring is not supported.\n");
4049                 goto out;
4050         }
4051
4052         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4053         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4054
4055         err = -EBUSY;
4056         if (!closing) {
4057                 if (atomic_read(&po->mapped))
4058                         goto out;
4059                 if (packet_read_pending(rb))
4060                         goto out;
4061         }
4062
4063         if (req->tp_block_nr) {
4064                 /* Sanity tests and some calculations */
4065                 err = -EBUSY;
4066                 if (unlikely(rb->pg_vec))
4067                         goto out;
4068
4069                 switch (po->tp_version) {
4070                 case TPACKET_V1:
4071                         po->tp_hdrlen = TPACKET_HDRLEN;
4072                         break;
4073                 case TPACKET_V2:
4074                         po->tp_hdrlen = TPACKET2_HDRLEN;
4075                         break;
4076                 case TPACKET_V3:
4077                         po->tp_hdrlen = TPACKET3_HDRLEN;
4078                         break;
4079                 }
4080
4081                 err = -EINVAL;
4082                 if (unlikely((int)req->tp_block_size <= 0))
4083                         goto out;
4084                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
4085                         goto out;
4086                 if (po->tp_version >= TPACKET_V3 &&
4087                     (int)(req->tp_block_size -
4088                           BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
4089                         goto out;
4090                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
4091                                         po->tp_reserve))
4092                         goto out;
4093                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4094                         goto out;
4095
4096                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
4097                 if (unlikely(rb->frames_per_block <= 0))
4098                         goto out;
4099                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4100                                         req->tp_frame_nr))
4101                         goto out;
4102
4103                 err = -ENOMEM;
4104                 order = get_order(req->tp_block_size);
4105                 pg_vec = alloc_pg_vec(req, order);
4106                 if (unlikely(!pg_vec))
4107                         goto out;
4108                 switch (po->tp_version) {
4109                 case TPACKET_V3:
4110                 /* Transmit path is not supported. We checked
4111                  * it above but just being paranoid
4112                  */
4113                         if (!tx_ring)
4114                                 init_prb_bdqc(po, rb, pg_vec, req_u);
4115                         break;
4116                 default:
4117                         break;
4118                 }
4119         }
4120         /* Done */
4121         else {
4122                 err = -EINVAL;
4123                 if (unlikely(req->tp_frame_nr))
4124                         goto out;
4125         }
4126
4127         lock_sock(sk);
4128
4129         /* Detach socket from network */
4130         spin_lock(&po->bind_lock);
4131         was_running = po->running;
4132         num = po->num;
4133         if (was_running) {
4134                 po->num = 0;
4135                 __unregister_prot_hook(sk, false);
4136         }
4137         spin_unlock(&po->bind_lock);
4138
4139         synchronize_net();
4140
4141         err = -EBUSY;
4142         mutex_lock(&po->pg_vec_lock);
4143         if (closing || atomic_read(&po->mapped) == 0) {
4144                 err = 0;
4145                 spin_lock_bh(&rb_queue->lock);
4146                 swap(rb->pg_vec, pg_vec);
4147                 rb->frame_max = (req->tp_frame_nr - 1);
4148                 rb->head = 0;
4149                 rb->frame_size = req->tp_frame_size;
4150                 spin_unlock_bh(&rb_queue->lock);
4151
4152                 swap(rb->pg_vec_order, order);
4153                 swap(rb->pg_vec_len, req->tp_block_nr);
4154
4155                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4156                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4157                                                 tpacket_rcv : packet_rcv;
4158                 skb_queue_purge(rb_queue);
