net/packet/af_packet.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              PACKET - implements raw packet sockets.
   7  *
   8  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  11  *
  12  * Fixes:
  13  *              Alan Cox        :       verify_area() now used correctly
  14  *              Alan Cox        :       new skbuff lists, look ma no backlogs!
  15  *              Alan Cox        :       tidied skbuff lists.
  16  *              Alan Cox        :       Now uses generic datagram routines I
  17  *                                      added. Also fixed the peek/read crash
  18  *                                      from all old Linux datagram code.
  19  *              Alan Cox        :       Uses the improved datagram code.
  20  *              Alan Cox        :       Added NULL's for socket options.
  21  *              Alan Cox        :       Re-commented the code.
  22  *              Alan Cox        :       Use new kernel side addressing
  23  *              Rob Janssen     :       Correct MTU usage.
  24  *              Dave Platt      :       Counter leaks caused by incorrect
  25  *                                      interrupt locking and some slightly
  26  *                                      dubious gcc output. Can you read
  27  *                                      compiler: it said _VOLATILE_
  28  *      Richard Kooijman        :       Timestamp fixes.
  29  *              Alan Cox        :       New buffers. Use sk->mac.raw.
  30  *              Alan Cox        :       sendmsg/recvmsg support.
  31  *              Alan Cox        :       Protocol setting support
  32  *      Alexey Kuznetsov        :       Untied from IPv4 stack.
  33  *      Cyrus Durgin            :       Fixed kerneld for kmod.
  34  *      Michal Ostrowski        :       Module initialization cleanup.
  35  *         Ulises Alonso        :       Frame number limit removal and
  36  *                                      packet_set_ring memory leak.
  37  *              Eric Biederman  :       Allow for > 8 byte hardware addresses.
  38  *                                      The convention is that longer addresses
  39  *                                      will simply extend the hardware address
  40  *                                      byte arrays at the end of sockaddr_ll
  41  *                                      and packet_mreq.
  42  *              Johann Baudy    :       Added TX RING.
  43  *              Chetan Loke     :       Implemented TPACKET_V3 block abstraction
  44  *                                      layer.
  45  *                                      Copyright (C) 2011, <lokec@ccs.neu.edu>
  46  *
  47  *
  48  *              This program is free software; you can redistribute it and/or
  49  *              modify it under the terms of the GNU General Public License
  50  *              as published by the Free Software Foundation; either version
  51  *              2 of the License, or (at your option) any later version.
  52  *
  53  */
  54
  55 #include <linux/types.h>
  56 #include <linux/mm.h>
  57 #include <linux/capability.h>
  58 #include <linux/fcntl.h>
  59 #include <linux/socket.h>
  60 #include <linux/in.h>
  61 #include <linux/inet.h>
  62 #include <linux/netdevice.h>
  63 #include <linux/if_packet.h>
  64 #include <linux/wireless.h>
  65 #include <linux/kernel.h>
  66 #include <linux/kmod.h>
  67 #include <linux/slab.h>
  68 #include <linux/vmalloc.h>
  69 #include <net/net_namespace.h>
  70 #include <net/ip.h>
  71 #include <net/protocol.h>
  72 #include <linux/skbuff.h>
  73 #include <net/sock.h>
  74 #include <linux/errno.h>
  75 #include <linux/timer.h>
  76 #include <asm/uaccess.h>
  77 #include <asm/ioctls.h>
  78 #include <asm/page.h>
  79 #include <asm/cacheflush.h>
  80 #include <asm/io.h>
  81 #include <linux/proc_fs.h>
  82 #include <linux/seq_file.h>
  83 #include <linux/poll.h>
  84 #include <linux/module.h>
  85 #include <linux/init.h>
  86 #include <linux/mutex.h>
  87 #include <linux/if_vlan.h>
  88 #include <linux/virtio_net.h>
  89 #include <linux/errqueue.h>
  90 #include <linux/net_tstamp.h>
  91 #include <linux/percpu.h>
  92 #ifdef CONFIG_INET
  93 #include <net/inet_common.h>
  94 #endif
  95 #include <linux/bpf.h>
  96
  97 #include "internal.h"
  98
  99 /*
 100    Assumptions:
 101    - if device has no dev->hard_header routine, it adds and removes ll header
 102      inside itself. In this case ll header is invisible outside of device,
 103      but higher levels still should reserve dev->hard_header_len.
 104      Some devices are enough clever to reallocate skb, when header
 105      will not fit to reserved space (tunnel), another ones are silly
 106      (PPP).
 107    - packet socket receives packets with pulled ll header,
 108      so that SOCK_RAW should push it back.
 109
 110 On receive:
 111 -----------
 112
 113 Incoming, dev->hard_header!=NULL
 114    mac_header -> ll header
 115    data       -> data
 116
 117 Outgoing, dev->hard_header!=NULL
 118    mac_header -> ll header
 119    data       -> ll header
 120
 121 Incoming, dev->hard_header==NULL
 122    mac_header -> UNKNOWN position. It is very likely, that it points to ll
 123                  header.  PPP makes it, that is wrong, because introduce
 124                  assymetry between rx and tx paths.
 125    data       -> data
 126
 127 Outgoing, dev->hard_header==NULL
 128    mac_header -> data. ll header is still not built!
 129    data       -> data
 130
 131 Resume
 132   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 133
 134
 135 On transmit:
 136 ------------
 137
 138 dev->hard_header != NULL
 139    mac_header -> ll header
 140    data       -> ll header
 141
 142 dev->hard_header == NULL (ll header is added by device, we cannot control it)
 143    mac_header -> data
 144    data       -> data
 145
 146    We should set nh.raw on output to correct posistion,
 147    packet classifier depends on it.
 148  */
 149
 150 /* Private packet socket structures. */
 151
 152 /* identical to struct packet_mreq except it has
 153  * a longer address field.
 154  */
 155 struct packet_mreq_max {
 156         int             mr_ifindex;
 157         unsigned short  mr_type;
 158         unsigned short  mr_alen;
 159         unsigned char   mr_address[MAX_ADDR_LEN];
 160 };
 161
 162 union tpacket_uhdr {
 163         struct tpacket_hdr  *h1;
 164         struct tpacket2_hdr *h2;
 165         struct tpacket3_hdr *h3;
 166         void *raw;
 167 };
 168
 169 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 170                 int closing, int tx_ring);
 171
 172 #define V3_ALIGNMENT    (8)
 173
 174 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 175
 176 #define BLK_PLUS_PRIV(sz_of_priv) \
 177         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 178
 179 #define PGV_FROM_VMALLOC 1
 180
 181 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
 182 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
 183 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
 184 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
 185 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
 186 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
 187 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
 188
 189 struct packet_sock;
 190 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
 191 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 192                        struct packet_type *pt, struct net_device *orig_dev);
 193
 194 static void *packet_previous_frame(struct packet_sock *po,
 195                 struct packet_ring_buffer *rb,
 196                 int status);
 197 static void packet_increment_head(struct packet_ring_buffer *buff);
 198 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
 199                         struct tpacket_block_desc *);
 200 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 201                         struct packet_sock *);
 202 static void prb_retire_current_block(struct tpacket_kbdq_core *,
 203                 struct packet_sock *, unsigned int status);
 204 static int prb_queue_frozen(struct tpacket_kbdq_core *);
 205 static void prb_open_block(struct tpacket_kbdq_core *,
 206                 struct tpacket_block_desc *);
 207 static void prb_retire_rx_blk_timer_expired(unsigned long);
 208 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 209 static void prb_init_blk_timer(struct packet_sock *,
 210                 struct tpacket_kbdq_core *,
 211                 void (*func) (unsigned long));
 212 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 213 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 214                 struct tpacket3_hdr *);
 215 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 216                 struct tpacket3_hdr *);
 217 static void packet_flush_mclist(struct sock *sk);
 218
 219 struct packet_skb_cb {
 220         union {
 221                 struct sockaddr_pkt pkt;
 222                 union {
 223                         /* Trick: alias skb original length with
 224                          * ll.sll_family and ll.protocol in order
 225                          * to save room.
 226                          */
 227                         unsigned int origlen;
 228                         struct sockaddr_ll ll;
 229                 };
 230         } sa;
 231 };
 232
 233 #define vio_le() virtio_legacy_is_little_endian()
 234
 235 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 236
 237 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 238 #define GET_PBLOCK_DESC(x, bid) \
 239         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 240 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 241         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 242 #define GET_NEXT_PRB_BLK_NUM(x) \
 243         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 244         ((x)->kactive_blk_num+1) : 0)
 245
 246 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 247 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 248
 249 static int packet_direct_xmit(struct sk_buff *skb)
 250 {
 251         struct net_device *dev = skb->dev;
 252         struct sk_buff *orig_skb = skb;
 253         struct netdev_queue *txq;
 254         int ret = NETDEV_TX_BUSY;
 255
 256         if (unlikely(!netif_running(dev) ||
 257                      !netif_carrier_ok(dev)))
 258                 goto drop;
 259
 260         skb = validate_xmit_skb_list(skb, dev);
 261         if (skb != orig_skb)
 262                 goto drop;
 263
 264         txq = skb_get_tx_queue(dev, skb);
 265
 266         local_bh_disable();
 267
 268         HARD_TX_LOCK(dev, txq, smp_processor_id());
 269         if (!netif_xmit_frozen_or_drv_stopped(txq))
 270                 ret = netdev_start_xmit(skb, dev, txq, false);
 271         HARD_TX_UNLOCK(dev, txq);
 272
 273         local_bh_enable();
 274
 275         if (!dev_xmit_complete(ret))
 276                 kfree_skb(skb);
 277
 278         return ret;
 279 drop:
 280         atomic_long_inc(&dev->tx_dropped);
 281         kfree_skb_list(skb);
 282         return NET_XMIT_DROP;
 283 }
 284
 285 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
 286 {
 287         struct net_device *dev;
 288
 289         rcu_read_lock();
 290         dev = rcu_dereference(po->cached_dev);
 291         if (likely(dev))
 292                 dev_hold(dev);
 293         rcu_read_unlock();
 294
 295         return dev;
 296 }
 297
 298 static void packet_cached_dev_assign(struct packet_sock *po,
 299                                      struct net_device *dev)
 300 {
 301         rcu_assign_pointer(po->cached_dev, dev);
 302 }
 303
 304 static void packet_cached_dev_reset(struct packet_sock *po)
 305 {
 306         RCU_INIT_POINTER(po->cached_dev, NULL);
 307 }
 308
 309 static bool packet_use_direct_xmit(const struct packet_sock *po)
 310 {
 311         return po->xmit == packet_direct_xmit;
 312 }
 313
 314 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
 315 {
 316         return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
 317 }
 318
 319 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
 320 {
 321         const struct net_device_ops *ops = dev->netdev_ops;
 322         u16 queue_index;
 323
 324         if (ops->ndo_select_queue) {
 325                 queue_index = ops->ndo_select_queue(dev, skb, NULL,
 326                                                     __packet_pick_tx_queue);
 327                 queue_index = netdev_cap_txqueue(dev, queue_index);
 328         } else {
 329                 queue_index = __packet_pick_tx_queue(dev, skb);
 330         }
 331
 332         skb_set_queue_mapping(skb, queue_index);
 333 }
 334
 335 /* register_prot_hook must be invoked with the po->bind_lock held,
 336  * or from a context in which asynchronous accesses to the packet
 337  * socket is not possible (packet_create()).
 338  */
 339 static void register_prot_hook(struct sock *sk)
 340 {
 341         struct packet_sock *po = pkt_sk(sk);
 342
 343         if (!po->running) {
 344                 if (po->fanout)
 345                         __fanout_link(sk, po);
 346                 else
 347                         dev_add_pack(&po->prot_hook);
 348
 349                 sock_hold(sk);
 350                 po->running = 1;
 351         }
 352 }
 353
 354 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
 355  * held.   If the sync parameter is true, we will temporarily drop
 356  * the po->bind_lock and do a synchronize_net to make sure no
 357  * asynchronous packet processing paths still refer to the elements
 358  * of po->prot_hook.  If the sync parameter is false, it is the
 359  * callers responsibility to take care of this.
 360  */
 361 static void __unregister_prot_hook(struct sock *sk, bool sync)
 362 {
 363         struct packet_sock *po = pkt_sk(sk);
 364
 365         po->running = 0;
 366
 367         if (po->fanout)
 368                 __fanout_unlink(sk, po);
 369         else
 370                 __dev_remove_pack(&po->prot_hook);
 371
 372         __sock_put(sk);
 373
 374         if (sync) {
 375                 spin_unlock(&po->bind_lock);
 376                 synchronize_net();
 377                 spin_lock(&po->bind_lock);
 378         }
 379 }
 380
 381 static void unregister_prot_hook(struct sock *sk, bool sync)
 382 {
 383         struct packet_sock *po = pkt_sk(sk);
 384
 385         if (po->running)
 386                 __unregister_prot_hook(sk, sync);
 387 }
 388
 389 static inline struct page * __pure pgv_to_page(void *addr)
 390 {
 391         if (is_vmalloc_addr(addr))
 392                 return vmalloc_to_page(addr);
 393         return virt_to_page(addr);
 394 }
 395
 396 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 397 {
 398         union tpacket_uhdr h;
 399
 400         h.raw = frame;
 401         switch (po->tp_version) {
 402         case TPACKET_V1:
 403                 h.h1->tp_status = status;
 404                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 405                 break;
 406         case TPACKET_V2:
 407                 h.h2->tp_status = status;
 408                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 409                 break;
 410         case TPACKET_V3:
 411         default:
 412                 WARN(1, "TPACKET version not supported.\n");
 413                 BUG();
 414         }
 415
 416         smp_wmb();
 417 }
 418
 419 static int __packet_get_status(struct packet_sock *po, void *frame)
 420 {
 421         union tpacket_uhdr h;
 422
 423         smp_rmb();
 424
 425         h.raw = frame;
 426         switch (po->tp_version) {
 427         case TPACKET_V1:
 428                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 429                 return h.h1->tp_status;
 430         case TPACKET_V2:
 431                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 432                 return h.h2->tp_status;
 433         case TPACKET_V3:
 434         default:
 435                 WARN(1, "TPACKET version not supported.\n");
 436                 BUG();
 437                 return 0;
 438         }
 439 }
 440
 441 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
 442                                    unsigned int flags)
 443 {
 444         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 445
 446         if (shhwtstamps &&
 447             (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
 448             ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
 449                 return TP_STATUS_TS_RAW_HARDWARE;
 450
 451         if (ktime_to_timespec_cond(skb->tstamp, ts))
 452                 return TP_STATUS_TS_SOFTWARE;
 453
 454         return 0;
 455 }
 456
 457 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
 458                                     struct sk_buff *skb)
 459 {
 460         union tpacket_uhdr h;
 461         struct timespec ts;
 462         __u32 ts_status;
 463
 464         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
 465                 return 0;
 466
 467         h.raw = frame;
 468         switch (po->tp_version) {
 469         case TPACKET_V1:
 470                 h.h1->tp_sec = ts.tv_sec;
 471                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
 472                 break;
 473         case TPACKET_V2:
 474                 h.h2->tp_sec = ts.tv_sec;
 475                 h.h2->tp_nsec = ts.tv_nsec;
 476                 break;
 477         case TPACKET_V3:
 478         default:
 479                 WARN(1, "TPACKET version not supported.\n");
 480                 BUG();
 481         }
 482
 483         /* one flush is safe, as both fields always lie on the same cacheline */
 484         flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
 485         smp_wmb();
 486
 487         return ts_status;
 488 }
 489
 490 static void *packet_lookup_frame(struct packet_sock *po,
 491                 struct packet_ring_buffer *rb,
 492                 unsigned int position,
 493                 int status)
 494 {
 495         unsigned int pg_vec_pos, frame_offset;
 496         union tpacket_uhdr h;
 497
 498         pg_vec_pos = position / rb->frames_per_block;
 499         frame_offset = position % rb->frames_per_block;
 500
 501         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 502                 (frame_offset * rb->frame_size);
 503
 504         if (status != __packet_get_status(po, h.raw))
 505                 return NULL;
 506
 507         return h.raw;
 508 }
 509
 510 static void *packet_current_frame(struct packet_sock *po,
 511                 struct packet_ring_buffer *rb,
 512                 int status)
 513 {
 514         return packet_lookup_frame(po, rb, rb->head, status);
 515 }
 516
 517 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 518 {
 519         del_timer_sync(&pkc->retire_blk_timer);
 520 }
 521
 522 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 523                 struct sk_buff_head *rb_queue)
 524 {
 525         struct tpacket_kbdq_core *pkc;
 526
 527         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 528
 529         spin_lock_bh(&rb_queue->lock);
 530         pkc->delete_blk_timer = 1;
 531         spin_unlock_bh(&rb_queue->lock);
 532
 533         prb_del_retire_blk_timer(pkc);
 534 }
 535
 536 static void prb_init_blk_timer(struct packet_sock *po,
 537                 struct tpacket_kbdq_core *pkc,
 538                 void (*func) (unsigned long))
 539 {
 540         init_timer(&pkc->retire_blk_timer);
 541         pkc->retire_blk_timer.data = (long)po;
 542         pkc->retire_blk_timer.function = func;
 543         pkc->retire_blk_timer.expires = jiffies;
 544 }
 545
 546 static void prb_setup_retire_blk_timer(struct packet_sock *po)
 547 {
 548         struct tpacket_kbdq_core *pkc;
 549
 550         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 551         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
 552 }
 553
 554 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 555                                 int blk_size_in_bytes)
 556 {
 557         struct net_device *dev;
 558         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 559         struct ethtool_cmd ecmd;
 560         int err;
 561         u32 speed;
 562
 563         rtnl_lock();
 564         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 565         if (unlikely(!dev)) {
 566                 rtnl_unlock();
 567                 return DEFAULT_PRB_RETIRE_TOV;
 568         }
 569         err = __ethtool_get_settings(dev, &ecmd);
 570         speed = ethtool_cmd_speed(&ecmd);
 571         rtnl_unlock();
 572         if (!err) {
 573                 /*
 574                  * If the link speed is so slow you don't really
 575                  * need to worry about perf anyways
 576                  */
 577                 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
 578                         return DEFAULT_PRB_RETIRE_TOV;
 579                 } else {
 580                         msec = 1;
 581                         div = speed / 1000;
 582                 }
 583         }
 584
 585         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 586
 587         if (div)
 588                 mbits /= div;
 589
 590         tmo = mbits * msec;
 591
 592         if (div)
 593                 return tmo+1;
 594         return tmo;
 595 }
 596
 597 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 598                         union tpacket_req_u *req_u)
 599 {
 600         p1->feature_req_word = req_u->req3.tp_feature_req_word;
 601 }
 602
 603 static void init_prb_bdqc(struct packet_sock *po,
 604                         struct packet_ring_buffer *rb,
 605                         struct pgv *pg_vec,
 606                         union tpacket_req_u *req_u)
 607 {
 608         struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
 609         struct tpacket_block_desc *pbd;
 610
 611         memset(p1, 0x0, sizeof(*p1));
 612
 613         p1->knxt_seq_num = 1;
 614         p1->pkbdq = pg_vec;
 615         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 616         p1->pkblk_start = pg_vec[0].buffer;
 617         p1->kblk_size = req_u->req3.tp_block_size;
 618         p1->knum_blocks = req_u->req3.tp_block_nr;
 619         p1->hdrlen = po->tp_hdrlen;
 620         p1->version = po->tp_version;
 621         p1->last_kactive_blk_num = 0;
 622         po->stats.stats3.tp_freeze_q_cnt = 0;
 623         if (req_u->req3.tp_retire_blk_tov)
 624                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 625         else
 626                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 627                                                 req_u->req3.tp_block_size);
 628         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 629         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 630
 631         p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
 632         prb_init_ft_ops(p1, req_u);
 633         prb_setup_retire_blk_timer(po);
 634         prb_open_block(p1, pbd);
 635 }
 636
 637 /*  Do NOT update the last_blk_num first.
