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 PGV_FROM_VMALLOC 1
 181
 182 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
 183 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
 184 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
 185 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
 186 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
 187 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
 188 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
 189
 190 struct packet_sock;
 191 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 192                        struct packet_type *pt, struct net_device *orig_dev);
 193
 194 static void *packet_previous_frame(struct packet_sock *po,
 195                 struct packet_ring_buffer *rb,
 196                 int status);
 197 static void packet_increment_head(struct packet_ring_buffer *buff);
 198 static int prb_curr_blk_in_use(struct tpacket_block_desc *);
 199 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 200                         struct packet_sock *);
 201 static void prb_retire_current_block(struct tpacket_kbdq_core *,
 202                 struct packet_sock *, unsigned int status);
 203 static int prb_queue_frozen(struct tpacket_kbdq_core *);
 204 static void prb_open_block(struct tpacket_kbdq_core *,
 205                 struct tpacket_block_desc *);
 206 static void prb_retire_rx_blk_timer_expired(unsigned long);
 207 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 208 static void prb_init_blk_timer(struct packet_sock *,
 209                 struct tpacket_kbdq_core *,
 210                 void (*func) (unsigned long));
 211 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 212 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 213                 struct tpacket3_hdr *);
 214 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 215                 struct tpacket3_hdr *);
 216 static void packet_flush_mclist(struct sock *sk);
 217 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb);
 218
 219 struct packet_skb_cb {
 220         union {
 221                 struct sockaddr_pkt pkt;
 222                 union {
 223                         /* Trick: alias skb original length with
 224                          * ll.sll_family and ll.protocol in order
 225                          * to save room.
 226                          */
 227                         unsigned int origlen;
 228                         struct sockaddr_ll ll;
 229                 };
 230         } sa;
 231 };
 232
 233 #define vio_le() virtio_legacy_is_little_endian()
 234
 235 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 236
 237 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 238 #define GET_PBLOCK_DESC(x, bid) \
 239         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 240 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 241         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 242 #define GET_NEXT_PRB_BLK_NUM(x) \
 243         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 244         ((x)->kactive_blk_num+1) : 0)
 245
 246 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 247 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 248
 249 static int packet_direct_xmit(struct sk_buff *skb)
 250 {
 251         struct net_device *dev = skb->dev;
 252         struct sk_buff *orig_skb = skb;
 253         struct netdev_queue *txq;
 254         int ret = NETDEV_TX_BUSY;
 255
 256         if (unlikely(!netif_running(dev) ||
 257                      !netif_carrier_ok(dev)))
 258                 goto drop;
 259
 260         skb = validate_xmit_skb_list(skb, dev);
 261         if (skb != orig_skb)
 262                 goto drop;
 263
 264         packet_pick_tx_queue(dev, skb);
 265         txq = skb_get_tx_queue(dev, skb);
 266
 267         local_bh_disable();
 268
 269         HARD_TX_LOCK(dev, txq, smp_processor_id());
 270         if (!netif_xmit_frozen_or_drv_stopped(txq))
 271                 ret = netdev_start_xmit(skb, dev, txq, false);
 272         HARD_TX_UNLOCK(dev, txq);
 273
 274         local_bh_enable();
 275
 276         if (!dev_xmit_complete(ret))
 277                 kfree_skb(skb);
 278
 279         return ret;
 280 drop:
 281         atomic_long_inc(&dev->tx_dropped);
 282         kfree_skb_list(skb);
 283         return NET_XMIT_DROP;
 284 }
 285
 286 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
 287 {
 288         struct net_device *dev;
 289
 290         rcu_read_lock();
 291         dev = rcu_dereference(po->cached_dev);
 292         if (likely(dev))
 293                 dev_hold(dev);
 294         rcu_read_unlock();
 295
 296         return dev;
 297 }
 298
 299 static void packet_cached_dev_assign(struct packet_sock *po,
 300                                      struct net_device *dev)
 301 {
 302         rcu_assign_pointer(po->cached_dev, dev);
 303 }
 304
 305 static void packet_cached_dev_reset(struct packet_sock *po)
 306 {
 307         RCU_INIT_POINTER(po->cached_dev, NULL);
 308 }
 309
 310 static bool packet_use_direct_xmit(const struct packet_sock *po)
 311 {
 312         return po->xmit == packet_direct_xmit;
 313 }
 314
 315 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
 316 {
 317         return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
 318 }
 319
 320 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
 321 {
 322         const struct net_device_ops *ops = dev->netdev_ops;
 323         u16 queue_index;
 324
 325         if (ops->ndo_select_queue) {
 326                 queue_index = ops->ndo_select_queue(dev, skb, NULL,
 327                                                     __packet_pick_tx_queue);
 328                 queue_index = netdev_cap_txqueue(dev, queue_index);
 329         } else {
 330                 queue_index = __packet_pick_tx_queue(dev, skb);
 331         }
 332
 333         skb_set_queue_mapping(skb, queue_index);
 334 }
 335
 336 /* register_prot_hook must be invoked with the po->bind_lock held,
 337  * or from a context in which asynchronous accesses to the packet
 338  * socket is not possible (packet_create()).
 339  */
 340 static void register_prot_hook(struct sock *sk)
 341 {
 342         struct packet_sock *po = pkt_sk(sk);
 343
 344         if (!po->running) {
 345                 if (po->fanout)
 346                         __fanout_link(sk, po);
 347                 else
 348                         dev_add_pack(&po->prot_hook);
 349
 350                 sock_hold(sk);
 351                 po->running = 1;
 352         }
 353 }
 354
 355 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
 356  * held.   If the sync parameter is true, we will temporarily drop
 357  * the po->bind_lock and do a synchronize_net to make sure no
 358  * asynchronous packet processing paths still refer to the elements
 359  * of po->prot_hook.  If the sync parameter is false, it is the
 360  * callers responsibility to take care of this.
 361  */
 362 static void __unregister_prot_hook(struct sock *sk, bool sync)
 363 {
 364         struct packet_sock *po = pkt_sk(sk);
 365
 366         po->running = 0;
 367
 368         if (po->fanout)
 369                 __fanout_unlink(sk, po);
 370         else
 371                 __dev_remove_pack(&po->prot_hook);
 372
 373         __sock_put(sk);
 374
 375         if (sync) {
 376                 spin_unlock(&po->bind_lock);
 377                 synchronize_net();
 378                 spin_lock(&po->bind_lock);
 379         }
 380 }
 381
 382 static void unregister_prot_hook(struct sock *sk, bool sync)
 383 {
 384         struct packet_sock *po = pkt_sk(sk);
 385
 386         if (po->running)
 387                 __unregister_prot_hook(sk, sync);
 388 }
 389
 390 static inline struct page * __pure pgv_to_page(void *addr)
 391 {
 392         if (is_vmalloc_addr(addr))
 393                 return vmalloc_to_page(addr);
 394         return virt_to_page(addr);
 395 }
 396
 397 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 398 {
 399         union tpacket_uhdr h;
 400
 401         h.raw = frame;
 402         switch (po->tp_version) {
 403         case TPACKET_V1:
 404                 h.h1->tp_status = status;
 405                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 406                 break;
 407         case TPACKET_V2:
 408                 h.h2->tp_status = status;
 409                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 410                 break;
 411         case TPACKET_V3:
 412                 h.h3->tp_status = status;
 413                 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
 414                 break;
 415         default:
 416                 WARN(1, "TPACKET version not supported.\n");
 417                 BUG();
 418         }
 419
 420         smp_wmb();
 421 }
 422
 423 static int __packet_get_status(struct packet_sock *po, void *frame)
 424 {
 425         union tpacket_uhdr h;
 426
 427         smp_rmb();
 428
 429         h.raw = frame;
 430         switch (po->tp_version) {
 431         case TPACKET_V1:
 432                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 433                 return h.h1->tp_status;
 434         case TPACKET_V2:
 435                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 436                 return h.h2->tp_status;
 437         case TPACKET_V3:
 438                 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
 439                 return h.h3->tp_status;
 440         default:
 441                 WARN(1, "TPACKET version not supported.\n");
 442                 BUG();
 443                 return 0;
 444         }
 445 }
 446
 447 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
 448                                    unsigned int flags)
 449 {
 450         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 451
 452         if (shhwtstamps &&
 453             (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
 454             ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
 455                 return TP_STATUS_TS_RAW_HARDWARE;
 456
 457         if (ktime_to_timespec_cond(skb->tstamp, ts))
 458                 return TP_STATUS_TS_SOFTWARE;
 459
 460         return 0;
 461 }
 462
 463 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
 464                                     struct sk_buff *skb)
 465 {
 466         union tpacket_uhdr h;
 467         struct timespec ts;
 468         __u32 ts_status;
 469
 470         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
 471                 return 0;
 472
 473         h.raw = frame;
 474         switch (po->tp_version) {
 475         case TPACKET_V1:
 476                 h.h1->tp_sec = ts.tv_sec;
 477                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
 478                 break;
 479         case TPACKET_V2:
 480                 h.h2->tp_sec = ts.tv_sec;
 481                 h.h2->tp_nsec = ts.tv_nsec;
 482                 break;
 483         case TPACKET_V3:
 484                 h.h3->tp_sec = ts.tv_sec;
 485                 h.h3->tp_nsec = ts.tv_nsec;
 486                 break;
 487         default:
 488                 WARN(1, "TPACKET version not supported.\n");
 489                 BUG();
 490         }
 491
 492         /* one flush is safe, as both fields always lie on the same cacheline */
 493         flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
 494         smp_wmb();
 495
 496         return ts_status;
 497 }
 498
 499 static void *packet_lookup_frame(struct packet_sock *po,
 500                 struct packet_ring_buffer *rb,
 501                 unsigned int position,
 502                 int status)
 503 {
 504         unsigned int pg_vec_pos, frame_offset;
 505         union tpacket_uhdr h;
 506
 507         pg_vec_pos = position / rb->frames_per_block;
 508         frame_offset = position % rb->frames_per_block;
 509
 510         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 511                 (frame_offset * rb->frame_size);
 512
 513         if (status != __packet_get_status(po, h.raw))
 514                 return NULL;
 515
 516         return h.raw;
 517 }
 518
 519 static void *packet_current_frame(struct packet_sock *po,
 520                 struct packet_ring_buffer *rb,
 521                 int status)
 522 {
 523         return packet_lookup_frame(po, rb, rb->head, status);
 524 }
 525
 526 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 527 {
 528         del_timer_sync(&pkc->retire_blk_timer);
 529 }
 530
 531 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 532                 struct sk_buff_head *rb_queue)
 533 {
 534         struct tpacket_kbdq_core *pkc;
 535
 536         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 537
 538         spin_lock_bh(&rb_queue->lock);
 539         pkc->delete_blk_timer = 1;
 540         spin_unlock_bh(&rb_queue->lock);
 541
 542         prb_del_retire_blk_timer(pkc);
 543 }
 544
 545 static void prb_init_blk_timer(struct packet_sock *po,
 546                 struct tpacket_kbdq_core *pkc,
 547                 void (*func) (unsigned long))
 548 {
 549         init_timer(&pkc->retire_blk_timer);
 550         pkc->retire_blk_timer.data = (long)po;
 551         pkc->retire_blk_timer.function = func;
 552         pkc->retire_blk_timer.expires = jiffies;
 553 }
 554
 555 static void prb_setup_retire_blk_timer(struct packet_sock *po)
 556 {
 557         struct tpacket_kbdq_core *pkc;
 558
 559         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 560         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
 561 }
 562
 563 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 564                                 int blk_size_in_bytes)
 565 {
 566         struct net_device *dev;
 567         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 568         struct ethtool_link_ksettings ecmd;
 569         int err;
 570
 571         rtnl_lock();
 572         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 573         if (unlikely(!dev)) {
 574                 rtnl_unlock();
 575                 return DEFAULT_PRB_RETIRE_TOV;
 576         }
 577         err = __ethtool_get_link_ksettings(dev, &ecmd);
 578         rtnl_unlock();
 579         if (!err) {
 580                 /*
 581                  * If the link speed is so slow you don't really
 582                  * need to worry about perf anyways
 583                  */
 584                 if (ecmd.base.speed < SPEED_1000 ||
 585                     ecmd.base.speed == SPEED_UNKNOWN) {
 586                         return DEFAULT_PRB_RETIRE_TOV;
 587                 } else {
 588                         msec = 1;
 589                         div = ecmd.base.speed / 1000;
 590                 }
 591         }
 592
 593         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 594
 595         if (div)
 596                 mbits /= div;
 597
 598         tmo = mbits * msec;
 599
 600         if (div)
 601                 return tmo+1;
 602         return tmo;
 603 }
 604
 605 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 606                         union tpacket_req_u *req_u)
 607 {
 608         p1->feature_req_word = req_u->req3.tp_feature_req_word;
 609 }
 610
 611 static void init_prb_bdqc(struct packet_sock *po,
 612                         struct packet_ring_buffer *rb,
 613                         struct pgv *pg_vec,
 614                         union tpacket_req_u *req_u)
 615 {
 616         struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
 617         struct tpacket_block_desc *pbd;
 618
 619         memset(p1, 0x0, sizeof(*p1));
 620
 621         p1->knxt_seq_num = 1;
 622         p1->pkbdq = pg_vec;
 623         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 624         p1->pkblk_start = pg_vec[0].buffer;
 625         p1->kblk_size = req_u->req3.tp_block_size;
 626         p1->knum_blocks = req_u->req3.tp_block_nr;
 627         p1->hdrlen = po->tp_hdrlen;
 628         p1->version = po->tp_version;
 629         p1->last_kactive_blk_num = 0;
 630         po->stats.stats3.tp_freeze_q_cnt = 0;
 631         if (req_u->req3.tp_retire_blk_tov)
 632                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 633         else
 634                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 635                                                 req_u->req3.tp_block_size);
 636         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 637         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 638
 639         p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
 640         prb_init_ft_ops(p1, req_u);
 641         prb_setup_retire_blk_timer(po);
 642         prb_open_block(p1, pbd);
 643 }
 644
 645 /*  Do NOT update the last_blk_num first.
 646  *  Assumes sk_buff_head lock is held.
 647  */
 648 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 649 {
 650         mod_timer(&pkc->retire_blk_timer,
 651                         jiffies + pkc->tov_in_jiffies);
 652         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 653 }
 654
 655 /*
 656  * Timer logic:
 657  * 1) We refresh the timer only when we open a block.
 658  *    By doing this we don't waste cycles refreshing the timer
 659  *        on packet-by-packet basis.
 660  *
 661  * With a 1MB block-size, on a 1Gbps line, it will take
 662  * i) ~8 ms to fill a block + ii) memcpy etc.
 663  * In this cut we are not accounting for the memcpy time.
 664  *
 665  * So, if the user sets the 'tmo' to 10ms then the timer
 666  * will never fire while the block is still getting filled
 667  * (which is what we want). However, the user could choose
 668  * to close a block early and that's fine.
 669  *
 670  * But when the timer does fire, we check whether or not to refresh it.
 671  * Since the tmo granularity is in msecs, it is not too expensive
 672  * to refresh the timer, lets say every '8' msecs.
 673  * Either the user can set the 'tmo' or we can derive it based on
 674  * a) line-speed and b) block-size.
 675  * prb_calc_retire_blk_tmo() calculates the tmo.
 676  *
 677  */
 678 static void prb_retire_rx_blk_timer_expired(unsigned long data)
 679 {
 680         struct packet_sock *po = (struct packet_sock *)data;
 681         struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 682         unsigned int frozen;
 683         struct tpacket_block_desc *pbd;
 684
 685         spin_lock(&po->sk.sk_receive_queue.lock);
 686
 687         frozen = prb_queue_frozen(pkc);
 688         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 689
 690         if (unlikely(pkc->delete_blk_timer))
 691                 goto out;
 692
 693         /* We only need to plug the race when the block is partially filled.
 694          * tpacket_rcv:
 695          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 696          *              copy_bits() is in progress ...
 697          *              timer fires on other cpu:
 698          *              we can't retire the current block because copy_bits
 699          *              is in progress.
 700          *
 701          */
 702         if (BLOCK_NUM_PKTS(pbd)) {
 703                 while (atomic_read(&pkc->blk_fill_in_prog)) {
 704                         /* Waiting for skb_copy_bits to finish... */
 705                         cpu_relax();
 706                 }
 707         }
 708
 709         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 710                 if (!frozen) {
 711                         if (!BLOCK_NUM_PKTS(pbd)) {
 712                                 /* An empty block. Just refresh the timer. */
 713                                 goto refresh_timer;
 714                         }
 715                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 716                         if (!prb_dispatch_next_block(pkc, po))
 717                                 goto refresh_timer;
 718                         else
 719                                 goto out;
 720                 } else {
 721                         /* Case 1. Queue was frozen because user-space was
 722                          *         lagging behind.
 723                          */
 724                         if (prb_curr_blk_in_use(pbd)) {
 725                                 /*
 726                                  * Ok, user-space is still behind.
