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 <linux/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 #include <net/compat.h>
  97
  98 #include "internal.h"
  99
 100 /*
 101    Assumptions:
 102    - if device has no dev->hard_header routine, it adds and removes ll header
 103      inside itself. In this case ll header is invisible outside of device,
 104      but higher levels still should reserve dev->hard_header_len.
 105      Some devices are enough clever to reallocate skb, when header
 106      will not fit to reserved space (tunnel), another ones are silly
 107      (PPP).
 108    - packet socket receives packets with pulled ll header,
 109      so that SOCK_RAW should push it back.
 110
 111 On receive:
 112 -----------
 113
 114 Incoming, dev->hard_header!=NULL
 115    mac_header -> ll header
 116    data       -> data
 117
 118 Outgoing, dev->hard_header!=NULL
 119    mac_header -> ll header
 120    data       -> ll header
 121
 122 Incoming, dev->hard_header==NULL
 123    mac_header -> UNKNOWN position. It is very likely, that it points to ll
 124                  header.  PPP makes it, that is wrong, because introduce
 125                  assymetry between rx and tx paths.
 126    data       -> data
 127
 128 Outgoing, dev->hard_header==NULL
 129    mac_header -> data. ll header is still not built!
 130    data       -> data
 131
 132 Resume
 133   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 134
 135
 136 On transmit:
 137 ------------
 138
 139 dev->hard_header != NULL
 140    mac_header -> ll header
 141    data       -> ll header
 142
 143 dev->hard_header == NULL (ll header is added by device, we cannot control it)
 144    mac_header -> data
 145    data       -> data
 146
 147    We should set nh.raw on output to correct posistion,
 148    packet classifier depends on it.
 149  */
 150
 151 /* Private packet socket structures. */
 152
 153 /* identical to struct packet_mreq except it has
 154  * a longer address field.
 155  */
 156 struct packet_mreq_max {
 157         int             mr_ifindex;
 158         unsigned short  mr_type;
 159         unsigned short  mr_alen;
 160         unsigned char   mr_address[MAX_ADDR_LEN];
 161 };
 162
 163 union tpacket_uhdr {
 164         struct tpacket_hdr  *h1;
 165         struct tpacket2_hdr *h2;
 166         struct tpacket3_hdr *h3;
 167         void *raw;
 168 };
 169
 170 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 171                 int closing, int tx_ring);
 172
 173 #define V3_ALIGNMENT    (8)
 174
 175 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 176
 177 #define BLK_PLUS_PRIV(sz_of_priv) \
 178         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 179
 180 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
 181 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
 182 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
 183 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
 184 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
 185 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
 186 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
 187
 188 struct packet_sock;
 189 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 190                        struct packet_type *pt, struct net_device *orig_dev);
 191
 192 static void *packet_previous_frame(struct packet_sock *po,
 193                 struct packet_ring_buffer *rb,
 194                 int status);
 195 static void packet_increment_head(struct packet_ring_buffer *buff);
 196 static int prb_curr_blk_in_use(struct tpacket_block_desc *);
 197 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 198                         struct packet_sock *);
 199 static void prb_retire_current_block(struct tpacket_kbdq_core *,
 200                 struct packet_sock *, unsigned int status);
 201 static int prb_queue_frozen(struct tpacket_kbdq_core *);
 202 static void prb_open_block(struct tpacket_kbdq_core *,
 203                 struct tpacket_block_desc *);
 204 static void prb_retire_rx_blk_timer_expired(unsigned long);
 205 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 206 static void prb_init_blk_timer(struct packet_sock *,
 207                 struct tpacket_kbdq_core *,
 208                 void (*func) (unsigned long));
 209 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 210 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 211                 struct tpacket3_hdr *);
 212 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 213                 struct tpacket3_hdr *);
 214 static void packet_flush_mclist(struct sock *sk);
 215 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb);
 216
 217 struct packet_skb_cb {
 218         union {
 219                 struct sockaddr_pkt pkt;
 220                 union {
 221                         /* Trick: alias skb original length with
 222                          * ll.sll_family and ll.protocol in order
 223                          * to save room.
 224                          */
 225                         unsigned int origlen;
 226                         struct sockaddr_ll ll;
 227                 };
 228         } sa;
 229 };
 230
 231 #define vio_le() virtio_legacy_is_little_endian()
 232
 233 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 234
 235 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 236 #define GET_PBLOCK_DESC(x, bid) \
 237         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 238 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 239         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 240 #define GET_NEXT_PRB_BLK_NUM(x) \
 241         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 242         ((x)->kactive_blk_num+1) : 0)
 243
 244 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 245 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 246
 247 static int packet_direct_xmit(struct sk_buff *skb)
 248 {
 249         struct net_device *dev = skb->dev;
 250         struct sk_buff *orig_skb = skb;
 251         struct netdev_queue *txq;
 252         int ret = NETDEV_TX_BUSY;
 253
 254         if (unlikely(!netif_running(dev) ||
 255                      !netif_carrier_ok(dev)))
 256                 goto drop;
 257
 258         skb = validate_xmit_skb_list(skb, dev);
 259         if (skb != orig_skb)
 260                 goto drop;
 261
 262         packet_pick_tx_queue(dev, skb);
 263         txq = skb_get_tx_queue(dev, skb);
 264
 265         local_bh_disable();
 266
 267         HARD_TX_LOCK(dev, txq, smp_processor_id());
 268         if (!netif_xmit_frozen_or_drv_stopped(txq))
 269                 ret = netdev_start_xmit(skb, dev, txq, false);
 270         HARD_TX_UNLOCK(dev, txq);
 271
 272         local_bh_enable();
 273
 274         if (!dev_xmit_complete(ret))
 275                 kfree_skb(skb);
 276
 277         return ret;
 278 drop:
 279         atomic_long_inc(&dev->tx_dropped);
 280         kfree_skb_list(skb);
 281         return NET_XMIT_DROP;
 282 }
 283
 284 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
 285 {
 286         struct net_device *dev;
 287
 288         rcu_read_lock();
 289         dev = rcu_dereference(po->cached_dev);
 290         if (likely(dev))
 291                 dev_hold(dev);
 292         rcu_read_unlock();
 293
 294         return dev;
 295 }
 296
 297 static void packet_cached_dev_assign(struct packet_sock *po,
 298                                      struct net_device *dev)
 299 {
 300         rcu_assign_pointer(po->cached_dev, dev);
 301 }
 302
 303 static void packet_cached_dev_reset(struct packet_sock *po)
 304 {
 305         RCU_INIT_POINTER(po->cached_dev, NULL);
 306 }
 307
 308 static bool packet_use_direct_xmit(const struct packet_sock *po)
 309 {
 310         return po->xmit == packet_direct_xmit;
 311 }
 312
 313 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
 314 {
 315         return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
 316 }
 317
 318 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
 319 {
 320         const struct net_device_ops *ops = dev->netdev_ops;
 321         u16 queue_index;
 322
 323         if (ops->ndo_select_queue) {
 324                 queue_index = ops->ndo_select_queue(dev, skb, NULL,
 325                                                     __packet_pick_tx_queue);
 326                 queue_index = netdev_cap_txqueue(dev, queue_index);
 327         } else {
 328                 queue_index = __packet_pick_tx_queue(dev, skb);
 329         }
 330
 331         skb_set_queue_mapping(skb, queue_index);
 332 }
 333
 334 /* register_prot_hook must be invoked with the po->bind_lock held,
 335  * or from a context in which asynchronous accesses to the packet
 336  * socket is not possible (packet_create()).
 337  */
 338 static void register_prot_hook(struct sock *sk)
 339 {
 340         struct packet_sock *po = pkt_sk(sk);
 341
 342         if (!po->running) {
 343                 if (po->fanout)
 344                         __fanout_link(sk, po);
 345                 else
 346                         dev_add_pack(&po->prot_hook);
 347
 348                 sock_hold(sk);
 349                 po->running = 1;
 350         }
 351 }
 352
 353 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
 354  * held.   If the sync parameter is true, we will temporarily drop
 355  * the po->bind_lock and do a synchronize_net to make sure no
 356  * asynchronous packet processing paths still refer to the elements
 357  * of po->prot_hook.  If the sync parameter is false, it is the
 358  * callers responsibility to take care of this.
 359  */
 360 static void __unregister_prot_hook(struct sock *sk, bool sync)
 361 {
 362         struct packet_sock *po = pkt_sk(sk);
 363
 364         po->running = 0;
 365
 366         if (po->fanout)
 367                 __fanout_unlink(sk, po);
 368         else
 369                 __dev_remove_pack(&po->prot_hook);
 370
 371         __sock_put(sk);
 372
 373         if (sync) {
 374                 spin_unlock(&po->bind_lock);
 375                 synchronize_net();
 376                 spin_lock(&po->bind_lock);
 377         }
 378 }
 379
 380 static void unregister_prot_hook(struct sock *sk, bool sync)
 381 {
 382         struct packet_sock *po = pkt_sk(sk);
 383
 384         if (po->running)
 385                 __unregister_prot_hook(sk, sync);
 386 }
 387
 388 static inline struct page * __pure pgv_to_page(void *addr)
 389 {
 390         if (is_vmalloc_addr(addr))
 391                 return vmalloc_to_page(addr);
 392         return virt_to_page(addr);
 393 }
 394
 395 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 396 {
 397         union tpacket_uhdr h;
 398
 399         h.raw = frame;
 400         switch (po->tp_version) {
 401         case TPACKET_V1:
 402                 h.h1->tp_status = status;
 403                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 404                 break;
 405         case TPACKET_V2:
 406                 h.h2->tp_status = status;
 407                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 408                 break;
 409         case TPACKET_V3:
 410                 h.h3->tp_status = status;
 411                 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
 412                 break;
 413         default:
 414                 WARN(1, "TPACKET version not supported.\n");
 415                 BUG();
 416         }
 417
 418         smp_wmb();
 419 }
 420
 421 static int __packet_get_status(struct packet_sock *po, void *frame)
 422 {
 423         union tpacket_uhdr h;
 424
 425         smp_rmb();
 426
 427         h.raw = frame;
 428         switch (po->tp_version) {
 429         case TPACKET_V1:
 430                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 431                 return h.h1->tp_status;
 432         case TPACKET_V2:
 433                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 434                 return h.h2->tp_status;
 435         case TPACKET_V3:
 436                 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
 437                 return h.h3->tp_status;
 438         default:
 439                 WARN(1, "TPACKET version not supported.\n");
 440                 BUG();
 441                 return 0;
 442         }
 443 }
 444
 445 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
 446                                    unsigned int flags)
 447 {
 448         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 449
 450         if (shhwtstamps &&
 451             (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
 452             ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
 453                 return TP_STATUS_TS_RAW_HARDWARE;
 454
 455         if (ktime_to_timespec_cond(skb->tstamp, ts))
 456                 return TP_STATUS_TS_SOFTWARE;
 457
 458         return 0;
 459 }
 460
 461 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
 462                                     struct sk_buff *skb)
 463 {
 464         union tpacket_uhdr h;
 465         struct timespec ts;
 466         __u32 ts_status;
 467
 468         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
 469                 return 0;
 470
 471         h.raw = frame;
 472         switch (po->tp_version) {
 473         case TPACKET_V1:
 474                 h.h1->tp_sec = ts.tv_sec;
 475                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
 476                 break;
 477         case TPACKET_V2:
 478                 h.h2->tp_sec = ts.tv_sec;
 479                 h.h2->tp_nsec = ts.tv_nsec;
 480                 break;
 481         case TPACKET_V3:
 482                 h.h3->tp_sec = ts.tv_sec;
 483                 h.h3->tp_nsec = ts.tv_nsec;
 484                 break;
 485         default:
 486                 WARN(1, "TPACKET version not supported.\n");
 487                 BUG();
 488         }
 489
 490         /* one flush is safe, as both fields always lie on the same cacheline */
 491         flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
 492         smp_wmb();
 493
 494         return ts_status;
 495 }
 496
 497 static void *packet_lookup_frame(struct packet_sock *po,
 498                 struct packet_ring_buffer *rb,
 499                 unsigned int position,
 500                 int status)
 501 {
 502         unsigned int pg_vec_pos, frame_offset;
 503         union tpacket_uhdr h;
 504
 505         pg_vec_pos = position / rb->frames_per_block;
 506         frame_offset = position % rb->frames_per_block;
 507
 508         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 509                 (frame_offset * rb->frame_size);
 510
 511         if (status != __packet_get_status(po, h.raw))
 512                 return NULL;
 513
 514         return h.raw;
 515 }
 516
 517 static void *packet_current_frame(struct packet_sock *po,
 518                 struct packet_ring_buffer *rb,
 519                 int status)
 520 {
 521         return packet_lookup_frame(po, rb, rb->head, status);
 522 }
 523
 524 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 525 {
 526         del_timer_sync(&pkc->retire_blk_timer);
 527 }
 528
 529 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 530                 struct sk_buff_head *rb_queue)
 531 {
 532         struct tpacket_kbdq_core *pkc;
 533
 534         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 535
 536         spin_lock_bh(&rb_queue->lock);
 537         pkc->delete_blk_timer = 1;
 538         spin_unlock_bh(&rb_queue->lock);
 539
 540         prb_del_retire_blk_timer(pkc);
 541 }
 542
 543 static void prb_init_blk_timer(struct packet_sock *po,
 544                 struct tpacket_kbdq_core *pkc,
 545                 void (*func) (unsigned long))
 546 {
 547         init_timer(&pkc->retire_blk_timer);
 548         pkc->retire_blk_timer.data = (long)po;
 549         pkc->retire_blk_timer.function = func;
 550         pkc->retire_blk_timer.expires = jiffies;
 551 }
 552
 553 static void prb_setup_retire_blk_timer(struct packet_sock *po)
 554 {
 555         struct tpacket_kbdq_core *pkc;
 556
 557         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 558         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
 559 }
 560
 561 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 562                                 int blk_size_in_bytes)
 563 {
 564         struct net_device *dev;
 565         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 566         struct ethtool_link_ksettings ecmd;
 567         int err;
 568
 569         rtnl_lock();
 570         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 571         if (unlikely(!dev)) {
 572                 rtnl_unlock();
 573                 return DEFAULT_PRB_RETIRE_TOV;
 574         }
 575         err = __ethtool_get_link_ksettings(dev, &ecmd);
 576         rtnl_unlock();
 577         if (!err) {
 578                 /*
 579                  * If the link speed is so slow you don't really
 580                  * need to worry about perf anyways
 581                  */
 582                 if (ecmd.base.speed < SPEED_1000 ||
 583                     ecmd.base.speed == SPEED_UNKNOWN) {
 584                         return DEFAULT_PRB_RETIRE_TOV;
 585                 } else {
 586                         msec = 1;
 587                         div = ecmd.base.speed / 1000;
 588                 }
 589         }
 590
 591         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 592
 593         if (div)
 594                 mbits /= div;
 595
 596         tmo = mbits * msec;
 597
 598         if (div)
 599                 return tmo+1;
 600         return tmo;
 601 }
 602
 603 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 604                         union tpacket_req_u *req_u)
 605 {
 606         p1->feature_req_word = req_u->req3.tp_feature_req_word;
 607 }
 608
 609 static void init_prb_bdqc(struct packet_sock *po,
 610                         struct packet_ring_buffer *rb,
 611                         struct pgv *pg_vec,
 612                         union tpacket_req_u *req_u)
 613 {
 614         struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
 615         struct tpacket_block_desc *pbd;
 616
 617         memset(p1, 0x0, sizeof(*p1));
 618
 619         p1->knxt_seq_num = 1;
 620         p1->pkbdq = pg_vec;
 621         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 622         p1->pkblk_start = pg_vec[0].buffer;
 623         p1->kblk_size = req_u->req3.tp_block_size;
 624         p1->knum_blocks = req_u->req3.tp_block_nr;
 625         p1->hdrlen = po->tp_hdrlen;
 626         p1->version = po->tp_version;
 627         p1->last_kactive_blk_num = 0;
 628         po->stats.stats3.tp_freeze_q_cnt = 0;
 629         if (req_u->req3.tp_retire_blk_tov)
 630                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 631         else
 632                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 633                                                 req_u->req3.tp_block_size);
 634         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 635         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 636
 637         p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
 638         prb_init_ft_ops(p1, req_u);
 639         prb_setup_retire_blk_timer(po);
 640         prb_open_block(p1, pbd);
 641 }
 642
 643 /*  Do NOT update the last_blk_num first.
 644  *  Assumes sk_buff_head lock is held.
 645  */
 646 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 647 {
 648         mod_timer(&pkc->retire_blk_timer,
 649                         jiffies + pkc->tov_in_jiffies);
 650         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 651 }
 652
 653 /*
 654  * Timer logic:
 655  * 1) We refresh the timer only when we open a block.
 656  *    By doing this we don't waste cycles refreshing the timer
 657  *        on packet-by-packet basis.
 658  *
 659  * With a 1MB block-size, on a 1Gbps line, it will take
 660  * i) ~8 ms to fill a block + ii) memcpy etc.
 661  * In this cut we are not accounting for the memcpy time.
 662  *
 663  * So, if the user sets the 'tmo' to 10ms then the timer
 664  * will never fire while the block is still getting filled
 665  * (which is what we want). However, the user could choose
 666  * to close a block early and that's fine.