 638  *  Assumes sk_buff_head lock is held.
 639  */
 640 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 641 {
 642         mod_timer(&pkc->retire_blk_timer,
 643                         jiffies + pkc->tov_in_jiffies);
 644         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 645 }
 646
 647 /*
 648  * Timer logic:
 649  * 1) We refresh the timer only when we open a block.
 650  *    By doing this we don't waste cycles refreshing the timer
 651  *        on packet-by-packet basis.
 652  *
 653  * With a 1MB block-size, on a 1Gbps line, it will take
 654  * i) ~8 ms to fill a block + ii) memcpy etc.
 655  * In this cut we are not accounting for the memcpy time.
 656  *
 657  * So, if the user sets the 'tmo' to 10ms then the timer
 658  * will never fire while the block is still getting filled
 659  * (which is what we want). However, the user could choose
 660  * to close a block early and that's fine.
 661  *
 662  * But when the timer does fire, we check whether or not to refresh it.
 663  * Since the tmo granularity is in msecs, it is not too expensive
 664  * to refresh the timer, lets say every '8' msecs.
 665  * Either the user can set the 'tmo' or we can derive it based on
 666  * a) line-speed and b) block-size.
 667  * prb_calc_retire_blk_tmo() calculates the tmo.
 668  *
 669  */
 670 static void prb_retire_rx_blk_timer_expired(unsigned long data)
 671 {
 672         struct packet_sock *po = (struct packet_sock *)data;
 673         struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 674         unsigned int frozen;
 675         struct tpacket_block_desc *pbd;
 676
 677         spin_lock(&po->sk.sk_receive_queue.lock);
 678
 679         frozen = prb_queue_frozen(pkc);
 680         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 681
 682         if (unlikely(pkc->delete_blk_timer))
 683                 goto out;
 684
 685         /* We only need to plug the race when the block is partially filled.
 686          * tpacket_rcv:
 687          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 688          *              copy_bits() is in progress ...
 689          *              timer fires on other cpu:
 690          *              we can't retire the current block because copy_bits
 691          *              is in progress.
 692          *
 693          */
 694         if (BLOCK_NUM_PKTS(pbd)) {
 695                 while (atomic_read(&pkc->blk_fill_in_prog)) {
 696                         /* Waiting for skb_copy_bits to finish... */
 697                         cpu_relax();
 698                 }
 699         }
 700
 701         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 702                 if (!frozen) {
 703                         if (!BLOCK_NUM_PKTS(pbd)) {
 704                                 /* An empty block. Just refresh the timer. */
 705                                 goto refresh_timer;
 706                         }
 707                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 708                         if (!prb_dispatch_next_block(pkc, po))
 709                                 goto refresh_timer;
 710                         else
 711                                 goto out;
 712                 } else {
 713                         /* Case 1. Queue was frozen because user-space was
 714                          *         lagging behind.
 715                          */
 716                         if (prb_curr_blk_in_use(pkc, pbd)) {
 717                                 /*
 718                                  * Ok, user-space is still behind.
 719                                  * So just refresh the timer.
 720                                  */
 721                                 goto refresh_timer;
 722                         } else {
 723                                /* Case 2. queue was frozen,user-space caught up,
 724                                 * now the link went idle && the timer fired.
 725                                 * We don't have a block to close.So we open this
 726                                 * block and restart the timer.
 727                                 * opening a block thaws the queue,restarts timer
 728                                 * Thawing/timer-refresh is a side effect.
 729                                 */
 730                                 prb_open_block(pkc, pbd);
 731                                 goto out;
 732                         }
 733                 }
 734         }
 735
 736 refresh_timer:
 737         _prb_refresh_rx_retire_blk_timer(pkc);
 738
 739 out:
 740         spin_unlock(&po->sk.sk_receive_queue.lock);
 741 }
 742
 743 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 744                 struct tpacket_block_desc *pbd1, __u32 status)
 745 {
 746         /* Flush everything minus the block header */
 747
 748 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 749         u8 *start, *end;
 750
 751         start = (u8 *)pbd1;
 752
 753         /* Skip the block header(we know header WILL fit in 4K) */
 754         start += PAGE_SIZE;
 755
 756         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 757         for (; start < end; start += PAGE_SIZE)
 758                 flush_dcache_page(pgv_to_page(start));
 759
 760         smp_wmb();
 761 #endif
 762
 763         /* Now update the block status. */
 764
 765         BLOCK_STATUS(pbd1) = status;
 766
 767         /* Flush the block header */
 768
 769 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 770         start = (u8 *)pbd1;
 771         flush_dcache_page(pgv_to_page(start));
 772
 773         smp_wmb();
 774 #endif
 775 }
 776
 777 /*
 778  * Side effect:
 779  *
 780  * 1) flush the block
 781  * 2) Increment active_blk_num
 782  *
 783  * Note:We DONT refresh the timer on purpose.
 784  *      Because almost always the next block will be opened.
 785  */
 786 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 787                 struct tpacket_block_desc *pbd1,
 788                 struct packet_sock *po, unsigned int stat)
 789 {
 790         __u32 status = TP_STATUS_USER | stat;
 791
 792         struct tpacket3_hdr *last_pkt;
 793         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 794         struct sock *sk = &po->sk;
 795
 796         if (po->stats.stats3.tp_drops)
 797                 status |= TP_STATUS_LOSING;
 798
 799         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 800         last_pkt->tp_next_offset = 0;
 801
 802         /* Get the ts of the last pkt */
 803         if (BLOCK_NUM_PKTS(pbd1)) {
 804                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 805                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 806         } else {
 807                 /* Ok, we tmo'd - so get the current time.
 808                  *
 809                  * It shouldn't really happen as we don't close empty
 810                  * blocks. See prb_retire_rx_blk_timer_expired().
 811                  */
 812                 struct timespec ts;
 813                 getnstimeofday(&ts);
 814                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 815                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 816         }
 817
 818         smp_wmb();
 819
 820         /* Flush the block */
 821         prb_flush_block(pkc1, pbd1, status);
 822
 823         sk->sk_data_ready(sk);
 824
 825         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 826 }
 827
 828 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 829 {
 830         pkc->reset_pending_on_curr_blk = 0;
 831 }
 832
 833 /*
 834  * Side effect of opening a block:
 835  *
 836  * 1) prb_queue is thawed.
 837  * 2) retire_blk_timer is refreshed.
 838  *
 839  */
 840 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 841         struct tpacket_block_desc *pbd1)
 842 {
 843         struct timespec ts;
 844         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 845
 846         smp_rmb();
 847
 848         /* We could have just memset this but we will lose the
 849          * flexibility of making the priv area sticky
 850          */
 851
 852         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 853         BLOCK_NUM_PKTS(pbd1) = 0;
 854         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 855
 856         getnstimeofday(&ts);
 857
 858         h1->ts_first_pkt.ts_sec = ts.tv_sec;
 859         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 860
 861         pkc1->pkblk_start = (char *)pbd1;
 862         pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 863
 864         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 865         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 866
 867         pbd1->version = pkc1->version;
 868         pkc1->prev = pkc1->nxt_offset;
 869         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 870
 871         prb_thaw_queue(pkc1);
 872         _prb_refresh_rx_retire_blk_timer(pkc1);
 873
 874         smp_wmb();
 875 }
 876
 877 /*
 878  * Queue freeze logic:
 879  * 1) Assume tp_block_nr = 8 blocks.
 880  * 2) At time 't0', user opens Rx ring.
 881  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 882  * 4) user-space is either sleeping or processing block '0'.
 883  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 884  *    it will close block-7,loop around and try to fill block '0'.
 885  *    call-flow:
 886  *    __packet_lookup_frame_in_block
 887  *      prb_retire_current_block()
 888  *      prb_dispatch_next_block()
 889  *        |->(BLOCK_STATUS == USER) evaluates to true
 890  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 891  * 6) Now there are two cases:
 892  *    6.1) Link goes idle right after the queue is frozen.
 893  *         But remember, the last open_block() refreshed the timer.
 894  *         When this timer expires,it will refresh itself so that we can
 895  *         re-open block-0 in near future.
 896  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 897  *         case and __packet_lookup_frame_in_block will check if block-0
 898  *         is free and can now be re-used.
 899  */
 900 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 901                                   struct packet_sock *po)
 902 {
 903         pkc->reset_pending_on_curr_blk = 1;
 904         po->stats.stats3.tp_freeze_q_cnt++;
 905 }
 906
 907 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 908
 909 /*
 910  * If the next block is free then we will dispatch it
 911  * and return a good offset.
 912  * Else, we will freeze the queue.
 913  * So, caller must check the return value.
 914  */
 915 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 916                 struct packet_sock *po)
 917 {
 918         struct tpacket_block_desc *pbd;
 919
 920         smp_rmb();
 921
 922         /* 1. Get current block num */
 923         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 924
 925         /* 2. If this block is currently in_use then freeze the queue */
 926         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 927                 prb_freeze_queue(pkc, po);
 928                 return NULL;
 929         }
 930
 931         /*
 932          * 3.
 933          * open this block and return the offset where the first packet
 934          * needs to get stored.
 935          */
 936         prb_open_block(pkc, pbd);
 937         return (void *)pkc->nxt_offset;
 938 }
 939
 940 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 941                 struct packet_sock *po, unsigned int status)
 942 {
 943         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 944
 945         /* retire/close the current block */
 946         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 947                 /*
 948                  * Plug the case where copy_bits() is in progress on
 949                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 950                  * have space to copy the pkt in the current block and
 951                  * called prb_retire_current_block()
 952                  *
 953                  * We don't need to worry about the TMO case because
 954                  * the timer-handler already handled this case.
 955                  */
 956                 if (!(status & TP_STATUS_BLK_TMO)) {
 957                         while (atomic_read(&pkc->blk_fill_in_prog)) {
 958                                 /* Waiting for skb_copy_bits to finish... */
 959                                 cpu_relax();
 960                         }
 961                 }
 962                 prb_close_block(pkc, pbd, po, status);
 963                 return;
 964         }
 965 }
 966
 967 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
 968                                       struct tpacket_block_desc *pbd)
 969 {
 970         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 971 }
 972
 973 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 974 {
 975         return pkc->reset_pending_on_curr_blk;
 976 }
 977
 978 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 979 {
 980         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 981         atomic_dec(&pkc->blk_fill_in_prog);
 982 }
 983
 984 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 985                         struct tpacket3_hdr *ppd)
 986 {
 987         ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
 988 }
 989
 990 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 991                         struct tpacket3_hdr *ppd)
 992 {
 993         ppd->hv1.tp_rxhash = 0;
 994 }
 995
 996 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
 997                         struct tpacket3_hdr *ppd)
 998 {
 999         if (skb_vlan_tag_present(pkc->skb)) {
1000                 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1001                 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1002                 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1003         } else {
1004                 ppd->hv1.tp_vlan_tci = 0;
1005                 ppd->hv1.tp_vlan_tpid = 0;
1006                 ppd->tp_status = TP_STATUS_AVAILABLE;
1007         }
1008 }
1009
1010 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1011                         struct tpacket3_hdr *ppd)
1012 {
1013         ppd->hv1.tp_padding = 0;
1014         prb_fill_vlan_info(pkc, ppd);
1015
1016         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1017                 prb_fill_rxhash(pkc, ppd);
1018         else
1019                 prb_clear_rxhash(pkc, ppd);
1020 }
1021
1022 static void prb_fill_curr_block(char *curr,
1023                                 struct tpacket_kbdq_core *pkc,
1024                                 struct tpacket_block_desc *pbd,
1025                                 unsigned int len)
1026 {
1027         struct tpacket3_hdr *ppd;
1028
1029         ppd  = (struct tpacket3_hdr *)curr;
1030         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1031         pkc->prev = curr;
1032         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1033         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1034         BLOCK_NUM_PKTS(pbd) += 1;
1035         atomic_inc(&pkc->blk_fill_in_prog);
1036         prb_run_all_ft_ops(pkc, ppd);
1037 }
1038
1039 /* Assumes caller has the sk->rx_queue.lock */
1040 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1041                                             struct sk_buff *skb,
1042                                                 int status,
1043                                             unsigned int len
1044                                             )
1045 {
1046         struct tpacket_kbdq_core *pkc;
1047         struct tpacket_block_desc *pbd;
1048         char *curr, *end;
1049
1050         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1051         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1052
1053         /* Queue is frozen when user space is lagging behind */
1054         if (prb_queue_frozen(pkc)) {
1055                 /*
1056                  * Check if that last block which caused the queue to freeze,
1057                  * is still in_use by user-space.
1058                  */
1059                 if (prb_curr_blk_in_use(pkc, pbd)) {
1060                         /* Can't record this packet */
1061                         return NULL;
1062                 } else {
1063                         /*
1064                          * Ok, the block was released by user-space.
1065                          * Now let's open that block.
1066                          * opening a block also thaws the queue.
1067                          * Thawing is a side effect.
1068                          */
1069                         prb_open_block(pkc, pbd);
1070                 }
1071         }
1072
1073         smp_mb();
1074         curr = pkc->nxt_offset;
1075         pkc->skb = skb;
1076         end = (char *)pbd + pkc->kblk_size;
1077
1078         /* first try the current block */
1079         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1080                 prb_fill_curr_block(curr, pkc, pbd, len);
1081                 return (void *)curr;
1082         }
1083
1084         /* Ok, close the current block */
1085         prb_retire_current_block(pkc, po, 0);
1086
1087         /* Now, try to dispatch the next block */
1088         curr = (char *)prb_dispatch_next_block(pkc, po);
1089         if (curr) {
1090                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1091                 prb_fill_curr_block(curr, pkc, pbd, len);
1092                 return (void *)curr;
1093         }
1094
1095         /*
1096          * No free blocks are available.user_space hasn't caught up yet.
1097          * Queue was just frozen and now this packet will get dropped.