 727                                  * So just refresh the timer.
 728                                  */
 729                                 goto refresh_timer;
 730                         } else {
 731                                /* Case 2. queue was frozen,user-space caught up,
 732                                 * now the link went idle && the timer fired.
 733                                 * We don't have a block to close.So we open this
 734                                 * block and restart the timer.
 735                                 * opening a block thaws the queue,restarts timer
 736                                 * Thawing/timer-refresh is a side effect.
 737                                 */
 738                                 prb_open_block(pkc, pbd);
 739                                 goto out;
 740                         }
 741                 }
 742         }
 743
 744 refresh_timer:
 745         _prb_refresh_rx_retire_blk_timer(pkc);
 746
 747 out:
 748         spin_unlock(&po->sk.sk_receive_queue.lock);
 749 }
 750
 751 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 752                 struct tpacket_block_desc *pbd1, __u32 status)
 753 {
 754         /* Flush everything minus the block header */
 755
 756 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 757         u8 *start, *end;
 758
 759         start = (u8 *)pbd1;
 760
 761         /* Skip the block header(we know header WILL fit in 4K) */
 762         start += PAGE_SIZE;
 763
 764         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 765         for (; start < end; start += PAGE_SIZE)
 766                 flush_dcache_page(pgv_to_page(start));
 767
 768         smp_wmb();
 769 #endif
 770
 771         /* Now update the block status. */
 772
 773         BLOCK_STATUS(pbd1) = status;
 774
 775         /* Flush the block header */
 776
 777 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 778         start = (u8 *)pbd1;
 779         flush_dcache_page(pgv_to_page(start));
 780
 781         smp_wmb();
 782 #endif
 783 }
 784
 785 /*
 786  * Side effect:
 787  *
 788  * 1) flush the block
 789  * 2) Increment active_blk_num
 790  *
 791  * Note:We DONT refresh the timer on purpose.
 792  *      Because almost always the next block will be opened.
 793  */
 794 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 795                 struct tpacket_block_desc *pbd1,
 796                 struct packet_sock *po, unsigned int stat)
 797 {
 798         __u32 status = TP_STATUS_USER | stat;
 799
 800         struct tpacket3_hdr *last_pkt;
 801         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 802         struct sock *sk = &po->sk;
 803
 804         if (po->stats.stats3.tp_drops)
 805                 status |= TP_STATUS_LOSING;
 806
 807         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 808         last_pkt->tp_next_offset = 0;
 809
 810         /* Get the ts of the last pkt */
 811         if (BLOCK_NUM_PKTS(pbd1)) {
 812                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 813                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 814         } else {
 815                 /* Ok, we tmo'd - so get the current time.
 816                  *
 817                  * It shouldn't really happen as we don't close empty
 818                  * blocks. See prb_retire_rx_blk_timer_expired().
 819                  */
 820                 struct timespec ts;
 821                 getnstimeofday(&ts);
 822                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 823                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 824         }
 825
 826         smp_wmb();
 827
 828         /* Flush the block */
 829         prb_flush_block(pkc1, pbd1, status);
 830
 831         sk->sk_data_ready(sk);
 832
 833         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 834 }
 835
 836 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 837 {
 838         pkc->reset_pending_on_curr_blk = 0;
 839 }
 840
 841 /*
 842  * Side effect of opening a block:
 843  *
 844  * 1) prb_queue is thawed.
 845  * 2) retire_blk_timer is refreshed.
 846  *
 847  */
 848 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 849         struct tpacket_block_desc *pbd1)
 850 {
 851         struct timespec ts;
 852         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 853
 854         smp_rmb();
 855
 856         /* We could have just memset this but we will lose the
 857          * flexibility of making the priv area sticky
 858          */
 859
 860         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 861         BLOCK_NUM_PKTS(pbd1) = 0;
 862         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 863
 864         getnstimeofday(&ts);
 865
 866         h1->ts_first_pkt.ts_sec = ts.tv_sec;
 867         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 868
 869         pkc1->pkblk_start = (char *)pbd1;
 870         pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 871
 872         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 873         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 874
 875         pbd1->version = pkc1->version;
 876         pkc1->prev = pkc1->nxt_offset;
 877         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 878
 879         prb_thaw_queue(pkc1);
 880         _prb_refresh_rx_retire_blk_timer(pkc1);
 881
 882         smp_wmb();
 883 }
 884
 885 /*
 886  * Queue freeze logic:
 887  * 1) Assume tp_block_nr = 8 blocks.
 888  * 2) At time 't0', user opens Rx ring.
 889  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 890  * 4) user-space is either sleeping or processing block '0'.
 891  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 892  *    it will close block-7,loop around and try to fill block '0'.
 893  *    call-flow:
 894  *    __packet_lookup_frame_in_block
 895  *      prb_retire_current_block()
 896  *      prb_dispatch_next_block()
 897  *        |->(BLOCK_STATUS == USER) evaluates to true
 898  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 899  * 6) Now there are two cases:
 900  *    6.1) Link goes idle right after the queue is frozen.
 901  *         But remember, the last open_block() refreshed the timer.
 902  *         When this timer expires,it will refresh itself so that we can
 903  *         re-open block-0 in near future.
 904  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 905  *         case and __packet_lookup_frame_in_block will check if block-0
 906  *         is free and can now be re-used.
 907  */
 908 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 909                                   struct packet_sock *po)
 910 {
 911         pkc->reset_pending_on_curr_blk = 1;
 912         po->stats.stats3.tp_freeze_q_cnt++;
 913 }
 914
 915 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 916
 917 /*
 918  * If the next block is free then we will dispatch it
 919  * and return a good offset.
 920  * Else, we will freeze the queue.
 921  * So, caller must check the return value.
 922  */
 923 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 924                 struct packet_sock *po)
 925 {
 926         struct tpacket_block_desc *pbd;
 927
 928         smp_rmb();
 929
 930         /* 1. Get current block num */
 931         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 932
 933         /* 2. If this block is currently in_use then freeze the queue */
 934         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 935                 prb_freeze_queue(pkc, po);
 936                 return NULL;
 937         }
 938
 939         /*
 940          * 3.
 941          * open this block and return the offset where the first packet
 942          * needs to get stored.
 943          */
 944         prb_open_block(pkc, pbd);
 945         return (void *)pkc->nxt_offset;
 946 }
 947
 948 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 949                 struct packet_sock *po, unsigned int status)
 950 {
 951         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 952
 953         /* retire/close the current block */
 954         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 955                 /*
 956                  * Plug the case where copy_bits() is in progress on
 957                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 958                  * have space to copy the pkt in the current block and
 959                  * called prb_retire_current_block()
 960                  *
 961                  * We don't need to worry about the TMO case because
 962                  * the timer-handler already handled this case.
 963                  */
 964                 if (!(status & TP_STATUS_BLK_TMO)) {
 965                         while (atomic_read(&pkc->blk_fill_in_prog)) {
 966                                 /* Waiting for skb_copy_bits to finish... */
 967                                 cpu_relax();
 968                         }
 969                 }
 970                 prb_close_block(pkc, pbd, po, status);
 971                 return;
 972         }
 973 }
 974
 975 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
 976 {
 977         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 978 }
 979
 980 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 981 {
 982         return pkc->reset_pending_on_curr_blk;
 983 }
 984
 985 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 986 {
 987         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 988         atomic_dec(&pkc->blk_fill_in_prog);
 989 }
 990
 991 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 992                         struct tpacket3_hdr *ppd)
 993 {
 994         ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
 995 }
 996
 997 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 998                         struct tpacket3_hdr *ppd)
 999 {
1000         ppd->hv1.tp_rxhash = 0;
1001 }
1002
1003 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1004                         struct tpacket3_hdr *ppd)
1005 {
1006         if (skb_vlan_tag_present(pkc->skb)) {
1007                 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1008                 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1009                 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1010         } else {
1011                 ppd->hv1.tp_vlan_tci = 0;
1012                 ppd->hv1.tp_vlan_tpid = 0;
1013                 ppd->tp_status = TP_STATUS_AVAILABLE;
1014         }
1015 }
1016
1017 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1018                         struct tpacket3_hdr *ppd)
1019 {
1020         ppd->hv1.tp_padding = 0;
1021         prb_fill_vlan_info(pkc, ppd);
1022
1023         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1024                 prb_fill_rxhash(pkc, ppd);
1025         else
1026                 prb_clear_rxhash(pkc, ppd);
1027 }
1028
1029 static void prb_fill_curr_block(char *curr,
1030                                 struct tpacket_kbdq_core *pkc,
1031                                 struct tpacket_block_desc *pbd,
1032                                 unsigned int len)
1033 {
1034         struct tpacket3_hdr *ppd;
1035
1036         ppd  = (struct tpacket3_hdr *)curr;
1037         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1038         pkc->prev = curr;
1039         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1040         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1041         BLOCK_NUM_PKTS(pbd) += 1;
1042         atomic_inc(&pkc->blk_fill_in_prog);
1043         prb_run_all_ft_ops(pkc, ppd);
1044 }
1045
1046 /* Assumes caller has the sk->rx_queue.lock */
1047 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1048                                             struct sk_buff *skb,
1049                                                 int status,
1050                                             unsigned int len
1051                                             )
1052 {
1053         struct tpacket_kbdq_core *pkc;
1054         struct tpacket_block_desc *pbd;
1055         char *curr, *end;
1056
1057         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1058         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1059
1060         /* Queue is frozen when user space is lagging behind */
1061         if (prb_queue_frozen(pkc)) {
1062                 /*
1063                  * Check if that last block which caused the queue to freeze,
1064                  * is still in_use by user-space.
1065                  */
1066                 if (prb_curr_blk_in_use(pbd)) {
1067                         /* Can't record this packet */
1068                         return NULL;
1069                 } else {
1070                         /*
1071                          * Ok, the block was released by user-space.
1072                          * Now let's open that block.
1073                          * opening a block also thaws the queue.
1074                          * Thawing is a side effect.
1075                          */
1076                         prb_open_block(pkc, pbd);
1077                 }
1078         }
1079
1080         smp_mb();
1081         curr = pkc->nxt_offset;
1082         pkc->skb = skb;
1083         end = (char *)pbd + pkc->kblk_size;
1084
1085         /* first try the current block */
1086         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1087                 prb_fill_curr_block(curr, pkc, pbd, len);
1088                 return (void *)curr;
1089         }
1090
1091         /* Ok, close the current block */
1092         prb_retire_current_block(pkc, po, 0);
1093
1094         /* Now, try to dispatch the next block */
1095         curr = (char *)prb_dispatch_next_block(pkc, po);
1096         if (curr) {
1097                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1098                 prb_fill_curr_block(curr, pkc, pbd, len);
1099                 return (void *)curr;
1100         }
1101
1102         /*
1103          * No free blocks are available.user_space hasn't caught up yet.
1104          * Queue was just frozen and now this packet will get dropped.
1105          */
1106         return NULL;
1107 }
1108
1109 static void *packet_current_rx_frame(struct packet_sock *po,
1110                                             struct sk_buff *skb,
1111                                             int status, unsigned int len)
1112 {
1113         char *curr = NULL;
1114         switch (po->tp_version) {
1115         case TPACKET_V1:
1116         case TPACKET_V2:
1117                 curr = packet_lookup_frame(po, &po->rx_ring,
1118                                         po->rx_ring.head, status);
1119                 return curr;
1120         case TPACKET_V3:
1121                 return __packet_lookup_frame_in_block(po, skb, status, len);
1122         default:
1123                 WARN(1, "TPACKET version not supported\n");
1124                 BUG();
1125                 return NULL;
1126         }
1127 }
1128
1129 static void *prb_lookup_block(struct packet_sock *po,
1130                                      struct packet_ring_buffer *rb,
1131                                      unsigned int idx,
1132                                      int status)
1133 {
1134         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1135         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1136
1137         if (status != BLOCK_STATUS(pbd))
1138                 return NULL;
1139         return pbd;
1140 }
1141
1142 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1143 {
1144         unsigned int prev;
1145         if (rb->prb_bdqc.kactive_blk_num)
1146                 prev = rb->prb_bdqc.kactive_blk_num-1;
1147         else
1148                 prev = rb->prb_bdqc.knum_blocks-1;
1149         return prev;
1150 }
1151
1152 /* Assumes caller has held the rx_queue.lock */
1153 static void *__prb_previous_block(struct packet_sock *po,
1154                                          struct packet_ring_buffer *rb,
1155                                          int status)
1156 {
1157         unsigned int previous = prb_previous_blk_num(rb);
1158         return prb_lookup_block(po, rb, previous, status);
1159 }
1160
1161 static void *packet_previous_rx_frame(struct packet_sock *po,
1162                                              struct packet_ring_buffer *rb,
1163                                              int status)
1164 {
1165         if (po->tp_version <= TPACKET_V2)
1166                 return packet_previous_frame(po, rb, status);
1167
1168         return __prb_previous_block(po, rb, status);
1169 }
1170
1171 static void packet_increment_rx_head(struct packet_sock *po,
1172                                             struct packet_ring_buffer *rb)
1173 {
1174         switch (po->tp_version) {
1175         case TPACKET_V1:
1176         case TPACKET_V2:
1177                 return packet_increment_head(rb);
1178         case TPACKET_V3:
1179         default:
1180                 WARN(1, "TPACKET version not supported.\n");
1181                 BUG();
1182                 return;
1183         }
1184 }
1185
1186 static void *packet_previous_frame(struct packet_sock *po,
1187                 struct packet_ring_buffer *rb,
1188                 int status)
1189 {
1190         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1191         return packet_lookup_frame(po, rb, previous, status);
1192 }
1193
1194 static void packet_increment_head(struct packet_ring_buffer *buff)
1195 {
1196         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1197 }
1198
1199 static void packet_inc_pending(struct packet_ring_buffer *rb)
1200 {
1201         this_cpu_inc(*rb->pending_refcnt);
1202 }
1203
1204 static void packet_dec_pending(struct packet_ring_buffer *rb)
1205 {
1206         this_cpu_dec(*rb->pending_refcnt);
1207 }
1208
1209 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1210 {
1211         unsigned int refcnt = 0;
1212         int cpu;
1213
1214         /* We don't use pending refcount in rx_ring. */
1215         if (rb->pending_refcnt == NULL)
1216                 return 0;
1217
1218         for_each_possible_cpu(cpu)
1219                 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1220
1221         return refcnt;
1222 }
1223
1224 static int packet_alloc_pending(struct packet_sock *po)
1225 {
1226         po->rx_ring.pending_refcnt = NULL;
1227
1228         po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1229         if (unlikely(po->tx_ring.pending_refcnt == NULL))
1230                 return -ENOBUFS;
1231
1232         return 0;
1233 }
1234
1235 static void packet_free_pending(struct packet_sock *po)
1236 {
1237         free_percpu(po->tx_ring.pending_refcnt);
1238 }
1239
1240 #define ROOM_POW_OFF    2
1241 #define ROOM_NONE       0x0
1242 #define ROOM_LOW        0x1
1243 #define ROOM_NORMAL     0x2
1244
1245 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1246 {
1247         int idx, len;
1248
1249         len = po->rx_ring.frame_max + 1;
1250         idx = po->rx_ring.head;
1251         if (pow_off)
1252                 idx += len >> pow_off;
1253         if (idx >= len)
1254                 idx -= len;
1255         return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1256 }
1257
1258 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1259 {
1260         int idx, len;
1261
1262         len = po->rx_ring.prb_bdqc.knum_blocks;
1263         idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1264         if (pow_off)
1265                 idx += len >> pow_off;
1266         if (idx >= len)
1267                 idx -= len;
1268         return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1269 }
1270
1271 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1272 {
1273         struct sock *sk = &po->sk;
1274         int ret = ROOM_NONE;
1275
1276         if (po->prot_hook.func != tpacket_rcv) {
1277                 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1278                                           - (skb ? skb->truesize : 0);
1279                 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1280                         return ROOM_NORMAL;
1281                 else if (avail > 0)
1282                         return ROOM_LOW;
1283                 else
1284                         return ROOM_NONE;
1285         }
1286
1287         if (po->tp_version == TPACKET_V3) {
1288                 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1289                         ret = ROOM_NORMAL;
1290                 else if (__tpacket_v3_has_room(po, 0))
1291                         ret = ROOM_LOW;
1292         } else {
1293                 if (__tpacket_has_room(po, ROOM_POW_OFF))
1294                         ret = ROOM_NORMAL;
1295                 else if (__tpacket_has_room(po, 0))
1296                         ret = ROOM_LOW;
1297         }
1298