4159                 if (atomic_read(&po->mapped))
4160                         pr_err("packet_mmap: vma is busy: %d\n",
4161                                atomic_read(&po->mapped));
4162         }
4163         mutex_unlock(&po->pg_vec_lock);
4164
4165         spin_lock(&po->bind_lock);
4166         if (was_running) {
4167                 po->num = num;
4168                 register_prot_hook(sk);
4169         }
4170         spin_unlock(&po->bind_lock);
4171         if (closing && (po->tp_version > TPACKET_V2)) {
4172                 /* Because we don't support block-based V3 on tx-ring */
4173                 if (!tx_ring)
4174                         prb_shutdown_retire_blk_timer(po, rb_queue);
4175         }
4176         release_sock(sk);
4177
4178         if (pg_vec)
4179                 free_pg_vec(pg_vec, order, req->tp_block_nr);
4180 out:
4181         return err;
4182 }
4183
4184 static int packet_mmap(struct file *file, struct socket *sock,
4185                 struct vm_area_struct *vma)
4186 {
4187         struct sock *sk = sock->sk;
4188         struct packet_sock *po = pkt_sk(sk);
4189         unsigned long size, expected_size;
4190         struct packet_ring_buffer *rb;
4191         unsigned long start;
4192         int err = -EINVAL;
4193         int i;
4194
4195         if (vma->vm_pgoff)
4196                 return -EINVAL;
4197
4198         mutex_lock(&po->pg_vec_lock);
4199
4200         expected_size = 0;
4201         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4202                 if (rb->pg_vec) {
4203                         expected_size += rb->pg_vec_len
4204                                                 * rb->pg_vec_pages
4205                                                 * PAGE_SIZE;
4206                 }
4207         }
4208
4209         if (expected_size == 0)
4210                 goto out;
4211
4212         size = vma->vm_end - vma->vm_start;
4213         if (size != expected_size)
4214                 goto out;
4215
4216         start = vma->vm_start;
4217         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4218                 if (rb->pg_vec == NULL)
4219                         continue;
4220
4221                 for (i = 0; i < rb->pg_vec_len; i++) {
4222                         struct page *page;
4223                         void *kaddr = rb->pg_vec[i].buffer;
4224                         int pg_num;
4225
4226                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4227                                 page = pgv_to_page(kaddr);
4228                                 err = vm_insert_page(vma, start, page);
4229                                 if (unlikely(err))
4230                                         goto out;
4231                                 start += PAGE_SIZE;
4232                                 kaddr += PAGE_SIZE;
4233                         }
4234                 }
4235         }
4236
4237         atomic_inc(&po->mapped);
4238         vma->vm_ops = &packet_mmap_ops;
4239         err = 0;
4240
4241 out:
4242         mutex_unlock(&po->pg_vec_lock);
4243         return err;
4244 }
4245
4246 static const struct proto_ops packet_ops_spkt = {
4247         .family =       PF_PACKET,
4248         .owner =        THIS_MODULE,
4249         .release =      packet_release,
4250         .bind =         packet_bind_spkt,
4251         .connect =      sock_no_connect,
4252         .socketpair =   sock_no_socketpair,
4253         .accept =       sock_no_accept,
4254         .getname =      packet_getname_spkt,
4255         .poll =         datagram_poll,
4256         .ioctl =        packet_ioctl,
4257         .listen =       sock_no_listen,
4258         .shutdown =     sock_no_shutdown,
4259         .setsockopt =   sock_no_setsockopt,
4260         .getsockopt =   sock_no_getsockopt,
4261         .sendmsg =      packet_sendmsg_spkt,
4262         .recvmsg =      packet_recvmsg,
4263         .mmap =         sock_no_mmap,
4264         .sendpage =     sock_no_sendpage,
4265 };
4266
4267 static const struct proto_ops packet_ops = {
4268         .family =       PF_PACKET,
4269         .owner =        THIS_MODULE,
4270         .release =      packet_release,
4271         .bind =         packet_bind,
4272         .connect =      sock_no_connect,
4273         .socketpair =   sock_no_socketpair,
4274         .accept =       sock_no_accept,
4275         .getname =      packet_getname,
4276         .poll =         packet_poll,