1098          */
1099         return NULL;
1100 }
1101
1102 static void *packet_current_rx_frame(struct packet_sock *po,
1103                                             struct sk_buff *skb,
1104                                             int status, unsigned int len)
1105 {
1106         char *curr = NULL;
1107         switch (po->tp_version) {
1108         case TPACKET_V1:
1109         case TPACKET_V2:
1110                 curr = packet_lookup_frame(po, &po->rx_ring,
1111                                         po->rx_ring.head, status);
1112                 return curr;
1113         case TPACKET_V3:
1114                 return __packet_lookup_frame_in_block(po, skb, status, len);
1115         default:
1116                 WARN(1, "TPACKET version not supported\n");
1117                 BUG();
1118                 return NULL;
1119         }
1120 }
1121
1122 static void *prb_lookup_block(struct packet_sock *po,
1123                                      struct packet_ring_buffer *rb,
1124                                      unsigned int idx,
1125                                      int status)
1126 {
1127         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1128         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1129
1130         if (status != BLOCK_STATUS(pbd))
1131                 return NULL;
1132         return pbd;
1133 }
1134
1135 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1136 {
1137         unsigned int prev;
1138         if (rb->prb_bdqc.kactive_blk_num)
1139                 prev = rb->prb_bdqc.kactive_blk_num-1;
1140         else
1141                 prev = rb->prb_bdqc.knum_blocks-1;
1142         return prev;
1143 }
1144
1145 /* Assumes caller has held the rx_queue.lock */
1146 static void *__prb_previous_block(struct packet_sock *po,
1147                                          struct packet_ring_buffer *rb,
1148                                          int status)
1149 {
1150         unsigned int previous = prb_previous_blk_num(rb);
1151         return prb_lookup_block(po, rb, previous, status);
1152 }
1153
1154 static void *packet_previous_rx_frame(struct packet_sock *po,
1155                                              struct packet_ring_buffer *rb,
1156                                              int status)
1157 {
1158         if (po->tp_version <= TPACKET_V2)
1159                 return packet_previous_frame(po, rb, status);
1160
1161         return __prb_previous_block(po, rb, status);
1162 }
1163
1164 static void packet_increment_rx_head(struct packet_sock *po,
1165                                             struct packet_ring_buffer *rb)
1166 {
1167         switch (po->tp_version) {
1168         case TPACKET_V1:
1169         case TPACKET_V2:
1170                 return packet_increment_head(rb);
1171         case TPACKET_V3:
1172         default:
1173                 WARN(1, "TPACKET version not supported.\n");
1174                 BUG();
1175                 return;
1176         }
1177 }
1178
1179 static void *packet_previous_frame(struct packet_sock *po,
1180                 struct packet_ring_buffer *rb,
1181                 int status)
1182 {
1183         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1184         return packet_lookup_frame(po, rb, previous, status);
1185 }
1186
1187 static void packet_increment_head(struct packet_ring_buffer *buff)
1188 {
1189         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1190 }
1191
1192 static void packet_inc_pending(struct packet_ring_buffer *rb)
1193 {
1194         this_cpu_inc(*rb->pending_refcnt);
1195 }
1196
1197 static void packet_dec_pending(struct packet_ring_buffer *rb)
1198 {
1199         this_cpu_dec(*rb->pending_refcnt);
1200 }
1201
1202 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1203 {
1204         unsigned int refcnt = 0;
1205         int cpu;
1206
1207         /* We don't use pending refcount in rx_ring. */
1208         if (rb->pending_refcnt == NULL)
1209                 return 0;
1210
1211         for_each_possible_cpu(cpu)
1212                 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1213
1214         return refcnt;
1215 }
1216
1217 static int packet_alloc_pending(struct packet_sock *po)
1218 {
1219         po->rx_ring.pending_refcnt = NULL;
1220
1221         po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1222         if (unlikely(po->tx_ring.pending_refcnt == NULL))
1223                 return -ENOBUFS;
1224
1225         return 0;
1226 }
1227
1228 static void packet_free_pending(struct packet_sock *po)
1229 {
1230         free_percpu(po->tx_ring.pending_refcnt);
1231 }
1232
1233 #define ROOM_POW_OFF    2
1234 #define ROOM_NONE       0x0
1235 #define ROOM_LOW        0x1
1236 #define ROOM_NORMAL     0x2
1237
1238 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1239 {
1240         int idx, len;
1241
1242         len = po->rx_ring.frame_max + 1;
1243         idx = po->rx_ring.head;
1244         if (pow_off)
1245                 idx += len >> pow_off;
1246         if (idx >= len)
1247                 idx -= len;
1248         return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1249 }
1250
1251 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1252 {
1253         int idx, len;
1254
1255         len = po->rx_ring.prb_bdqc.knum_blocks;
1256         idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1257         if (pow_off)
1258                 idx += len >> pow_off;
1259         if (idx >= len)
1260                 idx -= len;
1261         return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1262 }
1263
1264 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1265 {
1266         struct sock *sk = &po->sk;
1267         int ret = ROOM_NONE;
1268
1269         if (po->prot_hook.func != tpacket_rcv) {
1270                 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1271                                           - (skb ? skb->truesize : 0);
1272                 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1273                         return ROOM_NORMAL;
1274                 else if (avail > 0)
1275                         return ROOM_LOW;
1276                 else
1277                         return ROOM_NONE;
1278         }
1279
1280         if (po->tp_version == TPACKET_V3) {
1281                 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1282                         ret = ROOM_NORMAL;
1283                 else if (__tpacket_v3_has_room(po, 0))
1284                         ret = ROOM_LOW;
1285         } else {
1286                 if (__tpacket_has_room(po, ROOM_POW_OFF))
1287                         ret = ROOM_NORMAL;
1288                 else if (__tpacket_has_room(po, 0))
1289                         ret = ROOM_LOW;
1290         }
1291
1292         return ret;
1293 }
1294
1295 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1296 {
1297         int ret;
1298         bool has_room;
1299
1300         spin_lock_bh(&po->sk.sk_receive_queue.lock);
1301         ret = __packet_rcv_has_room(po, skb);
1302         has_room = ret == ROOM_NORMAL;
1303         if (po->pressure == has_room)
1304                 po->pressure = !has_room;
1305         spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1306
1307         return ret;
1308 }
1309
1310 static void packet_sock_destruct(struct sock *sk)
1311 {
1312         skb_queue_purge(&sk->sk_error_queue);
1313
1314         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1315         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1316
1317         if (!sock_flag(sk, SOCK_DEAD)) {
1318                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1319                 return;
1320         }
1321
1322         sk_refcnt_debug_dec(sk);
1323 }
1324
1325 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1326 {
1327         u32 rxhash;
1328         int i, count = 0;
1329
1330         rxhash = skb_get_hash(skb);
1331         for (i = 0; i < ROLLOVER_HLEN; i++)
1332                 if (po->rollover->history[i] == rxhash)
1333                         count++;
1334
1335         po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1336         return count > (ROLLOVER_HLEN >> 1);
1337 }
1338
1339 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1340                                       struct sk_buff *skb,
1341                                       unsigned int num)
1342 {
1343         return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1344 }
1345
1346 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1347                                     struct sk_buff *skb,
1348                                     unsigned int num)
1349 {
1350         unsigned int val = atomic_inc_return(&f->rr_cur);
1351
1352         return val % num;
1353 }
1354
1355 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1356                                      struct sk_buff *skb,
1357                                      unsigned int num)
1358 {
1359         return smp_processor_id() % num;
1360 }
1361
1362 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1363                                      struct sk_buff *skb,
1364                                      unsigned int num)
1365 {
1366         return prandom_u32_max(num);
1367 }
1368
1369 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1370                                           struct sk_buff *skb,
1371                                           unsigned int idx, bool try_self,
1372                                           unsigned int num)
1373 {
1374         struct packet_sock *po, *po_next, *po_skip = NULL;
1375         unsigned int i, j, room = ROOM_NONE;
1376
1377         po = pkt_sk(f->arr[idx]);
1378
1379         if (try_self) {
1380                 room = packet_rcv_has_room(po, skb);
1381                 if (room == ROOM_NORMAL ||
1382                     (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1383                         return idx;
1384                 po_skip = po;
1385         }
1386
1387         i = j = min_t(int, po->rollover->sock, num - 1);
1388         do {
1389                 po_next = pkt_sk(f->arr[i]);
1390                 if (po_next != po_skip && !po_next->pressure &&
1391                     packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1392                         if (i != j)
1393                                 po->rollover->sock = i;
1394                         atomic_long_inc(&po->rollover->num);
1395                         if (room == ROOM_LOW)
1396                                 atomic_long_inc(&po->rollover->num_huge);
1397                         return i;
1398                 }
1399
1400                 if (++i == num)
1401                         i = 0;
1402         } while (i != j);
1403
1404         atomic_long_inc(&po->rollover->num_failed);
1405         return idx;
1406 }
1407
1408 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1409                                     struct sk_buff *skb,
1410                                     unsigned int num)
1411 {
1412         return skb_get_queue_mapping(skb) % num;
1413 }
1414
1415 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1416                                      struct sk_buff *skb,
1417                                      unsigned int num)
1418 {
1419         struct bpf_prog *prog;
1420         unsigned int ret = 0;
1421
1422         rcu_read_lock();
1423         prog = rcu_dereference(f->bpf_prog);
1424         if (prog)
1425                 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1426         rcu_read_unlock();
1427
1428         return ret;
1429 }
1430
1431 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1432 {
1433         return f->flags & (flag >> 8);
1434 }
1435
1436 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1437                              struct packet_type *pt, struct net_device *orig_dev)
1438 {
1439         struct packet_fanout *f = pt->af_packet_priv;
1440         unsigned int num = READ_ONCE(f->num_members);
1441         struct net *net = read_pnet(&f->net);
1442         struct packet_sock *po;
1443         unsigned int idx;
1444
1445         if (!net_eq(dev_net(dev), net) || !num) {
1446                 kfree_skb(skb);
1447                 return 0;
1448         }
1449
1450         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1451                 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1452                 if (!skb)
1453                         return 0;
1454         }
1455         switch (f->type) {
1456         case PACKET_FANOUT_HASH:
1457         default:
1458                 idx = fanout_demux_hash(f, skb, num);
1459                 break;
1460         case PACKET_FANOUT_LB:
1461                 idx = fanout_demux_lb(f, skb, num);
1462                 break;
1463         case PACKET_FANOUT_CPU:
1464                 idx = fanout_demux_cpu(f, skb, num);
1465                 break;
1466         case PACKET_FANOUT_RND:
1467                 idx = fanout_demux_rnd(f, skb, num);
1468                 break;
1469         case PACKET_FANOUT_QM:
1470                 idx = fanout_demux_qm(f, skb, num);
1471                 break;
1472         case PACKET_FANOUT_ROLLOVER:
1473                 idx = fanout_demux_rollover(f, skb, 0, false, num);
1474                 break;
1475         case PACKET_FANOUT_CBPF:
1476         case PACKET_FANOUT_EBPF:
1477                 idx = fanout_demux_bpf(f, skb, num);
1478                 break;
1479         }
1480
1481         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1482                 idx = fanout_demux_rollover(f, skb, idx, true, num);
1483
1484         po = pkt_sk(f->arr[idx]);
1485         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1486 }
1487
1488 DEFINE_MUTEX(fanout_mutex);
1489 EXPORT_SYMBOL_GPL(fanout_mutex);
1490 static LIST_HEAD(fanout_list);
1491
1492 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1493 {
1494         struct packet_fanout *f = po->fanout;
1495
1496         spin_lock(&f->lock);
1497         f->arr[f->num_members] = sk;
1498         smp_wmb();
1499         f->num_members++;
1500         if (f->num_members == 1)
1501                 dev_add_pack(&f->prot_hook);
1502         spin_unlock(&f->lock);
1503 }
1504
1505 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1506 {
1507         struct packet_fanout *f = po->fanout;
1508         int i;
1509
1510         spin_lock(&f->lock);
1511         for (i = 0; i < f->num_members; i++) {
1512                 if (f->arr[i] == sk)
1513                         break;
1514         }
1515         BUG_ON(i >= f->num_members);
1516         f->arr[i] = f->arr[f->num_members - 1];
1517         f->num_members--;
1518         if (f->num_members == 0)
1519                 __dev_remove_pack(&f->prot_hook);
1520         spin_unlock(&f->lock);
1521 }
1522
1523 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1524 {
1525         if (sk->sk_family != PF_PACKET)
1526                 return false;
1527
1528         return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1529 }
1530
1531 static void fanout_init_data(struct packet_fanout *f)
1532 {
1533         switch (f->type) {
1534         case PACKET_FANOUT_LB:
1535                 atomic_set(&f->rr_cur, 0);
1536                 break;
1537         case PACKET_FANOUT_CBPF:
1538         case PACKET_FANOUT_EBPF:
1539                 RCU_INIT_POINTER(f->bpf_prog, NULL);
1540                 break;
1541         }
1542 }
1543
1544 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1545 {
1546         struct bpf_prog *old;
1547
1548         spin_lock(&f->lock);
1549         old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1550         rcu_assign_pointer(f->bpf_prog, new);
1551         spin_unlock(&f->lock);
1552
1553         if (old) {
1554                 synchronize_net();
1555                 bpf_prog_destroy(old);
1556         }
1557 }
1558
1559 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1560                                 unsigned int len)
1561 {
1562         struct bpf_prog *new;
1563         struct sock_fprog fprog;
1564         int ret;
1565
1566         if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1567                 return -EPERM;
1568         if (len != sizeof(fprog))
1569                 return -EINVAL;
1570         if (copy_from_user(&fprog, data, len))
1571                 return -EFAULT;
1572
1573         ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1574         if (ret)
1575                 return ret;
1576
1577         __fanout_set_data_bpf(po->fanout, new);
1578         return 0;
1579 }
1580
1581 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1582                                 unsigned int len)
1583 {
1584         struct bpf_prog *new;
1585         u32 fd;
1586
1587         if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1588                 return -EPERM;
1589         if (len != sizeof(fd))
1590                 return -EINVAL;
1591         if (copy_from_user(&fd, data, len))
1592                 return -EFAULT;
1593
1594         new = bpf_prog_get(fd);
1595         if (IS_ERR(new))
1596                 return PTR_ERR(new);
1597         if (new->type != BPF_PROG_TYPE_SOCKET_FILTER) {
1598                 bpf_prog_put(new);
1599                 return -EINVAL;
1600         }
1601
1602         __fanout_set_data_bpf(po->fanout, new);
1603         return 0;
1604 }
1605
1606 static int fanout_set_data(struct packet_sock *po, char __user *data,
1607                            unsigned int len)
1608 {
1609         switch (po->fanout->type) {
1610         case PACKET_FANOUT_CBPF:
1611                 return fanout_set_data_cbpf(po, data, len);
1612         case PACKET_FANOUT_EBPF:
1613                 return fanout_set_data_ebpf(po, data, len);
1614         default:
1615                 return -EINVAL;
1616         };
1617 }
1618
1619 static void fanout_release_data(struct packet_fanout *f)
1620 {
1621         switch (f->type) {
1622         case PACKET_FANOUT_CBPF:
1623         case PACKET_FANOUT_EBPF:
1624                 __fanout_set_data_bpf(f, NULL);
1625         };
1626 }
1627
1628 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1629 {
1630         struct packet_rollover *rollover = NULL;
1631         struct packet_sock *po = pkt_sk(sk);
1632         struct packet_fanout *f, *match;
1633         u8 type = type_flags & 0xff;
1634         u8 flags = type_flags >> 8;
1635         int err;
1636
1637         switch (type) {
1638         case PACKET_FANOUT_ROLLOVER:
1639                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1640                         return -EINVAL;
1641         case PACKET_FANOUT_HASH:
1642         case PACKET_FANOUT_LB:
1643         case PACKET_FANOUT_CPU:
1644         case PACKET_FANOUT_RND:
1645         case PACKET_FANOUT_QM:
1646         case PACKET_FANOUT_CBPF:
1647         case PACKET_FANOUT_EBPF:
1648                 break;
1649         default:
1650                 return -EINVAL;
1651         }
1652
1653         mutex_lock(&fanout_mutex);
1654
1655         err = -EINVAL;
1656         if (!po->running)
1657                 goto out;
1658
1659         err = -EALREADY;
1660         if (po->fanout)
1661                 goto out;
1662
1663         if (type == PACKET_FANOUT_ROLLOVER ||
1664             (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1665                 err = -ENOMEM;
1666                 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1667                 if (!rollover)
1668                         goto out;
1669                 atomic_long_set(&rollover->num, 0);
1670                 atomic_long_set(&rollover->num_huge, 0);
1671                 atomic_long_set(&rollover->num_failed, 0);
1672                 po->rollover = rollover;
1673         }
1674
1675         match = NULL;
1676         list_for_each_entry(f, &fanout_list, list) {
1677                 if (f->id == id &&
1678                     read_pnet(&f->net) == sock_net(sk)) {
1679                         match = f;
1680                         break;
1681                 }
1682         }
1683         err = -EINVAL;
1684         if (match && match->flags != flags)
1685                 goto out;
1686         if (!match) {
1687                 err = -ENOMEM;
1688                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1689                 if (!match)
1690                         goto out;
1691                 write_pnet(&match->net, sock_net(sk));
1692                 match->id = id;
1693                 match->type = type;
1694                 match->flags = flags;
1695                 INIT_LIST_HEAD(&match->list);
1696                 spin_lock_init(&match->lock);
1697                 atomic_set(&match->sk_ref, 0);
1698                 fanout_init_data(match);
1699                 match->prot_hook.type = po->prot_hook.type;
1700                 match->prot_hook.dev = po->prot_hook.dev;
1701                 match->prot_hook.func = packet_rcv_fanout;
1702                 match->prot_hook.af_packet_priv = match;
1703                 match->prot_hook.id_match = match_fanout_group;
1704                 list_add(&match->list, &fanout_list);
1705         }
1706         err = -EINVAL;
1707         if (match->type == type &&
1708             match->prot_hook.type == po->prot_hook.type &&
1709             match->prot_hook.dev == po->prot_hook.dev) {
1710                 err = -ENOSPC;
1711                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1712                         __dev_remove_pack(&po->prot_hook);
1713                         po->fanout = match;
1714                         atomic_inc(&match->sk_ref);
1715                         __fanout_link(sk, po);
1716                         err = 0;
1717                 }
1718         }
1719 out:
1720         if (err && rollover) {
1721                 kfree(rollover);
1722                 po->rollover = NULL;
1723         }
1724         mutex_unlock(&fanout_mutex);
1725         return err;
1726 }
1727
1728 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1729  * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1730  * It is the responsibility of the caller to call fanout_release_data() and
1731  * free the returned packet_fanout (after synchronize_net())
1732  */
1733 static struct packet_fanout *fanout_release(struct sock *sk)
1734 {
1735         struct packet_sock *po = pkt_sk(sk);
1736         struct packet_fanout *f;
1737
1738         mutex_lock(&fanout_mutex);
1739         f = po->fanout;
1740         if (f) {
1741                 po->fanout = NULL;
1742
1743                 if (atomic_dec_and_test(&f->sk_ref))
1744                         list_del(&f->list);
1745                 else
1746                         f = NULL;
1747
1748                 if (po->rollover)
1749                         kfree_rcu(po->rollover, rcu);
1750         }
1751         mutex_unlock(&fanout_mutex);
1752
1753         return f;
1754 }
1755
1756 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1757                                           struct sk_buff *skb)
1758 {
1759         /* Earlier code assumed this would be a VLAN pkt, double-check
1760          * this now that we have the actual packet in hand. We can only
1761          * do this check on Ethernet devices.