 667  *
 668  * But when the timer does fire, we check whether or not to refresh it.
 669  * Since the tmo granularity is in msecs, it is not too expensive
 670  * to refresh the timer, lets say every '8' msecs.
 671  * Either the user can set the 'tmo' or we can derive it based on
 672  * a) line-speed and b) block-size.
 673  * prb_calc_retire_blk_tmo() calculates the tmo.
 674  *
 675  */
 676 static void prb_retire_rx_blk_timer_expired(unsigned long data)
 677 {
 678         struct packet_sock *po = (struct packet_sock *)data;
 679         struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 680         unsigned int frozen;
 681         struct tpacket_block_desc *pbd;
 682
 683         spin_lock(&po->sk.sk_receive_queue.lock);
 684
 685         frozen = prb_queue_frozen(pkc);
 686         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 687
 688         if (unlikely(pkc->delete_blk_timer))
 689                 goto out;
 690
 691         /* We only need to plug the race when the block is partially filled.
 692          * tpacket_rcv:
 693          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 694          *              copy_bits() is in progress ...
 695          *              timer fires on other cpu:
 696          *              we can't retire the current block because copy_bits
 697          *              is in progress.
 698          *
 699          */
 700         if (BLOCK_NUM_PKTS(pbd)) {
 701                 while (atomic_read(&pkc->blk_fill_in_prog)) {
 702                         /* Waiting for skb_copy_bits to finish... */
 703                         cpu_relax();
 704                 }
 705         }
 706
 707         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 708                 if (!frozen) {
 709                         if (!BLOCK_NUM_PKTS(pbd)) {
 710                                 /* An empty block. Just refresh the timer. */
 711                                 goto refresh_timer;
 712                         }
 713                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 714                         if (!prb_dispatch_next_block(pkc, po))
 715                                 goto refresh_timer;
 716                         else
 717                                 goto out;
 718                 } else {
 719                         /* Case 1. Queue was frozen because user-space was
 720                          *         lagging behind.
 721                          */
 722                         if (prb_curr_blk_in_use(pbd)) {
 723                                 /*
 724                                  * Ok, user-space is still behind.
 725                                  * So just refresh the timer.
 726                                  */
 727                                 goto refresh_timer;
 728                         } else {
 729                                /* Case 2. queue was frozen,user-space caught up,
 730                                 * now the link went idle && the timer fired.
 731                                 * We don't have a block to close.So we open this
 732                                 * block and restart the timer.
 733                                 * opening a block thaws the queue,restarts timer
 734                                 * Thawing/timer-refresh is a side effect.
 735                                 */
 736                                 prb_open_block(pkc, pbd);
 737                                 goto out;
 738                         }
 739                 }
 740         }
 741
 742 refresh_timer:
 743         _prb_refresh_rx_retire_blk_timer(pkc);
 744
 745 out:
 746         spin_unlock(&po->sk.sk_receive_queue.lock);
 747 }
 748
 749 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 750                 struct tpacket_block_desc *pbd1, __u32 status)
 751 {
 752         /* Flush everything minus the block header */
 753
 754 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 755         u8 *start, *end;
 756
 757         start = (u8 *)pbd1;
 758
 759         /* Skip the block header(we know header WILL fit in 4K) */
 760         start += PAGE_SIZE;
 761
 762         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 763         for (; start < end; start += PAGE_SIZE)
 764                 flush_dcache_page(pgv_to_page(start));
 765
 766         smp_wmb();
 767 #endif
 768
 769         /* Now update the block status. */
 770
 771         BLOCK_STATUS(pbd1) = status;
 772
 773         /* Flush the block header */
 774
 775 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 776         start = (u8 *)pbd1;
 777         flush_dcache_page(pgv_to_page(start));
 778
 779         smp_wmb();
 780 #endif
 781 }
 782
 783 /*
 784  * Side effect:
 785  *
 786  * 1) flush the block
 787  * 2) Increment active_blk_num
 788  *
 789  * Note:We DONT refresh the timer on purpose.
 790  *      Because almost always the next block will be opened.
 791  */
 792 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 793                 struct tpacket_block_desc *pbd1,
 794                 struct packet_sock *po, unsigned int stat)
 795 {
 796         __u32 status = TP_STATUS_USER | stat;
 797
 798         struct tpacket3_hdr *last_pkt;
 799         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 800         struct sock *sk = &po->sk;
 801
 802         if (po->stats.stats3.tp_drops)
 803                 status |= TP_STATUS_LOSING;
 804
 805         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 806         last_pkt->tp_next_offset = 0;
 807
 808         /* Get the ts of the last pkt */
 809         if (BLOCK_NUM_PKTS(pbd1)) {
 810                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 811                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 812         } else {
 813                 /* Ok, we tmo'd - so get the current time.
 814                  *
 815                  * It shouldn't really happen as we don't close empty
 816                  * blocks. See prb_retire_rx_blk_timer_expired().
 817                  */
 818                 struct timespec ts;
 819                 getnstimeofday(&ts);
 820                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 821                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 822         }
 823
 824         smp_wmb();
 825
 826         /* Flush the block */
 827         prb_flush_block(pkc1, pbd1, status);
 828
 829         sk->sk_data_ready(sk);
 830
 831         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 832 }
 833
 834 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 835 {
 836         pkc->reset_pending_on_curr_blk = 0;
 837 }
 838
 839 /*
 840  * Side effect of opening a block:
 841  *
 842  * 1) prb_queue is thawed.
 843  * 2) retire_blk_timer is refreshed.
 844  *
 845  */
 846 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 847         struct tpacket_block_desc *pbd1)
 848 {
 849         struct timespec ts;
 850         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 851
 852         smp_rmb();
 853
 854         /* We could have just memset this but we will lose the
 855          * flexibility of making the priv area sticky
 856          */
 857
 858         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 859         BLOCK_NUM_PKTS(pbd1) = 0;
 860         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 861
 862         getnstimeofday(&ts);
 863
 864         h1->ts_first_pkt.ts_sec = ts.tv_sec;
 865         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 866
 867         pkc1->pkblk_start = (char *)pbd1;
 868         pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 869
 870         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 871         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 872
 873         pbd1->version = pkc1->version;
 874         pkc1->prev = pkc1->nxt_offset;
 875         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 876
 877         prb_thaw_queue(pkc1);
 878         _prb_refresh_rx_retire_blk_timer(pkc1);
 879
 880         smp_wmb();
 881 }
 882
 883 /*
 884  * Queue freeze logic:
 885  * 1) Assume tp_block_nr = 8 blocks.
 886  * 2) At time 't0', user opens Rx ring.
 887  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 888  * 4) user-space is either sleeping or processing block '0'.
 889  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 890  *    it will close block-7,loop around and try to fill block '0'.
 891  *    call-flow:
 892  *    __packet_lookup_frame_in_block
 893  *      prb_retire_current_block()
 894  *      prb_dispatch_next_block()
 895  *        |->(BLOCK_STATUS == USER) evaluates to true
 896  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 897  * 6) Now there are two cases:
 898  *    6.1) Link goes idle right after the queue is frozen.
 899  *         But remember, the last open_block() refreshed the timer.
 900  *         When this timer expires,it will refresh itself so that we can
 901  *         re-open block-0 in near future.
 902  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 903  *         case and __packet_lookup_frame_in_block will check if block-0
 904  *         is free and can now be re-used.
 905  */
 906 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 907                                   struct packet_sock *po)
 908 {
 909         pkc->reset_pending_on_curr_blk = 1;
 910         po->stats.stats3.tp_freeze_q_cnt++;
 911 }
 912
 913 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 914
 915 /*
 916  * If the next block is free then we will dispatch it
 917  * and return a good offset.
 918  * Else, we will freeze the queue.
 919  * So, caller must check the return value.
 920  */
 921 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 922                 struct packet_sock *po)
 923 {
 924         struct tpacket_block_desc *pbd;
 925
 926         smp_rmb();
 927
 928         /* 1. Get current block num */
 929         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 930
 931         /* 2. If this block is currently in_use then freeze the queue */
 932         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 933                 prb_freeze_queue(pkc, po);
 934                 return NULL;
 935         }
 936
 937         /*
 938          * 3.
 939          * open this block and return the offset where the first packet
 940          * needs to get stored.
 941          */
 942         prb_open_block(pkc, pbd);
 943         return (void *)pkc->nxt_offset;
 944 }
 945
 946 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 947                 struct packet_sock *po, unsigned int status)
 948 {
 949         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 950
 951         /* retire/close the current block */
 952         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 953                 /*
 954                  * Plug the case where copy_bits() is in progress on
 955                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 956                  * have space to copy the pkt in the current block and
 957                  * called prb_retire_current_block()
 958                  *
 959                  * We don't need to worry about the TMO case because
 960                  * the timer-handler already handled this case.
 961                  */
 962                 if (!(status & TP_STATUS_BLK_TMO)) {
 963                         while (atomic_read(&pkc->blk_fill_in_prog)) {
 964                                 /* Waiting for skb_copy_bits to finish... */
 965                                 cpu_relax();
 966                         }
 967                 }
 968                 prb_close_block(pkc, pbd, po, status);
 969                 return;
 970         }
 971 }
 972
 973 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
 974 {
 975         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 976 }
 977
 978 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 979 {
 980         return pkc->reset_pending_on_curr_blk;
 981 }
 982
 983 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 984 {
 985         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 986         atomic_dec(&pkc->blk_fill_in_prog);
 987 }
 988
 989 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 990                         struct tpacket3_hdr *ppd)
 991 {
 992         ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
 993 }
 994
 995 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 996                         struct tpacket3_hdr *ppd)
 997 {
 998         ppd->hv1.tp_rxhash = 0;
 999 }
1000
1001 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1002                         struct tpacket3_hdr *ppd)
1003 {
1004         if (skb_vlan_tag_present(pkc->skb)) {
1005                 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1006                 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1007                 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1008         } else {
1009                 ppd->hv1.tp_vlan_tci = 0;
1010                 ppd->hv1.tp_vlan_tpid = 0;
1011                 ppd->tp_status = TP_STATUS_AVAILABLE;
1012         }
1013 }
1014
1015 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1016                         struct tpacket3_hdr *ppd)
1017 {
1018         ppd->hv1.tp_padding = 0;
1019         prb_fill_vlan_info(pkc, ppd);
1020
1021         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1022                 prb_fill_rxhash(pkc, ppd);
1023         else
1024                 prb_clear_rxhash(pkc, ppd);
1025 }
1026
1027 static void prb_fill_curr_block(char *curr,
1028                                 struct tpacket_kbdq_core *pkc,
1029                                 struct tpacket_block_desc *pbd,
1030                                 unsigned int len)
1031 {
1032         struct tpacket3_hdr *ppd;
1033
1034         ppd  = (struct tpacket3_hdr *)curr;
1035         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1036         pkc->prev = curr;
1037         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1038         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1039         BLOCK_NUM_PKTS(pbd) += 1;
1040         atomic_inc(&pkc->blk_fill_in_prog);
1041         prb_run_all_ft_ops(pkc, ppd);
1042 }
1043
1044 /* Assumes caller has the sk->rx_queue.lock */
1045 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1046                                             struct sk_buff *skb,
1047                                                 int status,
1048                                             unsigned int len
1049                                             )
1050 {
1051         struct tpacket_kbdq_core *pkc;
1052         struct tpacket_block_desc *pbd;
1053         char *curr, *end;
1054
1055         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1056         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1057
1058         /* Queue is frozen when user space is lagging behind */
1059         if (prb_queue_frozen(pkc)) {
1060                 /*
1061                  * Check if that last block which caused the queue to freeze,
1062                  * is still in_use by user-space.
1063                  */
1064                 if (prb_curr_blk_in_use(pbd)) {
1065                         /* Can't record this packet */
1066                         return NULL;
1067                 } else {
1068                         /*
1069                          * Ok, the block was released by user-space.
1070                          * Now let's open that block.
1071                          * opening a block also thaws the queue.
1072                          * Thawing is a side effect.
1073                          */
1074                         prb_open_block(pkc, pbd);
1075                 }
1076         }
1077
1078         smp_mb();
1079         curr = pkc->nxt_offset;
1080         pkc->skb = skb;
1081         end = (char *)pbd + pkc->kblk_size;
1082
1083         /* first try the current block */
1084         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1085                 prb_fill_curr_block(curr, pkc, pbd, len);
1086                 return (void *)curr;
1087         }
1088
1089         /* Ok, close the current block */
1090         prb_retire_current_block(pkc, po, 0);
1091
1092         /* Now, try to dispatch the next block */
1093         curr = (char *)prb_dispatch_next_block(pkc, po);
1094         if (curr) {
1095                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1096                 prb_fill_curr_block(curr, pkc, pbd, len);
1097                 return (void *)curr;
1098         }
1099
1100         /*
1101          * No free blocks are available.user_space hasn't caught up yet.
1102          * Queue was just frozen and now this packet will get dropped.