1299         return ret;
1300 }
1301
1302 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1303 {
1304         int ret;
1305         bool has_room;
1306
1307         spin_lock_bh(&po->sk.sk_receive_queue.lock);
1308         ret = __packet_rcv_has_room(po, skb);
1309         has_room = ret == ROOM_NORMAL;
1310         if (po->pressure == has_room)
1311                 po->pressure = !has_room;
1312         spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1313
1314         return ret;
1315 }
1316
1317 static void packet_sock_destruct(struct sock *sk)
1318 {
1319         skb_queue_purge(&sk->sk_error_queue);
1320
1321         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1322         WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1323
1324         if (!sock_flag(sk, SOCK_DEAD)) {
1325                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1326                 return;
1327         }
1328
1329         sk_refcnt_debug_dec(sk);
1330 }
1331
1332 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1333 {
1334         u32 rxhash;
1335         int i, count = 0;
1336
1337         rxhash = skb_get_hash(skb);
1338         for (i = 0; i < ROLLOVER_HLEN; i++)
1339                 if (po->rollover->history[i] == rxhash)
1340                         count++;
1341
1342         po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1343         return count > (ROLLOVER_HLEN >> 1);
1344 }
1345
1346 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1347                                       struct sk_buff *skb,
1348                                       unsigned int num)
1349 {
1350         return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1351 }
1352
1353 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1354                                     struct sk_buff *skb,
1355                                     unsigned int num)
1356 {
1357         unsigned int val = atomic_inc_return(&f->rr_cur);
1358
1359         return val % num;
1360 }
1361
1362 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1363                                      struct sk_buff *skb,
1364                                      unsigned int num)
1365 {
1366         return smp_processor_id() % num;
1367 }
1368
1369 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1370                                      struct sk_buff *skb,
1371                                      unsigned int num)
1372 {
1373         return prandom_u32_max(num);
1374 }
1375
1376 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1377                                           struct sk_buff *skb,
1378                                           unsigned int idx, bool try_self,
1379                                           unsigned int num)
1380 {
1381         struct packet_sock *po, *po_next, *po_skip = NULL;
1382         unsigned int i, j, room = ROOM_NONE;
1383
1384         po = pkt_sk(f->arr[idx]);
1385
1386         if (try_self) {
1387                 room = packet_rcv_has_room(po, skb);
1388                 if (room == ROOM_NORMAL ||
1389                     (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1390                         return idx;
1391                 po_skip = po;
1392         }
1393
1394         i = j = min_t(int, po->rollover->sock, num - 1);
1395         do {
1396                 po_next = pkt_sk(f->arr[i]);
1397                 if (po_next != po_skip && !po_next->pressure &&
1398                     packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1399                         if (i != j)
1400                                 po->rollover->sock = i;
1401                         atomic_long_inc(&po->rollover->num);
1402                         if (room == ROOM_LOW)
1403                                 atomic_long_inc(&po->rollover->num_huge);
1404                         return i;
1405                 }
1406
1407                 if (++i == num)
1408                         i = 0;
1409         } while (i != j);
1410
1411         atomic_long_inc(&po->rollover->num_failed);
1412         return idx;
1413 }
1414
1415 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1416                                     struct sk_buff *skb,
1417                                     unsigned int num)
1418 {
1419         return skb_get_queue_mapping(skb) % num;
1420 }
1421
1422 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1423                                      struct sk_buff *skb,
1424                                      unsigned int num)
1425 {
1426         struct bpf_prog *prog;
1427         unsigned int ret = 0;
1428
1429         rcu_read_lock();
1430         prog = rcu_dereference(f->bpf_prog);
1431         if (prog)
1432                 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1433         rcu_read_unlock();
1434
1435         return ret;
1436 }
1437
1438 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1439 {
1440         return f->flags & (flag >> 8);
1441 }
1442
1443 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1444                              struct packet_type *pt, struct net_device *orig_dev)
1445 {
1446         struct packet_fanout *f = pt->af_packet_priv;
1447         unsigned int num = READ_ONCE(f->num_members);
1448         struct net *net = read_pnet(&f->net);
1449         struct packet_sock *po;
1450         unsigned int idx;
1451
1452         if (!net_eq(dev_net(dev), net) || !num) {
1453                 kfree_skb(skb);
1454                 return 0;
1455         }
1456
1457         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1458                 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1459                 if (!skb)
1460                         return 0;
1461         }
1462         switch (f->type) {
1463         case PACKET_FANOUT_HASH:
1464         default:
1465                 idx = fanout_demux_hash(f, skb, num);
1466                 break;
1467         case PACKET_FANOUT_LB:
1468                 idx = fanout_demux_lb(f, skb, num);
1469                 break;
1470         case PACKET_FANOUT_CPU:
1471                 idx = fanout_demux_cpu(f, skb, num);
1472                 break;
1473         case PACKET_FANOUT_RND:
1474                 idx = fanout_demux_rnd(f, skb, num);
1475                 break;
1476         case PACKET_FANOUT_QM:
1477                 idx = fanout_demux_qm(f, skb, num);
1478                 break;
1479         case PACKET_FANOUT_ROLLOVER:
1480                 idx = fanout_demux_rollover(f, skb, 0, false, num);
1481                 break;
1482         case PACKET_FANOUT_CBPF:
1483         case PACKET_FANOUT_EBPF:
1484                 idx = fanout_demux_bpf(f, skb, num);
1485                 break;
1486         }
1487
1488         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1489                 idx = fanout_demux_rollover(f, skb, idx, true, num);
1490
1491         po = pkt_sk(f->arr[idx]);
1492         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1493 }
1494
1495 DEFINE_MUTEX(fanout_mutex);
1496 EXPORT_SYMBOL_GPL(fanout_mutex);
1497 static LIST_HEAD(fanout_list);
1498 static u16 fanout_next_id;
1499
1500 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1501 {
1502         struct packet_fanout *f = po->fanout;
1503
1504         spin_lock(&f->lock);
1505         f->arr[f->num_members] = sk;
1506         smp_wmb();
1507         f->num_members++;
1508         if (f->num_members == 1)
1509                 dev_add_pack(&f->prot_hook);
1510         spin_unlock(&f->lock);
1511 }
1512
1513 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1514 {
1515         struct packet_fanout *f = po->fanout;
1516         int i;
1517
1518         spin_lock(&f->lock);
1519         for (i = 0; i < f->num_members; i++) {
1520                 if (f->arr[i] == sk)
1521                         break;
1522         }
1523         BUG_ON(i >= f->num_members);
1524         f->arr[i] = f->arr[f->num_members - 1];
1525         f->num_members--;
1526         if (f->num_members == 0)
1527                 __dev_remove_pack(&f->prot_hook);
1528         spin_unlock(&f->lock);
1529 }
1530
1531 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1532 {
1533         if (sk->sk_family != PF_PACKET)
1534                 return false;
1535
1536         return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1537 }
1538
1539 static void fanout_init_data(struct packet_fanout *f)
1540 {
1541         switch (f->type) {
1542         case PACKET_FANOUT_LB:
1543                 atomic_set(&f->rr_cur, 0);
1544                 break;
1545         case PACKET_FANOUT_CBPF:
1546         case PACKET_FANOUT_EBPF:
1547                 RCU_INIT_POINTER(f->bpf_prog, NULL);
1548                 break;
1549         }
1550 }
1551
1552 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1553 {
1554         struct bpf_prog *old;
1555
1556         spin_lock(&f->lock);
1557         old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1558         rcu_assign_pointer(f->bpf_prog, new);
1559         spin_unlock(&f->lock);
1560
1561         if (old) {
1562                 synchronize_net();
1563                 bpf_prog_destroy(old);
1564         }
1565 }
1566
1567 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1568                                 unsigned int len)
1569 {
1570         struct bpf_prog *new;
1571         struct sock_fprog fprog;
1572         int ret;
1573
1574         if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1575                 return -EPERM;
1576         if (len != sizeof(fprog))
1577                 return -EINVAL;
1578         if (copy_from_user(&fprog, data, len))
1579                 return -EFAULT;
1580
1581         ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1582         if (ret)
1583                 return ret;
1584
1585         __fanout_set_data_bpf(po->fanout, new);
1586         return 0;
1587 }
1588
1589 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1590                                 unsigned int len)
1591 {
1592         struct bpf_prog *new;
1593         u32 fd;
1594
1595         if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1596                 return -EPERM;
1597         if (len != sizeof(fd))
1598                 return -EINVAL;
1599         if (copy_from_user(&fd, data, len))
1600                 return -EFAULT;
1601
1602         new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1603         if (IS_ERR(new))
1604                 return PTR_ERR(new);
1605
1606         __fanout_set_data_bpf(po->fanout, new);
1607         return 0;
1608 }
1609
1610 static int fanout_set_data(struct packet_sock *po, char __user *data,
1611                            unsigned int len)
1612 {
1613         switch (po->fanout->type) {
1614         case PACKET_FANOUT_CBPF:
1615                 return fanout_set_data_cbpf(po, data, len);
1616         case PACKET_FANOUT_EBPF:
1617                 return fanout_set_data_ebpf(po, data, len);
1618         default:
1619                 return -EINVAL;
1620         };
1621 }
1622
1623 static void fanout_release_data(struct packet_fanout *f)
1624 {
1625         switch (f->type) {
1626         case PACKET_FANOUT_CBPF:
1627         case PACKET_FANOUT_EBPF:
1628                 __fanout_set_data_bpf(f, NULL);
1629         };
1630 }
1631
1632 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1633 {
1634         struct packet_fanout *f;
1635
1636         list_for_each_entry(f, &fanout_list, list) {
1637                 if (f->id == candidate_id &&
1638                     read_pnet(&f->net) == sock_net(sk)) {
1639                         return false;
1640                 }
1641         }
1642         return true;
1643 }
1644
1645 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1646 {
1647         u16 id = fanout_next_id;
1648
1649         do {
1650                 if (__fanout_id_is_free(sk, id)) {
1651                         *new_id = id;
1652                         fanout_next_id = id + 1;
1653                         return true;
1654                 }
1655
1656                 id++;
1657         } while (id != fanout_next_id);
1658
1659         return false;
1660 }
1661
1662 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1663 {
1664         struct packet_rollover *rollover = NULL;
1665         struct packet_sock *po = pkt_sk(sk);
1666         struct packet_fanout *f, *match;
1667         u8 type = type_flags & 0xff;
1668         u8 flags = type_flags >> 8;
1669         int err;
1670
1671         switch (type) {
1672         case PACKET_FANOUT_ROLLOVER:
1673                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1674                         return -EINVAL;
1675         case PACKET_FANOUT_HASH:
1676         case PACKET_FANOUT_LB:
1677         case PACKET_FANOUT_CPU:
1678         case PACKET_FANOUT_RND:
1679         case PACKET_FANOUT_QM:
1680         case PACKET_FANOUT_CBPF:
1681         case PACKET_FANOUT_EBPF:
1682                 break;
1683         default:
1684                 return -EINVAL;
1685         }
1686
1687         mutex_lock(&fanout_mutex);
1688
1689         err = -EINVAL;
1690         if (!po->running)
1691                 goto out;
1692
1693         err = -EALREADY;
1694         if (po->fanout)
1695                 goto out;
1696
1697         if (type == PACKET_FANOUT_ROLLOVER ||
1698             (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1699                 err = -ENOMEM;
1700                 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1701                 if (!rollover)
1702                         goto out;
1703                 atomic_long_set(&rollover->num, 0);
1704                 atomic_long_set(&rollover->num_huge, 0);
1705                 atomic_long_set(&rollover->num_failed, 0);
1706                 po->rollover = rollover;
1707         }
1708
1709         if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1710                 if (id != 0) {
1711                         err = -EINVAL;
1712                         goto out;
1713                 }
1714                 if (!fanout_find_new_id(sk, &id)) {
1715                         err = -ENOMEM;
1716                         goto out;
1717                 }
1718                 /* ephemeral flag for the first socket in the group: drop it */
1719                 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1720         }
1721
1722         match = NULL;
1723         list_for_each_entry(f, &fanout_list, list) {
1724                 if (f->id == id &&
1725                     read_pnet(&f->net) == sock_net(sk)) {
1726                         match = f;
1727                         break;
1728                 }
1729         }
1730         err = -EINVAL;
1731         if (match && match->flags != flags)
1732                 goto out;
1733         if (!match) {
1734                 err = -ENOMEM;
1735                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1736                 if (!match)
1737                         goto out;
1738                 write_pnet(&match->net, sock_net(sk));
1739                 match->id = id;
1740                 match->type = type;
1741                 match->flags = flags;
1742                 INIT_LIST_HEAD(&match->list);
1743                 spin_lock_init(&match->lock);
1744                 refcount_set(&match->sk_ref, 0);
1745                 fanout_init_data(match);
1746                 match->prot_hook.type = po->prot_hook.type;
1747                 match->prot_hook.dev = po->prot_hook.dev;
1748                 match->prot_hook.func = packet_rcv_fanout;
1749                 match->prot_hook.af_packet_priv = match;
1750                 match->prot_hook.id_match = match_fanout_group;
1751                 list_add(&match->list, &fanout_list);
1752         }
1753         err = -EINVAL;
1754         if (match->type == type &&
1755             match->prot_hook.type == po->prot_hook.type &&
1756             match->prot_hook.dev == po->prot_hook.dev) {
1757                 err = -ENOSPC;
1758                 if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1759                         __dev_remove_pack(&po->prot_hook);
1760                         po->fanout = match;
1761                         refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1762                         __fanout_link(sk, po);
1763                         err = 0;
1764                 }
1765         }
1766 out:
1767         if (err && rollover) {
1768                 kfree(rollover);
1769                 po->rollover = NULL;
1770         }
1771         mutex_unlock(&fanout_mutex);
1772         return err;
1773 }
1774
1775 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1776  * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1777  * It is the responsibility of the caller to call fanout_release_data() and
1778  * free the returned packet_fanout (after synchronize_net())
1779  */
1780 static struct packet_fanout *fanout_release(struct sock *sk)
1781 {
1782         struct packet_sock *po = pkt_sk(sk);
1783         struct packet_fanout *f;
1784
1785         mutex_lock(&fanout_mutex);
1786         f = po->fanout;
1787         if (f) {
1788                 po->fanout = NULL;
1789
1790                 if (refcount_dec_and_test(&f->sk_ref))
1791                         list_del(&f->list);
1792                 else
1793                         f = NULL;
1794
1795                 if (po->rollover)
1796                         kfree_rcu(po->rollover, rcu);
1797         }
1798         mutex_unlock(&fanout_mutex);
1799
1800         return f;
1801 }
1802
1803 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1804                                           struct sk_buff *skb)
1805 {
1806         /* Earlier code assumed this would be a VLAN pkt, double-check
1807          * this now that we have the actual packet in hand. We can only
1808          * do this check on Ethernet devices.
1809          */
1810         if (unlikely(dev->type != ARPHRD_ETHER))
1811                 return false;
1812
1813         skb_reset_mac_header(skb);
1814         return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1815 }
1816
1817 static const struct proto_ops packet_ops;
1818
1819 static const struct proto_ops packet_ops_spkt;
1820
1821 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1822                            struct packet_type *pt, struct net_device *orig_dev)
1823 {
1824         struct sock *sk;
1825         struct sockaddr_pkt *spkt;
1826
1827         /*
1828          *      When we registered the protocol we saved the socket in the data
1829          *      field for just this event.
1830          */
1831
1832         sk = pt->af_packet_priv;
1833
1834         /*
1835          *      Yank back the headers [hope the device set this
1836          *      right or kerboom...]
1837          *
1838          *      Incoming packets have ll header pulled,
1839          *      push it back.
1840          *
1841          *      For outgoing ones skb->data == skb_mac_header(skb)
1842          *      so that this procedure is noop.
1843          */
1844
1845         if (skb->pkt_type == PACKET_LOOPBACK)
1846                 goto out;
1847
1848         if (!net_eq(dev_net(dev), sock_net(sk)))
1849                 goto out;
1850
1851         skb = skb_share_check(skb, GFP_ATOMIC);
1852         if (skb == NULL)
1853                 goto oom;
1854
1855         /* drop any routing info */
1856         skb_dst_drop(skb);
1857
1858         /* drop conntrack reference */
1859         nf_reset(skb);
1860
1861         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1862
1863         skb_push(skb, skb->data - skb_mac_header(skb));
1864
1865         /*
1866          *      The SOCK_PACKET socket receives _all_ frames.
1867          */
1868
1869         spkt->spkt_family = dev->type;
1870         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1871         spkt->spkt_protocol = skb->protocol;
1872
1873         /*
1874          *      Charge the memory to the socket. This is done specifically
1875          *      to prevent sockets using all the memory up.
1876          */
1877
1878         if (sock_queue_rcv_skb(sk, skb) == 0)
1879                 return 0;
1880
1881 out:
1882         kfree_skb(skb);
1883 oom:
1884         return 0;
1885 }
1886
1887
1888 /*
1889  *      Output a raw packet to a device layer. This bypasses all the other
1890  *      protocol layers and you must therefore supply it with a complete frame
1891  */
1892
1893 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1894                                size_t len)
1895 {
1896         struct sock *sk = sock->sk;
1897         DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1898         struct sk_buff *skb = NULL;
1899         struct net_device *dev;
1900         struct sockcm_cookie sockc;
1901         __be16 proto = 0;
1902         int err;
1903         int extra_len = 0;
1904
1905         /*
1906          *      Get and verify the address.