1762          */
1763         if (unlikely(dev->type != ARPHRD_ETHER))
1764                 return false;
1765
1766         skb_reset_mac_header(skb);
1767         return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1768 }
1769
1770 static const struct proto_ops packet_ops;
1771
1772 static const struct proto_ops packet_ops_spkt;
1773
1774 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1775                            struct packet_type *pt, struct net_device *orig_dev)
1776 {
1777         struct sock *sk;
1778         struct sockaddr_pkt *spkt;
1779
1780         /*
1781          *      When we registered the protocol we saved the socket in the data
1782          *      field for just this event.
1783          */
1784
1785         sk = pt->af_packet_priv;
1786
1787         /*
1788          *      Yank back the headers [hope the device set this
1789          *      right or kerboom...]
1790          *
1791          *      Incoming packets have ll header pulled,
1792          *      push it back.
1793          *
1794          *      For outgoing ones skb->data == skb_mac_header(skb)
1795          *      so that this procedure is noop.
1796          */
1797
1798         if (skb->pkt_type == PACKET_LOOPBACK)
1799                 goto out;
1800
1801         if (!net_eq(dev_net(dev), sock_net(sk)))
1802                 goto out;
1803
1804         skb = skb_share_check(skb, GFP_ATOMIC);
1805         if (skb == NULL)
1806                 goto oom;
1807
1808         /* drop any routing info */
1809         skb_dst_drop(skb);
1810
1811         /* drop conntrack reference */
1812         nf_reset(skb);
1813
1814         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1815
1816         skb_push(skb, skb->data - skb_mac_header(skb));
1817
1818         /*
1819          *      The SOCK_PACKET socket receives _all_ frames.
1820          */
1821
1822         spkt->spkt_family = dev->type;
1823         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1824         spkt->spkt_protocol = skb->protocol;
1825
1826         /*
1827          *      Charge the memory to the socket. This is done specifically
1828          *      to prevent sockets using all the memory up.
1829          */
1830
1831         if (sock_queue_rcv_skb(sk, skb) == 0)
1832                 return 0;
1833
1834 out:
1835         kfree_skb(skb);
1836 oom:
1837         return 0;
1838 }
1839
1840
1841 /*
1842  *      Output a raw packet to a device layer. This bypasses all the other
1843  *      protocol layers and you must therefore supply it with a complete frame
1844  */
1845
1846 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1847                                size_t len)
1848 {
1849         struct sock *sk = sock->sk;
1850         DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1851         struct sk_buff *skb = NULL;
1852         struct net_device *dev;
1853         __be16 proto = 0;
1854         int err;
1855         int extra_len = 0;
1856
1857         /*
1858          *      Get and verify the address.
1859          */
1860
1861         if (saddr) {
1862                 if (msg->msg_namelen < sizeof(struct sockaddr))
1863                         return -EINVAL;
1864                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1865                         proto = saddr->spkt_protocol;
1866         } else
1867                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1868
1869         /*
1870          *      Find the device first to size check it
1871          */
1872
1873         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1874 retry:
1875         rcu_read_lock();
1876         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1877         err = -ENODEV;
1878         if (dev == NULL)
1879                 goto out_unlock;
1880
1881         err = -ENETDOWN;
1882         if (!(dev->flags & IFF_UP))
1883                 goto out_unlock;
1884
1885         /*
1886          * You may not queue a frame bigger than the mtu. This is the lowest level
1887          * raw protocol and you must do your own fragmentation at this level.
1888          */
1889
1890         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1891                 if (!netif_supports_nofcs(dev)) {
1892                         err = -EPROTONOSUPPORT;
1893                         goto out_unlock;
1894                 }
1895                 extra_len = 4; /* We're doing our own CRC */
1896         }
1897
1898         err = -EMSGSIZE;
1899         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1900                 goto out_unlock;
1901
1902         if (!skb) {
1903                 size_t reserved = LL_RESERVED_SPACE(dev);
1904                 int tlen = dev->needed_tailroom;
1905                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1906
1907                 rcu_read_unlock();
1908                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1909                 if (skb == NULL)
1910                         return -ENOBUFS;
1911                 /* FIXME: Save some space for broken drivers that write a hard
1912                  * header at transmission time by themselves. PPP is the notable
1913                  * one here. This should really be fixed at the driver level.
1914                  */
1915                 skb_reserve(skb, reserved);
1916                 skb_reset_network_header(skb);
1917
1918                 /* Try to align data part correctly */
1919                 if (hhlen) {
1920                         skb->data -= hhlen;
1921                         skb->tail -= hhlen;
1922                         if (len < hhlen)
1923                                 skb_reset_network_header(skb);
1924                 }
1925                 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1926                 if (err)
1927                         goto out_free;
1928                 goto retry;
1929         }
1930
1931         if (!dev_validate_header(dev, skb->data, len)) {
1932                 err = -EINVAL;
1933                 goto out_unlock;
1934         }
1935         if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1936             !packet_extra_vlan_len_allowed(dev, skb)) {
1937                 err = -EMSGSIZE;
1938                 goto out_unlock;
1939         }
1940
1941         skb->protocol = proto;
1942         skb->dev = dev;
1943         skb->priority = sk->sk_priority;
1944         skb->mark = sk->sk_mark;
1945
1946         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1947
1948         if (unlikely(extra_len == 4))
1949                 skb->no_fcs = 1;
1950
1951         skb_probe_transport_header(skb, 0);
1952
1953         dev_queue_xmit(skb);
1954         rcu_read_unlock();
1955         return len;
1956
1957 out_unlock:
1958         rcu_read_unlock();
1959 out_free:
1960         kfree_skb(skb);
1961         return err;
1962 }
1963
1964 static unsigned int run_filter(struct sk_buff *skb,
1965                                const struct sock *sk,
1966                                unsigned int res)
1967 {
1968         struct sk_filter *filter;
1969
1970         rcu_read_lock();
1971         filter = rcu_dereference(sk->sk_filter);
1972         if (filter != NULL)
1973                 res = bpf_prog_run_clear_cb(filter->prog, skb);
1974         rcu_read_unlock();
1975
1976         return res;
1977 }
1978
1979 /*
1980  * This function makes lazy skb cloning in hope that most of packets
1981  * are discarded by BPF.
1982  *
1983  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1984  * and skb->cb are mangled. It works because (and until) packets
1985  * falling here are owned by current CPU. Output packets are cloned
1986  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1987  * sequencially, so that if we return skb to original state on exit,
1988  * we will not harm anyone.
1989  */
1990
1991 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1992                       struct packet_type *pt, struct net_device *orig_dev)
1993 {
1994         struct sock *sk;
1995         struct sockaddr_ll *sll;
1996         struct packet_sock *po;
1997         u8 *skb_head = skb->data;
1998         int skb_len = skb->len;
1999         unsigned int snaplen, res;
2000
2001         if (skb->pkt_type == PACKET_LOOPBACK)
2002                 goto drop;
2003
2004         sk = pt->af_packet_priv;
2005         po = pkt_sk(sk);
2006
2007         if (!net_eq(dev_net(dev), sock_net(sk)))
2008                 goto drop;
2009
2010         skb->dev = dev;
2011
2012         if (dev->header_ops) {
2013                 /* The device has an explicit notion of ll header,
2014                  * exported to higher levels.
2015                  *
2016                  * Otherwise, the device hides details of its frame
2017                  * structure, so that corresponding packet head is
2018                  * never delivered to user.
2019                  */
2020                 if (sk->sk_type != SOCK_DGRAM)
2021                         skb_push(skb, skb->data - skb_mac_header(skb));
2022                 else if (skb->pkt_type == PACKET_OUTGOING) {
2023                         /* Special case: outgoing packets have ll header at head */
2024                         skb_pull(skb, skb_network_offset(skb));
2025                 }
2026         }
2027
2028         snaplen = skb->len;
2029
2030         res = run_filter(skb, sk, snaplen);
2031         if (!res)
2032                 goto drop_n_restore;
2033         if (snaplen > res)
2034                 snaplen = res;
2035
2036         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2037                 goto drop_n_acct;
2038
2039         if (skb_shared(skb)) {
2040                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2041                 if (nskb == NULL)
2042                         goto drop_n_acct;
2043
2044                 if (skb_head != skb->data) {
2045                         skb->data = skb_head;
2046                         skb->len = skb_len;
2047                 }
2048                 consume_skb(skb);
2049                 skb = nskb;
2050         }
2051
2052         sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2053
2054         sll = &PACKET_SKB_CB(skb)->sa.ll;
2055         sll->sll_hatype = dev->type;
2056         sll->sll_pkttype = skb->pkt_type;
2057         if (unlikely(po->origdev))
2058                 sll->sll_ifindex = orig_dev->ifindex;
2059         else
2060                 sll->sll_ifindex = dev->ifindex;
2061
2062         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2063
2064         /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2065          * Use their space for storing the original skb length.
2066          */
2067         PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2068
2069         if (pskb_trim(skb, snaplen))
2070                 goto drop_n_acct;
2071
2072         skb_set_owner_r(skb, sk);
2073         skb->dev = NULL;
2074         skb_dst_drop(skb);
2075
2076         /* drop conntrack reference */
2077         nf_reset(skb);
2078
2079         spin_lock(&sk->sk_receive_queue.lock);
2080         po->stats.stats1.tp_packets++;
2081         sock_skb_set_dropcount(sk, skb);
2082         __skb_queue_tail(&sk->sk_receive_queue, skb);
2083         spin_unlock(&sk->sk_receive_queue.lock);
2084         sk->sk_data_ready(sk);
2085         return 0;
2086
2087 drop_n_acct:
2088         spin_lock(&sk->sk_receive_queue.lock);
2089         po->stats.stats1.tp_drops++;
2090         atomic_inc(&sk->sk_drops);
2091         spin_unlock(&sk->sk_receive_queue.lock);
2092
2093 drop_n_restore:
2094         if (skb_head != skb->data && skb_shared(skb)) {
2095                 skb->data = skb_head;
2096                 skb->len = skb_len;
2097         }
2098 drop:
2099         consume_skb(skb);
2100         return 0;
2101 }
2102
2103 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2104                        struct packet_type *pt, struct net_device *orig_dev)
2105 {
2106         struct sock *sk;
2107         struct packet_sock *po;
2108         struct sockaddr_ll *sll;
2109         union tpacket_uhdr h;
2110         u8 *skb_head = skb->data;
2111         int skb_len = skb->len;
2112         unsigned int snaplen, res;
2113         unsigned long status = TP_STATUS_USER;
2114         unsigned short macoff, netoff, hdrlen;
2115         struct sk_buff *copy_skb = NULL;
2116         struct timespec ts;
2117         __u32 ts_status;
2118
2119         /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2120          * We may add members to them until current aligned size without forcing
2121          * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2122          */
2123         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2124         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2125
2126         if (skb->pkt_type == PACKET_LOOPBACK)
2127                 goto drop;
2128
2129         sk = pt->af_packet_priv;
2130         po = pkt_sk(sk);
2131
2132         if (!net_eq(dev_net(dev), sock_net(sk)))
2133                 goto drop;
2134
2135         if (dev->header_ops) {
2136                 if (sk->sk_type != SOCK_DGRAM)
2137                         skb_push(skb, skb->data - skb_mac_header(skb));
2138                 else if (skb->pkt_type == PACKET_OUTGOING) {
2139                         /* Special case: outgoing packets have ll header at head */
2140                         skb_pull(skb, skb_network_offset(skb));
2141                 }
2142         }
2143
2144         snaplen = skb->len;
2145
2146         res = run_filter(skb, sk, snaplen);
2147         if (!res)
2148                 goto drop_n_restore;
2149
2150         if (skb->ip_summed == CHECKSUM_PARTIAL)
2151                 status |= TP_STATUS_CSUMNOTREADY;
2152         else if (skb->pkt_type != PACKET_OUTGOING &&
2153                  (skb->ip_summed == CHECKSUM_COMPLETE ||
2154                   skb_csum_unnecessary(skb)))
2155                 status |= TP_STATUS_CSUM_VALID;
2156
2157         if (snaplen > res)
2158                 snaplen = res;
2159
2160         if (sk->sk_type == SOCK_DGRAM) {
2161                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2162                                   po->tp_reserve;
2163         } else {
2164                 unsigned int maclen = skb_network_offset(skb);
2165                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2166                                        (maclen < 16 ? 16 : maclen)) +
2167                         po->tp_reserve;
2168                 macoff = netoff - maclen;
2169         }
2170         if (po->tp_version <= TPACKET_V2) {
2171                 if (macoff + snaplen > po->rx_ring.frame_size) {
2172                         if (po->copy_thresh &&
2173                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2174                                 if (skb_shared(skb)) {
2175                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
2176                                 } else {
2177                                         copy_skb = skb_get(skb);
2178                                         skb_head = skb->data;
2179                                 }
2180                                 if (copy_skb)
2181                                         skb_set_owner_r(copy_skb, sk);
2182                         }
2183                         snaplen = po->rx_ring.frame_size - macoff;
2184                         if ((int)snaplen < 0)
2185                                 snaplen = 0;
2186                 }
2187         } else if (unlikely(macoff + snaplen >
2188                             GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2189                 u32 nval;
2190
2191                 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2192                 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2193                             snaplen, nval, macoff);
2194                 snaplen = nval;
2195                 if (unlikely((int)snaplen < 0)) {
2196                         snaplen = 0;
2197                         macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2198                 }
2199         }
2200         spin_lock(&sk->sk_receive_queue.lock);
2201         h.raw = packet_current_rx_frame(po, skb,
2202                                         TP_STATUS_KERNEL, (macoff+snaplen));
2203         if (!h.raw)
2204                 goto ring_is_full;
2205         if (po->tp_version <= TPACKET_V2) {
2206                 packet_increment_rx_head(po, &po->rx_ring);
2207         /*
2208          * LOSING will be reported till you read the stats,
2209          * because it's COR - Clear On Read.
2210          * Anyways, moving it for V1/V2 only as V3 doesn't need this
2211          * at packet level.
2212          */
2213                 if (po->stats.stats1.tp_drops)
2214                         status |= TP_STATUS_LOSING;
2215         }
2216         po->stats.stats1.tp_packets++;
2217         if (copy_skb) {
2218                 status |= TP_STATUS_COPY;
2219                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2220         }
2221         spin_unlock(&sk->sk_receive_queue.lock);
2222
2223         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2224
2225         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2226                 getnstimeofday(&ts);
2227
2228         status |= ts_status;
2229
2230         switch (po->tp_version) {
2231         case TPACKET_V1:
2232                 h.h1->tp_len = skb->len;
2233                 h.h1->tp_snaplen = snaplen;
2234                 h.h1->tp_mac = macoff;
2235                 h.h1->tp_net = netoff;
2236                 h.h1->tp_sec = ts.tv_sec;
2237                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2238                 hdrlen = sizeof(*h.h1);
2239                 break;
2240         case TPACKET_V2:
2241                 h.h2->tp_len = skb->len;
2242                 h.h2->tp_snaplen = snaplen;
2243                 h.h2->tp_mac = macoff;
2244                 h.h2->tp_net = netoff;
2245                 h.h2->tp_sec = ts.tv_sec;
2246                 h.h2->tp_nsec = ts.tv_nsec;
2247                 if (skb_vlan_tag_present(skb)) {
2248                         h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2249                         h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2250                         status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2251                 } else {
2252                         h.h2->tp_vlan_tci = 0;
2253                         h.h2->tp_vlan_tpid = 0;
2254                 }
2255                 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2256                 hdrlen = sizeof(*h.h2);
2257                 break;
2258         case TPACKET_V3:
2259                 /* tp_nxt_offset,vlan are already populated above.