4277         .ioctl =        packet_ioctl,
4278         .listen =       sock_no_listen,
4279         .shutdown =     sock_no_shutdown,
4280         .setsockopt =   packet_setsockopt,
4281         .getsockopt =   packet_getsockopt,
4282         .sendmsg =      packet_sendmsg,
4283         .recvmsg =      packet_recvmsg,
4284         .mmap =         packet_mmap,
4285         .sendpage =     sock_no_sendpage,
4286 };
4287
4288 static const struct net_proto_family packet_family_ops = {
4289         .family =       PF_PACKET,
4290         .create =       packet_create,
4291         .owner  =       THIS_MODULE,
4292 };
4293
4294 static struct notifier_block packet_netdev_notifier = {
4295         .notifier_call =        packet_notifier,
4296 };
4297
4298 #ifdef CONFIG_PROC_FS
4299
4300 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4301         __acquires(RCU)
4302 {
4303         struct net *net = seq_file_net(seq);
4304
4305         rcu_read_lock();
4306         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4307 }
4308
4309 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4310 {
4311         struct net *net = seq_file_net(seq);
4312         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4313 }
4314
4315 static void packet_seq_stop(struct seq_file *seq, void *v)
4316         __releases(RCU)
4317 {
4318         rcu_read_unlock();
4319 }
4320
4321 static int packet_seq_show(struct seq_file *seq, void *v)
4322 {
4323         if (v == SEQ_START_TOKEN)
4324                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
4325         else {
4326                 struct sock *s = sk_entry(v);
4327                 const struct packet_sock *po = pkt_sk(s);
4328
4329                 seq_printf(seq,
4330                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4331                            s,
4332                            atomic_read(&s->sk_refcnt),
4333                            s->sk_type,
4334                            ntohs(po->num),
4335                            po->ifindex,
4336                            po->running,
4337                            atomic_read(&s->sk_rmem_alloc),
4338                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4339                            sock_i_ino(s));
4340         }
4341
4342         return 0;
4343 }
4344
4345 static const struct seq_operations packet_seq_ops = {
4346         .start  = packet_seq_start,
4347         .next   = packet_seq_next,
4348         .stop   = packet_seq_stop,
4349         .show   = packet_seq_show,
4350 };
4351
4352 static int packet_seq_open(struct inode *inode, struct file *file)
4353 {
4354         return seq_open_net(inode, file, &packet_seq_ops,
4355                             sizeof(struct seq_net_private));
4356 }
4357
4358 static const struct file_operations packet_seq_fops = {
4359         .owner          = THIS_MODULE,
4360         .open           = packet_seq_open,
4361         .read           = seq_read,
4362         .llseek         = seq_lseek,
4363         .release        = seq_release_net,
4364 };
4365
4366 #endif
4367
4368 static int __net_init packet_net_init(struct net *net)
4369 {
4370         mutex_init(&net->packet.sklist_lock);
4371         INIT_HLIST_HEAD(&net->packet.sklist);
4372
4373         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4374                 return -ENOMEM;
4375
4376         return 0;
4377 }
4378
4379 static void __net_exit packet_net_exit(struct net *net)
4380 {
4381         remove_proc_entry("packet", net->proc_net);
4382 }
4383
4384 static struct pernet_operations packet_net_ops = {
4385         .init = packet_net_init,
4386         .exit = packet_net_exit,
4387 };
4388
4389
4390 static void __exit packet_exit(void)
4391 {
4392         unregister_netdevice_notifier(&packet_netdev_notifier);
4393         unregister_pernet_subsys(&packet_net_ops);
4394         sock_unregister(PF_PACKET);
4395         proto_unregister(&packet_proto);
4396 }
4397
4398 static int __init packet_init(void)
4399 {
4400         int rc = proto_register(&packet_proto, 0);
4401
4402         if (rc != 0)
4403                 goto out;
4404
4405         sock_register(&packet_family_ops);
4406         register_pernet_subsys(&packet_net_ops);
4407         register_netdevice_notifier(&packet_netdev_notifier);
4408 out:
4409         return rc;
4410 }
4411
4412 module_init(packet_init);
4413 module_exit(packet_exit);
4414 MODULE_LICENSE("GPL");
4415 MODULE_ALIAS_NETPROTO(PF_PACKET);