1907          */
1908
1909         if (saddr) {
1910                 if (msg->msg_namelen < sizeof(struct sockaddr))
1911                         return -EINVAL;
1912                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1913                         proto = saddr->spkt_protocol;
1914         } else
1915                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1916
1917         /*
1918          *      Find the device first to size check it
1919          */
1920
1921         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1922 retry:
1923         rcu_read_lock();
1924         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1925         err = -ENODEV;
1926         if (dev == NULL)
1927                 goto out_unlock;
1928
1929         err = -ENETDOWN;
1930         if (!(dev->flags & IFF_UP))
1931                 goto out_unlock;
1932
1933         /*
1934          * You may not queue a frame bigger than the mtu. This is the lowest level
1935          * raw protocol and you must do your own fragmentation at this level.
1936          */
1937
1938         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1939                 if (!netif_supports_nofcs(dev)) {
1940                         err = -EPROTONOSUPPORT;
1941                         goto out_unlock;
1942                 }
1943                 extra_len = 4; /* We're doing our own CRC */
1944         }
1945
1946         err = -EMSGSIZE;
1947         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1948                 goto out_unlock;
1949
1950         if (!skb) {
1951                 size_t reserved = LL_RESERVED_SPACE(dev);
1952                 int tlen = dev->needed_tailroom;
1953                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1954
1955                 rcu_read_unlock();
1956                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1957                 if (skb == NULL)
1958                         return -ENOBUFS;
1959                 /* FIXME: Save some space for broken drivers that write a hard
1960                  * header at transmission time by themselves. PPP is the notable
1961                  * one here. This should really be fixed at the driver level.
1962                  */
1963                 skb_reserve(skb, reserved);
1964                 skb_reset_network_header(skb);
1965
1966                 /* Try to align data part correctly */
1967                 if (hhlen) {
1968                         skb->data -= hhlen;
1969                         skb->tail -= hhlen;
1970                         if (len < hhlen)
1971                                 skb_reset_network_header(skb);
1972                 }
1973                 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1974                 if (err)
1975                         goto out_free;
1976                 goto retry;
1977         }
1978
1979         if (!dev_validate_header(dev, skb->data, len)) {
1980                 err = -EINVAL;
1981                 goto out_unlock;
1982         }
1983         if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1984             !packet_extra_vlan_len_allowed(dev, skb)) {
1985                 err = -EMSGSIZE;
1986                 goto out_unlock;
1987         }
1988
1989         sockc.tsflags = sk->sk_tsflags;
1990         if (msg->msg_controllen) {
1991                 err = sock_cmsg_send(sk, msg, &sockc);
1992                 if (unlikely(err))
1993                         goto out_unlock;
1994         }
1995
1996         skb->protocol = proto;
1997         skb->dev = dev;
1998         skb->priority = sk->sk_priority;
1999         skb->mark = sk->sk_mark;
2000
2001         sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
2002
2003         if (unlikely(extra_len == 4))
2004                 skb->no_fcs = 1;
2005
2006         skb_probe_transport_header(skb, 0);
2007
2008         dev_queue_xmit(skb);
2009         rcu_read_unlock();
2010         return len;
2011
2012 out_unlock:
2013         rcu_read_unlock();
2014 out_free:
2015         kfree_skb(skb);
2016         return err;
2017 }
2018
2019 static unsigned int run_filter(struct sk_buff *skb,
2020                                const struct sock *sk,
2021                                unsigned int res)
2022 {
2023         struct sk_filter *filter;
2024
2025         rcu_read_lock();
2026         filter = rcu_dereference(sk->sk_filter);
2027         if (filter != NULL)
2028                 res = bpf_prog_run_clear_cb(filter->prog, skb);
2029         rcu_read_unlock();
2030
2031         return res;
2032 }
2033
2034 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2035                            size_t *len)
2036 {
2037         struct virtio_net_hdr vnet_hdr;
2038
2039         if (*len < sizeof(vnet_hdr))
2040                 return -EINVAL;
2041         *len -= sizeof(vnet_hdr);
2042
2043         if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true))
2044                 return -EINVAL;
2045
2046         return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2047 }
2048
2049 /*
2050  * This function makes lazy skb cloning in hope that most of packets
2051  * are discarded by BPF.
2052  *
2053  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2054  * and skb->cb are mangled. It works because (and until) packets
2055  * falling here are owned by current CPU. Output packets are cloned
2056  * by dev_queue_xmit_nit(), input packets are processed by net_bh
2057  * sequencially, so that if we return skb to original state on exit,
2058  * we will not harm anyone.
2059  */
2060
2061 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2062                       struct packet_type *pt, struct net_device *orig_dev)
2063 {
2064         struct sock *sk;
2065         struct sockaddr_ll *sll;
2066         struct packet_sock *po;
2067         u8 *skb_head = skb->data;
2068         int skb_len = skb->len;
2069         unsigned int snaplen, res;
2070         bool is_drop_n_account = false;
2071
2072         if (skb->pkt_type == PACKET_LOOPBACK)
2073                 goto drop;
2074
2075         sk = pt->af_packet_priv;
2076         po = pkt_sk(sk);
2077
2078         if (!net_eq(dev_net(dev), sock_net(sk)))
2079                 goto drop;
2080
2081         skb->dev = dev;
2082
2083         if (dev->header_ops) {
2084                 /* The device has an explicit notion of ll header,
2085                  * exported to higher levels.
2086                  *
2087                  * Otherwise, the device hides details of its frame
2088                  * structure, so that corresponding packet head is
2089                  * never delivered to user.
2090                  */
2091                 if (sk->sk_type != SOCK_DGRAM)
2092                         skb_push(skb, skb->data - skb_mac_header(skb));
2093                 else if (skb->pkt_type == PACKET_OUTGOING) {
2094                         /* Special case: outgoing packets have ll header at head */
2095                         skb_pull(skb, skb_network_offset(skb));
2096                 }
2097         }
2098
2099         snaplen = skb->len;
2100
2101         res = run_filter(skb, sk, snaplen);
2102         if (!res)
2103                 goto drop_n_restore;
2104         if (snaplen > res)
2105                 snaplen = res;
2106
2107         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2108                 goto drop_n_acct;
2109
2110         if (skb_shared(skb)) {
2111                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2112                 if (nskb == NULL)
2113                         goto drop_n_acct;
2114
2115                 if (skb_head != skb->data) {
2116                         skb->data = skb_head;
2117                         skb->len = skb_len;
2118                 }
2119                 consume_skb(skb);
2120                 skb = nskb;
2121         }
2122
2123         sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2124
2125         sll = &PACKET_SKB_CB(skb)->sa.ll;
2126         sll->sll_hatype = dev->type;
2127         sll->sll_pkttype = skb->pkt_type;
2128         if (unlikely(po->origdev))
2129                 sll->sll_ifindex = orig_dev->ifindex;
2130         else
2131                 sll->sll_ifindex = dev->ifindex;
2132
2133         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2134
2135         /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2136          * Use their space for storing the original skb length.
2137          */
2138         PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2139
2140         if (pskb_trim(skb, snaplen))
2141                 goto drop_n_acct;
2142
2143         skb_set_owner_r(skb, sk);
2144         skb->dev = NULL;
2145         skb_dst_drop(skb);
2146
2147         /* drop conntrack reference */
2148         nf_reset(skb);
2149
2150         spin_lock(&sk->sk_receive_queue.lock);
2151         po->stats.stats1.tp_packets++;
2152         sock_skb_set_dropcount(sk, skb);
2153         __skb_queue_tail(&sk->sk_receive_queue, skb);
2154         spin_unlock(&sk->sk_receive_queue.lock);
2155         sk->sk_data_ready(sk);
2156         return 0;
2157
2158 drop_n_acct:
2159         is_drop_n_account = true;
2160         spin_lock(&sk->sk_receive_queue.lock);
2161         po->stats.stats1.tp_drops++;
2162         atomic_inc(&sk->sk_drops);
2163         spin_unlock(&sk->sk_receive_queue.lock);
2164
2165 drop_n_restore:
2166         if (skb_head != skb->data && skb_shared(skb)) {
2167                 skb->data = skb_head;
2168                 skb->len = skb_len;
2169         }
2170 drop:
2171         if (!is_drop_n_account)
2172                 consume_skb(skb);
2173         else
2174                 kfree_skb(skb);
2175         return 0;
2176 }
2177
2178 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2179                        struct packet_type *pt, struct net_device *orig_dev)
2180 {
2181         struct sock *sk;
2182         struct packet_sock *po;
2183         struct sockaddr_ll *sll;
2184         union tpacket_uhdr h;
2185         u8 *skb_head = skb->data;
2186         int skb_len = skb->len;
2187         unsigned int snaplen, res;
2188         unsigned long status = TP_STATUS_USER;
2189         unsigned short macoff, netoff, hdrlen;
2190         struct sk_buff *copy_skb = NULL;
2191         struct timespec ts;
2192         __u32 ts_status;
2193         bool is_drop_n_account = false;
2194         bool do_vnet = false;
2195
2196         /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2197          * We may add members to them until current aligned size without forcing
2198          * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2199          */
2200         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2201         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2202
2203         if (skb->pkt_type == PACKET_LOOPBACK)
2204                 goto drop;
2205
2206         sk = pt->af_packet_priv;
2207         po = pkt_sk(sk);
2208
2209         if (!net_eq(dev_net(dev), sock_net(sk)))
2210                 goto drop;
2211
2212         if (dev->header_ops) {
2213                 if (sk->sk_type != SOCK_DGRAM)
2214                         skb_push(skb, skb->data - skb_mac_header(skb));
2215                 else if (skb->pkt_type == PACKET_OUTGOING) {
2216                         /* Special case: outgoing packets have ll header at head */
2217                         skb_pull(skb, skb_network_offset(skb));
2218                 }
2219         }
2220
2221         snaplen = skb->len;
2222
2223         res = run_filter(skb, sk, snaplen);
2224         if (!res)
2225                 goto drop_n_restore;
2226
2227         if (skb->ip_summed == CHECKSUM_PARTIAL)
2228                 status |= TP_STATUS_CSUMNOTREADY;
2229         else if (skb->pkt_type != PACKET_OUTGOING &&
2230                  (skb->ip_summed == CHECKSUM_COMPLETE ||
2231                   skb_csum_unnecessary(skb)))
2232                 status |= TP_STATUS_CSUM_VALID;
2233
2234         if (snaplen > res)
2235                 snaplen = res;
2236
2237         if (sk->sk_type == SOCK_DGRAM) {
2238                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2239                                   po->tp_reserve;
2240         } else {
2241                 unsigned int maclen = skb_network_offset(skb);
2242                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2243                                        (maclen < 16 ? 16 : maclen)) +
2244                                        po->tp_reserve;
2245                 if (po->has_vnet_hdr) {
2246                         netoff += sizeof(struct virtio_net_hdr);
2247                         do_vnet = true;
2248                 }
2249                 macoff = netoff - maclen;
2250         }
2251         if (po->tp_version <= TPACKET_V2) {
2252                 if (macoff + snaplen > po->rx_ring.frame_size) {
2253                         if (po->copy_thresh &&
2254                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2255                                 if (skb_shared(skb)) {
2256                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
2257                                 } else {
2258                                         copy_skb = skb_get(skb);
2259                                         skb_head = skb->data;
2260                                 }
2261                                 if (copy_skb)
2262                                         skb_set_owner_r(copy_skb, sk);
2263                         }
2264                         snaplen = po->rx_ring.frame_size - macoff;
2265                         if ((int)snaplen < 0) {
2266                                 snaplen = 0;
2267                                 do_vnet = false;
2268                         }
2269                 }
2270         } else if (unlikely(macoff + snaplen >
2271                             GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2272                 u32 nval;
2273
2274                 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2275                 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2276                             snaplen, nval, macoff);
2277                 snaplen = nval;
2278                 if (unlikely((int)snaplen < 0)) {
2279                         snaplen = 0;
2280                         macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2281                         do_vnet = false;
2282                 }
2283         }
2284         spin_lock(&sk->sk_receive_queue.lock);
2285         h.raw = packet_current_rx_frame(po, skb,
2286                                         TP_STATUS_KERNEL, (macoff+snaplen));
2287         if (!h.raw)
2288                 goto drop_n_account;
2289         if (po->tp_version <= TPACKET_V2) {
2290                 packet_increment_rx_head(po, &po->rx_ring);
2291         /*
2292          * LOSING will be reported till you read the stats,
2293          * because it's COR - Clear On Read.
2294          * Anyways, moving it for V1/V2 only as V3 doesn't need this
2295          * at packet level.
2296          */
2297                 if (po->stats.stats1.tp_drops)
2298                         status |= TP_STATUS_LOSING;
2299         }
2300         po->stats.stats1.tp_packets++;
2301         if (copy_skb) {
2302                 status |= TP_STATUS_COPY;
2303                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2304         }
2305         spin_unlock(&sk->sk_receive_queue.lock);
2306
2307         if (do_vnet) {
2308                 if (virtio_net_hdr_from_skb(skb, h.raw + macoff -
2309                                             sizeof(struct virtio_net_hdr),
2310                                             vio_le(), true)) {
2311                         spin_lock(&sk->sk_receive_queue.lock);
2312                         goto drop_n_account;
2313                 }
2314         }
2315
2316         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2317
2318         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2319                 getnstimeofday(&ts);
2320
2321         status |= ts_status;
2322
2323         switch (po->tp_version) {
2324         case TPACKET_V1:
2325                 h.h1->tp_len = skb->len;
2326                 h.h1->tp_snaplen = snaplen;
2327                 h.h1->tp_mac = macoff;
2328                 h.h1->tp_net = netoff;
2329                 h.h1->tp_sec = ts.tv_sec;
2330                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2331                 hdrlen = sizeof(*h.h1);
2332                 break;
2333         case TPACKET_V2:
2334                 h.h2->tp_len = skb->len;
2335                 h.h2->tp_snaplen = snaplen;
2336                 h.h2->tp_mac = macoff;
2337                 h.h2->tp_net = netoff;
2338                 h.h2->tp_sec = ts.tv_sec;
2339                 h.h2->tp_nsec = ts.tv_nsec;
2340                 if (skb_vlan_tag_present(skb)) {
2341                         h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2342                         h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2343                         status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2344                 } else {
2345                         h.h2->tp_vlan_tci = 0;
2346                         h.h2->tp_vlan_tpid = 0;
2347                 }
2348                 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2349                 hdrlen = sizeof(*h.h2);
2350                 break;
2351         case TPACKET_V3:
2352                 /* tp_nxt_offset,vlan are already populated above.