2260                  * So DONT clear those fields here
2261                  */
2262                 h.h3->tp_status |= status;
2263                 h.h3->tp_len = skb->len;
2264                 h.h3->tp_snaplen = snaplen;
2265                 h.h3->tp_mac = macoff;
2266                 h.h3->tp_net = netoff;
2267                 h.h3->tp_sec  = ts.tv_sec;
2268                 h.h3->tp_nsec = ts.tv_nsec;
2269                 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2270                 hdrlen = sizeof(*h.h3);
2271                 break;
2272         default:
2273                 BUG();
2274         }
2275
2276         sll = h.raw + TPACKET_ALIGN(hdrlen);
2277         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2278         sll->sll_family = AF_PACKET;
2279         sll->sll_hatype = dev->type;
2280         sll->sll_protocol = skb->protocol;
2281         sll->sll_pkttype = skb->pkt_type;
2282         if (unlikely(po->origdev))
2283                 sll->sll_ifindex = orig_dev->ifindex;
2284         else
2285                 sll->sll_ifindex = dev->ifindex;
2286
2287         smp_mb();
2288
2289 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2290         if (po->tp_version <= TPACKET_V2) {
2291                 u8 *start, *end;
2292
2293                 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2294                                         macoff + snaplen);
2295
2296                 for (start = h.raw; start < end; start += PAGE_SIZE)
2297                         flush_dcache_page(pgv_to_page(start));
2298         }
2299         smp_wmb();
2300 #endif
2301
2302         if (po->tp_version <= TPACKET_V2) {
2303                 __packet_set_status(po, h.raw, status);
2304                 sk->sk_data_ready(sk);
2305         } else {
2306                 prb_clear_blk_fill_status(&po->rx_ring);
2307         }
2308
2309 drop_n_restore:
2310         if (skb_head != skb->data && skb_shared(skb)) {
2311                 skb->data = skb_head;
2312                 skb->len = skb_len;
2313         }
2314 drop:
2315         kfree_skb(skb);
2316         return 0;
2317
2318 ring_is_full:
2319         po->stats.stats1.tp_drops++;
2320         spin_unlock(&sk->sk_receive_queue.lock);
2321
2322         sk->sk_data_ready(sk);
2323         kfree_skb(copy_skb);
2324         goto drop_n_restore;
2325 }
2326
2327 static void tpacket_destruct_skb(struct sk_buff *skb)
2328 {
2329         struct packet_sock *po = pkt_sk(skb->sk);
2330
2331         if (likely(po->tx_ring.pg_vec)) {
2332                 void *ph;
2333                 __u32 ts;
2334
2335                 ph = skb_shinfo(skb)->destructor_arg;
2336                 packet_dec_pending(&po->tx_ring);
2337
2338                 ts = __packet_set_timestamp(po, ph, skb);
2339                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2340         }
2341
2342         sock_wfree(skb);
2343 }
2344
2345 static void tpacket_set_protocol(const struct net_device *dev,
2346                                  struct sk_buff *skb)
2347 {
2348         if (dev->type == ARPHRD_ETHER) {
2349                 skb_reset_mac_header(skb);
2350                 skb->protocol = eth_hdr(skb)->h_proto;
2351         }
2352 }
2353
2354 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2355                 void *frame, struct net_device *dev, int size_max,
2356                 __be16 proto, unsigned char *addr, int hlen)
2357 {
2358         union tpacket_uhdr ph;
2359         int to_write, offset, len, tp_len, nr_frags, len_max;
2360         struct socket *sock = po->sk.sk_socket;
2361         struct page *page;
2362         void *data;
2363         int err;
2364
2365         ph.raw = frame;
2366
2367         skb->protocol = proto;
2368         skb->dev = dev;
2369         skb->priority = po->sk.sk_priority;
2370         skb->mark = po->sk.sk_mark;
2371         sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2372         skb_shinfo(skb)->destructor_arg = ph.raw;
2373
2374         switch (po->tp_version) {
2375         case TPACKET_V2:
2376                 tp_len = ph.h2->tp_len;
2377                 break;
2378         default:
2379                 tp_len = ph.h1->tp_len;
2380                 break;
2381         }
2382         if (unlikely(tp_len > size_max)) {
2383                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2384                 return -EMSGSIZE;
2385         }
2386
2387         skb_reserve(skb, hlen);
2388         skb_reset_network_header(skb);
2389
2390         if (unlikely(po->tp_tx_has_off)) {
2391                 int off_min, off_max, off;
2392                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2393                 off_max = po->tx_ring.frame_size - tp_len;
2394                 if (sock->type == SOCK_DGRAM) {
2395                         switch (po->tp_version) {
2396                         case TPACKET_V2:
2397                                 off = ph.h2->tp_net;
2398                                 break;
2399                         default:
2400                                 off = ph.h1->tp_net;
2401                                 break;
2402                         }
2403                 } else {
2404                         switch (po->tp_version) {
2405                         case TPACKET_V2:
2406                                 off = ph.h2->tp_mac;
2407                                 break;
2408                         default:
2409                                 off = ph.h1->tp_mac;
2410                                 break;
2411                         }
2412                 }
2413                 if (unlikely((off < off_min) || (off_max < off)))
2414                         return -EINVAL;
2415                 data = ph.raw + off;
2416         } else {
2417                 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2418         }
2419         to_write = tp_len;
2420
2421         if (sock->type == SOCK_DGRAM) {
2422                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2423                                 NULL, tp_len);
2424                 if (unlikely(err < 0))
2425                         return -EINVAL;
2426         } else if (dev->hard_header_len) {
2427                 int hdrlen = min_t(int, dev->hard_header_len, tp_len);
2428
2429                 skb_push(skb, dev->hard_header_len);
2430                 err = skb_store_bits(skb, 0, data, hdrlen);
2431                 if (unlikely(err))
2432                         return err;
2433                 if (!dev_validate_header(dev, skb->data, hdrlen))
2434                         return -EINVAL;
2435                 if (!skb->protocol)
2436                         tpacket_set_protocol(dev, skb);
2437
2438                 data += hdrlen;
2439                 to_write -= hdrlen;
2440         }
2441
2442         offset = offset_in_page(data);
2443         len_max = PAGE_SIZE - offset;
2444         len = ((to_write > len_max) ? len_max : to_write);
2445
2446         skb->data_len = to_write;
2447         skb->len += to_write;
2448         skb->truesize += to_write;
2449         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2450
2451         while (likely(to_write)) {
2452                 nr_frags = skb_shinfo(skb)->nr_frags;
2453
2454                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2455                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2456                                MAX_SKB_FRAGS);
2457                         return -EFAULT;
2458                 }
2459
2460                 page = pgv_to_page(data);
2461                 data += len;
2462                 flush_dcache_page(page);
2463                 get_page(page);
2464                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2465                 to_write -= len;
2466                 offset = 0;
2467                 len_max = PAGE_SIZE;
2468                 len = ((to_write > len_max) ? len_max : to_write);
2469         }
2470
2471         skb_probe_transport_header(skb, 0);
2472
2473         return tp_len;
2474 }
2475
2476 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2477 {
2478         struct sk_buff *skb;
2479         struct net_device *dev;
2480         __be16 proto;
2481         int err, reserve = 0;
2482         void *ph;
2483         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2484         bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2485         int tp_len, size_max;
2486         unsigned char *addr;
2487         int len_sum = 0;
2488         int status = TP_STATUS_AVAILABLE;
2489         int hlen, tlen;
2490
2491         mutex_lock(&po->pg_vec_lock);
2492
2493         if (likely(saddr == NULL)) {
2494                 dev     = packet_cached_dev_get(po);
2495                 proto   = po->num;
2496                 addr    = NULL;
2497         } else {
2498                 err = -EINVAL;
2499                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2500                         goto out;
2501                 if (msg->msg_namelen < (saddr->sll_halen
2502                                         + offsetof(struct sockaddr_ll,
2503                                                 sll_addr)))
2504                         goto out;
2505                 proto   = saddr->sll_protocol;
2506                 addr    = saddr->sll_addr;
2507                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2508         }
2509
2510         err = -ENXIO;
2511         if (unlikely(dev == NULL))
2512                 goto out;
2513         err = -ENETDOWN;
2514         if (unlikely(!(dev->flags & IFF_UP)))
2515                 goto out_put;
2516
2517         if (po->sk.sk_socket->type == SOCK_RAW)
2518                 reserve = dev->hard_header_len;
2519         size_max = po->tx_ring.frame_size
2520                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2521
2522         if (size_max > dev->mtu + reserve + VLAN_HLEN)
2523                 size_max = dev->mtu + reserve + VLAN_HLEN;
2524
2525         do {
2526                 ph = packet_current_frame(po, &po->tx_ring,
2527                                           TP_STATUS_SEND_REQUEST);
2528                 if (unlikely(ph == NULL)) {
2529                         if (need_wait && need_resched())
2530                                 schedule();
2531                         continue;
2532                 }
2533
2534                 status = TP_STATUS_SEND_REQUEST;
2535                 hlen = LL_RESERVED_SPACE(dev);
2536                 tlen = dev->needed_tailroom;
2537                 skb = sock_alloc_send_skb(&po->sk,
2538                                 hlen + tlen + sizeof(struct sockaddr_ll),
2539                                 !need_wait, &err);
2540
2541                 if (unlikely(skb == NULL)) {
2542                         /* we assume the socket was initially writeable ... */
2543                         if (likely(len_sum > 0))
2544                                 err = len_sum;
2545                         goto out_status;
2546                 }
2547                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2548                                           addr, hlen);
2549                 if (likely(tp_len >= 0) &&
2550                     tp_len > dev->mtu + reserve &&
2551                     !packet_extra_vlan_len_allowed(dev, skb))
2552                         tp_len = -EMSGSIZE;
2553
2554                 if (unlikely(tp_len < 0)) {
2555                         if (po->tp_loss) {
2556                                 __packet_set_status(po, ph,
2557                                                 TP_STATUS_AVAILABLE);
2558                                 packet_increment_head(&po->tx_ring);
2559                                 kfree_skb(skb);
2560                                 continue;
2561                         } else {
2562                                 status = TP_STATUS_WRONG_FORMAT;
2563                                 err = tp_len;
2564                                 goto out_status;
2565                         }
2566                 }
2567
2568                 packet_pick_tx_queue(dev, skb);
2569
2570                 skb->destructor = tpacket_destruct_skb;
2571                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2572                 packet_inc_pending(&po->tx_ring);
2573
2574                 status = TP_STATUS_SEND_REQUEST;
2575                 err = po->xmit(skb);
2576                 if (unlikely(err > 0)) {
2577                         err = net_xmit_errno(err);
2578                         if (err && __packet_get_status(po, ph) ==
2579                                    TP_STATUS_AVAILABLE) {
2580                                 /* skb was destructed already */
2581                                 skb = NULL;
2582                                 goto out_status;
2583                         }
2584                         /*
2585                          * skb was dropped but not destructed yet;
2586                          * let's treat it like congestion or err < 0
2587                          */
2588                         err = 0;
2589                 }
2590                 packet_increment_head(&po->tx_ring);
2591                 len_sum += tp_len;
2592         } while (likely((ph != NULL) ||
2593                 /* Note: packet_read_pending() might be slow if we have
2594                  * to call it as it's per_cpu variable, but in fast-path
2595                  * we already short-circuit the loop with the first
2596                  * condition, and luckily don't have to go that path
2597                  * anyway.
2598                  */
2599                  (need_wait && packet_read_pending(&po->tx_ring))));
2600
2601         err = len_sum;
2602         goto out_put;
2603
2604 out_status:
2605         __packet_set_status(po, ph, status);
2606         kfree_skb(skb);
2607 out_put:
2608         dev_put(dev);
2609 out:
2610         mutex_unlock(&po->pg_vec_lock);
2611         return err;
2612 }
2613
2614 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2615                                         size_t reserve, size_t len,
2616                                         size_t linear, int noblock,
2617                                         int *err)
2618 {
2619         struct sk_buff *skb;
2620
2621         /* Under a page?  Don't bother with paged skb. */
2622         if (prepad + len < PAGE_SIZE || !linear)
2623                 linear = len;
2624
2625         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2626                                    err, 0);
2627         if (!skb)
2628                 return NULL;
2629
2630         skb_reserve(skb, reserve);
2631         skb_put(skb, linear);
2632         skb->data_len = len - linear;
2633         skb->len += len - linear;
2634
2635         return skb;
2636 }
2637
2638 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2639 {
2640         struct sock *sk = sock->sk;
2641         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2642         struct sk_buff *skb;
2643         struct net_device *dev;
2644         __be16 proto;
2645         unsigned char *addr;
2646         int err, reserve = 0;
2647         struct sockcm_cookie sockc;
2648         struct virtio_net_hdr vnet_hdr = { 0 };
2649         int offset = 0;
2650         int vnet_hdr_len;
2651         struct packet_sock *po = pkt_sk(sk);
2652         unsigned short gso_type = 0;
2653         int hlen, tlen, linear;
2654         int extra_len = 0;
2655         ssize_t n;
2656
2657         /*
2658          *      Get and verify the address.
2659          */
2660
2661         if (likely(saddr == NULL)) {
2662                 dev     = packet_cached_dev_get(po);
2663                 proto   = po->num;
2664                 addr    = NULL;
2665         } else {
2666                 err = -EINVAL;
2667                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2668                         goto out;
2669                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2670                         goto out;
2671                 proto   = saddr->sll_protocol;
2672                 addr    = saddr->sll_addr;
2673                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2674         }
2675
2676         err = -ENXIO;
2677         if (unlikely(dev == NULL))
2678                 goto out_unlock;
2679         err = -ENETDOWN;
2680         if (unlikely(!(dev->flags & IFF_UP)))
2681                 goto out_unlock;
2682
2683         sockc.mark = sk->sk_mark;
2684         if (msg->msg_controllen) {
2685                 err = sock_cmsg_send(sk, msg, &sockc);
2686                 if (unlikely(err))
2687                         goto out_unlock;
2688         }
2689
2690         if (sock->type == SOCK_RAW)
2691                 reserve = dev->hard_header_len;
2692         if (po->has_vnet_hdr) {
2693                 vnet_hdr_len = sizeof(vnet_hdr);
2694
2695                 err = -EINVAL;
2696                 if (len < vnet_hdr_len)
2697                         goto out_unlock;
2698
2699                 len -= vnet_hdr_len;
2700
2701                 err = -EFAULT;
2702                 n = copy_from_iter(&vnet_hdr, vnet_hdr_len, &msg->msg_iter);
2703                 if (n != vnet_hdr_len)
2704                         goto out_unlock;
2705
2706                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2707                     (__virtio16_to_cpu(vio_le(), vnet_hdr.csum_start) +
2708                      __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset) + 2 >
2709                       __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len)))
2710                         vnet_hdr.hdr_len = __cpu_to_virtio16(vio_le(),
2711                                  __virtio16_to_cpu(vio_le(), vnet_hdr.csum_start) +
2712                                 __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset) + 2);
2713
2714                 err = -EINVAL;
2715                 if (__virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len) > len)
2716                         goto out_unlock;
2717
2718                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2719                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2720                         case VIRTIO_NET_HDR_GSO_TCPV4:
2721                                 gso_type = SKB_GSO_TCPV4;
2722                                 break;
2723                         case VIRTIO_NET_HDR_GSO_TCPV6:
2724                                 gso_type = SKB_GSO_TCPV6;
2725                                 break;
2726                         case VIRTIO_NET_HDR_GSO_UDP:
2727                                 gso_type = SKB_GSO_UDP;
2728                                 break;
2729                         default:
2730                                 goto out_unlock;
2731                         }
2732
2733                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2734                                 gso_type |= SKB_GSO_TCP_ECN;
2735
2736                         if (vnet_hdr.gso_size == 0)
2737                                 goto out_unlock;
2738
2739                 }
2740         }
2741
2742         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2743                 if (!netif_supports_nofcs(dev)) {
2744                         err = -EPROTONOSUPPORT;
2745                         goto out_unlock;
2746                 }
2747                 extra_len = 4; /* We're doing our own CRC */
2748         }
2749
2750         err = -EMSGSIZE;
2751         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2752                 goto out_unlock;
2753
2754         err = -ENOBUFS;
2755         hlen = LL_RESERVED_SPACE(dev);
2756         tlen = dev->needed_tailroom;
2757         linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2758         linear = max(linear, min_t(int, len, dev->hard_header_len));
2759         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2760                                msg->msg_flags & MSG_DONTWAIT, &err);
2761         if (skb == NULL)
2762                 goto out_unlock;
2763
2764         skb_set_network_header(skb, reserve);
2765
2766         err = -EINVAL;
2767         if (sock->type == SOCK_DGRAM) {
2768                 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2769                 if (unlikely(offset < 0))
2770                         goto out_free;
2771         }
2772
2773         /* Returns -EFAULT on error */
2774         err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2775         if (err)
2776                 goto out_free;
2777
2778         if (sock->type == SOCK_RAW &&
2779             !dev_validate_header(dev, skb->data, len)) {
2780                 err = -EINVAL;
2781                 goto out_free;
2782         }
2783
2784         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2785
2786         if (!gso_type && (len > dev->mtu + reserve + extra_len) &&
2787             !packet_extra_vlan_len_allowed(dev, skb)) {
2788                 err = -EMSGSIZE;
2789                 goto out_free;
2790         }
2791
2792         skb->protocol = proto;
2793         skb->dev = dev;
2794         skb->priority = sk->sk_priority;
2795         skb->mark = sockc.mark;
2796
2797         packet_pick_tx_queue(dev, skb);
2798
2799         if (po->has_vnet_hdr) {
2800                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2801                         u16 s = __virtio16_to_cpu(vio_le(), vnet_hdr.csum_start);
2802                         u16 o = __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset);
2803                         if (!skb_partial_csum_set(skb, s, o)) {
2804                                 err = -EINVAL;
2805                                 goto out_free;
2806                         }
2807                 }
2808
2809                 skb_shinfo(skb)->gso_size =
2810                         __virtio16_to_cpu(vio_le(), vnet_hdr.gso_size);
2811                 skb_shinfo(skb)->gso_type = gso_type;
2812
2813                 /* Header must be checked, and gso_segs computed. */
2814                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2815                 skb_shinfo(skb)->gso_segs = 0;
2816
2817                 len += vnet_hdr_len;
2818         }
2819
2820         skb_probe_transport_header(skb, reserve);
2821
2822         if (unlikely(extra_len == 4))
2823                 skb->no_fcs = 1;
2824
2825         err = po->xmit(skb);
2826         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2827                 goto out_unlock;
2828
2829         dev_put(dev);
2830
2831         return len;
2832
2833 out_free:
2834         kfree_skb(skb);
2835 out_unlock:
2836         if (dev)
2837                 dev_put(dev);
2838 out:
2839         return err;
2840 }
2841
2842 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2843 {
2844         struct sock *sk = sock->sk;
2845         struct packet_sock *po = pkt_sk(sk);
2846
2847         if (po->tx_ring.pg_vec)
2848                 return tpacket_snd(po, msg);
2849         else
2850                 return packet_snd(sock, msg, len);
2851 }
2852
2853 /*
2854  *      Close a PACKET socket. This is fairly simple. We immediately go
2855  *      to 'closed' state and remove our protocol entry in the device list.