1103          */
1104         return NULL;
1105 }
1106
1107 static void *packet_current_rx_frame(struct packet_sock *po,
1108                                             struct sk_buff *skb,
1109                                             int status, unsigned int len)
1110 {
1111         char *curr = NULL;
1112         switch (po->tp_version) {
1113         case TPACKET_V1:
1114         case TPACKET_V2:
1115                 curr = packet_lookup_frame(po, &po->rx_ring,
1116                                         po->rx_ring.head, status);
1117                 return curr;
1118         case TPACKET_V3:
1119                 return __packet_lookup_frame_in_block(po, skb, status, len);
1120         default:
1121                 WARN(1, "TPACKET version not supported\n");
1122                 BUG();
1123                 return NULL;
1124         }
1125 }
1126
1127 static void *prb_lookup_block(struct packet_sock *po,
1128                                      struct packet_ring_buffer *rb,
1129                                      unsigned int idx,
1130                                      int status)
1131 {
1132         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1133         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1134
1135         if (status != BLOCK_STATUS(pbd))
1136                 return NULL;
1137         return pbd;
1138 }
1139
1140 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1141 {
1142         unsigned int prev;
1143         if (rb->prb_bdqc.kactive_blk_num)
1144                 prev = rb->prb_bdqc.kactive_blk_num-1;
1145         else
1146                 prev = rb->prb_bdqc.knum_blocks-1;
1147         return prev;
1148 }
1149
1150 /* Assumes caller has held the rx_queue.lock */
1151 static void *__prb_previous_block(struct packet_sock *po,
1152                                          struct packet_ring_buffer *rb,
1153                                          int status)
1154 {
1155         unsigned int previous = prb_previous_blk_num(rb);
1156         return prb_lookup_block(po, rb, previous, status);
1157 }
1158
1159 static void *packet_previous_rx_frame(struct packet_sock *po,
1160                                              struct packet_ring_buffer *rb,
1161                                              int status)
1162 {
1163         if (po->tp_version <= TPACKET_V2)
1164                 return packet_previous_frame(po, rb, status);
1165
1166         return __prb_previous_block(po, rb, status);
1167 }
1168
1169 static void packet_increment_rx_head(struct packet_sock *po,
1170                                             struct packet_ring_buffer *rb)
1171 {
1172         switch (po->tp_version) {
1173         case TPACKET_V1:
1174         case TPACKET_V2:
1175                 return packet_increment_head(rb);
1176         case TPACKET_V3:
1177         default:
1178                 WARN(1, "TPACKET version not supported.\n");
1179                 BUG();
1180                 return;
1181         }
1182 }
1183
1184 static void *packet_previous_frame(struct packet_sock *po,
1185                 struct packet_ring_buffer *rb,
1186                 int status)
1187 {
1188         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1189         return packet_lookup_frame(po, rb, previous, status);
1190 }
1191
1192 static void packet_increment_head(struct packet_ring_buffer *buff)
1193 {
1194         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1195 }
1196
1197 static void packet_inc_pending(struct packet_ring_buffer *rb)
1198 {
1199         this_cpu_inc(*rb->pending_refcnt);
1200 }
1201
1202 static void packet_dec_pending(struct packet_ring_buffer *rb)
1203 {
1204         this_cpu_dec(*rb->pending_refcnt);
1205 }
1206
1207 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1208 {
1209         unsigned int refcnt = 0;
1210         int cpu;
1211
1212         /* We don't use pending refcount in rx_ring. */
1213         if (rb->pending_refcnt == NULL)
1214                 return 0;
1215
1216         for_each_possible_cpu(cpu)
1217                 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1218
1219         return refcnt;
1220 }
1221
1222 static int packet_alloc_pending(struct packet_sock *po)
1223 {
1224         po->rx_ring.pending_refcnt = NULL;
1225
1226         po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1227         if (unlikely(po->tx_ring.pending_refcnt == NULL))
1228                 return -ENOBUFS;
1229
1230         return 0;
1231 }
1232
1233 static void packet_free_pending(struct packet_sock *po)
1234 {
1235         free_percpu(po->tx_ring.pending_refcnt);
1236 }
1237
1238 #define ROOM_POW_OFF    2
1239 #define ROOM_NONE       0x0
1240 #define ROOM_LOW        0x1
1241 #define ROOM_NORMAL     0x2
1242
1243 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1244 {
1245         int idx, len;
1246
1247         len = po->rx_ring.frame_max + 1;
1248         idx = po->rx_ring.head;
1249         if (pow_off)
1250                 idx += len >> pow_off;
1251         if (idx >= len)
1252                 idx -= len;
1253         return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1254 }
1255
1256 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1257 {
1258         int idx, len;
1259
1260         len = po->rx_ring.prb_bdqc.knum_blocks;
1261         idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1262         if (pow_off)
1263                 idx += len >> pow_off;
1264         if (idx >= len)
1265                 idx -= len;
1266         return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1267 }
1268
1269 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1270 {
1271         struct sock *sk = &po->sk;
1272         int ret = ROOM_NONE;
1273
1274         if (po->prot_hook.func != tpacket_rcv) {
1275                 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1276                                           - (skb ? skb->truesize : 0);
1277                 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1278                         return ROOM_NORMAL;
1279                 else if (avail > 0)
1280                         return ROOM_LOW;
1281                 else
1282                         return ROOM_NONE;
1283         }
1284
1285         if (po->tp_version == TPACKET_V3) {
1286                 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1287                         ret = ROOM_NORMAL;
1288                 else if (__tpacket_v3_has_room(po, 0))
1289                         ret = ROOM_LOW;
1290         } else {
1291                 if (__tpacket_has_room(po, ROOM_POW_OFF))
1292                         ret = ROOM_NORMAL;
1293                 else if (__tpacket_has_room(po, 0))
1294                         ret = ROOM_LOW;
1295         }
1296
1297         return ret;
1298 }
1299
1300 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1301 {
1302         int ret;
1303         bool has_room;
1304
1305         spin_lock_bh(&po->sk.sk_receive_queue.lock);
1306         ret = __packet_rcv_has_room(po, skb);
1307         has_room = ret == ROOM_NORMAL;
1308         if (po->pressure == has_room)
1309                 po->pressure = !has_room;
1310         spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1311
1312         return ret;
1313 }
1314
1315 static void packet_sock_destruct(struct sock *sk)
1316 {
1317         skb_queue_purge(&sk->sk_error_queue);
1318
1319         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1320         WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1321
1322         if (!sock_flag(sk, SOCK_DEAD)) {
1323                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1324                 return;
1325         }
1326
1327         sk_refcnt_debug_dec(sk);
1328 }
1329
1330 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1331 {
1332         u32 rxhash;
1333         int i, count = 0;
1334
1335         rxhash = skb_get_hash(skb);
1336         for (i = 0; i < ROLLOVER_HLEN; i++)
1337                 if (po->rollover->history[i] == rxhash)
1338                         count++;
1339
1340         po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1341         return count > (ROLLOVER_HLEN >> 1);
1342 }
1343
1344 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1345                                       struct sk_buff *skb,
1346                                       unsigned int num)
1347 {
1348         return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1349 }
1350
1351 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1352                                     struct sk_buff *skb,
1353                                     unsigned int num)
1354 {
1355         unsigned int val = atomic_inc_return(&f->rr_cur);
1356
1357         return val % num;
1358 }
1359
1360 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1361                                      struct sk_buff *skb,
1362                                      unsigned int num)
1363 {
1364         return smp_processor_id() % num;
1365 }
1366
1367 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1368                                      struct sk_buff *skb,
1369                                      unsigned int num)
1370 {
1371         return prandom_u32_max(num);
1372 }
1373
1374 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1375                                           struct sk_buff *skb,
1376                                           unsigned int idx, bool try_self,
1377                                           unsigned int num)
1378 {
1379         struct packet_sock *po, *po_next, *po_skip = NULL;
1380         unsigned int i, j, room = ROOM_NONE;
1381
1382         po = pkt_sk(f->arr[idx]);
1383
1384         if (try_self) {
1385                 room = packet_rcv_has_room(po, skb);
1386                 if (room == ROOM_NORMAL ||
1387                     (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1388                         return idx;
1389                 po_skip = po;
1390         }
1391
1392         i = j = min_t(int, po->rollover->sock, num - 1);
1393         do {
1394                 po_next = pkt_sk(f->arr[i]);
1395                 if (po_next != po_skip && !po_next->pressure &&
1396                     packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1397                         if (i != j)
1398                                 po->rollover->sock = i;
1399                         atomic_long_inc(&po->rollover->num);
1400                         if (room == ROOM_LOW)
1401                                 atomic_long_inc(&po->rollover->num_huge);
1402                         return i;
1403                 }
1404
1405                 if (++i == num)
1406                         i = 0;
1407         } while (i != j);
1408
1409         atomic_long_inc(&po->rollover->num_failed);
1410         return idx;
1411 }
1412
1413 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1414                                     struct sk_buff *skb,
1415                                     unsigned int num)
1416 {
1417         return skb_get_queue_mapping(skb) % num;
1418 }
1419
1420 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1421                                      struct sk_buff *skb,
1422                                      unsigned int num)
1423 {
1424         struct bpf_prog *prog;
1425         unsigned int ret = 0;
1426
1427         rcu_read_lock();
1428         prog = rcu_dereference(f->bpf_prog);
1429         if (prog)
1430                 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1431         rcu_read_unlock();
1432
1433         return ret;
1434 }
1435
1436 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1437 {
1438         return f->flags & (flag >> 8);
1439 }
1440
1441 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1442                              struct packet_type *pt, struct net_device *orig_dev)
1443 {
1444         struct packet_fanout *f = pt->af_packet_priv;
1445         unsigned int num = READ_ONCE(f->num_members);
1446         struct net *net = read_pnet(&f->net);
1447         struct packet_sock *po;
1448         unsigned int idx;
1449
1450         if (!net_eq(dev_net(dev), net) || !num) {
1451                 kfree_skb(skb);
1452                 return 0;
1453         }
1454
1455         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1456                 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1457                 if (!skb)
1458                         return 0;
1459         }
1460         switch (f->type) {
1461         case PACKET_FANOUT_HASH:
1462         default:
1463                 idx = fanout_demux_hash(f, skb, num);
1464                 break;
1465         case PACKET_FANOUT_LB:
1466                 idx = fanout_demux_lb(f, skb, num);
1467                 break;
1468         case PACKET_FANOUT_CPU:
1469                 idx = fanout_demux_cpu(f, skb, num);
1470                 break;
1471         case PACKET_FANOUT_RND:
1472                 idx = fanout_demux_rnd(f, skb, num);
1473                 break;
1474         case PACKET_FANOUT_QM:
1475                 idx = fanout_demux_qm(f, skb, num);
1476                 break;
1477         case PACKET_FANOUT_ROLLOVER:
1478                 idx = fanout_demux_rollover(f, skb, 0, false, num);
1479                 break;
1480         case PACKET_FANOUT_CBPF:
1481         case PACKET_FANOUT_EBPF:
1482                 idx = fanout_demux_bpf(f, skb, num);
1483                 break;
1484         }
1485
1486         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1487                 idx = fanout_demux_rollover(f, skb, idx, true, num);
1488
1489         po = pkt_sk(f->arr[idx]);
1490         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1491 }
1492
1493 DEFINE_MUTEX(fanout_mutex);
1494 EXPORT_SYMBOL_GPL(fanout_mutex);
1495 static LIST_HEAD(fanout_list);
1496 static u16 fanout_next_id;
1497
1498 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1499 {
1500         struct packet_fanout *f = po->fanout;
1501
1502         spin_lock(&f->lock);
1503         f->arr[f->num_members] = sk;
1504         smp_wmb();
1505         f->num_members++;
1506         if (f->num_members == 1)
1507                 dev_add_pack(&f->prot_hook);
1508         spin_unlock(&f->lock);
1509 }
1510
1511 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1512 {
1513         struct packet_fanout *f = po->fanout;
1514         int i;
1515
1516         spin_lock(&f->lock);
1517         for (i = 0; i < f->num_members; i++) {
1518                 if (f->arr[i] == sk)
1519                         break;
1520         }
1521         BUG_ON(i >= f->num_members);
1522         f->arr[i] = f->arr[f->num_members - 1];
1523         f->num_members--;
1524         if (f->num_members == 0)
1525                 __dev_remove_pack(&f->prot_hook);
1526         spin_unlock(&f->lock);
1527 }
1528
1529 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1530 {
1531         if (sk->sk_family != PF_PACKET)
1532                 return false;
1533
1534         return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1535 }
1536
1537 static void fanout_init_data(struct packet_fanout *f)
1538 {
1539         switch (f->type) {
1540         case PACKET_FANOUT_LB:
1541                 atomic_set(&f->rr_cur, 0);
1542                 break;
1543         case PACKET_FANOUT_CBPF:
1544         case PACKET_FANOUT_EBPF:
1545                 RCU_INIT_POINTER(f->bpf_prog, NULL);
1546                 break;
1547         }
1548 }
1549
1550 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1551 {
1552         struct bpf_prog *old;
1553
1554         spin_lock(&f->lock);
1555         old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1556         rcu_assign_pointer(f->bpf_prog, new);
1557         spin_unlock(&f->lock);
1558
1559         if (old) {
1560                 synchronize_net();
1561                 bpf_prog_destroy(old);
1562         }
1563 }
1564
1565 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1566                                 unsigned int len)
1567 {
1568         struct bpf_prog *new;
1569         struct sock_fprog fprog;
1570         int ret;
1571
1572         if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1573                 return -EPERM;
1574         if (len != sizeof(fprog))
1575                 return -EINVAL;
1576         if (copy_from_user(&fprog, data, len))
1577                 return -EFAULT;
1578
1579         ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1580         if (ret)
1581                 return ret;
1582
1583         __fanout_set_data_bpf(po->fanout, new);
1584         return 0;
1585 }
1586
1587 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1588                                 unsigned int len)
1589 {
1590         struct bpf_prog *new;
1591         u32 fd;
1592
1593         if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1594                 return -EPERM;
1595         if (len != sizeof(fd))
1596                 return -EINVAL;
1597         if (copy_from_user(&fd, data, len))
1598                 return -EFAULT;
1599
1600         new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1601         if (IS_ERR(new))
1602                 return PTR_ERR(new);
1603
1604         __fanout_set_data_bpf(po->fanout, new);
1605         return 0;
1606 }
1607
1608 static int fanout_set_data(struct packet_sock *po, char __user *data,
1609                            unsigned int len)
1610 {
1611         switch (po->fanout->type) {
1612         case PACKET_FANOUT_CBPF:
1613                 return fanout_set_data_cbpf(po, data, len);
1614         case PACKET_FANOUT_EBPF:
1615                 return fanout_set_data_ebpf(po, data, len);
1616         default:
1617                 return -EINVAL;
1618         };
1619 }
1620
1621 static void fanout_release_data(struct packet_fanout *f)
1622 {
1623         switch (f->type) {
1624         case PACKET_FANOUT_CBPF:
1625         case PACKET_FANOUT_EBPF:
1626                 __fanout_set_data_bpf(f, NULL);
1627         };
1628 }
1629
1630 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1631 {
1632         struct packet_fanout *f;
1633
1634         list_for_each_entry(f, &fanout_list, list) {
1635                 if (f->id == candidate_id &&
1636                     read_pnet(&f->net) == sock_net(sk)) {
1637                         return false;
1638                 }
1639         }
1640         return true;
1641 }
1642
1643 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1644 {
1645         u16 id = fanout_next_id;
1646
1647         do {
1648                 if (__fanout_id_is_free(sk, id)) {
1649                         *new_id = id;
1650                         fanout_next_id = id + 1;
1651                         return true;
1652                 }
1653
1654                 id++;
1655         } while (id != fanout_next_id);
1656
1657         return false;
1658 }
1659
1660 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1661 {
1662         struct packet_rollover *rollover = NULL;
1663         struct packet_sock *po = pkt_sk(sk);
1664         struct packet_fanout *f, *match;
1665         u8 type = type_flags & 0xff;
1666         u8 flags = type_flags >> 8;
1667         int err;
1668
1669         switch (type) {
1670         case PACKET_FANOUT_ROLLOVER:
1671                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1672                         return -EINVAL;
1673         case PACKET_FANOUT_HASH:
1674         case PACKET_FANOUT_LB:
1675         case PACKET_FANOUT_CPU:
1676         case PACKET_FANOUT_RND:
1677         case PACKET_FANOUT_QM:
1678         case PACKET_FANOUT_CBPF:
1679         case PACKET_FANOUT_EBPF:
1680                 break;
1681         default:
1682                 return -EINVAL;
1683         }
1684
1685         mutex_lock(&fanout_mutex);
1686
1687         err = -EALREADY;
1688         if (po->fanout)
1689                 goto out;
1690
1691         if (type == PACKET_FANOUT_ROLLOVER ||
1692             (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1693                 err = -ENOMEM;
1694                 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1695                 if (!rollover)
1696                         goto out;
1697                 atomic_long_set(&rollover->num, 0);
1698                 atomic_long_set(&rollover->num_huge, 0);
1699                 atomic_long_set(&rollover->num_failed, 0);
1700                 po->rollover = rollover;
1701         }
1702
1703         if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1704                 if (id != 0) {
1705                         err = -EINVAL;
1706                         goto out;
1707                 }
1708                 if (!fanout_find_new_id(sk, &id)) {
1709                         err = -ENOMEM;
1710                         goto out;
1711                 }
1712                 /* ephemeral flag for the first socket in the group: drop it */
1713                 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1714         }
1715
1716         match = NULL;
1717         list_for_each_entry(f, &fanout_list, list) {
1718                 if (f->id == id &&
1719                     read_pnet(&f->net) == sock_net(sk)) {
1720                         match = f;
1721                         break;
1722                 }
1723         }
1724         err = -EINVAL;
1725         if (match && match->flags != flags)
1726                 goto out;
1727         if (!match) {
1728                 err = -ENOMEM;
1729                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1730                 if (!match)
1731                         goto out;
1732                 write_pnet(&match->net, sock_net(sk));
1733                 match->id = id;
1734                 match->type = type;
1735                 match->flags = flags;
1736                 INIT_LIST_HEAD(&match->list);
1737                 spin_lock_init(&match->lock);
1738                 refcount_set(&match->sk_ref, 0);
1739                 fanout_init_data(match);
1740                 match->prot_hook.type = po->prot_hook.type;
1741                 match->prot_hook.dev = po->prot_hook.dev;
1742                 match->prot_hook.func = packet_rcv_fanout;
1743                 match->prot_hook.af_packet_priv = match;
1744                 match->prot_hook.id_match = match_fanout_group;
1745                 list_add(&match->list, &fanout_list);
1746         }
1747         err = -EINVAL;
1748
1749         spin_lock(&po->bind_lock);
1750         if (po->running &&
1751             match->type == type &&
1752             match->prot_hook.type == po->prot_hook.type &&
1753             match->prot_hook.dev == po->prot_hook.dev) {
1754                 err = -ENOSPC;
1755                 if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1756                         __dev_remove_pack(&po->prot_hook);
1757                         po->fanout = match;
1758                         refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1759                         __fanout_link(sk, po);
1760                         err = 0;
1761                 }
1762         }
1763         spin_unlock(&po->bind_lock);
1764
1765         if (err && !refcount_read(&match->sk_ref)) {
1766                 list_del(&match->list);
1767                 kfree(match);
1768         }
1769
1770 out:
1771         if (err && rollover) {
1772                 kfree(rollover);
1773                 po->rollover = NULL;
1774         }
1775         mutex_unlock(&fanout_mutex);
1776         return err;
1777 }
1778
1779 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1780  * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1781  * It is the responsibility of the caller to call fanout_release_data() and
1782  * free the returned packet_fanout (after synchronize_net())
1783  */
1784 static struct packet_fanout *fanout_release(struct sock *sk)
1785 {
1786         struct packet_sock *po = pkt_sk(sk);
1787         struct packet_fanout *f;
1788
1789         mutex_lock(&fanout_mutex);
1790         f = po->fanout;
1791         if (f) {
1792                 po->fanout = NULL;
1793
1794                 if (refcount_dec_and_test(&f->sk_ref))
1795                         list_del(&f->list);
1796                 else
1797                         f = NULL;
1798
1799                 if (po->rollover)
1800                         kfree_rcu(po->rollover, rcu);
1801         }
1802         mutex_unlock(&fanout_mutex);
1803
1804         return f;
1805 }
1806
1807 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1808                                           struct sk_buff *skb)
1809 {
1810         /* Earlier code assumed this would be a VLAN pkt, double-check
1811          * this now that we have the actual packet in hand. We can only
1812          * do this check on Ethernet devices.
1813          */
1814         if (unlikely(dev->type != ARPHRD_ETHER))
1815                 return false;
1816
1817         skb_reset_mac_header(skb);
1818         return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1819 }
1820
1821 static const struct proto_ops packet_ops;
1822
1823 static const struct proto_ops packet_ops_spkt;
1824
1825 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1826                            struct packet_type *pt, struct net_device *orig_dev)
1827 {
1828         struct sock *sk;
1829         struct sockaddr_pkt *spkt;
1830
1831         /*
1832          *      When we registered the protocol we saved the socket in the data
1833          *      field for just this event.
1834          */
1835
1836         sk = pt->af_packet_priv;
1837
1838         /*
1839          *      Yank back the headers [hope the device set this
1840          *      right or kerboom...]