2353                  * So DONT clear those fields here
2354                  */
2355                 h.h3->tp_status |= status;
2356                 h.h3->tp_len = skb->len;
2357                 h.h3->tp_snaplen = snaplen;
2358                 h.h3->tp_mac = macoff;
2359                 h.h3->tp_net = netoff;
2360                 h.h3->tp_sec  = ts.tv_sec;
2361                 h.h3->tp_nsec = ts.tv_nsec;
2362                 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2363                 hdrlen = sizeof(*h.h3);
2364                 break;
2365         default:
2366                 BUG();
2367         }
2368
2369         sll = h.raw + TPACKET_ALIGN(hdrlen);
2370         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2371         sll->sll_family = AF_PACKET;
2372         sll->sll_hatype = dev->type;
2373         sll->sll_protocol = skb->protocol;
2374         sll->sll_pkttype = skb->pkt_type;
2375         if (unlikely(po->origdev))
2376                 sll->sll_ifindex = orig_dev->ifindex;
2377         else
2378                 sll->sll_ifindex = dev->ifindex;
2379
2380         smp_mb();
2381
2382 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2383         if (po->tp_version <= TPACKET_V2) {
2384                 u8 *start, *end;
2385
2386                 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2387                                         macoff + snaplen);
2388
2389                 for (start = h.raw; start < end; start += PAGE_SIZE)
2390                         flush_dcache_page(pgv_to_page(start));
2391         }
2392         smp_wmb();
2393 #endif
2394
2395         if (po->tp_version <= TPACKET_V2) {
2396                 __packet_set_status(po, h.raw, status);
2397                 sk->sk_data_ready(sk);
2398         } else {
2399                 prb_clear_blk_fill_status(&po->rx_ring);
2400         }
2401
2402 drop_n_restore:
2403         if (skb_head != skb->data && skb_shared(skb)) {
2404                 skb->data = skb_head;
2405                 skb->len = skb_len;
2406         }
2407 drop:
2408         if (!is_drop_n_account)
2409                 consume_skb(skb);
2410         else
2411                 kfree_skb(skb);
2412         return 0;
2413
2414 drop_n_account:
2415         is_drop_n_account = true;
2416         po->stats.stats1.tp_drops++;
2417         spin_unlock(&sk->sk_receive_queue.lock);
2418
2419         sk->sk_data_ready(sk);
2420         kfree_skb(copy_skb);
2421         goto drop_n_restore;
2422 }
2423
2424 static void tpacket_destruct_skb(struct sk_buff *skb)
2425 {
2426         struct packet_sock *po = pkt_sk(skb->sk);
2427
2428         if (likely(po->tx_ring.pg_vec)) {
2429                 void *ph;
2430                 __u32 ts;
2431
2432                 ph = skb_shinfo(skb)->destructor_arg;
2433                 packet_dec_pending(&po->tx_ring);
2434
2435                 ts = __packet_set_timestamp(po, ph, skb);
2436                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2437         }
2438
2439         sock_wfree(skb);
2440 }
2441
2442 static void tpacket_set_protocol(const struct net_device *dev,
2443                                  struct sk_buff *skb)
2444 {
2445         if (dev->type == ARPHRD_ETHER) {
2446                 skb_reset_mac_header(skb);
2447                 skb->protocol = eth_hdr(skb)->h_proto;
2448         }
2449 }
2450
2451 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2452 {
2453         if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2454             (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2455              __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2456               __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2457                 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2458                          __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2459                         __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2460
2461         if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2462                 return -EINVAL;
2463
2464         return 0;
2465 }
2466
2467 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2468                                  struct virtio_net_hdr *vnet_hdr)
2469 {
2470         if (*len < sizeof(*vnet_hdr))
2471                 return -EINVAL;
2472         *len -= sizeof(*vnet_hdr);
2473
2474         if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2475                 return -EFAULT;
2476
2477         return __packet_snd_vnet_parse(vnet_hdr, *len);
2478 }
2479
2480 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2481                 void *frame, struct net_device *dev, void *data, int tp_len,
2482                 __be16 proto, unsigned char *addr, int hlen, int copylen,
2483                 const struct sockcm_cookie *sockc)
2484 {
2485         union tpacket_uhdr ph;
2486         int to_write, offset, len, nr_frags, len_max;
2487         struct socket *sock = po->sk.sk_socket;
2488         struct page *page;
2489         int err;
2490
2491         ph.raw = frame;
2492
2493         skb->protocol = proto;
2494         skb->dev = dev;
2495         skb->priority = po->sk.sk_priority;
2496         skb->mark = po->sk.sk_mark;
2497         sock_tx_timestamp(&po->sk, sockc->tsflags, &skb_shinfo(skb)->tx_flags);
2498         skb_shinfo(skb)->destructor_arg = ph.raw;
2499
2500         skb_reserve(skb, hlen);
2501         skb_reset_network_header(skb);
2502
2503         to_write = tp_len;
2504
2505         if (sock->type == SOCK_DGRAM) {
2506                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2507                                 NULL, tp_len);
2508                 if (unlikely(err < 0))
2509                         return -EINVAL;
2510         } else if (copylen) {
2511                 int hdrlen = min_t(int, copylen, tp_len);
2512
2513                 skb_push(skb, dev->hard_header_len);
2514                 skb_put(skb, copylen - dev->hard_header_len);
2515                 err = skb_store_bits(skb, 0, data, hdrlen);
2516                 if (unlikely(err))
2517                         return err;
2518                 if (!dev_validate_header(dev, skb->data, hdrlen))
2519                         return -EINVAL;
2520                 if (!skb->protocol)
2521                         tpacket_set_protocol(dev, skb);
2522
2523                 data += hdrlen;
2524                 to_write -= hdrlen;
2525         }
2526
2527         offset = offset_in_page(data);
2528         len_max = PAGE_SIZE - offset;
2529         len = ((to_write > len_max) ? len_max : to_write);
2530
2531         skb->data_len = to_write;
2532         skb->len += to_write;
2533         skb->truesize += to_write;
2534         refcount_add(to_write, &po->sk.sk_wmem_alloc);
2535
2536         while (likely(to_write)) {
2537                 nr_frags = skb_shinfo(skb)->nr_frags;
2538
2539                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2540                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2541                                MAX_SKB_FRAGS);
2542                         return -EFAULT;
2543                 }
2544
2545                 page = pgv_to_page(data);
2546                 data += len;
2547                 flush_dcache_page(page);
2548                 get_page(page);
2549                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2550                 to_write -= len;
2551                 offset = 0;
2552                 len_max = PAGE_SIZE;
2553                 len = ((to_write > len_max) ? len_max : to_write);
2554         }
2555
2556         skb_probe_transport_header(skb, 0);
2557
2558         return tp_len;
2559 }
2560
2561 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2562                                 int size_max, void **data)
2563 {
2564         union tpacket_uhdr ph;
2565         int tp_len, off;
2566
2567         ph.raw = frame;
2568
2569         switch (po->tp_version) {
2570         case TPACKET_V3:
2571                 if (ph.h3->tp_next_offset != 0) {
2572                         pr_warn_once("variable sized slot not supported");
2573                         return -EINVAL;
2574                 }
2575                 tp_len = ph.h3->tp_len;
2576                 break;
2577         case TPACKET_V2:
2578                 tp_len = ph.h2->tp_len;
2579                 break;
2580         default:
2581                 tp_len = ph.h1->tp_len;
2582                 break;
2583         }
2584         if (unlikely(tp_len > size_max)) {
2585                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2586                 return -EMSGSIZE;
2587         }
2588
2589         if (unlikely(po->tp_tx_has_off)) {
2590                 int off_min, off_max;
2591
2592                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2593                 off_max = po->tx_ring.frame_size - tp_len;
2594                 if (po->sk.sk_type == SOCK_DGRAM) {
2595                         switch (po->tp_version) {
2596                         case TPACKET_V3:
2597                                 off = ph.h3->tp_net;
2598                                 break;
2599                         case TPACKET_V2:
2600                                 off = ph.h2->tp_net;
2601                                 break;
2602                         default:
2603                                 off = ph.h1->tp_net;
2604                                 break;
2605                         }
2606                 } else {
2607                         switch (po->tp_version) {
2608                         case TPACKET_V3:
2609                                 off = ph.h3->tp_mac;
2610                                 break;
2611                         case TPACKET_V2:
2612                                 off = ph.h2->tp_mac;
2613                                 break;
2614                         default:
2615                                 off = ph.h1->tp_mac;
2616                                 break;
2617                         }
2618                 }
2619                 if (unlikely((off < off_min) || (off_max < off)))
2620                         return -EINVAL;
2621         } else {
2622                 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2623         }
2624
2625         *data = frame + off;
2626         return tp_len;
2627 }
2628
2629 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2630 {
2631         struct sk_buff *skb;
2632         struct net_device *dev;
2633         struct virtio_net_hdr *vnet_hdr = NULL;
2634         struct sockcm_cookie sockc;
2635         __be16 proto;
2636         int err, reserve = 0;
2637         void *ph;
2638         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2639         bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2640         int tp_len, size_max;
2641         unsigned char *addr;
2642         void *data;
2643         int len_sum = 0;
2644         int status = TP_STATUS_AVAILABLE;
2645         int hlen, tlen, copylen = 0;
2646
2647         mutex_lock(&po->pg_vec_lock);
2648
2649         if (likely(saddr == NULL)) {
2650                 dev     = packet_cached_dev_get(po);
2651                 proto   = po->num;
2652                 addr    = NULL;
2653         } else {
2654                 err = -EINVAL;
2655                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2656                         goto out;
2657                 if (msg->msg_namelen < (saddr->sll_halen
2658                                         + offsetof(struct sockaddr_ll,
2659                                                 sll_addr)))
2660                         goto out;
2661                 proto   = saddr->sll_protocol;
2662                 addr    = saddr->sll_addr;
2663                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2664         }
2665
2666         err = -ENXIO;
2667         if (unlikely(dev == NULL))
2668                 goto out;
2669         err = -ENETDOWN;
2670         if (unlikely(!(dev->flags & IFF_UP)))
2671                 goto out_put;
2672
2673         sockc.tsflags = po->sk.sk_tsflags;
2674         if (msg->msg_controllen) {
2675                 err = sock_cmsg_send(&po->sk, msg, &sockc);
2676                 if (unlikely(err))
2677                         goto out_put;
2678         }
2679
2680         if (po->sk.sk_socket->type == SOCK_RAW)
2681                 reserve = dev->hard_header_len;
2682         size_max = po->tx_ring.frame_size
2683                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2684
2685         if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2686                 size_max = dev->mtu + reserve + VLAN_HLEN;
2687
2688         do {
2689                 ph = packet_current_frame(po, &po->tx_ring,
2690                                           TP_STATUS_SEND_REQUEST);
2691                 if (unlikely(ph == NULL)) {
2692                         if (need_wait && need_resched())
2693                                 schedule();
2694                         continue;
2695                 }
2696
2697                 skb = NULL;
2698                 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2699                 if (tp_len < 0)
2700                         goto tpacket_error;
2701
2702                 status = TP_STATUS_SEND_REQUEST;
2703                 hlen = LL_RESERVED_SPACE(dev);
2704                 tlen = dev->needed_tailroom;
2705                 if (po->has_vnet_hdr) {
2706                         vnet_hdr = data;
2707                         data += sizeof(*vnet_hdr);
2708                         tp_len -= sizeof(*vnet_hdr);
2709                         if (tp_len < 0 ||
2710                             __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2711                                 tp_len = -EINVAL;
2712                                 goto tpacket_error;
2713                         }
2714                         copylen = __virtio16_to_cpu(vio_le(),
2715                                                     vnet_hdr->hdr_len);
2716                 }
2717                 copylen = max_t(int, copylen, dev->hard_header_len);
2718                 skb = sock_alloc_send_skb(&po->sk,
2719                                 hlen + tlen + sizeof(struct sockaddr_ll) +
2720                                 (copylen - dev->hard_header_len),
2721                                 !need_wait, &err);
2722
2723                 if (unlikely(skb == NULL)) {
2724                         /* we assume the socket was initially writeable ... */
2725                         if (likely(len_sum > 0))
2726                                 err = len_sum;
2727                         goto out_status;
2728                 }
2729                 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2730                                           addr, hlen, copylen, &sockc);
2731                 if (likely(tp_len >= 0) &&
2732                     tp_len > dev->mtu + reserve &&
2733                     !po->has_vnet_hdr &&
2734                     !packet_extra_vlan_len_allowed(dev, skb))
2735                         tp_len = -EMSGSIZE;
2736
2737                 if (unlikely(tp_len < 0)) {
2738 tpacket_error:
2739                         if (po->tp_loss) {
2740                                 __packet_set_status(po, ph,
2741                                                 TP_STATUS_AVAILABLE);
2742                                 packet_increment_head(&po->tx_ring);
2743                                 kfree_skb(skb);
2744                                 continue;
2745                         } else {
2746                                 status = TP_STATUS_WRONG_FORMAT;
2747                                 err = tp_len;
2748                                 goto out_status;
2749                         }
2750                 }
2751
2752                 if (po->has_vnet_hdr && virtio_net_hdr_to_skb(skb, vnet_hdr,
2753                                                               vio_le())) {
2754                         tp_len = -EINVAL;
2755                         goto tpacket_error;
2756                 }
2757
2758                 skb->destructor = tpacket_destruct_skb;
2759                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2760                 packet_inc_pending(&po->tx_ring);
2761
2762                 status = TP_STATUS_SEND_REQUEST;
2763                 err = po->xmit(skb);
2764                 if (unlikely(err > 0)) {
2765                         err = net_xmit_errno(err);
2766                         if (err && __packet_get_status(po, ph) ==
2767                                    TP_STATUS_AVAILABLE) {
2768                                 /* skb was destructed already */
2769                                 skb = NULL;
2770                                 goto out_status;
2771                         }
2772                         /*
2773                          * skb was dropped but not destructed yet;
2774                          * let's treat it like congestion or err < 0
2775                          */
2776                         err = 0;
2777                 }
2778                 packet_increment_head(&po->tx_ring);
2779                 len_sum += tp_len;
2780         } while (likely((ph != NULL) ||
2781                 /* Note: packet_read_pending() might be slow if we have
2782                  * to call it as it's per_cpu variable, but in fast-path
2783                  * we already short-circuit the loop with the first
2784                  * condition, and luckily don't have to go that path
2785                  * anyway.
2786                  */
2787                  (need_wait && packet_read_pending(&po->tx_ring))));
2788
2789         err = len_sum;
2790         goto out_put;
2791
2792 out_status:
2793         __packet_set_status(po, ph, status);
2794         kfree_skb(skb);
2795 out_put:
2796         dev_put(dev);
2797 out:
2798         mutex_unlock(&po->pg_vec_lock);
2799         return err;
2800 }
2801
2802 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2803                                         size_t reserve, size_t len,
2804                                         size_t linear, int noblock,
2805                                         int *err)
2806 {
2807         struct sk_buff *skb;
2808
2809         /* Under a page?  Don't bother with paged skb. */
2810         if (prepad + len < PAGE_SIZE || !linear)
2811                 linear = len;
2812
2813         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2814                                    err, 0);
2815         if (!skb)
2816                 return NULL;
2817
2818         skb_reserve(skb, reserve);
2819         skb_put(skb, linear);
2820         skb->data_len = len - linear;
2821         skb->len += len - linear;
2822
2823         return skb;
2824 }
2825
2826 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2827 {
2828         struct sock *sk = sock->sk;
2829         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2830         struct sk_buff *skb;
2831         struct net_device *dev;
2832         __be16 proto;
2833         unsigned char *addr;
2834         int err, reserve = 0;
2835         struct sockcm_cookie sockc;
2836         struct virtio_net_hdr vnet_hdr = { 0 };
2837         int offset = 0;
2838         struct packet_sock *po = pkt_sk(sk);
2839         int hlen, tlen, linear;
2840         int extra_len = 0;
2841
2842         /*
2843          *      Get and verify the address.
2844          */
2845
2846         if (likely(saddr == NULL)) {
2847                 dev     = packet_cached_dev_get(po);
2848                 proto   = po->num;
2849                 addr    = NULL;
2850         } else {
2851                 err = -EINVAL;
2852                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2853                         goto out;
2854                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2855                         goto out;
2856                 proto   = saddr->sll_protocol;
2857                 addr    = saddr->sll_addr;
2858                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2859         }
2860
2861         err = -ENXIO;
2862         if (unlikely(dev == NULL))
2863                 goto out_unlock;
2864         err = -ENETDOWN;
2865         if (unlikely(!(dev->flags & IFF_UP)))
2866                 goto out_unlock;
2867
2868         sockc.tsflags = sk->sk_tsflags;
2869         sockc.mark = sk->sk_mark;
2870         if (msg->msg_controllen) {
2871                 err = sock_cmsg_send(sk, msg, &sockc);
2872                 if (unlikely(err))
2873                         goto out_unlock;
2874         }
2875
2876         if (sock->type == SOCK_RAW)
2877                 reserve = dev->hard_header_len;
2878         if (po->has_vnet_hdr) {
2879                 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2880                 if (err)
2881                         goto out_unlock;
2882         }
2883
2884         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2885                 if (!netif_supports_nofcs(dev)) {
2886                         err = -EPROTONOSUPPORT;
2887                         goto out_unlock;
2888                 }
2889                 extra_len = 4; /* We're doing our own CRC */
2890         }
2891
2892         err = -EMSGSIZE;
2893         if (!vnet_hdr.gso_type &&
2894             (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2895                 goto out_unlock;
2896
2897         err = -ENOBUFS;
2898         hlen = LL_RESERVED_SPACE(dev);
2899         tlen = dev->needed_tailroom;
2900         linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2901         linear = max(linear, min_t(int, len, dev->hard_header_len));
2902         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2903                                msg->msg_flags & MSG_DONTWAIT, &err);
2904         if (skb == NULL)
2905                 goto out_unlock;
2906
2907         skb_set_network_header(skb, reserve);
2908
2909         err = -EINVAL;
2910         if (sock->type == SOCK_DGRAM) {
2911                 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2912                 if (unlikely(offset < 0))
2913                         goto out_free;
2914         }
2915
2916         /* Returns -EFAULT on error */
2917         err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2918         if (err)
2919                 goto out_free;
2920
2921         if (sock->type == SOCK_RAW &&
2922             !dev_validate_header(dev, skb->data, len)) {
2923                 err = -EINVAL;
2924                 goto out_free;
2925         }
2926
2927         sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
2928
2929         if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2930             !packet_extra_vlan_len_allowed(dev, skb)) {
2931                 err = -EMSGSIZE;
2932                 goto out_free;
2933         }
2934
2935         skb->protocol = proto;
2936         skb->dev = dev;
2937         skb->priority = sk->sk_priority;
2938         skb->mark = sockc.mark;
2939
2940         if (po->has_vnet_hdr) {
2941                 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2942                 if (err)
2943                         goto out_free;
2944                 len += sizeof(vnet_hdr);
2945         }
2946
2947         skb_probe_transport_header(skb, reserve);
2948
2949         if (unlikely(extra_len == 4))
2950                 skb->no_fcs = 1;
2951
2952         err = po->xmit(skb);
2953         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2954                 goto out_unlock;
2955
2956         dev_put(dev);
2957
2958         return len;
2959
2960 out_free:
2961         kfree_skb(skb);
2962 out_unlock:
2963         if (dev)
2964                 dev_put(dev);
2965 out:
2966         return err;
2967 }
2968
2969 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2970 {
2971         struct sock *sk = sock->sk;
2972         struct packet_sock *po = pkt_sk(sk);
2973
2974         if (po->tx_ring.pg_vec)
2975                 return tpacket_snd(po, msg);
2976         else
2977                 return packet_snd(sock, msg, len);
2978 }
2979
2980 /*
2981  *      Close a PACKET socket. This is fairly simple. We immediately go
2982  *      to 'closed' state and remove our protocol entry in the device list.