2856  */
2857
2858 static int packet_release(struct socket *sock)
2859 {
2860         struct sock *sk = sock->sk;
2861         struct packet_sock *po;
2862         struct packet_fanout *f;
2863         struct net *net;
2864         union tpacket_req_u req_u;
2865
2866         if (!sk)
2867                 return 0;
2868
2869         net = sock_net(sk);
2870         po = pkt_sk(sk);
2871
2872         mutex_lock(&net->packet.sklist_lock);
2873         sk_del_node_init_rcu(sk);
2874         mutex_unlock(&net->packet.sklist_lock);
2875
2876         preempt_disable();
2877         sock_prot_inuse_add(net, sk->sk_prot, -1);
2878         preempt_enable();
2879
2880         spin_lock(&po->bind_lock);
2881         unregister_prot_hook(sk, false);
2882         packet_cached_dev_reset(po);
2883
2884         if (po->prot_hook.dev) {
2885                 dev_put(po->prot_hook.dev);
2886                 po->prot_hook.dev = NULL;
2887         }
2888         spin_unlock(&po->bind_lock);
2889
2890         packet_flush_mclist(sk);
2891
2892         if (po->rx_ring.pg_vec) {
2893                 memset(&req_u, 0, sizeof(req_u));
2894                 packet_set_ring(sk, &req_u, 1, 0);
2895         }
2896
2897         if (po->tx_ring.pg_vec) {
2898                 memset(&req_u, 0, sizeof(req_u));
2899                 packet_set_ring(sk, &req_u, 1, 1);
2900         }
2901
2902         f = fanout_release(sk);
2903
2904         synchronize_net();
2905
2906         if (f) {
2907                 fanout_release_data(f);
2908                 kfree(f);
2909         }
2910         /*
2911          *      Now the socket is dead. No more input will appear.
2912          */
2913         sock_orphan(sk);
2914         sock->sk = NULL;
2915
2916         /* Purge queues */
2917
2918         skb_queue_purge(&sk->sk_receive_queue);
2919         packet_free_pending(po);
2920         sk_refcnt_debug_release(sk);
2921
2922         sock_put(sk);
2923         return 0;
2924 }
2925
2926 /*
2927  *      Attach a packet hook.
2928  */
2929
2930 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
2931                           __be16 proto)
2932 {
2933         struct packet_sock *po = pkt_sk(sk);
2934         struct net_device *dev_curr;
2935         __be16 proto_curr;
2936         bool need_rehook;
2937         struct net_device *dev = NULL;
2938         int ret = 0;
2939         bool unlisted = false;
2940
2941         if (po->fanout)
2942                 return -EINVAL;
2943
2944         lock_sock(sk);
2945         spin_lock(&po->bind_lock);
2946         rcu_read_lock();
2947
2948         if (name) {
2949                 dev = dev_get_by_name_rcu(sock_net(sk), name);
2950                 if (!dev) {
2951                         ret = -ENODEV;
2952                         goto out_unlock;
2953                 }
2954         } else if (ifindex) {
2955                 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
2956                 if (!dev) {
2957                         ret = -ENODEV;
2958                         goto out_unlock;
2959                 }
2960         }
2961
2962         if (dev)
2963                 dev_hold(dev);
2964
2965         proto_curr = po->prot_hook.type;
2966         dev_curr = po->prot_hook.dev;
2967
2968         need_rehook = proto_curr != proto || dev_curr != dev;
2969
2970         if (need_rehook) {
2971                 if (po->running) {
2972                         rcu_read_unlock();
2973                         __unregister_prot_hook(sk, true);
2974                         rcu_read_lock();
2975                         dev_curr = po->prot_hook.dev;
2976                         if (dev)
2977                                 unlisted = !dev_get_by_index_rcu(sock_net(sk),
2978                                                                  dev->ifindex);
2979                 }
2980
2981                 po->num = proto;
2982                 po->prot_hook.type = proto;
2983
2984                 if (unlikely(unlisted)) {
2985                         dev_put(dev);
2986                         po->prot_hook.dev = NULL;
2987                         po->ifindex = -1;
2988                         packet_cached_dev_reset(po);
2989                 } else {
2990                         po->prot_hook.dev = dev;
2991                         po->ifindex = dev ? dev->ifindex : 0;
2992                         packet_cached_dev_assign(po, dev);
2993                 }
2994         }
2995         if (dev_curr)
2996                 dev_put(dev_curr);
2997
2998         if (proto == 0 || !need_rehook)
2999                 goto out_unlock;
3000
3001         if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3002                 register_prot_hook(sk);
3003         } else {
3004                 sk->sk_err = ENETDOWN;
3005                 if (!sock_flag(sk, SOCK_DEAD))
3006                         sk->sk_error_report(sk);
3007         }
3008
3009 out_unlock:
3010         rcu_read_unlock();
3011         spin_unlock(&po->bind_lock);
3012         release_sock(sk);
3013         return ret;
3014 }
3015
3016 /*
3017  *      Bind a packet socket to a device
3018  */
3019
3020 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3021                             int addr_len)
3022 {
3023         struct sock *sk = sock->sk;
3024         char name[15];
3025
3026         /*
3027          *      Check legality
3028          */
3029
3030         if (addr_len != sizeof(struct sockaddr))
3031                 return -EINVAL;
3032         strlcpy(name, uaddr->sa_data, sizeof(name));
3033
3034         return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3035 }
3036
3037 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3038 {
3039         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3040         struct sock *sk = sock->sk;
3041
3042         /*
3043          *      Check legality
3044          */
3045
3046         if (addr_len < sizeof(struct sockaddr_ll))
3047                 return -EINVAL;
3048         if (sll->sll_family != AF_PACKET)
3049                 return -EINVAL;
3050
3051         return packet_do_bind(sk, NULL, sll->sll_ifindex,
3052                               sll->sll_protocol ? : pkt_sk(sk)->num);
3053 }
3054
3055 static struct proto packet_proto = {
3056         .name     = "PACKET",
3057         .owner    = THIS_MODULE,
3058         .obj_size = sizeof(struct packet_sock),
3059 };
3060
3061 /*
3062  *      Create a packet of type SOCK_PACKET.
3063  */
3064
3065 static int packet_create(struct net *net, struct socket *sock, int protocol,
3066                          int kern)
3067 {
3068         struct sock *sk;
3069         struct packet_sock *po;
3070         __be16 proto = (__force __be16)protocol; /* weird, but documented */
3071         int err;
3072
3073         if (!ns_capable(net->user_ns, CAP_NET_RAW))
3074                 return -EPERM;
3075         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3076             sock->type != SOCK_PACKET)
3077                 return -ESOCKTNOSUPPORT;
3078
3079         sock->state = SS_UNCONNECTED;
3080
3081         err = -ENOBUFS;
3082         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3083         if (sk == NULL)
3084                 goto out;
3085
3086         sock->ops = &packet_ops;
3087         if (sock->type == SOCK_PACKET)
3088                 sock->ops = &packet_ops_spkt;
3089
3090         sock_init_data(sock, sk);
3091
3092         po = pkt_sk(sk);
3093         sk->sk_family = PF_PACKET;
3094         po->num = proto;
3095         po->xmit = dev_queue_xmit;
3096
3097         err = packet_alloc_pending(po);
3098         if (err)
3099                 goto out2;
3100
3101         packet_cached_dev_reset(po);
3102
3103         sk->sk_destruct = packet_sock_destruct;
3104         sk_refcnt_debug_inc(sk);
3105
3106         /*
3107          *      Attach a protocol block
3108          */
3109
3110         spin_lock_init(&po->bind_lock);
3111         mutex_init(&po->pg_vec_lock);
3112         po->rollover = NULL;
3113         po->prot_hook.func = packet_rcv;
3114
3115         if (sock->type == SOCK_PACKET)
3116                 po->prot_hook.func = packet_rcv_spkt;
3117
3118         po->prot_hook.af_packet_priv = sk;
3119
3120         if (proto) {
3121                 po->prot_hook.type = proto;
3122                 register_prot_hook(sk);
3123         }
3124
3125         mutex_lock(&net->packet.sklist_lock);
3126         sk_add_node_rcu(sk, &net->packet.sklist);
3127         mutex_unlock(&net->packet.sklist_lock);
3128
3129         preempt_disable();
3130         sock_prot_inuse_add(net, &packet_proto, 1);
3131         preempt_enable();
3132
3133         return 0;
3134 out2:
3135         sk_free(sk);
3136 out:
3137         return err;
3138 }
3139
3140 /*
3141  *      Pull a packet from our receive queue and hand it to the user.
3142  *      If necessary we block.
3143  */
3144
3145 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3146                           int flags)
3147 {
3148         struct sock *sk = sock->sk;
3149         struct sk_buff *skb;
3150         int copied, err;
3151         int vnet_hdr_len = 0;
3152         unsigned int origlen = 0;
3153
3154         err = -EINVAL;
3155         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3156                 goto out;
3157
3158 #if 0
3159         /* What error should we return now? EUNATTACH? */
3160         if (pkt_sk(sk)->ifindex < 0)
3161                 return -ENODEV;
3162 #endif
3163
3164         if (flags & MSG_ERRQUEUE) {
3165                 err = sock_recv_errqueue(sk, msg, len,
3166                                          SOL_PACKET, PACKET_TX_TIMESTAMP);
3167                 goto out;
3168         }
3169
3170         /*
3171          *      Call the generic datagram receiver. This handles all sorts
3172          *      of horrible races and re-entrancy so we can forget about it
3173          *      in the protocol layers.
3174          *
3175          *      Now it will return ENETDOWN, if device have just gone down,
3176          *      but then it will block.
3177          */
3178
3179         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3180
3181         /*
3182          *      An error occurred so return it. Because skb_recv_datagram()
3183          *      handles the blocking we don't see and worry about blocking
3184          *      retries.
3185          */
3186
3187         if (skb == NULL)
3188                 goto out;
3189
3190         if (pkt_sk(sk)->pressure)
3191                 packet_rcv_has_room(pkt_sk(sk), NULL);
3192
3193         if (pkt_sk(sk)->has_vnet_hdr) {
3194                 struct virtio_net_hdr vnet_hdr = { 0 };
3195
3196                 err = -EINVAL;
3197                 vnet_hdr_len = sizeof(vnet_hdr);
3198                 if (len < vnet_hdr_len)
3199                         goto out_free;
3200
3201                 len -= vnet_hdr_len;
3202
3203                 if (skb_is_gso(skb)) {
3204                         struct skb_shared_info *sinfo = skb_shinfo(skb);
3205
3206                         /* This is a hint as to how much should be linear. */
3207                         vnet_hdr.hdr_len =
3208                                 __cpu_to_virtio16(vio_le(), skb_headlen(skb));
3209                         vnet_hdr.gso_size =
3210                                 __cpu_to_virtio16(vio_le(), sinfo->gso_size);
3211                         if (sinfo->gso_type & SKB_GSO_TCPV4)
3212                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
3213                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
3214                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
3215                         else if (sinfo->gso_type & SKB_GSO_UDP)
3216                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
3217                         else if (sinfo->gso_type & SKB_GSO_FCOE)
3218                                 goto out_free;
3219                         else
3220                                 BUG();
3221                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
3222                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
3223                 } else
3224                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
3225
3226                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3227                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
3228                         vnet_hdr.csum_start = __cpu_to_virtio16(vio_le(),
3229                                           skb_checksum_start_offset(skb));
3230                         vnet_hdr.csum_offset = __cpu_to_virtio16(vio_le(),
3231                                                          skb->csum_offset);
3232                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3233                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
3234                 } /* else everything is zero */
3235
3236                 err = memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_len);
3237                 if (err < 0)
3238                         goto out_free;
3239         }
3240
3241         /* You lose any data beyond the buffer you gave. If it worries
3242          * a user program they can ask the device for its MTU
3243          * anyway.
3244          */
3245         copied = skb->len;
3246         if (copied > len) {
3247                 copied = len;
3248                 msg->msg_flags |= MSG_TRUNC;
3249         }
3250
3251         err = skb_copy_datagram_msg(skb, 0, msg, copied);
3252         if (err)
3253                 goto out_free;
3254
3255         if (sock->type != SOCK_PACKET) {
3256                 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3257
3258                 /* Original length was stored in sockaddr_ll fields */
3259                 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3260                 sll->sll_family = AF_PACKET;
3261                 sll->sll_protocol = skb->protocol;
3262         }
3263
3264         sock_recv_ts_and_drops(msg, sk, skb);
3265
3266         if (msg->msg_name) {
3267                 /* If the address length field is there to be filled
3268                  * in, we fill it in now.
3269                  */
3270                 if (sock->type == SOCK_PACKET) {
3271                         __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3272                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
3273                 } else {
3274                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3275
3276                         msg->msg_namelen = sll->sll_halen +
3277                                 offsetof(struct sockaddr_ll, sll_addr);
3278                 }
3279                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3280                        msg->msg_namelen);
3281         }
3282
3283         if (pkt_sk(sk)->auxdata) {
3284                 struct tpacket_auxdata aux;
3285
3286                 aux.tp_status = TP_STATUS_USER;
3287                 if (skb->ip_summed == CHECKSUM_PARTIAL)
3288                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3289                 else if (skb->pkt_type != PACKET_OUTGOING &&
3290                          (skb->ip_summed == CHECKSUM_COMPLETE ||
3291                           skb_csum_unnecessary(skb)))
3292                         aux.tp_status |= TP_STATUS_CSUM_VALID;
3293
3294                 aux.tp_len = origlen;
3295                 aux.tp_snaplen = skb->len;
3296                 aux.tp_mac = 0;
3297                 aux.tp_net = skb_network_offset(skb);
3298                 if (skb_vlan_tag_present(skb)) {
3299                         aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3300                         aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3301                         aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3302                 } else {
3303                         aux.tp_vlan_tci = 0;
3304                         aux.tp_vlan_tpid = 0;
3305                 }
3306                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3307         }
3308
3309         /*
3310          *      Free or return the buffer as appropriate. Again this
3311          *      hides all the races and re-entrancy issues from us.