1841          *
1842          *      Incoming packets have ll header pulled,
1843          *      push it back.
1844          *
1845          *      For outgoing ones skb->data == skb_mac_header(skb)
1846          *      so that this procedure is noop.
1847          */
1848
1849         if (skb->pkt_type == PACKET_LOOPBACK)
1850                 goto out;
1851
1852         if (!net_eq(dev_net(dev), sock_net(sk)))
1853                 goto out;
1854
1855         skb = skb_share_check(skb, GFP_ATOMIC);
1856         if (skb == NULL)
1857                 goto oom;
1858
1859         /* drop any routing info */
1860         skb_dst_drop(skb);
1861
1862         /* drop conntrack reference */
1863         nf_reset(skb);
1864
1865         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1866
1867         skb_push(skb, skb->data - skb_mac_header(skb));
1868
1869         /*
1870          *      The SOCK_PACKET socket receives _all_ frames.
1871          */
1872
1873         spkt->spkt_family = dev->type;
1874         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1875         spkt->spkt_protocol = skb->protocol;
1876
1877         /*
1878          *      Charge the memory to the socket. This is done specifically
1879          *      to prevent sockets using all the memory up.
1880          */
1881
1882         if (sock_queue_rcv_skb(sk, skb) == 0)
1883                 return 0;
1884
1885 out:
1886         kfree_skb(skb);
1887 oom:
1888         return 0;
1889 }
1890
1891
1892 /*
1893  *      Output a raw packet to a device layer. This bypasses all the other
1894  *      protocol layers and you must therefore supply it with a complete frame
1895  */
1896
1897 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1898                                size_t len)
1899 {
1900         struct sock *sk = sock->sk;
1901         DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1902         struct sk_buff *skb = NULL;
1903         struct net_device *dev;
1904         struct sockcm_cookie sockc;
1905         __be16 proto = 0;
1906         int err;
1907         int extra_len = 0;
1908
1909         /*
1910          *      Get and verify the address.
1911          */
1912
1913         if (saddr) {
1914                 if (msg->msg_namelen < sizeof(struct sockaddr))
1915                         return -EINVAL;
1916                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1917                         proto = saddr->spkt_protocol;
1918         } else
1919                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1920
1921         /*
1922          *      Find the device first to size check it
1923          */
1924
1925         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1926 retry:
1927         rcu_read_lock();
1928         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1929         err = -ENODEV;
1930         if (dev == NULL)
1931                 goto out_unlock;
1932
1933         err = -ENETDOWN;
1934         if (!(dev->flags & IFF_UP))
1935                 goto out_unlock;
1936
1937         /*
1938          * You may not queue a frame bigger than the mtu. This is the lowest level
1939          * raw protocol and you must do your own fragmentation at this level.
1940          */
1941
1942         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1943                 if (!netif_supports_nofcs(dev)) {
1944                         err = -EPROTONOSUPPORT;
1945                         goto out_unlock;
1946                 }
1947                 extra_len = 4; /* We're doing our own CRC */
1948         }
1949
1950         err = -EMSGSIZE;
1951         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1952                 goto out_unlock;
1953
1954         if (!skb) {
1955                 size_t reserved = LL_RESERVED_SPACE(dev);
1956                 int tlen = dev->needed_tailroom;
1957                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1958
1959                 rcu_read_unlock();
1960                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1961                 if (skb == NULL)
1962                         return -ENOBUFS;
1963                 /* FIXME: Save some space for broken drivers that write a hard
1964                  * header at transmission time by themselves. PPP is the notable
1965                  * one here. This should really be fixed at the driver level.
1966                  */
1967                 skb_reserve(skb, reserved);
1968                 skb_reset_network_header(skb);
1969
1970                 /* Try to align data part correctly */
1971                 if (hhlen) {
1972                         skb->data -= hhlen;
1973                         skb->tail -= hhlen;
1974                         if (len < hhlen)
1975                                 skb_reset_network_header(skb);
1976                 }
1977                 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1978                 if (err)
1979                         goto out_free;
1980                 goto retry;
1981         }
1982
1983         if (!dev_validate_header(dev, skb->data, len)) {
1984                 err = -EINVAL;
1985                 goto out_unlock;
1986         }
1987         if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1988             !packet_extra_vlan_len_allowed(dev, skb)) {
1989                 err = -EMSGSIZE;
1990                 goto out_unlock;
1991         }
1992
1993         sockc.tsflags = sk->sk_tsflags;
1994         if (msg->msg_controllen) {
1995                 err = sock_cmsg_send(sk, msg, &sockc);
1996                 if (unlikely(err))
1997                         goto out_unlock;
1998         }
1999
2000         skb->protocol = proto;
2001         skb->dev = dev;
2002         skb->priority = sk->sk_priority;
2003         skb->mark = sk->sk_mark;
2004
2005         sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
2006
2007         if (unlikely(extra_len == 4))
2008                 skb->no_fcs = 1;
2009
2010         skb_probe_transport_header(skb, 0);
2011
2012         dev_queue_xmit(skb);
2013         rcu_read_unlock();
2014         return len;
2015
2016 out_unlock:
2017         rcu_read_unlock();
2018 out_free:
2019         kfree_skb(skb);
2020         return err;
2021 }
2022
2023 static unsigned int run_filter(struct sk_buff *skb,
2024                                const struct sock *sk,
2025                                unsigned int res)
2026 {
2027         struct sk_filter *filter;
2028
2029         rcu_read_lock();
2030         filter = rcu_dereference(sk->sk_filter);
2031         if (filter != NULL)
2032                 res = bpf_prog_run_clear_cb(filter->prog, skb);
2033         rcu_read_unlock();
2034
2035         return res;
2036 }
2037
2038 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2039                            size_t *len)
2040 {
2041         struct virtio_net_hdr vnet_hdr;
2042
2043         if (*len < sizeof(vnet_hdr))
2044                 return -EINVAL;
2045         *len -= sizeof(vnet_hdr);
2046
2047         if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true))
2048                 return -EINVAL;
2049
2050         return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2051 }
2052
2053 /*
2054  * This function makes lazy skb cloning in hope that most of packets
2055  * are discarded by BPF.
2056  *
2057  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2058  * and skb->cb are mangled. It works because (and until) packets
2059  * falling here are owned by current CPU. Output packets are cloned
2060  * by dev_queue_xmit_nit(), input packets are processed by net_bh
2061  * sequencially, so that if we return skb to original state on exit,
2062  * we will not harm anyone.
2063  */
2064
2065 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2066                       struct packet_type *pt, struct net_device *orig_dev)
2067 {
2068         struct sock *sk;
2069         struct sockaddr_ll *sll;
2070         struct packet_sock *po;
2071         u8 *skb_head = skb->data;
2072         int skb_len = skb->len;
2073         unsigned int snaplen, res;
2074         bool is_drop_n_account = false;
2075
2076         if (skb->pkt_type == PACKET_LOOPBACK)
2077                 goto drop;
2078
2079         sk = pt->af_packet_priv;
2080         po = pkt_sk(sk);
2081
2082         if (!net_eq(dev_net(dev), sock_net(sk)))
2083                 goto drop;
2084
2085         skb->dev = dev;
2086
2087         if (dev->header_ops) {
2088                 /* The device has an explicit notion of ll header,
2089                  * exported to higher levels.
2090                  *
2091                  * Otherwise, the device hides details of its frame
2092                  * structure, so that corresponding packet head is
2093                  * never delivered to user.
2094                  */
2095                 if (sk->sk_type != SOCK_DGRAM)
2096                         skb_push(skb, skb->data - skb_mac_header(skb));
2097                 else if (skb->pkt_type == PACKET_OUTGOING) {
2098                         /* Special case: outgoing packets have ll header at head */
2099                         skb_pull(skb, skb_network_offset(skb));
2100                 }
2101         }
2102
2103         snaplen = skb->len;
2104
2105         res = run_filter(skb, sk, snaplen);
2106         if (!res)
2107                 goto drop_n_restore;
2108         if (snaplen > res)
2109                 snaplen = res;
2110
2111         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2112                 goto drop_n_acct;
2113
2114         if (skb_shared(skb)) {
2115                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2116                 if (nskb == NULL)
2117                         goto drop_n_acct;
2118
2119                 if (skb_head != skb->data) {
2120                         skb->data = skb_head;
2121                         skb->len = skb_len;
2122                 }
2123                 consume_skb(skb);
2124                 skb = nskb;
2125         }
2126
2127         sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2128
2129         sll = &PACKET_SKB_CB(skb)->sa.ll;
2130         sll->sll_hatype = dev->type;
2131         sll->sll_pkttype = skb->pkt_type;
2132         if (unlikely(po->origdev))
2133                 sll->sll_ifindex = orig_dev->ifindex;
2134         else
2135                 sll->sll_ifindex = dev->ifindex;
2136
2137         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2138
2139         /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2140          * Use their space for storing the original skb length.
2141          */
2142         PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2143
2144         if (pskb_trim(skb, snaplen))
2145                 goto drop_n_acct;
2146
2147         skb_set_owner_r(skb, sk);
2148         skb->dev = NULL;
2149         skb_dst_drop(skb);
2150
2151         /* drop conntrack reference */
2152         nf_reset(skb);
2153
2154         spin_lock(&sk->sk_receive_queue.lock);
2155         po->stats.stats1.tp_packets++;
2156         sock_skb_set_dropcount(sk, skb);
2157         __skb_queue_tail(&sk->sk_receive_queue, skb);
2158         spin_unlock(&sk->sk_receive_queue.lock);
2159         sk->sk_data_ready(sk);
2160         return 0;
2161
2162 drop_n_acct:
2163         is_drop_n_account = true;
2164         spin_lock(&sk->sk_receive_queue.lock);
2165         po->stats.stats1.tp_drops++;
2166         atomic_inc(&sk->sk_drops);
2167         spin_unlock(&sk->sk_receive_queue.lock);
2168
2169 drop_n_restore:
2170         if (skb_head != skb->data && skb_shared(skb)) {
2171                 skb->data = skb_head;
2172                 skb->len = skb_len;
2173         }
2174 drop:
2175         if (!is_drop_n_account)
2176                 consume_skb(skb);
2177         else
2178                 kfree_skb(skb);
2179         return 0;
2180 }
2181
2182 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2183                        struct packet_type *pt, struct net_device *orig_dev)
2184 {
2185         struct sock *sk;
2186         struct packet_sock *po;
2187         struct sockaddr_ll *sll;
2188         union tpacket_uhdr h;
2189         u8 *skb_head = skb->data;
2190         int skb_len = skb->len;
2191         unsigned int snaplen, res;
2192         unsigned long status = TP_STATUS_USER;
2193         unsigned short macoff, netoff, hdrlen;
2194         struct sk_buff *copy_skb = NULL;
2195         struct timespec ts;
2196         __u32 ts_status;
2197         bool is_drop_n_account = false;
2198         bool do_vnet = false;
2199
2200         /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2201          * We may add members to them until current aligned size without forcing
2202          * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2203          */
2204         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2205         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2206
2207         if (skb->pkt_type == PACKET_LOOPBACK)
2208                 goto drop;
2209
2210         sk = pt->af_packet_priv;
2211         po = pkt_sk(sk);
2212
2213         if (!net_eq(dev_net(dev), sock_net(sk)))
2214                 goto drop;
2215
2216         if (dev->header_ops) {
2217                 if (sk->sk_type != SOCK_DGRAM)
2218                         skb_push(skb, skb->data - skb_mac_header(skb));
2219                 else if (skb->pkt_type == PACKET_OUTGOING) {
2220                         /* Special case: outgoing packets have ll header at head */
2221                         skb_pull(skb, skb_network_offset(skb));
2222                 }
2223         }
2224
2225         snaplen = skb->len;
2226
2227         res = run_filter(skb, sk, snaplen);
2228         if (!res)
2229                 goto drop_n_restore;
2230
2231         if (skb->ip_summed == CHECKSUM_PARTIAL)
2232                 status |= TP_STATUS_CSUMNOTREADY;
2233         else if (skb->pkt_type != PACKET_OUTGOING &&
2234                  (skb->ip_summed == CHECKSUM_COMPLETE ||
2235                   skb_csum_unnecessary(skb)))
2236                 status |= TP_STATUS_CSUM_VALID;
2237
2238         if (snaplen > res)
2239                 snaplen = res;
2240
2241         if (sk->sk_type == SOCK_DGRAM) {
2242                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2243                                   po->tp_reserve;
2244         } else {
2245                 unsigned int maclen = skb_network_offset(skb);
2246                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2247                                        (maclen < 16 ? 16 : maclen)) +
2248                                        po->tp_reserve;
2249                 if (po->has_vnet_hdr) {
2250                         netoff += sizeof(struct virtio_net_hdr);
2251                         do_vnet = true;
2252                 }
2253                 macoff = netoff - maclen;
2254         }
2255         if (po->tp_version <= TPACKET_V2) {
2256                 if (macoff + snaplen > po->rx_ring.frame_size) {
2257                         if (po->copy_thresh &&
2258                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2259                                 if (skb_shared(skb)) {
2260                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
2261                                 } else {
2262                                         copy_skb = skb_get(skb);
2263                                         skb_head = skb->data;
2264                                 }
2265                                 if (copy_skb)
2266                                         skb_set_owner_r(copy_skb, sk);
2267                         }
2268                         snaplen = po->rx_ring.frame_size - macoff;
2269                         if ((int)snaplen < 0) {
2270                                 snaplen = 0;
2271                                 do_vnet = false;
2272                         }
2273                 }
2274         } else if (unlikely(macoff + snaplen >
2275                             GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2276                 u32 nval;
2277
2278                 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2279                 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2280                             snaplen, nval, macoff);
2281                 snaplen = nval;
2282                 if (unlikely((int)snaplen < 0)) {
2283                         snaplen = 0;
2284                         macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2285                         do_vnet = false;
2286                 }
2287         }
2288         spin_lock(&sk->sk_receive_queue.lock);
2289         h.raw = packet_current_rx_frame(po, skb,
2290                                         TP_STATUS_KERNEL, (macoff+snaplen));
2291         if (!h.raw)
2292                 goto drop_n_account;
2293         if (po->tp_version <= TPACKET_V2) {
2294                 packet_increment_rx_head(po, &po->rx_ring);
2295         /*
2296          * LOSING will be reported till you read the stats,
2297          * because it's COR - Clear On Read.
2298          * Anyways, moving it for V1/V2 only as V3 doesn't need this
2299          * at packet level.
2300          */
2301                 if (po->stats.stats1.tp_drops)
2302                         status |= TP_STATUS_LOSING;
2303         }
2304         po->stats.stats1.tp_packets++;
2305         if (copy_skb) {
2306                 status |= TP_STATUS_COPY;
2307                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2308         }
2309         spin_unlock(&sk->sk_receive_queue.lock);
2310
2311         if (do_vnet) {
2312                 if (virtio_net_hdr_from_skb(skb, h.raw + macoff -
2313                                             sizeof(struct virtio_net_hdr),
2314                                             vio_le(), true)) {
2315                         spin_lock(&sk->sk_receive_queue.lock);
2316                         goto drop_n_account;
2317                 }
2318         }
2319
2320         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2321
2322         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2323                 getnstimeofday(&ts);
2324
2325         status |= ts_status;
2326
2327         switch (po->tp_version) {
2328         case TPACKET_V1:
2329                 h.h1->tp_len = skb->len;
2330                 h.h1->tp_snaplen = snaplen;
2331                 h.h1->tp_mac = macoff;
2332                 h.h1->tp_net = netoff;
2333                 h.h1->tp_sec = ts.tv_sec;
2334                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2335                 hdrlen = sizeof(*h.h1);
2336                 break;
2337         case TPACKET_V2:
2338                 h.h2->tp_len = skb->len;
2339                 h.h2->tp_snaplen = snaplen;
2340                 h.h2->tp_mac = macoff;
2341                 h.h2->tp_net = netoff;
2342                 h.h2->tp_sec = ts.tv_sec;
2343                 h.h2->tp_nsec = ts.tv_nsec;
2344                 if (skb_vlan_tag_present(skb)) {
2345                         h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2346                         h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2347                         status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2348                 } else {
2349                         h.h2->tp_vlan_tci = 0;
2350                         h.h2->tp_vlan_tpid = 0;
2351                 }
2352                 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2353                 hdrlen = sizeof(*h.h2);
2354                 break;
2355         case TPACKET_V3:
2356                 /* tp_nxt_offset,vlan are already populated above.