2983  */
2984
2985 static int packet_release(struct socket *sock)
2986 {
2987         struct sock *sk = sock->sk;
2988         struct packet_sock *po;
2989         struct packet_fanout *f;
2990         struct net *net;
2991         union tpacket_req_u req_u;
2992
2993         if (!sk)
2994                 return 0;
2995
2996         net = sock_net(sk);
2997         po = pkt_sk(sk);
2998
2999         mutex_lock(&net->packet.sklist_lock);
3000         sk_del_node_init_rcu(sk);
3001         mutex_unlock(&net->packet.sklist_lock);
3002
3003         preempt_disable();
3004         sock_prot_inuse_add(net, sk->sk_prot, -1);
3005         preempt_enable();
3006
3007         spin_lock(&po->bind_lock);
3008         unregister_prot_hook(sk, false);
3009         packet_cached_dev_reset(po);
3010
3011         if (po->prot_hook.dev) {
3012                 dev_put(po->prot_hook.dev);
3013                 po->prot_hook.dev = NULL;
3014         }
3015         spin_unlock(&po->bind_lock);
3016
3017         packet_flush_mclist(sk);
3018
3019         if (po->rx_ring.pg_vec) {
3020                 memset(&req_u, 0, sizeof(req_u));
3021                 packet_set_ring(sk, &req_u, 1, 0);
3022         }
3023
3024         if (po->tx_ring.pg_vec) {
3025                 memset(&req_u, 0, sizeof(req_u));
3026                 packet_set_ring(sk, &req_u, 1, 1);
3027         }
3028
3029         f = fanout_release(sk);
3030
3031         synchronize_net();
3032
3033         if (f) {
3034                 fanout_release_data(f);
3035                 kfree(f);
3036         }
3037         /*
3038          *      Now the socket is dead. No more input will appear.
3039          */
3040         sock_orphan(sk);
3041         sock->sk = NULL;
3042
3043         /* Purge queues */
3044
3045         skb_queue_purge(&sk->sk_receive_queue);
3046         packet_free_pending(po);
3047         sk_refcnt_debug_release(sk);
3048
3049         sock_put(sk);
3050         return 0;
3051 }
3052
3053 /*
3054  *      Attach a packet hook.
3055  */
3056
3057 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3058                           __be16 proto)
3059 {
3060         struct packet_sock *po = pkt_sk(sk);
3061         struct net_device *dev_curr;
3062         __be16 proto_curr;
3063         bool need_rehook;
3064         struct net_device *dev = NULL;
3065         int ret = 0;
3066         bool unlisted = false;
3067
3068         if (po->fanout)
3069                 return -EINVAL;
3070
3071         lock_sock(sk);
3072         spin_lock(&po->bind_lock);
3073         rcu_read_lock();
3074
3075         if (name) {
3076                 dev = dev_get_by_name_rcu(sock_net(sk), name);
3077                 if (!dev) {
3078                         ret = -ENODEV;
3079                         goto out_unlock;
3080                 }
3081         } else if (ifindex) {
3082                 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3083                 if (!dev) {
3084                         ret = -ENODEV;
3085                         goto out_unlock;
3086                 }
3087         }
3088
3089         if (dev)
3090                 dev_hold(dev);
3091
3092         proto_curr = po->prot_hook.type;
3093         dev_curr = po->prot_hook.dev;
3094
3095         need_rehook = proto_curr != proto || dev_curr != dev;
3096
3097         if (need_rehook) {
3098                 if (po->running) {
3099                         rcu_read_unlock();
3100                         __unregister_prot_hook(sk, true);
3101                         rcu_read_lock();
3102                         dev_curr = po->prot_hook.dev;
3103                         if (dev)
3104                                 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3105                                                                  dev->ifindex);
3106                 }
3107
3108                 po->num = proto;
3109                 po->prot_hook.type = proto;
3110
3111                 if (unlikely(unlisted)) {
3112                         dev_put(dev);
3113                         po->prot_hook.dev = NULL;
3114                         po->ifindex = -1;
3115                         packet_cached_dev_reset(po);
3116                 } else {
3117                         po->prot_hook.dev = dev;
3118                         po->ifindex = dev ? dev->ifindex : 0;
3119                         packet_cached_dev_assign(po, dev);
3120                 }
3121         }
3122         if (dev_curr)
3123                 dev_put(dev_curr);
3124
3125         if (proto == 0 || !need_rehook)
3126                 goto out_unlock;
3127
3128         if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3129                 register_prot_hook(sk);
3130         } else {
3131                 sk->sk_err = ENETDOWN;
3132                 if (!sock_flag(sk, SOCK_DEAD))
3133                         sk->sk_error_report(sk);
3134         }
3135
3136 out_unlock:
3137         rcu_read_unlock();
3138         spin_unlock(&po->bind_lock);
3139         release_sock(sk);
3140         return ret;
3141 }
3142
3143 /*
3144  *      Bind a packet socket to a device
3145  */
3146
3147 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3148                             int addr_len)
3149 {
3150         struct sock *sk = sock->sk;
3151         char name[sizeof(uaddr->sa_data) + 1];
3152
3153         /*
3154          *      Check legality
3155          */
3156
3157         if (addr_len != sizeof(struct sockaddr))
3158                 return -EINVAL;
3159         /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3160          * zero-terminated.
3161          */
3162         memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3163         name[sizeof(uaddr->sa_data)] = 0;
3164
3165         return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3166 }
3167
3168 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3169 {
3170         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3171         struct sock *sk = sock->sk;
3172
3173         /*
3174          *      Check legality
3175          */
3176
3177         if (addr_len < sizeof(struct sockaddr_ll))
3178                 return -EINVAL;
3179         if (sll->sll_family != AF_PACKET)
3180                 return -EINVAL;
3181
3182         return packet_do_bind(sk, NULL, sll->sll_ifindex,
3183                               sll->sll_protocol ? : pkt_sk(sk)->num);
3184 }
3185
3186 static struct proto packet_proto = {
3187         .name     = "PACKET",
3188         .owner    = THIS_MODULE,
3189         .obj_size = sizeof(struct packet_sock),
3190 };
3191
3192 /*
3193  *      Create a packet of type SOCK_PACKET.
3194  */
3195
3196 static int packet_create(struct net *net, struct socket *sock, int protocol,
3197                          int kern)
3198 {
3199         struct sock *sk;
3200         struct packet_sock *po;
3201         __be16 proto = (__force __be16)protocol; /* weird, but documented */
3202         int err;
3203
3204         if (!ns_capable(net->user_ns, CAP_NET_RAW))
3205                 return -EPERM;
3206         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3207             sock->type != SOCK_PACKET)
3208                 return -ESOCKTNOSUPPORT;
3209
3210         sock->state = SS_UNCONNECTED;
3211
3212         err = -ENOBUFS;
3213         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3214         if (sk == NULL)
3215                 goto out;
3216
3217         sock->ops = &packet_ops;
3218         if (sock->type == SOCK_PACKET)
3219                 sock->ops = &packet_ops_spkt;
3220
3221         sock_init_data(sock, sk);
3222
3223         po = pkt_sk(sk);
3224         sk->sk_family = PF_PACKET;
3225         po->num = proto;
3226         po->xmit = dev_queue_xmit;
3227
3228         err = packet_alloc_pending(po);
3229         if (err)
3230                 goto out2;
3231
3232         packet_cached_dev_reset(po);
3233
3234         sk->sk_destruct = packet_sock_destruct;
3235         sk_refcnt_debug_inc(sk);
3236
3237         /*
3238          *      Attach a protocol block
3239          */
3240
3241         spin_lock_init(&po->bind_lock);
3242         mutex_init(&po->pg_vec_lock);
3243         po->rollover = NULL;
3244         po->prot_hook.func = packet_rcv;
3245
3246         if (sock->type == SOCK_PACKET)
3247                 po->prot_hook.func = packet_rcv_spkt;
3248
3249         po->prot_hook.af_packet_priv = sk;
3250
3251         if (proto) {
3252                 po->prot_hook.type = proto;
3253                 register_prot_hook(sk);
3254         }
3255
3256         mutex_lock(&net->packet.sklist_lock);
3257         sk_add_node_rcu(sk, &net->packet.sklist);
3258         mutex_unlock(&net->packet.sklist_lock);
3259
3260         preempt_disable();
3261         sock_prot_inuse_add(net, &packet_proto, 1);
3262         preempt_enable();
3263
3264         return 0;
3265 out2:
3266         sk_free(sk);
3267 out:
3268         return err;
3269 }
3270
3271 /*
3272  *      Pull a packet from our receive queue and hand it to the user.
3273  *      If necessary we block.
3274  */
3275
3276 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3277                           int flags)
3278 {
3279         struct sock *sk = sock->sk;
3280         struct sk_buff *skb;
3281         int copied, err;
3282         int vnet_hdr_len = 0;
3283         unsigned int origlen = 0;
3284
3285         err = -EINVAL;
3286         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3287                 goto out;
3288
3289 #if 0
3290         /* What error should we return now? EUNATTACH? */
3291         if (pkt_sk(sk)->ifindex < 0)
3292                 return -ENODEV;
3293 #endif
3294
3295         if (flags & MSG_ERRQUEUE) {
3296                 err = sock_recv_errqueue(sk, msg, len,
3297                                          SOL_PACKET, PACKET_TX_TIMESTAMP);
3298                 goto out;
3299         }
3300
3301         /*
3302          *      Call the generic datagram receiver. This handles all sorts
3303          *      of horrible races and re-entrancy so we can forget about it
3304          *      in the protocol layers.
3305          *
3306          *      Now it will return ENETDOWN, if device have just gone down,
3307          *      but then it will block.
3308          */
3309
3310         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3311
3312         /*
3313          *      An error occurred so return it. Because skb_recv_datagram()
3314          *      handles the blocking we don't see and worry about blocking
3315          *      retries.
3316          */
3317
3318         if (skb == NULL)
3319                 goto out;
3320
3321         if (pkt_sk(sk)->pressure)
3322                 packet_rcv_has_room(pkt_sk(sk), NULL);
3323
3324         if (pkt_sk(sk)->has_vnet_hdr) {
3325                 err = packet_rcv_vnet(msg, skb, &len);
3326                 if (err)
3327                         goto out_free;
3328                 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3329         }
3330
3331         /* You lose any data beyond the buffer you gave. If it worries
3332          * a user program they can ask the device for its MTU
3333          * anyway.
3334          */
3335         copied = skb->len;
3336         if (copied > len) {
3337                 copied = len;
3338                 msg->msg_flags |= MSG_TRUNC;
3339         }
3340
3341         err = skb_copy_datagram_msg(skb, 0, msg, copied);
3342         if (err)
3343                 goto out_free;
3344
3345         if (sock->type != SOCK_PACKET) {
3346                 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3347
3348                 /* Original length was stored in sockaddr_ll fields */
3349                 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3350                 sll->sll_family = AF_PACKET;
3351                 sll->sll_protocol = skb->protocol;
3352         }
3353
3354         sock_recv_ts_and_drops(msg, sk, skb);
3355
3356         if (msg->msg_name) {
3357                 /* If the address length field is there to be filled
3358                  * in, we fill it in now.
3359                  */
3360                 if (sock->type == SOCK_PACKET) {
3361                         __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3362                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
3363                 } else {
3364                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3365
3366                         msg->msg_namelen = sll->sll_halen +
3367                                 offsetof(struct sockaddr_ll, sll_addr);
3368                 }
3369                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3370                        msg->msg_namelen);
3371         }
3372
3373         if (pkt_sk(sk)->auxdata) {
3374                 struct tpacket_auxdata aux;
3375
3376                 aux.tp_status = TP_STATUS_USER;
3377                 if (skb->ip_summed == CHECKSUM_PARTIAL)
3378                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3379                 else if (skb->pkt_type != PACKET_OUTGOING &&
3380                          (skb->ip_summed == CHECKSUM_COMPLETE ||
3381                           skb_csum_unnecessary(skb)))
3382                         aux.tp_status |= TP_STATUS_CSUM_VALID;
3383
3384                 aux.tp_len = origlen;
3385                 aux.tp_snaplen = skb->len;
3386                 aux.tp_mac = 0;
3387                 aux.tp_net = skb_network_offset(skb);
3388                 if (skb_vlan_tag_present(skb)) {
3389                         aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3390                         aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3391                         aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3392                 } else {
3393                         aux.tp_vlan_tci = 0;
3394                         aux.tp_vlan_tpid = 0;
3395                 }
3396                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3397         }
3398
3399         /*
3400          *      Free or return the buffer as appropriate. Again this
3401          *      hides all the races and re-entrancy issues from us.