3312          */
3313         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3314
3315 out_free:
3316         skb_free_datagram(sk, skb);
3317 out:
3318         return err;
3319 }
3320
3321 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3322                                int *uaddr_len, int peer)
3323 {
3324         struct net_device *dev;
3325         struct sock *sk = sock->sk;
3326
3327         if (peer)
3328                 return -EOPNOTSUPP;
3329
3330         uaddr->sa_family = AF_PACKET;
3331         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3332         rcu_read_lock();
3333         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3334         if (dev)
3335                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3336         rcu_read_unlock();
3337         *uaddr_len = sizeof(*uaddr);
3338
3339         return 0;
3340 }
3341
3342 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3343                           int *uaddr_len, int peer)
3344 {
3345         struct net_device *dev;
3346         struct sock *sk = sock->sk;
3347         struct packet_sock *po = pkt_sk(sk);
3348         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3349
3350         if (peer)
3351                 return -EOPNOTSUPP;
3352
3353         sll->sll_family = AF_PACKET;
3354         sll->sll_ifindex = po->ifindex;
3355         sll->sll_protocol = po->num;
3356         sll->sll_pkttype = 0;
3357         rcu_read_lock();
3358         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3359         if (dev) {
3360                 sll->sll_hatype = dev->type;
3361                 sll->sll_halen = dev->addr_len;
3362                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3363         } else {
3364                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
3365                 sll->sll_halen = 0;
3366         }
3367         rcu_read_unlock();
3368         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3369
3370         return 0;
3371 }
3372
3373 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3374                          int what)
3375 {
3376         switch (i->type) {
3377         case PACKET_MR_MULTICAST:
3378                 if (i->alen != dev->addr_len)
3379                         return -EINVAL;
3380                 if (what > 0)
3381                         return dev_mc_add(dev, i->addr);
3382                 else
3383                         return dev_mc_del(dev, i->addr);
3384                 break;
3385         case PACKET_MR_PROMISC:
3386                 return dev_set_promiscuity(dev, what);
3387         case PACKET_MR_ALLMULTI:
3388                 return dev_set_allmulti(dev, what);
3389         case PACKET_MR_UNICAST:
3390                 if (i->alen != dev->addr_len)
3391                         return -EINVAL;
3392                 if (what > 0)
3393                         return dev_uc_add(dev, i->addr);
3394                 else
3395                         return dev_uc_del(dev, i->addr);
3396                 break;
3397         default:
3398                 break;
3399         }
3400         return 0;
3401 }
3402
3403 static void packet_dev_mclist_delete(struct net_device *dev,
3404                                      struct packet_mclist **mlp)
3405 {
3406         struct packet_mclist *ml;
3407
3408         while ((ml = *mlp) != NULL) {
3409                 if (ml->ifindex == dev->ifindex) {
3410                         packet_dev_mc(dev, ml, -1);
3411                         *mlp = ml->next;
3412                         kfree(ml);
3413                 } else
3414                         mlp = &ml->next;
3415         }
3416 }
3417
3418 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3419 {
3420         struct packet_sock *po = pkt_sk(sk);
3421         struct packet_mclist *ml, *i;
3422         struct net_device *dev;
3423         int err;
3424
3425         rtnl_lock();
3426
3427         err = -ENODEV;
3428         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3429         if (!dev)
3430                 goto done;
3431
3432         err = -EINVAL;
3433         if (mreq->mr_alen > dev->addr_len)
3434                 goto done;
3435
3436         err = -ENOBUFS;
3437         i = kmalloc(sizeof(*i), GFP_KERNEL);
3438         if (i == NULL)
3439                 goto done;
3440
3441         err = 0;
3442         for (ml = po->mclist; ml; ml = ml->next) {
3443                 if (ml->ifindex == mreq->mr_ifindex &&
3444                     ml->type == mreq->mr_type &&
3445                     ml->alen == mreq->mr_alen &&
3446                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3447                         ml->count++;
3448                         /* Free the new element ... */
3449                         kfree(i);
3450                         goto done;
3451                 }
3452         }
3453
3454         i->type = mreq->mr_type;
3455         i->ifindex = mreq->mr_ifindex;
3456         i->alen = mreq->mr_alen;
3457         memcpy(i->addr, mreq->mr_address, i->alen);
3458         memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3459         i->count = 1;
3460         i->next = po->mclist;
3461         po->mclist = i;
3462         err = packet_dev_mc(dev, i, 1);
3463         if (err) {
3464                 po->mclist = i->next;
3465                 kfree(i);
3466         }
3467
3468 done:
3469         rtnl_unlock();
3470         return err;
3471 }
3472
3473 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3474 {
3475         struct packet_mclist *ml, **mlp;
3476
3477         rtnl_lock();
3478
3479         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3480                 if (ml->ifindex == mreq->mr_ifindex &&
3481                     ml->type == mreq->mr_type &&
3482                     ml->alen == mreq->mr_alen &&
3483                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3484                         if (--ml->count == 0) {
3485                                 struct net_device *dev;
3486                                 *mlp = ml->next;
3487                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3488                                 if (dev)
3489                                         packet_dev_mc(dev, ml, -1);
3490                                 kfree(ml);
3491                         }
3492                         break;
3493                 }
3494         }
3495         rtnl_unlock();
3496         return 0;
3497 }
3498
3499 static void packet_flush_mclist(struct sock *sk)
3500 {
3501         struct packet_sock *po = pkt_sk(sk);
3502         struct packet_mclist *ml;
3503
3504         if (!po->mclist)
3505                 return;
3506
3507         rtnl_lock();
3508         while ((ml = po->mclist) != NULL) {
3509                 struct net_device *dev;
3510
3511                 po->mclist = ml->next;
3512                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3513                 if (dev != NULL)
3514                         packet_dev_mc(dev, ml, -1);
3515                 kfree(ml);
3516         }
3517         rtnl_unlock();
3518 }
3519
3520 static int
3521 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3522 {
3523         struct sock *sk = sock->sk;
3524         struct packet_sock *po = pkt_sk(sk);
3525         int ret;
3526
3527         if (level != SOL_PACKET)
3528                 return -ENOPROTOOPT;
3529
3530         switch (optname) {
3531         case PACKET_ADD_MEMBERSHIP:
3532         case PACKET_DROP_MEMBERSHIP:
3533         {
3534                 struct packet_mreq_max mreq;
3535                 int len = optlen;
3536                 memset(&mreq, 0, sizeof(mreq));
3537                 if (len < sizeof(struct packet_mreq))
3538                         return -EINVAL;
3539                 if (len > sizeof(mreq))
3540                         len = sizeof(mreq);
3541                 if (copy_from_user(&mreq, optval, len))
3542                         return -EFAULT;
3543                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3544                         return -EINVAL;
3545                 if (optname == PACKET_ADD_MEMBERSHIP)
3546                         ret = packet_mc_add(sk, &mreq);
3547                 else
3548                         ret = packet_mc_drop(sk, &mreq);
3549                 return ret;
3550         }
3551
3552         case PACKET_RX_RING:
3553         case PACKET_TX_RING:
3554         {
3555                 union tpacket_req_u req_u;
3556                 int len;
3557
3558                 switch (po->tp_version) {
3559                 case TPACKET_V1:
3560                 case TPACKET_V2:
3561                         len = sizeof(req_u.req);
3562                         break;
3563                 case TPACKET_V3:
3564                 default:
3565                         len = sizeof(req_u.req3);
3566                         break;
3567                 }
3568                 if (optlen < len)
3569                         return -EINVAL;
3570                 if (pkt_sk(sk)->has_vnet_hdr)
3571                         return -EINVAL;
3572                 if (copy_from_user(&req_u.req, optval, len))
3573                         return -EFAULT;
3574                 return packet_set_ring(sk, &req_u, 0,
3575                         optname == PACKET_TX_RING);
3576         }
3577         case PACKET_COPY_THRESH:
3578         {
3579                 int val;
3580
3581                 if (optlen != sizeof(val))
3582                         return -EINVAL;
3583                 if (copy_from_user(&val, optval, sizeof(val)))
3584                         return -EFAULT;
3585
3586                 pkt_sk(sk)->copy_thresh = val;
3587                 return 0;
3588         }
3589         case PACKET_VERSION:
3590         {
3591                 int val;
3592
3593                 if (optlen != sizeof(val))
3594                         return -EINVAL;
3595                 if (copy_from_user(&val, optval, sizeof(val)))
3596                         return -EFAULT;
3597                 switch (val) {
3598                 case TPACKET_V1:
3599                 case TPACKET_V2:
3600                 case TPACKET_V3:
3601                         break;
3602                 default:
3603                         return -EINVAL;
3604                 }
3605                 lock_sock(sk);
3606                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3607                         ret = -EBUSY;
3608                 } else {
3609                         po->tp_version = val;
3610                         ret = 0;
3611                 }
3612                 release_sock(sk);
3613                 return ret;
3614         }
3615         case PACKET_RESERVE:
3616         {
3617                 unsigned int val;
3618
3619                 if (optlen != sizeof(val))
3620                         return -EINVAL;
3621                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3622                         return -EBUSY;
3623                 if (copy_from_user(&val, optval, sizeof(val)))
3624                         return -EFAULT;
3625                 po->tp_reserve = val;
3626                 return 0;
3627         }
3628         case PACKET_LOSS:
3629         {
3630                 unsigned int val;
3631
3632                 if (optlen != sizeof(val))
3633                         return -EINVAL;
3634                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3635                         return -EBUSY;
3636                 if (copy_from_user(&val, optval, sizeof(val)))
3637                         return -EFAULT;
3638                 po->tp_loss = !!val;
3639                 return 0;
3640         }
3641         case PACKET_AUXDATA:
3642         {
3643                 int val;
3644
3645                 if (optlen < sizeof(val))
3646                         return -EINVAL;
3647                 if (copy_from_user(&val, optval, sizeof(val)))
3648                         return -EFAULT;
3649
3650                 po->auxdata = !!val;
3651                 return 0;
3652         }
3653         case PACKET_ORIGDEV:
3654         {
3655                 int val;
3656
3657                 if (optlen < sizeof(val))
3658                         return -EINVAL;
3659                 if (copy_from_user(&val, optval, sizeof(val)))
3660                         return -EFAULT;
3661
3662                 po->origdev = !!val;
3663                 return 0;
3664         }
3665         case PACKET_VNET_HDR:
3666         {
3667                 int val;
3668
3669                 if (sock->type != SOCK_RAW)
3670                         return -EINVAL;
3671                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3672                         return -EBUSY;
3673                 if (optlen < sizeof(val))
3674                         return -EINVAL;
3675                 if (copy_from_user(&val, optval, sizeof(val)))
3676                         return -EFAULT;
3677
3678                 po->has_vnet_hdr = !!val;
3679                 return 0;
3680         }
3681         case PACKET_TIMESTAMP:
3682         {
3683                 int val;
3684
3685                 if (optlen != sizeof(val))
3686                         return -EINVAL;
3687                 if (copy_from_user(&val, optval, sizeof(val)))
3688                         return -EFAULT;
3689
3690                 po->tp_tstamp = val;
3691                 return 0;
3692         }
3693         case PACKET_FANOUT:
3694         {
3695                 int val;
3696
3697                 if (optlen != sizeof(val))
3698                         return -EINVAL;
3699                 if (copy_from_user(&val, optval, sizeof(val)))
3700                         return -EFAULT;
3701
3702                 return fanout_add(sk, val & 0xffff, val >> 16);
3703         }
3704         case PACKET_FANOUT_DATA:
3705         {
3706                 if (!po->fanout)
3707                         return -EINVAL;
3708
3709                 return fanout_set_data(po, optval, optlen);
3710         }
3711         case PACKET_TX_HAS_OFF:
3712         {
3713                 unsigned int val;
3714
3715                 if (optlen != sizeof(val))
3716                         return -EINVAL;
3717                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3718                         return -EBUSY;
3719                 if (copy_from_user(&val, optval, sizeof(val)))
3720                         return -EFAULT;
3721                 po->tp_tx_has_off = !!val;
3722                 return 0;
3723         }
3724         case PACKET_QDISC_BYPASS:
3725         {
3726                 int val;
3727
3728                 if (optlen != sizeof(val))
3729                         return -EINVAL;
3730                 if (copy_from_user(&val, optval, sizeof(val)))
3731                         return -EFAULT;
3732
3733                 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3734                 return 0;
3735         }
3736         default:
3737                 return -ENOPROTOOPT;
3738         }
3739 }
3740
3741 static int packet_getsockopt(struct socket *sock, int level, int optname,
3742                              char __user *optval, int __user *optlen)
3743 {
3744         int len;
3745         int val, lv = sizeof(val);
3746         struct sock *sk = sock->sk;
3747         struct packet_sock *po = pkt_sk(sk);
3748         void *data = &val;
3749         union tpacket_stats_u st;
3750         struct tpacket_rollover_stats rstats;
3751
3752         if (level != SOL_PACKET)
3753                 return -ENOPROTOOPT;
3754
3755         if (get_user(len, optlen))
3756                 return -EFAULT;
3757
3758         if (len < 0)
3759                 return -EINVAL;
3760
3761         switch (optname) {
3762         case PACKET_STATISTICS:
3763                 spin_lock_bh(&sk->sk_receive_queue.lock);
3764                 memcpy(&st, &po->stats, sizeof(st));
3765                 memset(&po->stats, 0, sizeof(po->stats));
3766                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3767
3768                 if (po->tp_version == TPACKET_V3) {
3769                         lv = sizeof(struct tpacket_stats_v3);
3770                         st.stats3.tp_packets += st.stats3.tp_drops;
3771                         data = &st.stats3;
3772                 } else {
3773                         lv = sizeof(struct tpacket_stats);
3774                         st.stats1.tp_packets += st.stats1.tp_drops;
3775                         data = &st.stats1;
3776                 }
3777
3778                 break;
3779         case PACKET_AUXDATA:
3780                 val = po->auxdata;
3781                 break;
3782         case PACKET_ORIGDEV:
3783                 val = po->origdev;
3784                 break;
3785         case PACKET_VNET_HDR:
3786                 val = po->has_vnet_hdr;
3787                 break;
3788         case PACKET_VERSION:
3789                 val = po->tp_version;
3790                 break;
3791         case PACKET_HDRLEN:
3792                 if (len > sizeof(int))
3793                         len = sizeof(int);
3794                 if (copy_from_user(&val, optval, len))
3795                         return -EFAULT;
3796                 switch (val) {
3797                 case TPACKET_V1:
3798                         val = sizeof(struct tpacket_hdr);
3799                         break;
3800                 case TPACKET_V2:
3801                         val = sizeof(struct tpacket2_hdr);
3802                         break;
3803                 case TPACKET_V3:
3804                         val = sizeof(struct tpacket3_hdr);
3805                         break;
3806                 default:
3807                         return -EINVAL;
3808                 }
3809                 break;
3810         case PACKET_RESERVE:
3811                 val = po->tp_reserve;
3812                 break;
3813         case PACKET_LOSS:
3814                 val = po->tp_loss;
3815                 break;
3816         case PACKET_TIMESTAMP:
3817                 val = po->tp_tstamp;
3818                 break;
3819         case PACKET_FANOUT:
3820                 val = (po->fanout ?
3821                        ((u32)po->fanout->id |
3822                         ((u32)po->fanout->type << 16) |
3823                         ((u32)po->fanout->flags << 24)) :
3824                        0);
3825                 break;
3826         case PACKET_ROLLOVER_STATS:
3827                 if (!po->rollover)
3828                         return -EINVAL;
3829                 rstats.tp_all = atomic_long_read(&po->rollover->num);
3830                 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3831                 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3832                 data = &rstats;
3833                 lv = sizeof(rstats);
3834                 break;
3835         case PACKET_TX_HAS_OFF:
3836                 val = po->tp_tx_has_off;
3837                 break;
3838         case PACKET_QDISC_BYPASS:
3839                 val = packet_use_direct_xmit(po);
3840                 break;
3841         default:
3842                 return -ENOPROTOOPT;
3843         }
3844
3845         if (len > lv)
3846                 len = lv;
3847         if (put_user(len, optlen))
3848                 return -EFAULT;
3849         if (copy_to_user(optval, data, len))
3850                 return -EFAULT;
3851         return 0;
3852 }
3853
3854
3855 static int packet_notifier(struct notifier_block *this,
3856                            unsigned long msg, void *ptr)
3857 {
3858         struct sock *sk;
3859         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3860         struct net *net = dev_net(dev);
3861
3862         rcu_read_lock();
3863         sk_for_each_rcu(sk, &net->packet.sklist) {
3864                 struct packet_sock *po = pkt_sk(sk);
3865
3866                 switch (msg) {
3867                 case NETDEV_UNREGISTER:
3868                         if (po->mclist)
3869                                 packet_dev_mclist_delete(dev, &po->mclist);
3870                         /* fallthrough */
3871
3872                 case NETDEV_DOWN:
3873                         if (dev->ifindex == po->ifindex) {
3874                                 spin_lock(&po->bind_lock);
3875                                 if (po->running) {
3876                                         __unregister_prot_hook(sk, false);
3877                                         sk->sk_err = ENETDOWN;
3878                                         if (!sock_flag(sk, SOCK_DEAD))
3879                                                 sk->sk_error_report(sk);
3880                                 }
3881                                 if (msg == NETDEV_UNREGISTER) {
3882                                         packet_cached_dev_reset(po);
3883                                         po->ifindex = -1;
3884                                         if (po->prot_hook.dev)
3885                                                 dev_put(po->prot_hook.dev);
3886                                         po->prot_hook.dev = NULL;
3887                                 }
3888                                 spin_unlock(&po->bind_lock);
3889                         }
3890                         break;
3891                 case NETDEV_UP:
3892                         if (dev->ifindex == po->ifindex) {
3893                                 spin_lock(&po->bind_lock);
3894                                 if (po->num)
3895                                         register_prot_hook(sk);
3896                                 spin_unlock(&po->bind_lock);
3897                         }
3898                         break;
3899                 }
3900         }
3901         rcu_read_unlock();
3902         return NOTIFY_DONE;
3903 }
3904
3905
3906 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3907                         unsigned long arg)
3908 {
3909         struct sock *sk = sock->sk;
3910
3911         switch (cmd) {
3912         case SIOCOUTQ:
3913         {
3914                 int amount = sk_wmem_alloc_get(sk);
3915
3916                 return put_user(amount, (int __user *)arg);
3917         }
3918         case SIOCINQ:
3919         {
3920                 struct sk_buff *skb;
3921                 int amount = 0;
3922
3923                 spin_lock_bh(&sk->sk_receive_queue.lock);
3924                 skb = skb_peek(&sk->sk_receive_queue);
3925                 if (skb)
3926                         amount = skb->len;
3927                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3928                 return put_user(amount, (int __user *)arg);
3929         }
3930         case SIOCGSTAMP:
3931                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3932         case SIOCGSTAMPNS:
3933                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3934
3935 #ifdef CONFIG_INET
3936         case SIOCADDRT:
3937         case SIOCDELRT:
3938         case SIOCDARP:
3939         case SIOCGARP:
3940         case SIOCSARP:
3941         case SIOCGIFADDR:
3942         case SIOCSIFADDR:
3943         case SIOCGIFBRDADDR:
3944         case SIOCSIFBRDADDR:
3945         case SIOCGIFNETMASK:
3946         case SIOCSIFNETMASK:
3947         case SIOCGIFDSTADDR:
3948         case SIOCSIFDSTADDR:
3949         case SIOCSIFFLAGS:
3950                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3951 #endif
3952
3953         default:
3954                 return -ENOIOCTLCMD;
3955         }
3956         return 0;
3957 }
3958
3959 static unsigned int packet_poll(struct file *file, struct socket *sock,
3960                                 poll_table *wait)
3961 {
3962         struct sock *sk = sock->sk;
3963         struct packet_sock *po = pkt_sk(sk);
3964         unsigned int mask = datagram_poll(file, sock, wait);
3965
3966         spin_lock_bh(&sk->sk_receive_queue.lock);
3967         if (po->rx_ring.pg_vec) {
3968                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3969                         TP_STATUS_KERNEL))
3970                         mask |= POLLIN | POLLRDNORM;
3971         }
3972         if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
3973                 po->pressure = 0;
3974         spin_unlock_bh(&sk->sk_receive_queue.lock);
3975         spin_lock_bh(&sk->sk_write_queue.lock);
3976         if (po->tx_ring.pg_vec) {
3977                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3978                         mask |= POLLOUT | POLLWRNORM;
3979         }
3980         spin_unlock_bh(&sk->sk_write_queue.lock);
3981         return mask;
3982 }
3983
3984
3985 /* Dirty? Well, I still did not learn better way to account
3986  * for user mmaps.