2357                  * So DONT clear those fields here
2358                  */
2359                 h.h3->tp_status |= status;
2360                 h.h3->tp_len = skb->len;
2361                 h.h3->tp_snaplen = snaplen;
2362                 h.h3->tp_mac = macoff;
2363                 h.h3->tp_net = netoff;
2364                 h.h3->tp_sec  = ts.tv_sec;
2365                 h.h3->tp_nsec = ts.tv_nsec;
2366                 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2367                 hdrlen = sizeof(*h.h3);
2368                 break;
2369         default:
2370                 BUG();
2371         }
2372
2373         sll = h.raw + TPACKET_ALIGN(hdrlen);
2374         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2375         sll->sll_family = AF_PACKET;
2376         sll->sll_hatype = dev->type;
2377         sll->sll_protocol = skb->protocol;
2378         sll->sll_pkttype = skb->pkt_type;
2379         if (unlikely(po->origdev))
2380                 sll->sll_ifindex = orig_dev->ifindex;
2381         else
2382                 sll->sll_ifindex = dev->ifindex;
2383
2384         smp_mb();
2385
2386 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2387         if (po->tp_version <= TPACKET_V2) {
2388                 u8 *start, *end;
2389
2390                 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2391                                         macoff + snaplen);
2392
2393                 for (start = h.raw; start < end; start += PAGE_SIZE)
2394                         flush_dcache_page(pgv_to_page(start));
2395         }
2396         smp_wmb();
2397 #endif
2398
2399         if (po->tp_version <= TPACKET_V2) {
2400                 __packet_set_status(po, h.raw, status);
2401                 sk->sk_data_ready(sk);
2402         } else {
2403                 prb_clear_blk_fill_status(&po->rx_ring);
2404         }
2405
2406 drop_n_restore:
2407         if (skb_head != skb->data && skb_shared(skb)) {
2408                 skb->data = skb_head;
2409                 skb->len = skb_len;
2410         }
2411 drop:
2412         if (!is_drop_n_account)
2413                 consume_skb(skb);
2414         else
2415                 kfree_skb(skb);
2416         return 0;
2417
2418 drop_n_account:
2419         is_drop_n_account = true;
2420         po->stats.stats1.tp_drops++;
2421         spin_unlock(&sk->sk_receive_queue.lock);
2422
2423         sk->sk_data_ready(sk);
2424         kfree_skb(copy_skb);
2425         goto drop_n_restore;
2426 }
2427
2428 static void tpacket_destruct_skb(struct sk_buff *skb)
2429 {
2430         struct packet_sock *po = pkt_sk(skb->sk);
2431
2432         if (likely(po->tx_ring.pg_vec)) {
2433                 void *ph;
2434                 __u32 ts;
2435
2436                 ph = skb_shinfo(skb)->destructor_arg;
2437                 packet_dec_pending(&po->tx_ring);
2438
2439                 ts = __packet_set_timestamp(po, ph, skb);
2440                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2441         }
2442
2443         sock_wfree(skb);
2444 }
2445
2446 static void tpacket_set_protocol(const struct net_device *dev,
2447                                  struct sk_buff *skb)
2448 {
2449         if (dev->type == ARPHRD_ETHER) {
2450                 skb_reset_mac_header(skb);
2451                 skb->protocol = eth_hdr(skb)->h_proto;
2452         }
2453 }
2454
2455 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2456 {
2457         if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2458             (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2459              __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2460               __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2461                 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2462                          __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2463                         __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2464
2465         if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2466                 return -EINVAL;
2467
2468         return 0;
2469 }
2470
2471 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2472                                  struct virtio_net_hdr *vnet_hdr)
2473 {
2474         if (*len < sizeof(*vnet_hdr))
2475                 return -EINVAL;
2476         *len -= sizeof(*vnet_hdr);
2477
2478         if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2479                 return -EFAULT;
2480
2481         return __packet_snd_vnet_parse(vnet_hdr, *len);
2482 }
2483
2484 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2485                 void *frame, struct net_device *dev, void *data, int tp_len,
2486                 __be16 proto, unsigned char *addr, int hlen, int copylen,
2487                 const struct sockcm_cookie *sockc)
2488 {
2489         union tpacket_uhdr ph;
2490         int to_write, offset, len, nr_frags, len_max;
2491         struct socket *sock = po->sk.sk_socket;
2492         struct page *page;
2493         int err;
2494
2495         ph.raw = frame;
2496
2497         skb->protocol = proto;
2498         skb->dev = dev;
2499         skb->priority = po->sk.sk_priority;
2500         skb->mark = po->sk.sk_mark;
2501         sock_tx_timestamp(&po->sk, sockc->tsflags, &skb_shinfo(skb)->tx_flags);
2502         skb_shinfo(skb)->destructor_arg = ph.raw;
2503
2504         skb_reserve(skb, hlen);
2505         skb_reset_network_header(skb);
2506
2507         to_write = tp_len;
2508
2509         if (sock->type == SOCK_DGRAM) {
2510                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2511                                 NULL, tp_len);
2512                 if (unlikely(err < 0))
2513                         return -EINVAL;
2514         } else if (copylen) {
2515                 int hdrlen = min_t(int, copylen, tp_len);
2516
2517                 skb_push(skb, dev->hard_header_len);
2518                 skb_put(skb, copylen - dev->hard_header_len);
2519                 err = skb_store_bits(skb, 0, data, hdrlen);
2520                 if (unlikely(err))
2521                         return err;
2522                 if (!dev_validate_header(dev, skb->data, hdrlen))
2523                         return -EINVAL;
2524                 if (!skb->protocol)
2525                         tpacket_set_protocol(dev, skb);
2526
2527                 data += hdrlen;
2528                 to_write -= hdrlen;
2529         }
2530
2531         offset = offset_in_page(data);
2532         len_max = PAGE_SIZE - offset;
2533         len = ((to_write > len_max) ? len_max : to_write);
2534
2535         skb->data_len = to_write;
2536         skb->len += to_write;
2537         skb->truesize += to_write;
2538         refcount_add(to_write, &po->sk.sk_wmem_alloc);
2539
2540         while (likely(to_write)) {
2541                 nr_frags = skb_shinfo(skb)->nr_frags;
2542
2543                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2544                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2545                                MAX_SKB_FRAGS);
2546                         return -EFAULT;
2547                 }
2548
2549                 page = pgv_to_page(data);
2550                 data += len;
2551                 flush_dcache_page(page);
2552                 get_page(page);
2553                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2554                 to_write -= len;
2555                 offset = 0;
2556                 len_max = PAGE_SIZE;
2557                 len = ((to_write > len_max) ? len_max : to_write);
2558         }
2559
2560         skb_probe_transport_header(skb, 0);
2561
2562         return tp_len;
2563 }
2564
2565 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2566                                 int size_max, void **data)
2567 {
2568         union tpacket_uhdr ph;
2569         int tp_len, off;
2570
2571         ph.raw = frame;
2572
2573         switch (po->tp_version) {
2574         case TPACKET_V3:
2575                 if (ph.h3->tp_next_offset != 0) {
2576                         pr_warn_once("variable sized slot not supported");
2577                         return -EINVAL;
2578                 }
2579                 tp_len = ph.h3->tp_len;
2580                 break;
2581         case TPACKET_V2:
2582                 tp_len = ph.h2->tp_len;
2583                 break;
2584         default:
2585                 tp_len = ph.h1->tp_len;
2586                 break;
2587         }
2588         if (unlikely(tp_len > size_max)) {
2589                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2590                 return -EMSGSIZE;
2591         }
2592
2593         if (unlikely(po->tp_tx_has_off)) {
2594                 int off_min, off_max;
2595
2596                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2597                 off_max = po->tx_ring.frame_size - tp_len;
2598                 if (po->sk.sk_type == SOCK_DGRAM) {
2599                         switch (po->tp_version) {
2600                         case TPACKET_V3:
2601                                 off = ph.h3->tp_net;
2602                                 break;
2603                         case TPACKET_V2:
2604                                 off = ph.h2->tp_net;
2605                                 break;
2606                         default:
2607                                 off = ph.h1->tp_net;
2608                                 break;
2609                         }
2610                 } else {
2611                         switch (po->tp_version) {
2612                         case TPACKET_V3:
2613                                 off = ph.h3->tp_mac;
2614                                 break;
2615                         case TPACKET_V2:
2616                                 off = ph.h2->tp_mac;
2617                                 break;
2618                         default:
2619                                 off = ph.h1->tp_mac;
2620                                 break;
2621                         }
2622                 }
2623                 if (unlikely((off < off_min) || (off_max < off)))
2624                         return -EINVAL;
2625         } else {
2626                 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2627         }
2628
2629         *data = frame + off;
2630         return tp_len;
2631 }
2632
2633 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2634 {
2635         struct sk_buff *skb;
2636         struct net_device *dev;
2637         struct virtio_net_hdr *vnet_hdr = NULL;
2638         struct sockcm_cookie sockc;
2639         __be16 proto;
2640         int err, reserve = 0;
2641         void *ph;
2642         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2643         bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2644         int tp_len, size_max;
2645         unsigned char *addr;
2646         void *data;
2647         int len_sum = 0;
2648         int status = TP_STATUS_AVAILABLE;
2649         int hlen, tlen, copylen = 0;
2650
2651         mutex_lock(&po->pg_vec_lock);
2652
2653         if (likely(saddr == NULL)) {
2654                 dev     = packet_cached_dev_get(po);
2655                 proto   = po->num;
2656                 addr    = NULL;
2657         } else {
2658                 err = -EINVAL;
2659                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2660                         goto out;
2661                 if (msg->msg_namelen < (saddr->sll_halen
2662                                         + offsetof(struct sockaddr_ll,
2663                                                 sll_addr)))
2664                         goto out;
2665                 proto   = saddr->sll_protocol;
2666                 addr    = saddr->sll_addr;
2667                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2668         }
2669
2670         err = -ENXIO;
2671         if (unlikely(dev == NULL))
2672                 goto out;
2673         err = -ENETDOWN;
2674         if (unlikely(!(dev->flags & IFF_UP)))
2675                 goto out_put;
2676
2677         sockc.tsflags = po->sk.sk_tsflags;
2678         if (msg->msg_controllen) {
2679                 err = sock_cmsg_send(&po->sk, msg, &sockc);
2680                 if (unlikely(err))
2681                         goto out_put;
2682         }
2683
2684         if (po->sk.sk_socket->type == SOCK_RAW)
2685                 reserve = dev->hard_header_len;
2686         size_max = po->tx_ring.frame_size
2687                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2688
2689         if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2690                 size_max = dev->mtu + reserve + VLAN_HLEN;
2691
2692         do {
2693                 ph = packet_current_frame(po, &po->tx_ring,
2694                                           TP_STATUS_SEND_REQUEST);
2695                 if (unlikely(ph == NULL)) {
2696                         if (need_wait && need_resched())
2697                                 schedule();
2698                         continue;
2699                 }
2700
2701                 skb = NULL;
2702                 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2703                 if (tp_len < 0)
2704                         goto tpacket_error;
2705
2706                 status = TP_STATUS_SEND_REQUEST;
2707                 hlen = LL_RESERVED_SPACE(dev);
2708                 tlen = dev->needed_tailroom;
2709                 if (po->has_vnet_hdr) {
2710                         vnet_hdr = data;
2711                         data += sizeof(*vnet_hdr);
2712                         tp_len -= sizeof(*vnet_hdr);
2713                         if (tp_len < 0 ||
2714                             __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2715                                 tp_len = -EINVAL;
2716                                 goto tpacket_error;
2717                         }
2718                         copylen = __virtio16_to_cpu(vio_le(),
2719                                                     vnet_hdr->hdr_len);
2720                 }
2721                 copylen = max_t(int, copylen, dev->hard_header_len);
2722                 skb = sock_alloc_send_skb(&po->sk,
2723                                 hlen + tlen + sizeof(struct sockaddr_ll) +
2724                                 (copylen - dev->hard_header_len),
2725                                 !need_wait, &err);
2726
2727                 if (unlikely(skb == NULL)) {
2728                         /* we assume the socket was initially writeable ... */
2729                         if (likely(len_sum > 0))
2730                                 err = len_sum;
2731                         goto out_status;
2732                 }
2733                 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2734                                           addr, hlen, copylen, &sockc);
2735                 if (likely(tp_len >= 0) &&
2736                     tp_len > dev->mtu + reserve &&
2737                     !po->has_vnet_hdr &&
2738                     !packet_extra_vlan_len_allowed(dev, skb))
2739                         tp_len = -EMSGSIZE;
2740
2741                 if (unlikely(tp_len < 0)) {
2742 tpacket_error:
2743                         if (po->tp_loss) {
2744                                 __packet_set_status(po, ph,
2745                                                 TP_STATUS_AVAILABLE);
2746                                 packet_increment_head(&po->tx_ring);
2747                                 kfree_skb(skb);
2748                                 continue;
2749                         } else {
2750                                 status = TP_STATUS_WRONG_FORMAT;
2751                                 err = tp_len;
2752                                 goto out_status;
2753                         }
2754                 }
2755
2756                 if (po->has_vnet_hdr && virtio_net_hdr_to_skb(skb, vnet_hdr,
2757                                                               vio_le())) {
2758                         tp_len = -EINVAL;
2759                         goto tpacket_error;
2760                 }
2761
2762                 skb->destructor = tpacket_destruct_skb;
2763                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2764                 packet_inc_pending(&po->tx_ring);
2765
2766                 status = TP_STATUS_SEND_REQUEST;
2767                 err = po->xmit(skb);
2768                 if (unlikely(err > 0)) {
2769                         err = net_xmit_errno(err);
2770                         if (err && __packet_get_status(po, ph) ==
2771                                    TP_STATUS_AVAILABLE) {
2772                                 /* skb was destructed already */
2773                                 skb = NULL;
2774                                 goto out_status;
2775                         }
2776                         /*
2777                          * skb was dropped but not destructed yet;
2778                          * let's treat it like congestion or err < 0
2779                          */
2780                         err = 0;
2781                 }
2782                 packet_increment_head(&po->tx_ring);
2783                 len_sum += tp_len;
2784         } while (likely((ph != NULL) ||
2785                 /* Note: packet_read_pending() might be slow if we have
2786                  * to call it as it's per_cpu variable, but in fast-path
2787                  * we already short-circuit the loop with the first
2788                  * condition, and luckily don't have to go that path
2789                  * anyway.
2790                  */
2791                  (need_wait && packet_read_pending(&po->tx_ring))));
2792
2793         err = len_sum;
2794         goto out_put;
2795
2796 out_status:
2797         __packet_set_status(po, ph, status);
2798         kfree_skb(skb);
2799 out_put:
2800         dev_put(dev);
2801 out:
2802         mutex_unlock(&po->pg_vec_lock);
2803         return err;
2804 }
2805
2806 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2807                                         size_t reserve, size_t len,
2808                                         size_t linear, int noblock,
2809                                         int *err)
2810 {
2811         struct sk_buff *skb;
2812
2813         /* Under a page?  Don't bother with paged skb. */
2814         if (prepad + len < PAGE_SIZE || !linear)
2815                 linear = len;
2816
2817         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2818                                    err, 0);
2819         if (!skb)
2820                 return NULL;
2821
2822         skb_reserve(skb, reserve);
2823         skb_put(skb, linear);
2824         skb->data_len = len - linear;
2825         skb->len += len - linear;
2826
2827         return skb;
2828 }
2829
2830 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2831 {
2832         struct sock *sk = sock->sk;
2833         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2834         struct sk_buff *skb;
2835         struct net_device *dev;
2836         __be16 proto;
2837         unsigned char *addr;
2838         int err, reserve = 0;
2839         struct sockcm_cookie sockc;
2840         struct virtio_net_hdr vnet_hdr = { 0 };
2841         int offset = 0;
2842         struct packet_sock *po = pkt_sk(sk);
2843         int hlen, tlen, linear;
2844         int extra_len = 0;
2845
2846         /*
2847          *      Get and verify the address.
2848          */
2849
2850         if (likely(saddr == NULL)) {
2851                 dev     = packet_cached_dev_get(po);
2852                 proto   = po->num;
2853                 addr    = NULL;
2854         } else {
2855                 err = -EINVAL;
2856                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2857                         goto out;
2858                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2859                         goto out;
2860                 proto   = saddr->sll_protocol;
2861                 addr    = saddr->sll_addr;
2862                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2863         }
2864
2865         err = -ENXIO;
2866         if (unlikely(dev == NULL))
2867                 goto out_unlock;
2868         err = -ENETDOWN;
2869         if (unlikely(!(dev->flags & IFF_UP)))
2870                 goto out_unlock;
2871
2872         sockc.tsflags = sk->sk_tsflags;
2873         sockc.mark = sk->sk_mark;
2874         if (msg->msg_controllen) {
2875                 err = sock_cmsg_send(sk, msg, &sockc);
2876                 if (unlikely(err))
2877                         goto out_unlock;
2878         }
2879
2880         if (sock->type == SOCK_RAW)
2881                 reserve = dev->hard_header_len;
2882         if (po->has_vnet_hdr) {
2883                 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2884                 if (err)
2885                         goto out_unlock;
2886         }
2887
2888         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2889                 if (!netif_supports_nofcs(dev)) {
2890                         err = -EPROTONOSUPPORT;
2891                         goto out_unlock;
2892                 }
2893                 extra_len = 4; /* We're doing our own CRC */
2894         }
2895
2896         err = -EMSGSIZE;
2897         if (!vnet_hdr.gso_type &&
2898             (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2899                 goto out_unlock;
2900
2901         err = -ENOBUFS;
2902         hlen = LL_RESERVED_SPACE(dev);
2903         tlen = dev->needed_tailroom;
2904         linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2905         linear = max(linear, min_t(int, len, dev->hard_header_len));
2906         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2907                                msg->msg_flags & MSG_DONTWAIT, &err);
2908         if (skb == NULL)
2909                 goto out_unlock;
2910
2911         skb_set_network_header(skb, reserve);
2912
2913         err = -EINVAL;
2914         if (sock->type == SOCK_DGRAM) {
2915                 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2916                 if (unlikely(offset < 0))
2917                         goto out_free;
2918         }
2919
2920         /* Returns -EFAULT on error */
2921         err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2922         if (err)
2923                 goto out_free;
2924
2925         if (sock->type == SOCK_RAW &&
2926             !dev_validate_header(dev, skb->data, len)) {
2927                 err = -EINVAL;
2928                 goto out_free;
2929         }
2930
2931         sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
2932
2933         if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2934             !packet_extra_vlan_len_allowed(dev, skb)) {
2935                 err = -EMSGSIZE;
2936                 goto out_free;
2937         }
2938
2939         skb->protocol = proto;
2940         skb->dev = dev;
2941         skb->priority = sk->sk_priority;
2942         skb->mark = sockc.mark;
2943
2944         if (po->has_vnet_hdr) {
2945                 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2946                 if (err)
2947                         goto out_free;
2948                 len += sizeof(vnet_hdr);
2949         }
2950
2951         skb_probe_transport_header(skb, reserve);
2952
2953         if (unlikely(extra_len == 4))
2954                 skb->no_fcs = 1;
2955
2956         err = po->xmit(skb);
2957         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2958                 goto out_unlock;
2959
2960         dev_put(dev);
2961
2962         return len;
2963
2964 out_free:
2965         kfree_skb(skb);
2966 out_unlock:
2967         if (dev)
2968                 dev_put(dev);
2969 out:
2970         return err;
2971 }
2972
2973 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2974 {
2975         struct sock *sk = sock->sk;
2976         struct packet_sock *po = pkt_sk(sk);
2977
2978         if (po->tx_ring.pg_vec)
2979                 return tpacket_snd(po, msg);
2980         else
2981                 return packet_snd(sock, msg, len);
2982 }
2983
2984 /*
2985  *      Close a PACKET socket. This is fairly simple. We immediately go
2986  *      to 'closed' state and remove our protocol entry in the device list.