3402          */
3403         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3404
3405 out_free:
3406         skb_free_datagram(sk, skb);
3407 out:
3408         return err;
3409 }
3410
3411 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3412                                int *uaddr_len, int peer)
3413 {
3414         struct net_device *dev;
3415         struct sock *sk = sock->sk;
3416
3417         if (peer)
3418                 return -EOPNOTSUPP;
3419
3420         uaddr->sa_family = AF_PACKET;
3421         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3422         rcu_read_lock();
3423         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3424         if (dev)
3425                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3426         rcu_read_unlock();
3427         *uaddr_len = sizeof(*uaddr);
3428
3429         return 0;
3430 }
3431
3432 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3433                           int *uaddr_len, int peer)
3434 {
3435         struct net_device *dev;
3436         struct sock *sk = sock->sk;
3437         struct packet_sock *po = pkt_sk(sk);
3438         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3439
3440         if (peer)
3441                 return -EOPNOTSUPP;
3442
3443         sll->sll_family = AF_PACKET;
3444         sll->sll_ifindex = po->ifindex;
3445         sll->sll_protocol = po->num;
3446         sll->sll_pkttype = 0;
3447         rcu_read_lock();
3448         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3449         if (dev) {
3450                 sll->sll_hatype = dev->type;
3451                 sll->sll_halen = dev->addr_len;
3452                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3453         } else {
3454                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
3455                 sll->sll_halen = 0;
3456         }
3457         rcu_read_unlock();
3458         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3459
3460         return 0;
3461 }
3462
3463 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3464                          int what)
3465 {
3466         switch (i->type) {
3467         case PACKET_MR_MULTICAST:
3468                 if (i->alen != dev->addr_len)
3469                         return -EINVAL;
3470                 if (what > 0)
3471                         return dev_mc_add(dev, i->addr);
3472                 else
3473                         return dev_mc_del(dev, i->addr);
3474                 break;
3475         case PACKET_MR_PROMISC:
3476                 return dev_set_promiscuity(dev, what);
3477         case PACKET_MR_ALLMULTI:
3478                 return dev_set_allmulti(dev, what);
3479         case PACKET_MR_UNICAST:
3480                 if (i->alen != dev->addr_len)
3481                         return -EINVAL;
3482                 if (what > 0)
3483                         return dev_uc_add(dev, i->addr);
3484                 else
3485                         return dev_uc_del(dev, i->addr);
3486                 break;
3487         default:
3488                 break;
3489         }
3490         return 0;
3491 }
3492
3493 static void packet_dev_mclist_delete(struct net_device *dev,
3494                                      struct packet_mclist **mlp)
3495 {
3496         struct packet_mclist *ml;
3497
3498         while ((ml = *mlp) != NULL) {
3499                 if (ml->ifindex == dev->ifindex) {
3500                         packet_dev_mc(dev, ml, -1);
3501                         *mlp = ml->next;
3502                         kfree(ml);
3503                 } else
3504                         mlp = &ml->next;
3505         }
3506 }
3507
3508 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3509 {
3510         struct packet_sock *po = pkt_sk(sk);
3511         struct packet_mclist *ml, *i;
3512         struct net_device *dev;
3513         int err;
3514
3515         rtnl_lock();
3516
3517         err = -ENODEV;
3518         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3519         if (!dev)
3520                 goto done;
3521
3522         err = -EINVAL;
3523         if (mreq->mr_alen > dev->addr_len)
3524                 goto done;
3525
3526         err = -ENOBUFS;
3527         i = kmalloc(sizeof(*i), GFP_KERNEL);
3528         if (i == NULL)
3529                 goto done;
3530
3531         err = 0;
3532         for (ml = po->mclist; ml; ml = ml->next) {
3533                 if (ml->ifindex == mreq->mr_ifindex &&
3534                     ml->type == mreq->mr_type &&
3535                     ml->alen == mreq->mr_alen &&
3536                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3537                         ml->count++;
3538                         /* Free the new element ... */
3539                         kfree(i);
3540                         goto done;
3541                 }
3542         }
3543
3544         i->type = mreq->mr_type;
3545         i->ifindex = mreq->mr_ifindex;
3546         i->alen = mreq->mr_alen;
3547         memcpy(i->addr, mreq->mr_address, i->alen);
3548         memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3549         i->count = 1;
3550         i->next = po->mclist;
3551         po->mclist = i;
3552         err = packet_dev_mc(dev, i, 1);
3553         if (err) {
3554                 po->mclist = i->next;
3555                 kfree(i);
3556         }
3557
3558 done:
3559         rtnl_unlock();
3560         return err;
3561 }
3562
3563 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3564 {
3565         struct packet_mclist *ml, **mlp;
3566
3567         rtnl_lock();
3568
3569         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3570                 if (ml->ifindex == mreq->mr_ifindex &&
3571                     ml->type == mreq->mr_type &&
3572                     ml->alen == mreq->mr_alen &&
3573                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3574                         if (--ml->count == 0) {
3575                                 struct net_device *dev;
3576                                 *mlp = ml->next;
3577                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3578                                 if (dev)
3579                                         packet_dev_mc(dev, ml, -1);
3580                                 kfree(ml);
3581                         }
3582                         break;
3583                 }
3584         }
3585         rtnl_unlock();
3586         return 0;
3587 }
3588
3589 static void packet_flush_mclist(struct sock *sk)
3590 {
3591         struct packet_sock *po = pkt_sk(sk);
3592         struct packet_mclist *ml;
3593
3594         if (!po->mclist)
3595                 return;
3596
3597         rtnl_lock();
3598         while ((ml = po->mclist) != NULL) {
3599                 struct net_device *dev;
3600
3601                 po->mclist = ml->next;
3602                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3603                 if (dev != NULL)
3604                         packet_dev_mc(dev, ml, -1);
3605                 kfree(ml);
3606         }
3607         rtnl_unlock();
3608 }
3609
3610 static int
3611 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3612 {
3613         struct sock *sk = sock->sk;
3614         struct packet_sock *po = pkt_sk(sk);
3615         int ret;
3616
3617         if (level != SOL_PACKET)
3618                 return -ENOPROTOOPT;
3619
3620         switch (optname) {
3621         case PACKET_ADD_MEMBERSHIP:
3622         case PACKET_DROP_MEMBERSHIP:
3623         {
3624                 struct packet_mreq_max mreq;
3625                 int len = optlen;
3626                 memset(&mreq, 0, sizeof(mreq));
3627                 if (len < sizeof(struct packet_mreq))
3628                         return -EINVAL;
3629                 if (len > sizeof(mreq))
3630                         len = sizeof(mreq);
3631                 if (copy_from_user(&mreq, optval, len))
3632                         return -EFAULT;
3633                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3634                         return -EINVAL;
3635                 if (optname == PACKET_ADD_MEMBERSHIP)
3636                         ret = packet_mc_add(sk, &mreq);
3637                 else
3638                         ret = packet_mc_drop(sk, &mreq);
3639                 return ret;
3640         }
3641
3642         case PACKET_RX_RING:
3643         case PACKET_TX_RING:
3644         {
3645                 union tpacket_req_u req_u;
3646                 int len;
3647
3648                 switch (po->tp_version) {
3649                 case TPACKET_V1:
3650                 case TPACKET_V2:
3651                         len = sizeof(req_u.req);
3652                         break;
3653                 case TPACKET_V3:
3654                 default:
3655                         len = sizeof(req_u.req3);
3656                         break;
3657                 }
3658                 if (optlen < len)
3659                         return -EINVAL;
3660                 if (copy_from_user(&req_u.req, optval, len))
3661                         return -EFAULT;
3662                 return packet_set_ring(sk, &req_u, 0,
3663                         optname == PACKET_TX_RING);
3664         }
3665         case PACKET_COPY_THRESH:
3666         {
3667                 int val;
3668
3669                 if (optlen != sizeof(val))
3670                         return -EINVAL;
3671                 if (copy_from_user(&val, optval, sizeof(val)))
3672                         return -EFAULT;
3673
3674                 pkt_sk(sk)->copy_thresh = val;
3675                 return 0;
3676         }
3677         case PACKET_VERSION:
3678         {
3679                 int val;
3680
3681                 if (optlen != sizeof(val))
3682                         return -EINVAL;
3683                 if (copy_from_user(&val, optval, sizeof(val)))
3684                         return -EFAULT;
3685                 switch (val) {
3686                 case TPACKET_V1:
3687                 case TPACKET_V2:
3688                 case TPACKET_V3:
3689                         break;
3690                 default:
3691                         return -EINVAL;
3692                 }
3693                 lock_sock(sk);
3694                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3695                         ret = -EBUSY;
3696                 } else {
3697                         po->tp_version = val;
3698                         ret = 0;
3699                 }
3700                 release_sock(sk);
3701                 return ret;
3702         }
3703         case PACKET_RESERVE:
3704         {
3705                 unsigned int val;
3706
3707                 if (optlen != sizeof(val))
3708                         return -EINVAL;
3709                 if (copy_from_user(&val, optval, sizeof(val)))
3710                         return -EFAULT;
3711                 if (val > INT_MAX)
3712                         return -EINVAL;
3713                 lock_sock(sk);
3714                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3715                         ret = -EBUSY;
3716                 } else {
3717                         po->tp_reserve = val;
3718                         ret = 0;
3719                 }
3720                 release_sock(sk);
3721                 return ret;
3722         }
3723         case PACKET_LOSS:
3724         {
3725                 unsigned int val;
3726
3727                 if (optlen != sizeof(val))
3728                         return -EINVAL;
3729                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3730                         return -EBUSY;
3731                 if (copy_from_user(&val, optval, sizeof(val)))
3732                         return -EFAULT;
3733                 po->tp_loss = !!val;
3734                 return 0;
3735         }
3736         case PACKET_AUXDATA:
3737         {
3738                 int val;
3739
3740                 if (optlen < sizeof(val))
3741                         return -EINVAL;
3742                 if (copy_from_user(&val, optval, sizeof(val)))
3743                         return -EFAULT;
3744
3745                 po->auxdata = !!val;
3746                 return 0;
3747         }
3748         case PACKET_ORIGDEV:
3749         {
3750                 int val;
3751
3752                 if (optlen < sizeof(val))
3753                         return -EINVAL;
3754                 if (copy_from_user(&val, optval, sizeof(val)))
3755                         return -EFAULT;
3756
3757                 po->origdev = !!val;
3758                 return 0;
3759         }
3760         case PACKET_VNET_HDR:
3761         {
3762                 int val;
3763
3764                 if (sock->type != SOCK_RAW)
3765                         return -EINVAL;
3766                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3767                         return -EBUSY;
3768                 if (optlen < sizeof(val))
3769                         return -EINVAL;
3770                 if (copy_from_user(&val, optval, sizeof(val)))
3771                         return -EFAULT;
3772
3773                 po->has_vnet_hdr = !!val;
3774                 return 0;
3775         }
3776         case PACKET_TIMESTAMP:
3777         {
3778                 int val;
3779
3780                 if (optlen != sizeof(val))
3781                         return -EINVAL;
3782                 if (copy_from_user(&val, optval, sizeof(val)))
3783                         return -EFAULT;
3784
3785                 po->tp_tstamp = val;
3786                 return 0;
3787         }
3788         case PACKET_FANOUT:
3789         {
3790                 int val;
3791
3792                 if (optlen != sizeof(val))
3793                         return -EINVAL;
3794                 if (copy_from_user(&val, optval, sizeof(val)))
3795                         return -EFAULT;
3796
3797                 return fanout_add(sk, val & 0xffff, val >> 16);
3798         }
3799         case PACKET_FANOUT_DATA:
3800         {
3801                 if (!po->fanout)
3802                         return -EINVAL;
3803
3804                 return fanout_set_data(po, optval, optlen);
3805         }
3806         case PACKET_TX_HAS_OFF:
3807         {
3808                 unsigned int val;
3809
3810                 if (optlen != sizeof(val))
3811                         return -EINVAL;
3812                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3813                         return -EBUSY;
3814                 if (copy_from_user(&val, optval, sizeof(val)))
3815                         return -EFAULT;
3816                 po->tp_tx_has_off = !!val;
3817                 return 0;
3818         }
3819         case PACKET_QDISC_BYPASS:
3820         {
3821                 int val;
3822
3823                 if (optlen != sizeof(val))
3824                         return -EINVAL;
3825                 if (copy_from_user(&val, optval, sizeof(val)))
3826                         return -EFAULT;
3827
3828                 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3829                 return 0;
3830         }
3831         default:
3832                 return -ENOPROTOOPT;
3833         }
3834 }
3835
3836 static int packet_getsockopt(struct socket *sock, int level, int optname,
3837                              char __user *optval, int __user *optlen)
3838 {
3839         int len;
3840         int val, lv = sizeof(val);
3841         struct sock *sk = sock->sk;
3842         struct packet_sock *po = pkt_sk(sk);
3843         void *data = &val;
3844         union tpacket_stats_u st;
3845         struct tpacket_rollover_stats rstats;
3846
3847         if (level != SOL_PACKET)
3848                 return -ENOPROTOOPT;
3849
3850         if (get_user(len, optlen))
3851                 return -EFAULT;
3852
3853         if (len < 0)
3854                 return -EINVAL;
3855
3856         switch (optname) {
3857         case PACKET_STATISTICS:
3858                 spin_lock_bh(&sk->sk_receive_queue.lock);
3859                 memcpy(&st, &po->stats, sizeof(st));
3860                 memset(&po->stats, 0, sizeof(po->stats));
3861                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3862
3863                 if (po->tp_version == TPACKET_V3) {
3864                         lv = sizeof(struct tpacket_stats_v3);
3865                         st.stats3.tp_packets += st.stats3.tp_drops;
3866                         data = &st.stats3;
3867                 } else {
3868                         lv = sizeof(struct tpacket_stats);
3869                         st.stats1.tp_packets += st.stats1.tp_drops;
3870                         data = &st.stats1;
3871                 }
3872
3873                 break;
3874         case PACKET_AUXDATA:
3875                 val = po->auxdata;
3876                 break;
3877         case PACKET_ORIGDEV:
3878                 val = po->origdev;
3879                 break;
3880         case PACKET_VNET_HDR:
3881                 val = po->has_vnet_hdr;
3882                 break;
3883         case PACKET_VERSION:
3884                 val = po->tp_version;
3885                 break;
3886         case PACKET_HDRLEN:
3887                 if (len > sizeof(int))
3888                         len = sizeof(int);
3889                 if (len < sizeof(int))
3890                         return -EINVAL;
3891                 if (copy_from_user(&val, optval, len))
3892                         return -EFAULT;
3893                 switch (val) {
3894                 case TPACKET_V1:
3895                         val = sizeof(struct tpacket_hdr);
3896                         break;
3897                 case TPACKET_V2:
3898                         val = sizeof(struct tpacket2_hdr);
3899                         break;
3900                 case TPACKET_V3:
3901                         val = sizeof(struct tpacket3_hdr);
3902                         break;
3903                 default:
3904                         return -EINVAL;
3905                 }
3906                 break;
3907         case PACKET_RESERVE:
3908                 val = po->tp_reserve;
3909                 break;
3910         case PACKET_LOSS:
3911                 val = po->tp_loss;
3912                 break;
3913         case PACKET_TIMESTAMP:
3914                 val = po->tp_tstamp;
3915                 break;
3916         case PACKET_FANOUT:
3917                 val = (po->fanout ?
3918                        ((u32)po->fanout->id |
3919                         ((u32)po->fanout->type << 16) |
3920                         ((u32)po->fanout->flags << 24)) :
3921                        0);
3922                 break;
3923         case PACKET_ROLLOVER_STATS:
3924                 if (!po->rollover)
3925                         return -EINVAL;
3926                 rstats.tp_all = atomic_long_read(&po->rollover->num);
3927                 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3928                 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3929                 data = &rstats;
3930                 lv = sizeof(rstats);
3931                 break;
3932         case PACKET_TX_HAS_OFF:
3933                 val = po->tp_tx_has_off;
3934                 break;
3935         case PACKET_QDISC_BYPASS:
3936                 val = packet_use_direct_xmit(po);
3937                 break;
3938         default:
3939                 return -ENOPROTOOPT;
3940         }
3941
3942         if (len > lv)
3943                 len = lv;
3944         if (put_user(len, optlen))
3945                 return -EFAULT;
3946         if (copy_to_user(optval, data, len))
3947                 return -EFAULT;
3948         return 0;
3949 }
3950
3951
3952 #ifdef CONFIG_COMPAT
3953 static int compat_packet_setsockopt(struct socket *sock, int level, int optname,
3954                                     char __user *optval, unsigned int optlen)
3955 {
3956         struct packet_sock *po = pkt_sk(sock->sk);
3957
3958         if (level != SOL_PACKET)
3959                 return -ENOPROTOOPT;
3960
3961         if (optname == PACKET_FANOUT_DATA &&
3962             po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) {
3963                 optval = (char __user *)get_compat_bpf_fprog(optval);
3964                 if (!optval)
3965                         return -EFAULT;
3966                 optlen = sizeof(struct sock_fprog);
3967         }
3968
3969         return packet_setsockopt(sock, level, optname, optval, optlen);
3970 }
3971 #endif
3972
3973 static int packet_notifier(struct notifier_block *this,
3974                            unsigned long msg, void *ptr)
3975 {
3976         struct sock *sk;
3977         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3978         struct net *net = dev_net(dev);
3979
3980         rcu_read_lock();
3981         sk_for_each_rcu(sk, &net->packet.sklist) {
3982                 struct packet_sock *po = pkt_sk(sk);
3983
3984                 switch (msg) {
3985                 case NETDEV_UNREGISTER:
3986                         if (po->mclist)
3987                                 packet_dev_mclist_delete(dev, &po->mclist);
3988                         /* fallthrough */
3989
3990                 case NETDEV_DOWN:
3991                         if (dev->ifindex == po->ifindex) {
3992                                 spin_lock(&po->bind_lock);
3993                                 if (po->running) {
3994                                         __unregister_prot_hook(sk, false);
3995                                         sk->sk_err = ENETDOWN;
3996                                         if (!sock_flag(sk, SOCK_DEAD))
3997                                                 sk->sk_error_report(sk);
3998                                 }
3999                                 if (msg == NETDEV_UNREGISTER) {
4000                                         packet_cached_dev_reset(po);
4001                                         po->ifindex = -1;
4002                                         if (po->prot_hook.dev)
4003                                                 dev_put(po->prot_hook.dev);
4004                                         po->prot_hook.dev = NULL;
4005                                 }
4006                                 spin_unlock(&po->bind_lock);
4007                         }
4008                         break;
4009                 case NETDEV_UP:
4010                         if (dev->ifindex == po->ifindex) {
4011                                 spin_lock(&po->bind_lock);
4012                                 if (po->num)
4013                                         register_prot_hook(sk);
4014                                 spin_unlock(&po->bind_lock);
4015                         }
4016                         break;
4017                 }
4018         }
4019         rcu_read_unlock();
4020         return NOTIFY_DONE;
4021 }
4022
4023
4024 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4025                         unsigned long arg)
4026 {
4027         struct sock *sk = sock->sk;
4028
4029         switch (cmd) {
4030         case SIOCOUTQ:
4031         {
4032                 int amount = sk_wmem_alloc_get(sk);
4033
4034                 return put_user(amount, (int __user *)arg);
4035         }
4036         case SIOCINQ:
4037         {
4038                 struct sk_buff *skb;
4039                 int amount = 0;
4040
4041                 spin_lock_bh(&sk->sk_receive_queue.lock);
4042                 skb = skb_peek(&sk->sk_receive_queue);
4043                 if (skb)
4044                         amount = skb->len;
4045                 spin_unlock_bh(&sk->sk_receive_queue.lock);
4046                 return put_user(amount, (int __user *)arg);
4047         }
4048         case SIOCGSTAMP:
4049                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
4050         case SIOCGSTAMPNS:
4051                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
4052
4053 #ifdef CONFIG_INET
4054         case SIOCADDRT:
4055         case SIOCDELRT:
4056         case SIOCDARP:
4057         case SIOCGARP:
4058         case SIOCSARP:
4059         case SIOCGIFADDR:
4060         case SIOCSIFADDR:
4061         case SIOCGIFBRDADDR:
4062         case SIOCSIFBRDADDR:
4063         case SIOCGIFNETMASK:
4064         case SIOCSIFNETMASK:
4065         case SIOCGIFDSTADDR:
4066         case SIOCSIFDSTADDR:
4067         case SIOCSIFFLAGS:
4068                 return inet_dgram_ops.ioctl(sock, cmd, arg);
4069 #endif
4070
4071         default:
4072                 return -ENOIOCTLCMD;
4073         }
4074         return 0;
4075 }
4076
4077 static unsigned int packet_poll(struct file *file, struct socket *sock,
4078                                 poll_table *wait)
4079 {
4080         struct sock *sk = sock->sk;
4081         struct packet_sock *po = pkt_sk(sk);
4082         unsigned int mask = datagram_poll(file, sock, wait);
4083
4084         spin_lock_bh(&sk->sk_receive_queue.lock);
4085         if (po->rx_ring.pg_vec) {
4086                 if (!packet_previous_rx_frame(po, &po->rx_ring,
4087                         TP_STATUS_KERNEL))
4088                         mask |= POLLIN | POLLRDNORM;
4089         }
4090         if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
4091                 po->pressure = 0;
4092         spin_unlock_bh(&sk->sk_receive_queue.lock);
4093         spin_lock_bh(&sk->sk_write_queue.lock);
4094         if (po->tx_ring.pg_vec) {
4095                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4096                         mask |= POLLOUT | POLLWRNORM;
4097         }
4098         spin_unlock_bh(&sk->sk_write_queue.lock);
4099         return mask;
4100 }
4101
4102
4103 /* Dirty? Well, I still did not learn better way to account
4104  * for user mmaps.