3987  */
3988
3989 static void packet_mm_open(struct vm_area_struct *vma)
3990 {
3991         struct file *file = vma->vm_file;
3992         struct socket *sock = file->private_data;
3993         struct sock *sk = sock->sk;
3994
3995         if (sk)
3996                 atomic_inc(&pkt_sk(sk)->mapped);
3997 }
3998
3999 static void packet_mm_close(struct vm_area_struct *vma)
4000 {
4001         struct file *file = vma->vm_file;
4002         struct socket *sock = file->private_data;
4003         struct sock *sk = sock->sk;
4004
4005         if (sk)
4006                 atomic_dec(&pkt_sk(sk)->mapped);
4007 }
4008
4009 static const struct vm_operations_struct packet_mmap_ops = {
4010         .open   =       packet_mm_open,
4011         .close  =       packet_mm_close,
4012 };
4013
4014 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4015                         unsigned int len)
4016 {
4017         int i;
4018
4019         for (i = 0; i < len; i++) {
4020                 if (likely(pg_vec[i].buffer)) {
4021                         if (is_vmalloc_addr(pg_vec[i].buffer))
4022                                 vfree(pg_vec[i].buffer);
4023                         else
4024                                 free_pages((unsigned long)pg_vec[i].buffer,
4025                                            order);
4026                         pg_vec[i].buffer = NULL;
4027                 }
4028         }
4029         kfree(pg_vec);
4030 }
4031
4032 static char *alloc_one_pg_vec_page(unsigned long order)
4033 {
4034         char *buffer;
4035         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4036                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4037
4038         buffer = (char *) __get_free_pages(gfp_flags, order);
4039         if (buffer)
4040                 return buffer;
4041
4042         /* __get_free_pages failed, fall back to vmalloc */
4043         buffer = vzalloc((1 << order) * PAGE_SIZE);
4044         if (buffer)
4045                 return buffer;
4046
4047         /* vmalloc failed, lets dig into swap here */
4048         gfp_flags &= ~__GFP_NORETRY;
4049         buffer = (char *) __get_free_pages(gfp_flags, order);
4050         if (buffer)
4051                 return buffer;
4052
4053         /* complete and utter failure */
4054         return NULL;
4055 }
4056
4057 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4058 {
4059         unsigned int block_nr = req->tp_block_nr;
4060         struct pgv *pg_vec;
4061         int i;
4062
4063         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4064         if (unlikely(!pg_vec))
4065                 goto out;
4066
4067         for (i = 0; i < block_nr; i++) {
4068                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4069                 if (unlikely(!pg_vec[i].buffer))
4070                         goto out_free_pgvec;
4071         }
4072
4073 out:
4074         return pg_vec;
4075
4076 out_free_pgvec:
4077         free_pg_vec(pg_vec, order, block_nr);
4078         pg_vec = NULL;
4079         goto out;
4080 }
4081
4082 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4083                 int closing, int tx_ring)
4084 {
4085         struct pgv *pg_vec = NULL;
4086         struct packet_sock *po = pkt_sk(sk);
4087         int was_running, order = 0;
4088         struct packet_ring_buffer *rb;
4089         struct sk_buff_head *rb_queue;
4090         __be16 num;
4091         int err = -EINVAL;
4092         /* Added to avoid minimal code churn */
4093         struct tpacket_req *req = &req_u->req;
4094
4095         lock_sock(sk);
4096         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
4097         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
4098                 WARN(1, "Tx-ring is not supported.\n");
4099                 goto out;
4100         }
4101
4102         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4103         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4104
4105         err = -EBUSY;
4106         if (!closing) {
4107                 if (atomic_read(&po->mapped))
4108                         goto out;
4109                 if (packet_read_pending(rb))
4110                         goto out;
4111         }
4112
4113         if (req->tp_block_nr) {
4114                 /* Sanity tests and some calculations */
4115                 err = -EBUSY;
4116                 if (unlikely(rb->pg_vec))
4117                         goto out;
4118
4119                 switch (po->tp_version) {
4120                 case TPACKET_V1:
4121                         po->tp_hdrlen = TPACKET_HDRLEN;
4122                         break;
4123                 case TPACKET_V2:
4124                         po->tp_hdrlen = TPACKET2_HDRLEN;
4125                         break;
4126                 case TPACKET_V3:
4127                         po->tp_hdrlen = TPACKET3_HDRLEN;
4128                         break;
4129                 }
4130
4131                 err = -EINVAL;
4132                 if (unlikely((int)req->tp_block_size <= 0))
4133                         goto out;
4134                 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4135                         goto out;
4136                 if (po->tp_version >= TPACKET_V3 &&
4137                     (int)(req->tp_block_size -
4138                           BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
4139                         goto out;
4140                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
4141                                         po->tp_reserve))
4142                         goto out;
4143                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4144                         goto out;
4145
4146                 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4147                 if (unlikely(rb->frames_per_block == 0))
4148                         goto out;
4149                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4150                                         req->tp_frame_nr))
4151                         goto out;
4152
4153                 err = -ENOMEM;
4154                 order = get_order(req->tp_block_size);
4155                 pg_vec = alloc_pg_vec(req, order);
4156                 if (unlikely(!pg_vec))
4157                         goto out;
4158                 switch (po->tp_version) {
4159                 case TPACKET_V3:
4160                 /* Transmit path is not supported. We checked
4161                  * it above but just being paranoid
4162                  */
4163                         if (!tx_ring)
4164                                 init_prb_bdqc(po, rb, pg_vec, req_u);
4165                         break;
4166                 default:
4167                         break;
4168                 }
4169         }
4170         /* Done */
4171         else {
4172                 err = -EINVAL;
4173                 if (unlikely(req->tp_frame_nr))
4174                         goto out;
4175         }
4176
4177
4178         /* Detach socket from network */
4179         spin_lock(&po->bind_lock);
4180         was_running = po->running;
4181         num = po->num;
4182         if (was_running) {
4183                 po->num = 0;
4184                 __unregister_prot_hook(sk, false);
4185         }
4186         spin_unlock(&po->bind_lock);
4187
4188         synchronize_net();
4189
4190         err = -EBUSY;
4191         mutex_lock(&po->pg_vec_lock);
4192         if (closing || atomic_read(&po->mapped) == 0) {
4193                 err = 0;
4194                 spin_lock_bh(&rb_queue->lock);
4195                 swap(rb->pg_vec, pg_vec);
4196                 rb->frame_max = (req->tp_frame_nr - 1);
4197                 rb->head = 0;
4198                 rb->frame_size = req->tp_frame_size;
4199                 spin_unlock_bh(&rb_queue->lock);
4200
4201                 swap(rb->pg_vec_order, order);
4202                 swap(rb->pg_vec_len, req->tp_block_nr);
4203
4204                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4205                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4206                                                 tpacket_rcv : packet_rcv;
4207                 skb_queue_purge(rb_queue);
4208                 if (atomic_read(&po->mapped))
4209                         pr_err("packet_mmap: vma is busy: %d\n",
4210                                atomic_read(&po->mapped));
4211         }
4212         mutex_unlock(&po->pg_vec_lock);
4213
4214         spin_lock(&po->bind_lock);
4215         if (was_running) {
4216                 po->num = num;
4217                 register_prot_hook(sk);
4218         }
4219         spin_unlock(&po->bind_lock);
4220         if (closing && (po->tp_version > TPACKET_V2)) {
4221                 /* Because we don't support block-based V3 on tx-ring */
4222                 if (!tx_ring)
4223                         prb_shutdown_retire_blk_timer(po, rb_queue);
4224         }
4225
4226         if (pg_vec)
4227                 free_pg_vec(pg_vec, order, req->tp_block_nr);
4228 out:
4229         release_sock(sk);
4230         return err;
4231 }
4232
4233 static int packet_mmap(struct file *file, struct socket *sock,
4234                 struct vm_area_struct *vma)
4235 {
4236         struct sock *sk = sock->sk;
4237         struct packet_sock *po = pkt_sk(sk);
4238         unsigned long size, expected_size;
4239         struct packet_ring_buffer *rb;
4240         unsigned long start;
4241         int err = -EINVAL;
4242         int i;
4243
4244         if (vma->vm_pgoff)
4245                 return -EINVAL;
4246
4247         mutex_lock(&po->pg_vec_lock);
4248
4249         expected_size = 0;
4250         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4251                 if (rb->pg_vec) {
4252                         expected_size += rb->pg_vec_len
4253                                                 * rb->pg_vec_pages
4254                                                 * PAGE_SIZE;
4255                 }
4256         }
4257
4258         if (expected_size == 0)
4259                 goto out;
4260
4261         size = vma->vm_end - vma->vm_start;
4262         if (size != expected_size)
4263                 goto out;
4264
4265         start = vma->vm_start;
4266         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4267                 if (rb->pg_vec == NULL)
4268                         continue;
4269
4270                 for (i = 0; i < rb->pg_vec_len; i++) {
4271                         struct page *page;
4272                         void *kaddr = rb->pg_vec[i].buffer;
4273                         int pg_num;
4274
4275                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4276                                 page = pgv_to_page(kaddr);
4277                                 err = vm_insert_page(vma, start, page);
4278                                 if (unlikely(err))
4279                                         goto out;
4280                                 start += PAGE_SIZE;
4281                                 kaddr += PAGE_SIZE;
4282                         }
4283                 }
4284         }
4285
4286         atomic_inc(&po->mapped);
4287         vma->vm_ops = &packet_mmap_ops;
4288         err = 0;
4289
4290 out:
4291         mutex_unlock(&po->pg_vec_lock);
4292         return err;
4293 }
4294
4295 static const struct proto_ops packet_ops_spkt = {
4296         .family =       PF_PACKET,
4297         .owner =        THIS_MODULE,
4298         .release =      packet_release,
4299         .bind =         packet_bind_spkt,
4300         .connect =      sock_no_connect,
4301         .socketpair =   sock_no_socketpair,
4302         .accept =       sock_no_accept,
4303         .getname =      packet_getname_spkt,
4304         .poll =         datagram_poll,
4305         .ioctl =        packet_ioctl,
4306         .listen =       sock_no_listen,
4307         .shutdown =     sock_no_shutdown,
4308         .setsockopt =   sock_no_setsockopt,
4309         .getsockopt =   sock_no_getsockopt,
4310         .sendmsg =      packet_sendmsg_spkt,
4311         .recvmsg =      packet_recvmsg,
4312         .mmap =         sock_no_mmap,
4313         .sendpage =     sock_no_sendpage,
4314 };
4315
4316 static const struct proto_ops packet_ops = {
4317         .family =       PF_PACKET,
4318         .owner =        THIS_MODULE,
4319         .release =      packet_release,
4320         .bind =         packet_bind,
4321         .connect =      sock_no_connect,
4322         .socketpair =   sock_no_socketpair,
4323         .accept =       sock_no_accept,
4324         .getname =      packet_getname,
4325         .poll =         packet_poll,
4326         .ioctl =        packet_ioctl,
4327         .listen =       sock_no_listen,
4328         .shutdown =     sock_no_shutdown,
4329         .setsockopt =   packet_setsockopt,
4330         .getsockopt =   packet_getsockopt,
4331         .sendmsg =      packet_sendmsg,
4332         .recvmsg =      packet_recvmsg,
4333         .mmap =         packet_mmap,
4334         .sendpage =     sock_no_sendpage,
4335 };
4336
4337 static const struct net_proto_family packet_family_ops = {
4338         .family =       PF_PACKET,
4339         .create =       packet_create,
4340         .owner  =       THIS_MODULE,
4341 };
4342
4343 static struct notifier_block packet_netdev_notifier = {
4344         .notifier_call =        packet_notifier,
4345 };
4346
4347 #ifdef CONFIG_PROC_FS
4348
4349 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4350         __acquires(RCU)
4351 {
4352         struct net *net = seq_file_net(seq);
4353
4354         rcu_read_lock();
4355         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4356 }
4357
4358 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4359 {
4360         struct net *net = seq_file_net(seq);
4361         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4362 }
4363
4364 static void packet_seq_stop(struct seq_file *seq, void *v)
4365         __releases(RCU)
4366 {
4367         rcu_read_unlock();
4368 }
4369
4370 static int packet_seq_show(struct seq_file *seq, void *v)
4371 {
4372         if (v == SEQ_START_TOKEN)
4373                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
4374         else {
4375                 struct sock *s = sk_entry(v);
4376                 const struct packet_sock *po = pkt_sk(s);
4377
4378                 seq_printf(seq,
4379                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4380                            s,
4381                            atomic_read(&s->sk_refcnt),
4382                            s->sk_type,
4383                            ntohs(po->num),
4384                            po->ifindex,
4385                            po->running,
4386                            atomic_read(&s->sk_rmem_alloc),
4387                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4388                            sock_i_ino(s));
4389         }
4390
4391         return 0;
4392 }
4393
4394 static const struct seq_operations packet_seq_ops = {
4395         .start  = packet_seq_start,
4396         .next   = packet_seq_next,
4397         .stop   = packet_seq_stop,
4398         .show   = packet_seq_show,
4399 };
4400
4401 static int packet_seq_open(struct inode *inode, struct file *file)
4402 {
4403         return seq_open_net(inode, file, &packet_seq_ops,
4404                             sizeof(struct seq_net_private));
4405 }
4406
4407 static const struct file_operations packet_seq_fops = {
4408         .owner          = THIS_MODULE,
4409         .open           = packet_seq_open,
4410         .read           = seq_read,
4411         .llseek         = seq_lseek,
4412         .release        = seq_release_net,
4413 };
4414
4415 #endif
4416
4417 static int __net_init packet_net_init(struct net *net)
4418 {
4419         mutex_init(&net->packet.sklist_lock);
4420         INIT_HLIST_HEAD(&net->packet.sklist);
4421
4422         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4423                 return -ENOMEM;
4424
4425         return 0;
4426 }
4427
4428 static void __net_exit packet_net_exit(struct net *net)
4429 {
4430         remove_proc_entry("packet", net->proc_net);
4431 }
4432
4433 static struct pernet_operations packet_net_ops = {
4434         .init = packet_net_init,
4435         .exit = packet_net_exit,
4436 };
4437
4438
4439 static void __exit packet_exit(void)
4440 {
4441         unregister_netdevice_notifier(&packet_netdev_notifier);
4442         unregister_pernet_subsys(&packet_net_ops);
4443         sock_unregister(PF_PACKET);
4444         proto_unregister(&packet_proto);
4445 }
4446
4447 static int __init packet_init(void)
4448 {
4449         int rc = proto_register(&packet_proto, 0);
4450
4451         if (rc != 0)
4452                 goto out;
4453
4454         sock_register(&packet_family_ops);
4455         register_pernet_subsys(&packet_net_ops);
4456         register_netdevice_notifier(&packet_netdev_notifier);
4457 out:
4458         return rc;
4459 }
4460
4461 module_init(packet_init);
4462 module_exit(packet_exit);
4463 MODULE_LICENSE("GPL");
4464 MODULE_ALIAS_NETPROTO(PF_PACKET);