2987  */
2988
2989 static int packet_release(struct socket *sock)
2990 {
2991         struct sock *sk = sock->sk;
2992         struct packet_sock *po;
2993         struct packet_fanout *f;
2994         struct net *net;
2995         union tpacket_req_u req_u;
2996
2997         if (!sk)
2998                 return 0;
2999
3000         net = sock_net(sk);
3001         po = pkt_sk(sk);
3002
3003         mutex_lock(&net->packet.sklist_lock);
3004         sk_del_node_init_rcu(sk);
3005         mutex_unlock(&net->packet.sklist_lock);
3006
3007         preempt_disable();
3008         sock_prot_inuse_add(net, sk->sk_prot, -1);
3009         preempt_enable();
3010
3011         spin_lock(&po->bind_lock);
3012         unregister_prot_hook(sk, false);
3013         packet_cached_dev_reset(po);
3014
3015         if (po->prot_hook.dev) {
3016                 dev_put(po->prot_hook.dev);
3017                 po->prot_hook.dev = NULL;
3018         }
3019         spin_unlock(&po->bind_lock);
3020
3021         packet_flush_mclist(sk);
3022
3023         if (po->rx_ring.pg_vec) {
3024                 memset(&req_u, 0, sizeof(req_u));
3025                 packet_set_ring(sk, &req_u, 1, 0);
3026         }
3027
3028         if (po->tx_ring.pg_vec) {
3029                 memset(&req_u, 0, sizeof(req_u));
3030                 packet_set_ring(sk, &req_u, 1, 1);
3031         }
3032
3033         f = fanout_release(sk);
3034
3035         synchronize_net();
3036
3037         if (f) {
3038                 fanout_release_data(f);
3039                 kfree(f);
3040         }
3041         /*
3042          *      Now the socket is dead. No more input will appear.
3043          */
3044         sock_orphan(sk);
3045         sock->sk = NULL;
3046
3047         /* Purge queues */
3048
3049         skb_queue_purge(&sk->sk_receive_queue);
3050         packet_free_pending(po);
3051         sk_refcnt_debug_release(sk);
3052
3053         sock_put(sk);
3054         return 0;
3055 }
3056
3057 /*
3058  *      Attach a packet hook.
3059  */
3060
3061 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3062                           __be16 proto)
3063 {
3064         struct packet_sock *po = pkt_sk(sk);
3065         struct net_device *dev_curr;
3066         __be16 proto_curr;
3067         bool need_rehook;
3068         struct net_device *dev = NULL;
3069         int ret = 0;
3070         bool unlisted = false;
3071
3072         if (po->fanout)
3073                 return -EINVAL;
3074
3075         lock_sock(sk);
3076         spin_lock(&po->bind_lock);
3077         rcu_read_lock();
3078
3079         if (name) {
3080                 dev = dev_get_by_name_rcu(sock_net(sk), name);
3081                 if (!dev) {
3082                         ret = -ENODEV;
3083                         goto out_unlock;
3084                 }
3085         } else if (ifindex) {
3086                 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3087                 if (!dev) {
3088                         ret = -ENODEV;
3089                         goto out_unlock;
3090                 }
3091         }
3092
3093         if (dev)
3094                 dev_hold(dev);
3095
3096         proto_curr = po->prot_hook.type;
3097         dev_curr = po->prot_hook.dev;
3098
3099         need_rehook = proto_curr != proto || dev_curr != dev;
3100
3101         if (need_rehook) {
3102                 if (po->running) {
3103                         rcu_read_unlock();
3104                         __unregister_prot_hook(sk, true);
3105                         rcu_read_lock();
3106                         dev_curr = po->prot_hook.dev;
3107                         if (dev)
3108                                 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3109                                                                  dev->ifindex);
3110                 }
3111
3112                 po->num = proto;
3113                 po->prot_hook.type = proto;
3114
3115                 if (unlikely(unlisted)) {
3116                         dev_put(dev);
3117                         po->prot_hook.dev = NULL;
3118                         po->ifindex = -1;
3119                         packet_cached_dev_reset(po);
3120                 } else {
3121                         po->prot_hook.dev = dev;
3122                         po->ifindex = dev ? dev->ifindex : 0;
3123                         packet_cached_dev_assign(po, dev);
3124                 }
3125         }
3126         if (dev_curr)
3127                 dev_put(dev_curr);
3128
3129         if (proto == 0 || !need_rehook)
3130                 goto out_unlock;
3131
3132         if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3133                 register_prot_hook(sk);
3134         } else {
3135                 sk->sk_err = ENETDOWN;
3136                 if (!sock_flag(sk, SOCK_DEAD))
3137                         sk->sk_error_report(sk);
3138         }
3139
3140 out_unlock:
3141         rcu_read_unlock();
3142         spin_unlock(&po->bind_lock);
3143         release_sock(sk);
3144         return ret;
3145 }
3146
3147 /*
3148  *      Bind a packet socket to a device
3149  */
3150
3151 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3152                             int addr_len)
3153 {
3154         struct sock *sk = sock->sk;
3155         char name[sizeof(uaddr->sa_data) + 1];
3156
3157         /*
3158          *      Check legality
3159          */
3160
3161         if (addr_len != sizeof(struct sockaddr))
3162                 return -EINVAL;
3163         /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3164          * zero-terminated.
3165          */
3166         memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3167         name[sizeof(uaddr->sa_data)] = 0;
3168
3169         return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3170 }
3171
3172 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3173 {
3174         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3175         struct sock *sk = sock->sk;
3176
3177         /*
3178          *      Check legality
3179          */
3180
3181         if (addr_len < sizeof(struct sockaddr_ll))
3182                 return -EINVAL;
3183         if (sll->sll_family != AF_PACKET)
3184                 return -EINVAL;
3185
3186         return packet_do_bind(sk, NULL, sll->sll_ifindex,
3187                               sll->sll_protocol ? : pkt_sk(sk)->num);
3188 }
3189
3190 static struct proto packet_proto = {
3191         .name     = "PACKET",
3192         .owner    = THIS_MODULE,
3193         .obj_size = sizeof(struct packet_sock),
3194 };
3195
3196 /*
3197  *      Create a packet of type SOCK_PACKET.
3198  */
3199
3200 static int packet_create(struct net *net, struct socket *sock, int protocol,
3201                          int kern)
3202 {
3203         struct sock *sk;
3204         struct packet_sock *po;
3205         __be16 proto = (__force __be16)protocol; /* weird, but documented */
3206         int err;
3207
3208         if (!ns_capable(net->user_ns, CAP_NET_RAW))
3209                 return -EPERM;
3210         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3211             sock->type != SOCK_PACKET)
3212                 return -ESOCKTNOSUPPORT;
3213
3214         sock->state = SS_UNCONNECTED;
3215
3216         err = -ENOBUFS;
3217         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3218         if (sk == NULL)
3219                 goto out;
3220
3221         sock->ops = &packet_ops;
3222         if (sock->type == SOCK_PACKET)
3223                 sock->ops = &packet_ops_spkt;
3224
3225         sock_init_data(sock, sk);
3226
3227         po = pkt_sk(sk);
3228         sk->sk_family = PF_PACKET;
3229         po->num = proto;
3230         po->xmit = dev_queue_xmit;
3231
3232         err = packet_alloc_pending(po);
3233         if (err)
3234                 goto out2;
3235
3236         packet_cached_dev_reset(po);
3237
3238         sk->sk_destruct = packet_sock_destruct;
3239         sk_refcnt_debug_inc(sk);
3240
3241         /*
3242          *      Attach a protocol block
3243          */
3244
3245         spin_lock_init(&po->bind_lock);
3246         mutex_init(&po->pg_vec_lock);
3247         po->rollover = NULL;
3248         po->prot_hook.func = packet_rcv;
3249
3250         if (sock->type == SOCK_PACKET)
3251                 po->prot_hook.func = packet_rcv_spkt;
3252
3253         po->prot_hook.af_packet_priv = sk;
3254
3255         if (proto) {
3256                 po->prot_hook.type = proto;
3257                 register_prot_hook(sk);
3258         }
3259
3260         mutex_lock(&net->packet.sklist_lock);
3261         sk_add_node_rcu(sk, &net->packet.sklist);
3262         mutex_unlock(&net->packet.sklist_lock);
3263
3264         preempt_disable();
3265         sock_prot_inuse_add(net, &packet_proto, 1);
3266         preempt_enable();
3267
3268         return 0;
3269 out2:
3270         sk_free(sk);
3271 out:
3272         return err;
3273 }
3274
3275 /*
3276  *      Pull a packet from our receive queue and hand it to the user.
3277  *      If necessary we block.
3278  */
3279
3280 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3281                           int flags)
3282 {
3283         struct sock *sk = sock->sk;
3284         struct sk_buff *skb;
3285         int copied, err;
3286         int vnet_hdr_len = 0;
3287         unsigned int origlen = 0;
3288
3289         err = -EINVAL;
3290         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3291                 goto out;
3292
3293 #if 0
3294         /* What error should we return now? EUNATTACH? */
3295         if (pkt_sk(sk)->ifindex < 0)
3296                 return -ENODEV;
3297 #endif
3298
3299         if (flags & MSG_ERRQUEUE) {
3300                 err = sock_recv_errqueue(sk, msg, len,
3301                                          SOL_PACKET, PACKET_TX_TIMESTAMP);
3302                 goto out;
3303         }
3304
3305         /*
3306          *      Call the generic datagram receiver. This handles all sorts
3307          *      of horrible races and re-entrancy so we can forget about it
3308          *      in the protocol layers.
3309          *
3310          *      Now it will return ENETDOWN, if device have just gone down,
3311          *      but then it will block.
3312          */
3313
3314         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3315
3316         /*
3317          *      An error occurred so return it. Because skb_recv_datagram()
3318          *      handles the blocking we don't see and worry about blocking
3319          *      retries.
3320          */
3321
3322         if (skb == NULL)
3323                 goto out;
3324
3325         if (pkt_sk(sk)->pressure)
3326                 packet_rcv_has_room(pkt_sk(sk), NULL);
3327
3328         if (pkt_sk(sk)->has_vnet_hdr) {
3329                 err = packet_rcv_vnet(msg, skb, &len);
3330                 if (err)
3331                         goto out_free;
3332                 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3333         }
3334
3335         /* You lose any data beyond the buffer you gave. If it worries
3336          * a user program they can ask the device for its MTU
3337          * anyway.
3338          */
3339         copied = skb->len;
3340         if (copied > len) {
3341                 copied = len;
3342                 msg->msg_flags |= MSG_TRUNC;
3343         }
3344
3345         err = skb_copy_datagram_msg(skb, 0, msg, copied);
3346         if (err)
3347                 goto out_free;
3348
3349         if (sock->type != SOCK_PACKET) {
3350                 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3351
3352                 /* Original length was stored in sockaddr_ll fields */
3353                 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3354                 sll->sll_family = AF_PACKET;
3355                 sll->sll_protocol = skb->protocol;
3356         }
3357
3358         sock_recv_ts_and_drops(msg, sk, skb);
3359
3360         if (msg->msg_name) {
3361                 /* If the address length field is there to be filled
3362                  * in, we fill it in now.
3363                  */
3364                 if (sock->type == SOCK_PACKET) {
3365                         __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3366                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
3367                 } else {
3368                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3369
3370                         msg->msg_namelen = sll->sll_halen +
3371                                 offsetof(struct sockaddr_ll, sll_addr);
3372                 }
3373                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3374                        msg->msg_namelen);
3375         }
3376
3377         if (pkt_sk(sk)->auxdata) {
3378                 struct tpacket_auxdata aux;
3379
3380                 aux.tp_status = TP_STATUS_USER;
3381                 if (skb->ip_summed == CHECKSUM_PARTIAL)
3382                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3383                 else if (skb->pkt_type != PACKET_OUTGOING &&
3384                          (skb->ip_summed == CHECKSUM_COMPLETE ||
3385                           skb_csum_unnecessary(skb)))
3386                         aux.tp_status |= TP_STATUS_CSUM_VALID;
3387
3388                 aux.tp_len = origlen;
3389                 aux.tp_snaplen = skb->len;
3390                 aux.tp_mac = 0;
3391                 aux.tp_net = skb_network_offset(skb);
3392                 if (skb_vlan_tag_present(skb)) {
3393                         aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3394                         aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3395                         aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3396                 } else {
3397                         aux.tp_vlan_tci = 0;
3398                         aux.tp_vlan_tpid = 0;
3399                 }
3400                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3401         }
3402
3403         /*
3404          *      Free or return the buffer as appropriate. Again this
3405          *      hides all the races and re-entrancy issues from us.