4105  */
4106
4107 static void packet_mm_open(struct vm_area_struct *vma)
4108 {
4109         struct file *file = vma->vm_file;
4110         struct socket *sock = file->private_data;
4111         struct sock *sk = sock->sk;
4112
4113         if (sk)
4114                 atomic_inc(&pkt_sk(sk)->mapped);
4115 }
4116
4117 static void packet_mm_close(struct vm_area_struct *vma)
4118 {
4119         struct file *file = vma->vm_file;
4120         struct socket *sock = file->private_data;
4121         struct sock *sk = sock->sk;
4122
4123         if (sk)
4124                 atomic_dec(&pkt_sk(sk)->mapped);
4125 }
4126
4127 static const struct vm_operations_struct packet_mmap_ops = {
4128         .open   =       packet_mm_open,
4129         .close  =       packet_mm_close,
4130 };
4131
4132 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4133                         unsigned int len)
4134 {
4135         int i;
4136
4137         for (i = 0; i < len; i++) {
4138                 if (likely(pg_vec[i].buffer)) {
4139                         if (is_vmalloc_addr(pg_vec[i].buffer))
4140                                 vfree(pg_vec[i].buffer);
4141                         else
4142                                 free_pages((unsigned long)pg_vec[i].buffer,
4143                                            order);
4144                         pg_vec[i].buffer = NULL;
4145                 }
4146         }
4147         kfree(pg_vec);
4148 }
4149
4150 static char *alloc_one_pg_vec_page(unsigned long order)
4151 {
4152         char *buffer;
4153         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4154                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4155
4156         buffer = (char *) __get_free_pages(gfp_flags, order);
4157         if (buffer)
4158                 return buffer;
4159
4160         /* __get_free_pages failed, fall back to vmalloc */
4161         buffer = vzalloc((1 << order) * PAGE_SIZE);
4162         if (buffer)
4163                 return buffer;
4164
4165         /* vmalloc failed, lets dig into swap here */
4166         gfp_flags &= ~__GFP_NORETRY;
4167         buffer = (char *) __get_free_pages(gfp_flags, order);
4168         if (buffer)
4169                 return buffer;
4170
4171         /* complete and utter failure */
4172         return NULL;
4173 }
4174
4175 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4176 {
4177         unsigned int block_nr = req->tp_block_nr;
4178         struct pgv *pg_vec;
4179         int i;
4180
4181         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4182         if (unlikely(!pg_vec))
4183                 goto out;
4184
4185         for (i = 0; i < block_nr; i++) {
4186                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4187                 if (unlikely(!pg_vec[i].buffer))
4188                         goto out_free_pgvec;
4189         }
4190
4191 out:
4192         return pg_vec;
4193
4194 out_free_pgvec:
4195         free_pg_vec(pg_vec, order, block_nr);
4196         pg_vec = NULL;
4197         goto out;
4198 }
4199
4200 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4201                 int closing, int tx_ring)
4202 {
4203         struct pgv *pg_vec = NULL;
4204         struct packet_sock *po = pkt_sk(sk);
4205         int was_running, order = 0;
4206         struct packet_ring_buffer *rb;
4207         struct sk_buff_head *rb_queue;
4208         __be16 num;
4209         int err = -EINVAL;
4210         /* Added to avoid minimal code churn */
4211         struct tpacket_req *req = &req_u->req;
4212
4213         lock_sock(sk);
4214
4215         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4216         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4217
4218         err = -EBUSY;
4219         if (!closing) {
4220                 if (atomic_read(&po->mapped))
4221                         goto out;
4222                 if (packet_read_pending(rb))
4223                         goto out;
4224         }
4225
4226         if (req->tp_block_nr) {
4227                 /* Sanity tests and some calculations */
4228                 err = -EBUSY;
4229                 if (unlikely(rb->pg_vec))
4230                         goto out;
4231
4232                 switch (po->tp_version) {
4233                 case TPACKET_V1:
4234                         po->tp_hdrlen = TPACKET_HDRLEN;
4235                         break;
4236                 case TPACKET_V2:
4237                         po->tp_hdrlen = TPACKET2_HDRLEN;
4238                         break;
4239                 case TPACKET_V3:
4240                         po->tp_hdrlen = TPACKET3_HDRLEN;
4241                         break;
4242                 }
4243
4244                 err = -EINVAL;
4245                 if (unlikely((int)req->tp_block_size <= 0))
4246                         goto out;
4247                 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4248                         goto out;
4249                 if (po->tp_version >= TPACKET_V3 &&
4250                     req->tp_block_size <=
4251                           BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv))
4252                         goto out;
4253                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
4254                                         po->tp_reserve))
4255                         goto out;
4256                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4257                         goto out;
4258
4259                 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4260                 if (unlikely(rb->frames_per_block == 0))
4261                         goto out;
4262                 if (unlikely(req->tp_block_size > UINT_MAX / req->tp_block_nr))
4263                         goto out;
4264                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4265                                         req->tp_frame_nr))
4266                         goto out;
4267
4268                 err = -ENOMEM;
4269                 order = get_order(req->tp_block_size);
4270                 pg_vec = alloc_pg_vec(req, order);
4271                 if (unlikely(!pg_vec))
4272                         goto out;
4273                 switch (po->tp_version) {
4274                 case TPACKET_V3:
4275                         /* Block transmit is not supported yet */
4276                         if (!tx_ring) {
4277                                 init_prb_bdqc(po, rb, pg_vec, req_u);
4278                         } else {
4279                                 struct tpacket_req3 *req3 = &req_u->req3;
4280
4281                                 if (req3->tp_retire_blk_tov ||
4282                                     req3->tp_sizeof_priv ||
4283                                     req3->tp_feature_req_word) {
4284                                         err = -EINVAL;
4285                                         goto out;
4286                                 }
4287                         }
4288                         break;
4289                 default:
4290                         break;
4291                 }
4292         }
4293         /* Done */
4294         else {
4295                 err = -EINVAL;
4296                 if (unlikely(req->tp_frame_nr))
4297                         goto out;
4298         }
4299
4300
4301         /* Detach socket from network */
4302         spin_lock(&po->bind_lock);
4303         was_running = po->running;
4304         num = po->num;
4305         if (was_running) {
4306                 po->num = 0;
4307                 __unregister_prot_hook(sk, false);
4308         }
4309         spin_unlock(&po->bind_lock);
4310
4311         synchronize_net();
4312
4313         err = -EBUSY;
4314         mutex_lock(&po->pg_vec_lock);
4315         if (closing || atomic_read(&po->mapped) == 0) {
4316                 err = 0;
4317                 spin_lock_bh(&rb_queue->lock);
4318                 swap(rb->pg_vec, pg_vec);
4319                 rb->frame_max = (req->tp_frame_nr - 1);
4320                 rb->head = 0;
4321                 rb->frame_size = req->tp_frame_size;
4322                 spin_unlock_bh(&rb_queue->lock);
4323
4324                 swap(rb->pg_vec_order, order);
4325                 swap(rb->pg_vec_len, req->tp_block_nr);
4326
4327                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4328                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4329                                                 tpacket_rcv : packet_rcv;
4330                 skb_queue_purge(rb_queue);
4331                 if (atomic_read(&po->mapped))
4332                         pr_err("packet_mmap: vma is busy: %d\n",
4333                                atomic_read(&po->mapped));
4334         }
4335         mutex_unlock(&po->pg_vec_lock);
4336
4337         spin_lock(&po->bind_lock);
4338         if (was_running) {
4339                 po->num = num;
4340                 register_prot_hook(sk);
4341         }
4342         spin_unlock(&po->bind_lock);
4343         if (pg_vec && (po->tp_version > TPACKET_V2)) {
4344                 /* Because we don't support block-based V3 on tx-ring */
4345                 if (!tx_ring)
4346                         prb_shutdown_retire_blk_timer(po, rb_queue);
4347         }
4348
4349         if (pg_vec)
4350                 free_pg_vec(pg_vec, order, req->tp_block_nr);
4351 out:
4352         release_sock(sk);
4353         return err;
4354 }
4355
4356 static int packet_mmap(struct file *file, struct socket *sock,
4357                 struct vm_area_struct *vma)
4358 {
4359         struct sock *sk = sock->sk;
4360         struct packet_sock *po = pkt_sk(sk);
4361         unsigned long size, expected_size;
4362         struct packet_ring_buffer *rb;
4363         unsigned long start;
4364         int err = -EINVAL;
4365         int i;
4366
4367         if (vma->vm_pgoff)
4368                 return -EINVAL;
4369
4370         mutex_lock(&po->pg_vec_lock);
4371
4372         expected_size = 0;
4373         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4374                 if (rb->pg_vec) {
4375                         expected_size += rb->pg_vec_len
4376                                                 * rb->pg_vec_pages
4377                                                 * PAGE_SIZE;
4378                 }
4379         }
4380
4381         if (expected_size == 0)
4382                 goto out;
4383
4384         size = vma->vm_end - vma->vm_start;
4385         if (size != expected_size)
4386                 goto out;
4387
4388         start = vma->vm_start;
4389         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4390                 if (rb->pg_vec == NULL)
4391                         continue;
4392
4393                 for (i = 0; i < rb->pg_vec_len; i++) {
4394                         struct page *page;
4395                         void *kaddr = rb->pg_vec[i].buffer;
4396                         int pg_num;
4397
4398                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4399                                 page = pgv_to_page(kaddr);
4400                                 err = vm_insert_page(vma, start, page);
4401                                 if (unlikely(err))
4402                                         goto out;
4403                                 start += PAGE_SIZE;
4404                                 kaddr += PAGE_SIZE;
4405                         }
4406                 }
4407         }
4408
4409         atomic_inc(&po->mapped);
4410         vma->vm_ops = &packet_mmap_ops;
4411         err = 0;
4412
4413 out:
4414         mutex_unlock(&po->pg_vec_lock);
4415         return err;
4416 }
4417
4418 static const struct proto_ops packet_ops_spkt = {
4419         .family =       PF_PACKET,
4420         .owner =        THIS_MODULE,
4421         .release =      packet_release,
4422         .bind =         packet_bind_spkt,
4423         .connect =      sock_no_connect,
4424         .socketpair =   sock_no_socketpair,
4425         .accept =       sock_no_accept,
4426         .getname =      packet_getname_spkt,
4427         .poll =         datagram_poll,
4428         .ioctl =        packet_ioctl,
4429         .listen =       sock_no_listen,
4430         .shutdown =     sock_no_shutdown,
4431         .setsockopt =   sock_no_setsockopt,
4432         .getsockopt =   sock_no_getsockopt,
4433         .sendmsg =      packet_sendmsg_spkt,
4434         .recvmsg =      packet_recvmsg,
4435         .mmap =         sock_no_mmap,
4436         .sendpage =     sock_no_sendpage,
4437 };
4438
4439 static const struct proto_ops packet_ops = {
4440         .family =       PF_PACKET,
4441         .owner =        THIS_MODULE,
4442         .release =      packet_release,
4443         .bind =         packet_bind,
4444         .connect =      sock_no_connect,
4445         .socketpair =   sock_no_socketpair,
4446         .accept =       sock_no_accept,
4447         .getname =      packet_getname,
4448         .poll =         packet_poll,
4449         .ioctl =        packet_ioctl,
4450         .listen =       sock_no_listen,
4451         .shutdown =     sock_no_shutdown,
4452         .setsockopt =   packet_setsockopt,
4453         .getsockopt =   packet_getsockopt,
4454 #ifdef CONFIG_COMPAT
4455         .compat_setsockopt = compat_packet_setsockopt,
4456 #endif
4457         .sendmsg =      packet_sendmsg,
4458         .recvmsg =      packet_recvmsg,
4459         .mmap =         packet_mmap,
4460         .sendpage =     sock_no_sendpage,
4461 };
4462
4463 static const struct net_proto_family packet_family_ops = {
4464         .family =       PF_PACKET,
4465         .create =       packet_create,
4466         .owner  =       THIS_MODULE,
4467 };
4468
4469 static struct notifier_block packet_netdev_notifier = {
4470         .notifier_call =        packet_notifier,
4471 };
4472
4473 #ifdef CONFIG_PROC_FS
4474
4475 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4476         __acquires(RCU)
4477 {
4478         struct net *net = seq_file_net(seq);
4479
4480         rcu_read_lock();
4481         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4482 }
4483
4484 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4485 {
4486         struct net *net = seq_file_net(seq);
4487         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4488 }
4489
4490 static void packet_seq_stop(struct seq_file *seq, void *v)
4491         __releases(RCU)
4492 {
4493         rcu_read_unlock();
4494 }
4495
4496 static int packet_seq_show(struct seq_file *seq, void *v)
4497 {
4498         if (v == SEQ_START_TOKEN)
4499                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
4500         else {
4501                 struct sock *s = sk_entry(v);
4502                 const struct packet_sock *po = pkt_sk(s);
4503
4504                 seq_printf(seq,
4505                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4506                            s,
4507                            refcount_read(&s->sk_refcnt),
4508                            s->sk_type,
4509                            ntohs(po->num),
4510                            po->ifindex,
4511                            po->running,
4512                            atomic_read(&s->sk_rmem_alloc),
4513                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4514                            sock_i_ino(s));
4515         }
4516
4517         return 0;
4518 }
4519
4520 static const struct seq_operations packet_seq_ops = {
4521         .start  = packet_seq_start,
4522         .next   = packet_seq_next,
4523         .stop   = packet_seq_stop,
4524         .show   = packet_seq_show,
4525 };
4526
4527 static int packet_seq_open(struct inode *inode, struct file *file)
4528 {
4529         return seq_open_net(inode, file, &packet_seq_ops,
4530                             sizeof(struct seq_net_private));
4531 }
4532
4533 static const struct file_operations packet_seq_fops = {
4534         .owner          = THIS_MODULE,
4535         .open           = packet_seq_open,
4536         .read           = seq_read,
4537         .llseek         = seq_lseek,
4538         .release        = seq_release_net,
4539 };
4540
4541 #endif
4542
4543 static int __net_init packet_net_init(struct net *net)
4544 {
4545         mutex_init(&net->packet.sklist_lock);
4546         INIT_HLIST_HEAD(&net->packet.sklist);
4547
4548         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4549                 return -ENOMEM;
4550
4551         return 0;
4552 }
4553
4554 static void __net_exit packet_net_exit(struct net *net)
4555 {
4556         remove_proc_entry("packet", net->proc_net);
4557 }
4558
4559 static struct pernet_operations packet_net_ops = {
4560         .init = packet_net_init,
4561         .exit = packet_net_exit,
4562 };
4563
4564
4565 static void __exit packet_exit(void)
4566 {
4567         unregister_netdevice_notifier(&packet_netdev_notifier);
4568         unregister_pernet_subsys(&packet_net_ops);
4569         sock_unregister(PF_PACKET);
4570         proto_unregister(&packet_proto);
4571 }
4572
4573 static int __init packet_init(void)
4574 {
4575         int rc = proto_register(&packet_proto, 0);
4576
4577         if (rc != 0)
4578                 goto out;
4579
4580         sock_register(&packet_family_ops);
4581         register_pernet_subsys(&packet_net_ops);
4582         register_netdevice_notifier(&packet_netdev_notifier);
4583 out:
4584         return rc;
4585 }
4586
4587 module_init(packet_init);
4588 module_exit(packet_exit);
4589 MODULE_LICENSE("GPL");
4590 MODULE_ALIAS_NETPROTO(PF_PACKET);