3406          */
3407         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3408
3409 out_free:
3410         skb_free_datagram(sk, skb);
3411 out:
3412         return err;
3413 }
3414
3415 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3416                                int *uaddr_len, int peer)
3417 {
3418         struct net_device *dev;
3419         struct sock *sk = sock->sk;
3420
3421         if (peer)
3422                 return -EOPNOTSUPP;
3423
3424         uaddr->sa_family = AF_PACKET;
3425         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3426         rcu_read_lock();
3427         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3428         if (dev)
3429                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3430         rcu_read_unlock();
3431         *uaddr_len = sizeof(*uaddr);
3432
3433         return 0;
3434 }
3435
3436 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3437                           int *uaddr_len, int peer)
3438 {
3439         struct net_device *dev;
3440         struct sock *sk = sock->sk;
3441         struct packet_sock *po = pkt_sk(sk);
3442         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3443
3444         if (peer)
3445                 return -EOPNOTSUPP;
3446
3447         sll->sll_family = AF_PACKET;
3448         sll->sll_ifindex = po->ifindex;
3449         sll->sll_protocol = po->num;
3450         sll->sll_pkttype = 0;
3451         rcu_read_lock();
3452         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3453         if (dev) {
3454                 sll->sll_hatype = dev->type;
3455                 sll->sll_halen = dev->addr_len;
3456                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3457         } else {
3458                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
3459                 sll->sll_halen = 0;
3460         }
3461         rcu_read_unlock();
3462         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3463
3464         return 0;
3465 }
3466
3467 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3468                          int what)
3469 {
3470         switch (i->type) {
3471         case PACKET_MR_MULTICAST:
3472                 if (i->alen != dev->addr_len)
3473                         return -EINVAL;
3474                 if (what > 0)
3475                         return dev_mc_add(dev, i->addr);
3476                 else
3477                         return dev_mc_del(dev, i->addr);
3478                 break;
3479         case PACKET_MR_PROMISC:
3480                 return dev_set_promiscuity(dev, what);
3481         case PACKET_MR_ALLMULTI:
3482                 return dev_set_allmulti(dev, what);
3483         case PACKET_MR_UNICAST:
3484                 if (i->alen != dev->addr_len)
3485                         return -EINVAL;
3486                 if (what > 0)
3487                         return dev_uc_add(dev, i->addr);
3488                 else
3489                         return dev_uc_del(dev, i->addr);
3490                 break;
3491         default:
3492                 break;
3493         }
3494         return 0;
3495 }
3496
3497 static void packet_dev_mclist_delete(struct net_device *dev,
3498                                      struct packet_mclist **mlp)
3499 {
3500         struct packet_mclist *ml;
3501
3502         while ((ml = *mlp) != NULL) {
3503                 if (ml->ifindex == dev->ifindex) {
3504                         packet_dev_mc(dev, ml, -1);
3505                         *mlp = ml->next;
3506                         kfree(ml);
3507                 } else
3508                         mlp = &ml->next;
3509         }
3510 }
3511
3512 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3513 {
3514         struct packet_sock *po = pkt_sk(sk);
3515         struct packet_mclist *ml, *i;
3516         struct net_device *dev;
3517         int err;
3518
3519         rtnl_lock();
3520
3521         err = -ENODEV;
3522         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3523         if (!dev)
3524                 goto done;
3525
3526         err = -EINVAL;
3527         if (mreq->mr_alen > dev->addr_len)
3528                 goto done;
3529
3530         err = -ENOBUFS;
3531         i = kmalloc(sizeof(*i), GFP_KERNEL);
3532         if (i == NULL)
3533                 goto done;
3534
3535         err = 0;
3536         for (ml = po->mclist; ml; ml = ml->next) {
3537                 if (ml->ifindex == mreq->mr_ifindex &&
3538                     ml->type == mreq->mr_type &&
3539                     ml->alen == mreq->mr_alen &&
3540                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3541                         ml->count++;
3542                         /* Free the new element ... */
3543                         kfree(i);
3544                         goto done;
3545                 }
3546         }
3547
3548         i->type = mreq->mr_type;
3549         i->ifindex = mreq->mr_ifindex;
3550         i->alen = mreq->mr_alen;
3551         memcpy(i->addr, mreq->mr_address, i->alen);
3552         memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3553         i->count = 1;
3554         i->next = po->mclist;
3555         po->mclist = i;
3556         err = packet_dev_mc(dev, i, 1);
3557         if (err) {
3558                 po->mclist = i->next;
3559                 kfree(i);
3560         }
3561
3562 done:
3563         rtnl_unlock();
3564         return err;
3565 }
3566
3567 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3568 {
3569         struct packet_mclist *ml, **mlp;
3570
3571         rtnl_lock();
3572
3573         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3574                 if (ml->ifindex == mreq->mr_ifindex &&
3575                     ml->type == mreq->mr_type &&
3576                     ml->alen == mreq->mr_alen &&
3577                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3578                         if (--ml->count == 0) {
3579                                 struct net_device *dev;
3580                                 *mlp = ml->next;
3581                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3582                                 if (dev)
3583                                         packet_dev_mc(dev, ml, -1);
3584                                 kfree(ml);
3585                         }
3586                         break;
3587                 }
3588         }
3589         rtnl_unlock();
3590         return 0;
3591 }
3592
3593 static void packet_flush_mclist(struct sock *sk)
3594 {
3595         struct packet_sock *po = pkt_sk(sk);
3596         struct packet_mclist *ml;
3597
3598         if (!po->mclist)
3599                 return;
3600
3601         rtnl_lock();
3602         while ((ml = po->mclist) != NULL) {
3603                 struct net_device *dev;
3604
3605                 po->mclist = ml->next;
3606                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3607                 if (dev != NULL)
3608                         packet_dev_mc(dev, ml, -1);
3609                 kfree(ml);
3610         }
3611         rtnl_unlock();
3612 }
3613
3614 static int
3615 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3616 {
3617         struct sock *sk = sock->sk;
3618         struct packet_sock *po = pkt_sk(sk);
3619         int ret;
3620
3621         if (level != SOL_PACKET)
3622                 return -ENOPROTOOPT;
3623
3624         switch (optname) {
3625         case PACKET_ADD_MEMBERSHIP:
3626         case PACKET_DROP_MEMBERSHIP:
3627         {
3628                 struct packet_mreq_max mreq;
3629                 int len = optlen;
3630                 memset(&mreq, 0, sizeof(mreq));
3631                 if (len < sizeof(struct packet_mreq))
3632                         return -EINVAL;
3633                 if (len > sizeof(mreq))
3634                         len = sizeof(mreq);
3635                 if (copy_from_user(&mreq, optval, len))
3636                         return -EFAULT;
3637                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3638                         return -EINVAL;
3639                 if (optname == PACKET_ADD_MEMBERSHIP)
3640                         ret = packet_mc_add(sk, &mreq);
3641                 else
3642                         ret = packet_mc_drop(sk, &mreq);
3643                 return ret;
3644         }
3645
3646         case PACKET_RX_RING:
3647         case PACKET_TX_RING:
3648         {
3649                 union tpacket_req_u req_u;
3650                 int len;
3651
3652                 switch (po->tp_version) {
3653                 case TPACKET_V1:
3654                 case TPACKET_V2:
3655                         len = sizeof(req_u.req);
3656                         break;
3657                 case TPACKET_V3:
3658                 default:
3659                         len = sizeof(req_u.req3);
3660                         break;
3661                 }
3662                 if (optlen < len)
3663                         return -EINVAL;
3664                 if (copy_from_user(&req_u.req, optval, len))
3665                         return -EFAULT;
3666                 return packet_set_ring(sk, &req_u, 0,
3667                         optname == PACKET_TX_RING);
3668         }
3669         case PACKET_COPY_THRESH:
3670         {
3671                 int val;
3672
3673                 if (optlen != sizeof(val))
3674                         return -EINVAL;
3675                 if (copy_from_user(&val, optval, sizeof(val)))
3676                         return -EFAULT;
3677
3678                 pkt_sk(sk)->copy_thresh = val;
3679                 return 0;
3680         }
3681         case PACKET_VERSION:
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                 switch (val) {
3690                 case TPACKET_V1:
3691                 case TPACKET_V2:
3692                 case TPACKET_V3:
3693                         break;
3694                 default:
3695                         return -EINVAL;
3696                 }
3697                 lock_sock(sk);
3698                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3699                         ret = -EBUSY;
3700                 } else {
3701                         po->tp_version = val;
3702                         ret = 0;
3703                 }
3704                 release_sock(sk);
3705                 return ret;
3706         }
3707         case PACKET_RESERVE:
3708         {
3709                 unsigned int val;
3710
3711                 if (optlen != sizeof(val))
3712                         return -EINVAL;
3713                 if (copy_from_user(&val, optval, sizeof(val)))
3714                         return -EFAULT;
3715                 if (val > INT_MAX)
3716                         return -EINVAL;
3717                 lock_sock(sk);
3718                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3719                         ret = -EBUSY;
3720                 } else {
3721                         po->tp_reserve = val;
3722                         ret = 0;
3723                 }
3724                 release_sock(sk);
3725                 return ret;
3726         }
3727         case PACKET_LOSS:
3728         {
3729                 unsigned int val;
3730
3731                 if (optlen != sizeof(val))
3732                         return -EINVAL;
3733                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3734                         return -EBUSY;
3735                 if (copy_from_user(&val, optval, sizeof(val)))
3736                         return -EFAULT;
3737                 po->tp_loss = !!val;
3738                 return 0;
3739         }
3740         case PACKET_AUXDATA:
3741         {
3742                 int val;
3743
3744                 if (optlen < sizeof(val))
3745                         return -EINVAL;
3746                 if (copy_from_user(&val, optval, sizeof(val)))
3747                         return -EFAULT;
3748
3749                 po->auxdata = !!val;
3750                 return 0;
3751         }
3752         case PACKET_ORIGDEV:
3753         {
3754                 int val;
3755
3756                 if (optlen < sizeof(val))
3757                         return -EINVAL;
3758                 if (copy_from_user(&val, optval, sizeof(val)))
3759                         return -EFAULT;
3760
3761                 po->origdev = !!val;
3762                 return 0;
3763         }
3764         case PACKET_VNET_HDR:
3765         {
3766                 int val;
3767
3768                 if (sock->type != SOCK_RAW)
3769                         return -EINVAL;
3770                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3771                         return -EBUSY;
3772                 if (optlen < sizeof(val))
3773                         return -EINVAL;
3774                 if (copy_from_user(&val, optval, sizeof(val)))
3775                         return -EFAULT;
3776
3777                 po->has_vnet_hdr = !!val;
3778                 return 0;
3779         }
3780         case PACKET_TIMESTAMP:
3781         {
3782                 int val;
3783
3784                 if (optlen != sizeof(val))
3785                         return -EINVAL;
3786                 if (copy_from_user(&val, optval, sizeof(val)))
3787                         return -EFAULT;
3788
3789                 po->tp_tstamp = val;
3790                 return 0;
3791         }
3792         case PACKET_FANOUT:
3793         {
3794                 int val;
3795
3796                 if (optlen != sizeof(val))
3797                         return -EINVAL;
3798                 if (copy_from_user(&val, optval, sizeof(val)))
3799                         return -EFAULT;
3800
3801                 return fanout_add(sk, val & 0xffff, val >> 16);
3802         }
3803         case PACKET_FANOUT_DATA:
3804         {
3805                 if (!po->fanout)
3806                         return -EINVAL;
3807
3808                 return fanout_set_data(po, optval, optlen);
3809         }
3810         case PACKET_TX_HAS_OFF:
3811         {
3812                 unsigned int val;
3813
3814                 if (optlen != sizeof(val))
3815                         return -EINVAL;
3816                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3817                         return -EBUSY;
3818                 if (copy_from_user(&val, optval, sizeof(val)))
3819                         return -EFAULT;
3820                 po->tp_tx_has_off = !!val;
3821                 return 0;
3822         }
3823         case PACKET_QDISC_BYPASS:
3824         {
3825                 int val;
3826
3827                 if (optlen != sizeof(val))
3828                         return -EINVAL;
3829                 if (copy_from_user(&val, optval, sizeof(val)))
3830                         return -EFAULT;
3831
3832                 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3833                 return 0;
3834         }
3835         default:
3836                 return -ENOPROTOOPT;
3837         }
3838 }
3839
3840 static int packet_getsockopt(struct socket *sock, int level, int optname,
3841                              char __user *optval, int __user *optlen)
3842 {
3843         int len;
3844         int val, lv = sizeof(val);
3845         struct sock *sk = sock->sk;
3846         struct packet_sock *po = pkt_sk(sk);
3847         void *data = &val;
3848         union tpacket_stats_u st;
3849         struct tpacket_rollover_stats rstats;
3850
3851         if (level != SOL_PACKET)
3852                 return -ENOPROTOOPT;
3853
3854         if (get_user(len, optlen))
3855                 return -EFAULT;
3856
3857         if (len < 0)
3858                 return -EINVAL;
3859
3860         switch (optname) {
3861         case PACKET_STATISTICS:
3862                 spin_lock_bh(&sk->sk_receive_queue.lock);
3863                 memcpy(&st, &po->stats, sizeof(st));
3864                 memset(&po->stats, 0, sizeof(po->stats));
3865                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3866
3867                 if (po->tp_version == TPACKET_V3) {
3868                         lv = sizeof(struct tpacket_stats_v3);
3869                         st.stats3.tp_packets += st.stats3.tp_drops;
3870                         data = &st.stats3;
3871                 } else {
3872                         lv = sizeof(struct tpacket_stats);
3873                         st.stats1.tp_packets += st.stats1.tp_drops;
3874                         data = &st.stats1;
3875                 }
3876
3877                 break;
3878         case PACKET_AUXDATA:
3879                 val = po->auxdata;
3880                 break;
3881         case PACKET_ORIGDEV:
3882                 val = po->origdev;
3883                 break;
3884         case PACKET_VNET_HDR:
3885                 val = po->has_vnet_hdr;
3886                 break;
3887         case PACKET_VERSION:
3888                 val = po->tp_version;
3889                 break;
3890         case PACKET_HDRLEN:
3891                 if (len > sizeof(int))
3892                         len = sizeof(int);
3893                 if (len < sizeof(int))
3894                         return -EINVAL;
3895                 if (copy_from_user(&val, optval, len))
3896                         return -EFAULT;
3897                 switch (val) {
3898                 case TPACKET_V1:
3899                         val = sizeof(struct tpacket_hdr);
3900                         break;
3901                 case TPACKET_V2:
3902                         val = sizeof(struct tpacket2_hdr);
3903                         break;
3904                 case TPACKET_V3:
3905                         val = sizeof(struct tpacket3_hdr);
3906                         break;
3907                 default:
3908                         return -EINVAL;
3909                 }
3910                 break;
3911         case PACKET_RESERVE:
3912                 val = po->tp_reserve;
3913                 break;
3914         case PACKET_LOSS:
3915                 val = po->tp_loss;
3916                 break;
3917         case PACKET_TIMESTAMP:
3918                 val = po->tp_tstamp;
3919                 break;
3920         case PACKET_FANOUT:
3921                 val = (po->fanout ?
3922                        ((u32)po->fanout->id |
3923                         ((u32)po->fanout->type << 16) |
3924                         ((u32)po->fanout->flags << 24)) :
3925                        0);
3926                 break;
3927         case PACKET_ROLLOVER_STATS:
3928                 if (!po->rollover)
3929                         return -EINVAL;
3930                 rstats.tp_all = atomic_long_read(&po->rollover->num);
3931                 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3932                 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3933                 data = &rstats;
3934                 lv = sizeof(rstats);
3935                 break;
3936         case PACKET_TX_HAS_OFF:
3937                 val = po->tp_tx_has_off;
3938                 break;
3939         case PACKET_QDISC_BYPASS:
3940                 val = packet_use_direct_xmit(po);
3941                 break;
3942         default:
3943                 return -ENOPROTOOPT;
3944         }
3945
3946         if (len > lv)
3947                 len = lv;
3948         if (put_user(len, optlen))
3949                 return -EFAULT;
3950         if (copy_to_user(optval, data, len))
3951                 return -EFAULT;
3952         return 0;
3953 }
3954
3955
3956 #ifdef CONFIG_COMPAT
3957 static int compat_packet_setsockopt(struct socket *sock, int level, int optname,
3958                                     char __user *optval, unsigned int optlen)
3959 {
3960         struct packet_sock *po = pkt_sk(sock->sk);
3961
3962         if (level != SOL_PACKET)
3963                 return -ENOPROTOOPT;
3964
3965         if (optname == PACKET_FANOUT_DATA &&
3966             po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) {
3967                 optval = (char __user *)get_compat_bpf_fprog(optval);
3968                 if (!optval)
3969                         return -EFAULT;
3970                 optlen = sizeof(struct sock_fprog);
3971         }
3972
3973         return packet_setsockopt(sock, level, optname, optval, optlen);
3974 }
3975 #endif
3976
3977 static int packet_notifier(struct notifier_block *this,
3978                            unsigned long msg, void *ptr)
3979 {
3980         struct sock *sk;
3981         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3982         struct net *net = dev_net(dev);
3983
3984         rcu_read_lock();
3985         sk_for_each_rcu(sk, &net->packet.sklist) {
3986                 struct packet_sock *po = pkt_sk(sk);
3987
3988                 switch (msg) {
3989                 case NETDEV_UNREGISTER:
3990                         if (po->mclist)
3991                                 packet_dev_mclist_delete(dev, &po->mclist);
3992                         /* fallthrough */
3993
3994                 case NETDEV_DOWN:
3995                         if (dev->ifindex == po->ifindex) {
3996                                 spin_lock(&po->bind_lock);
3997                                 if (po->running) {
3998                                         __unregister_prot_hook(sk, false);
3999                                         sk->sk_err = ENETDOWN;
4000                                         if (!sock_flag(sk, SOCK_DEAD))
4001                                                 sk->sk_error_report(sk);
4002                                 }
4003                                 if (msg == NETDEV_UNREGISTER) {
4004                                         packet_cached_dev_reset(po);
4005                                         po->ifindex = -1;
4006                                         if (po->prot_hook.dev)
4007                                                 dev_put(po->prot_hook.dev);
4008                                         po->prot_hook.dev = NULL;
4009                                 }
4010                                 spin_unlock(&po->bind_lock);
4011                         }
4012                         break;
4013                 case NETDEV_UP:
4014                         if (dev->ifindex == po->ifindex) {
4015                                 spin_lock(&po->bind_lock);
4016                                 if (po->num)
4017                                         register_prot_hook(sk);
4018                                 spin_unlock(&po->bind_lock);
4019                         }
4020                         break;
4021                 }
4022         }
4023         rcu_read_unlock();
4024         return NOTIFY_DONE;
4025 }
4026
4027
4028 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4029                         unsigned long arg)
4030 {
4031         struct sock *sk = sock->sk;
4032
4033         switch (cmd) {
4034         case SIOCOUTQ:
4035         {
4036                 int amount = sk_wmem_alloc_get(sk);
4037
4038                 return put_user(amount, (int __user *)arg);
4039         }
4040         case SIOCINQ:
4041         {
4042                 struct sk_buff *skb;
4043                 int amount = 0;
4044
4045                 spin_lock_bh(&sk->sk_receive_queue.lock);
4046                 skb = skb_peek(&sk->sk_receive_queue);
4047                 if (skb)
4048                         amount = skb->len;
4049                 spin_unlock_bh(&sk->sk_receive_queue.lock);
4050                 return put_user(amount, (int __user *)arg);
4051         }
4052         case SIOCGSTAMP:
4053                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
4054         case SIOCGSTAMPNS:
4055                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
4056
4057 #ifdef CONFIG_INET
4058         case SIOCADDRT:
4059         case SIOCDELRT:
4060         case SIOCDARP:
4061         case SIOCGARP:
4062         case SIOCSARP:
4063         case SIOCGIFADDR:
4064         case SIOCSIFADDR:
4065         case SIOCGIFBRDADDR:
4066         case SIOCSIFBRDADDR:
4067         case SIOCGIFNETMASK:
4068         case SIOCSIFNETMASK:
4069         case SIOCGIFDSTADDR:
4070         case SIOCSIFDSTADDR:
4071         case SIOCSIFFLAGS:
4072                 return inet_dgram_ops.ioctl(sock, cmd, arg);
4073 #endif
4074
4075         default:
4076                 return -ENOIOCTLCMD;
4077         }
4078         return 0;
4079 }
4080
4081 static unsigned int packet_poll(struct file *file, struct socket *sock,
4082                                 poll_table *wait)
4083 {
4084         struct sock *sk = sock->sk;
4085         struct packet_sock *po = pkt_sk(sk);
4086         unsigned int mask = datagram_poll(file, sock, wait);
4087
4088         spin_lock_bh(&sk->sk_receive_queue.lock);
4089         if (po->rx_ring.pg_vec) {
4090                 if (!packet_previous_rx_frame(po, &po->rx_ring,
4091                         TP_STATUS_KERNEL))
4092                         mask |= POLLIN | POLLRDNORM;
4093         }
4094         if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
4095                 po->pressure = 0;
4096         spin_unlock_bh(&sk->sk_receive_queue.lock);
4097         spin_lock_bh(&sk->sk_write_queue.lock);
4098         if (po->tx_ring.pg_vec) {
4099                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4100                         mask |= POLLOUT | POLLWRNORM;
4101         }
4102         spin_unlock_bh(&sk->sk_write_queue.lock);
4103         return mask;
4104 }
4105
4106
4107 /* Dirty? Well, I still did not learn better way to account
4108  * for user mmaps.
4109  */
4110
4111 static void packet_mm_open(struct vm_area_struct *vma)
4112 {
4113         struct file *file = vma->vm_file;
4114         struct socket *sock = file->private_data;
4115         struct sock *sk = sock->sk;
4116
4117         if (sk)
4118                 atomic_inc(&pkt_sk(sk)->mapped);
4119 }
4120
4121 static void packet_mm_close(struct vm_area_struct *vma)
4122 {
4123         struct file *file = vma->vm_file;
4124         struct socket *sock = file->private_data;
4125         struct sock *sk = sock->sk;
4126
4127         if (sk)
4128                 atomic_dec(&pkt_sk(sk)->mapped);
4129 }
4130
4131 static const struct vm_operations_struct packet_mmap_ops = {
4132         .open   =       packet_mm_open,
4133         .close  =       packet_mm_close,
4134 };
4135
4136 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4137                         unsigned int len)
4138 {
4139         int i;
4140
4141         for (i = 0; i < len; i++) {
4142                 if (likely(pg_vec[i].buffer)) {
4143                         if (is_vmalloc_addr(pg_vec[i].buffer))
4144                                 vfree(pg_vec[i].buffer);
4145                         else
4146                                 free_pages((unsigned long)pg_vec[i].buffer,
4147                                            order);
4148                         pg_vec[i].buffer = NULL;
4149                 }
4150         }
4151         kfree(pg_vec);
4152 }
4153
4154 static char *alloc_one_pg_vec_page(unsigned long order)
4155 {
4156         char *buffer;
4157         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4158                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4159
4160         buffer = (char *) __get_free_pages(gfp_flags, order);
4161         if (buffer)
4162                 return buffer;
4163
4164         /* __get_free_pages failed, fall back to vmalloc */
4165         buffer = vzalloc((1 << order) * PAGE_SIZE);
4166         if (buffer)
4167                 return buffer;
4168
4169         /* vmalloc failed, lets dig into swap here */
4170         gfp_flags &= ~__GFP_NORETRY;
4171         buffer = (char *) __get_free_pages(gfp_flags, order);
4172         if (buffer)
4173                 return buffer;
4174
4175         /* complete and utter failure */
4176         return NULL;
4177 }
4178
4179 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4180 {
4181         unsigned int block_nr = req->tp_block_nr;
4182         struct pgv *pg_vec;
4183         int i;
4184
4185         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4186         if (unlikely(!pg_vec))
4187                 goto out;
4188
4189         for (i = 0; i < block_nr; i++) {
4190                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4191                 if (unlikely(!pg_vec[i].buffer))
4192                         goto out_free_pgvec;
4193         }
4194
4195 out:
4196         return pg_vec;
4197
4198 out_free_pgvec:
4199         free_pg_vec(pg_vec, order, block_nr);
4200         pg_vec = NULL;
4201         goto out;
4202 }
4203
4204 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4205                 int closing, int tx_ring)
4206 {
4207         struct pgv *pg_vec = NULL;
4208         struct packet_sock *po = pkt_sk(sk);
4209         int was_running, order = 0;
4210         struct packet_ring_buffer *rb;
4211         struct sk_buff_head *rb_queue;
4212         __be16 num;
4213         int err = -EINVAL;
4214         /* Added to avoid minimal code churn */
4215         struct tpacket_req *req = &req_u->req;
4216
4217         lock_sock(sk);
4218
4219         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4220         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4221
4222         err = -EBUSY;
4223         if (!closing) {
4224                 if (atomic_read(&po->mapped))
4225                         goto out;
4226                 if (packet_read_pending(rb))
4227                         goto out;
4228         }
4229
4230         if (req->tp_block_nr) {
4231                 /* Sanity tests and some calculations */
4232                 err = -EBUSY;
4233                 if (unlikely(rb->pg_vec))
4234                         goto out;
4235
4236                 switch (po->tp_version) {
4237                 case TPACKET_V1:
4238                         po->tp_hdrlen = TPACKET_HDRLEN;
4239                         break;
4240                 case TPACKET_V2:
4241                         po->tp_hdrlen = TPACKET2_HDRLEN;
4242                         break;
4243                 case TPACKET_V3:
4244                         po->tp_hdrlen = TPACKET3_HDRLEN;
4245                         break;
4246                 }
4247
4248                 err = -EINVAL;
4249                 if (unlikely((int)req->tp_block_size <= 0))
4250                         goto out;
4251                 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4252                         goto out;
4253                 if (po->tp_version >= TPACKET_V3 &&
4254                     req->tp_block_size <=
4255                           BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv))
4256                         goto out;
4257                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
4258                                         po->tp_reserve))
4259                         goto out;
4260                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4261                         goto out;
4262
4263                 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4264                 if (unlikely(rb->frames_per_block == 0))
4265                         goto out;
4266                 if (unlikely(req->tp_block_size > UINT_MAX / req->tp_block_nr))
4267                         goto out;
4268                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4269                                         req->tp_frame_nr))
4270                         goto out;
4271
4272                 err = -ENOMEM;
4273                 order = get_order(req->tp_block_size);
4274                 pg_vec = alloc_pg_vec(req, order);
4275                 if (unlikely(!pg_vec))
4276                         goto out;
4277                 switch (po->tp_version) {
4278                 case TPACKET_V3:
4279                         /* Block transmit is not supported yet */
4280                         if (!tx_ring) {
4281                                 init_prb_bdqc(po, rb, pg_vec, req_u);
4282                         } else {
4283                                 struct tpacket_req3 *req3 = &req_u->req3;
4284
4285                                 if (req3->tp_retire_blk_tov ||
4286                                     req3->tp_sizeof_priv ||
4287                                     req3->tp_feature_req_word) {
4288                                         err = -EINVAL;
4289                                         goto out;
4290                                 }
4291                         }
4292                         break;
4293                 default:
4294                         break;
4295                 }
4296         }
4297         /* Done */
4298         else {
4299                 err = -EINVAL;
4300                 if (unlikely(req->tp_frame_nr))
4301                         goto out;
4302         }
4303
4304
4305         /* Detach socket from network */
4306         spin_lock(&po->bind_lock);
4307         was_running = po->running;
4308         num = po->num;
4309         if (was_running) {
4310                 po->num = 0;
4311                 __unregister_prot_hook(sk, false);
4312         }
4313         spin_unlock(&po->bind_lock);
4314
4315         synchronize_net();
4316
4317         err = -EBUSY;
4318         mutex_lock(&po->pg_vec_lock);
4319         if (closing || atomic_read(&po->mapped) == 0) {
4320                 err = 0;
4321                 spin_lock_bh(&rb_queue->lock);
4322                 swap(rb->pg_vec, pg_vec);
4323                 rb->frame_max = (req->tp_frame_nr - 1);
4324                 rb->head = 0;
4325                 rb->frame_size = req->tp_frame_size;
4326                 spin_unlock_bh(&rb_queue->lock);
4327
4328                 swap(rb->pg_vec_order, order);
4329                 swap(rb->pg_vec_len, req->tp_block_nr);
4330
4331                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4332                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4333                                                 tpacket_rcv : packet_rcv;
4334                 skb_queue_purge(rb_queue);
4335                 if (atomic_read(&po->mapped))
4336                         pr_err("packet_mmap: vma is busy: %d\n",
4337                                atomic_read(&po->mapped));
4338         }
4339         mutex_unlock(&po->pg_vec_lock);
4340
4341         spin_lock(&po->bind_lock);
4342         if (was_running) {
4343                 po->num = num;
4344                 register_prot_hook(sk);
4345         }
4346         spin_unlock(&po->bind_lock);
4347         if (pg_vec && (po->tp_version > TPACKET_V2)) {
4348                 /* Because we don't support block-based V3 on tx-ring */
4349                 if (!tx_ring)
4350                         prb_shutdown_retire_blk_timer(po, rb_queue);
4351         }
4352
4353         if (pg_vec)
4354                 free_pg_vec(pg_vec, order, req->tp_block_nr);
4355 out:
4356         release_sock(sk);
4357         return err;
4358 }
4359
4360 static int packet_mmap(struct file *file, struct socket *sock,
4361                 struct vm_area_struct *vma)
4362 {
4363         struct sock *sk = sock->sk;
4364         struct packet_sock *po = pkt_sk(sk);
4365         unsigned long size, expected_size;
4366         struct packet_ring_buffer *rb;
4367         unsigned long start;
4368         int err = -EINVAL;
4369         int i;
4370
4371         if (vma->vm_pgoff)
4372                 return -EINVAL;
4373
4374         mutex_lock(&po->pg_vec_lock);
4375
4376         expected_size = 0;
4377         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4378                 if (rb->pg_vec) {
4379                         expected_size += rb->pg_vec_len
4380                                                 * rb->pg_vec_pages
4381                                                 * PAGE_SIZE;
4382                 }
4383         }
4384
4385         if (expected_size == 0)
4386                 goto out;
4387
4388         size = vma->vm_end - vma->vm_start;
4389         if (size != expected_size)
4390                 goto out;
4391
4392         start = vma->vm_start;
4393         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4394                 if (rb->pg_vec == NULL)
4395                         continue;
4396
4397                 for (i = 0; i < rb->pg_vec_len; i++) {
4398                         struct page *page;
4399                         void *kaddr = rb->pg_vec[i].buffer;
4400                         int pg_num;
4401
4402                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4403                                 page = pgv_to_page(kaddr);
4404                                 err = vm_insert_page(vma, start, page);
4405                                 if (unlikely(err))
4406                                         goto out;
4407                                 start += PAGE_SIZE;
4408                                 kaddr += PAGE_SIZE;
4409                         }
4410                 }
4411         }
4412
4413         atomic_inc(&po->mapped);
4414         vma->vm_ops = &packet_mmap_ops;
4415         err = 0;
4416
4417 out:
4418         mutex_unlock(&po->pg_vec_lock);
4419         return err;
4420 }
4421
4422 static const struct proto_ops packet_ops_spkt = {
4423         .family =       PF_PACKET,
4424         .owner =        THIS_MODULE,
4425         .release =      packet_release,
4426         .bind =         packet_bind_spkt,
4427         .connect =      sock_no_connect,
4428         .socketpair =   sock_no_socketpair,
4429         .accept =       sock_no_accept,
4430         .getname =      packet_getname_spkt,
4431         .poll =         datagram_poll,
4432         .ioctl =        packet_ioctl,
4433         .listen =       sock_no_listen,
4434         .shutdown =     sock_no_shutdown,
4435         .setsockopt =   sock_no_setsockopt,
4436         .getsockopt =   sock_no_getsockopt,
4437         .sendmsg =      packet_sendmsg_spkt,
4438         .recvmsg =      packet_recvmsg,
4439         .mmap =         sock_no_mmap,
4440         .sendpage =     sock_no_sendpage,
4441 };
4442
4443 static const struct proto_ops packet_ops = {
4444         .family =       PF_PACKET,
4445         .owner =        THIS_MODULE,
4446         .release =      packet_release,
4447         .bind =         packet_bind,
4448         .connect =      sock_no_connect,
4449         .socketpair =   sock_no_socketpair,
4450         .accept =       sock_no_accept,
4451         .getname =      packet_getname,
4452         .poll =         packet_poll,
4453         .ioctl =        packet_ioctl,
4454         .listen =       sock_no_listen,
4455         .shutdown =     sock_no_shutdown,
4456         .setsockopt =   packet_setsockopt,
4457         .getsockopt =   packet_getsockopt,
4458 #ifdef CONFIG_COMPAT
4459         .compat_setsockopt = compat_packet_setsockopt,
4460 #endif
4461         .sendmsg =      packet_sendmsg,
4462         .recvmsg =      packet_recvmsg,
4463         .mmap =         packet_mmap,
4464         .sendpage =     sock_no_sendpage,
4465 };
4466
4467 static const struct net_proto_family packet_family_ops = {
4468         .family =       PF_PACKET,
4469         .create =       packet_create,
4470         .owner  =       THIS_MODULE,
4471 };
4472
4473 static struct notifier_block packet_netdev_notifier = {
4474         .notifier_call =        packet_notifier,
4475 };
4476
4477 #ifdef CONFIG_PROC_FS
4478
4479 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4480         __acquires(RCU)
4481 {
4482         struct net *net = seq_file_net(seq);
4483
4484         rcu_read_lock();
4485         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4486 }
4487
4488 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4489 {
4490         struct net *net = seq_file_net(seq);
4491         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4492 }
4493
4494 static void packet_seq_stop(struct seq_file *seq, void *v)
4495         __releases(RCU)
4496 {
4497         rcu_read_unlock();
4498 }
4499
4500 static int packet_seq_show(struct seq_file *seq, void *v)
4501 {
4502         if (v == SEQ_START_TOKEN)
4503                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
4504         else {
4505                 struct sock *s = sk_entry(v);
4506                 const struct packet_sock *po = pkt_sk(s);
4507
4508                 seq_printf(seq,
4509                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4510                            s,
4511                            refcount_read(&s->sk_refcnt),
4512                            s->sk_type,
4513                            ntohs(po->num),
4514                            po->ifindex,
4515                            po->running,
4516                            atomic_read(&s->sk_rmem_alloc),
4517                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4518                            sock_i_ino(s));
4519         }
4520
4521         return 0;
4522 }
4523
4524 static const struct seq_operations packet_seq_ops = {
4525         .start  = packet_seq_start,
4526         .next   = packet_seq_next,
4527         .stop   = packet_seq_stop,
4528         .show   = packet_seq_show,
4529 };
4530
4531 static int packet_seq_open(struct inode *inode, struct file *file)
4532 {
4533         return seq_open_net(inode, file, &packet_seq_ops,
4534                             sizeof(struct seq_net_private));
4535 }
4536
4537 static const struct file_operations packet_seq_fops = {
4538         .owner          = THIS_MODULE,
4539         .open           = packet_seq_open,
4540         .read           = seq_read,
4541         .llseek         = seq_lseek,
4542         .release        = seq_release_net,
4543 };
4544
4545 #endif
4546
4547 static int __net_init packet_net_init(struct net *net)
4548 {
4549         mutex_init(&net->packet.sklist_lock);
4550         INIT_HLIST_HEAD(&net->packet.sklist);
4551
4552         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4553                 return -ENOMEM;
4554
4555         return 0;
4556 }
4557
4558 static void __net_exit packet_net_exit(struct net *net)
4559 {
4560         remove_proc_entry("packet", net->proc_net);
4561 }
4562
4563 static struct pernet_operations packet_net_ops = {
4564         .init = packet_net_init,
4565         .exit = packet_net_exit,
4566 };
4567
4568
4569 static void __exit packet_exit(void)
4570 {
4571         unregister_netdevice_notifier(&packet_netdev_notifier);
4572         unregister_pernet_subsys(&packet_net_ops);
4573         sock_unregister(PF_PACKET);
4574         proto_unregister(&packet_proto);
4575 }
4576
4577 static int __init packet_init(void)
4578 {
4579         int rc = proto_register(&packet_proto, 0);
4580
4581         if (rc != 0)
4582                 goto out;
4583
4584         sock_register(&packet_family_ops);
4585         register_pernet_subsys(&packet_net_ops);
4586         register_netdevice_notifier(&packet_netdev_notifier);
4587 out:
4588         return rc;
4589 }
4590
4591 module_init(packet_init);
4592 module_exit(packet_exit);
4593 MODULE_LICENSE("GPL");
4594 MODULE_ALIAS_NETPROTO(PF_PACKET);