net/packet/af_packet.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              PACKET - implements raw packet sockets.
   7  *
   8  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  11  *
  12  * Fixes:
  13  *              Alan Cox        :       verify_area() now used correctly
  14  *              Alan Cox        :       new skbuff lists, look ma no backlogs!
  15  *              Alan Cox        :       tidied skbuff lists.
  16  *              Alan Cox        :       Now uses generic datagram routines I
  17  *                                      added. Also fixed the peek/read crash
  18  *                                      from all old Linux datagram code.
  19  *              Alan Cox        :       Uses the improved datagram code.
  20  *              Alan Cox        :       Added NULL's for socket options.
  21  *              Alan Cox        :       Re-commented the code.
  22  *              Alan Cox        :       Use new kernel side addressing
  23  *              Rob Janssen     :       Correct MTU usage.
  24  *              Dave Platt      :       Counter leaks caused by incorrect
  25  *                                      interrupt locking and some slightly
  26  *                                      dubious gcc output. Can you read
  27  *                                      compiler: it said _VOLATILE_
  28  *      Richard Kooijman        :       Timestamp fixes.
  29  *              Alan Cox        :       New buffers. Use sk->mac.raw.
  30  *              Alan Cox        :       sendmsg/recvmsg support.
  31  *              Alan Cox        :       Protocol setting support
  32  *      Alexey Kuznetsov        :       Untied from IPv4 stack.
  33  *      Cyrus Durgin            :       Fixed kerneld for kmod.
  34  *      Michal Ostrowski        :       Module initialization cleanup.
  35  *         Ulises Alonso        :       Frame number limit removal and
  36  *                                      packet_set_ring memory leak.
  37  *              Eric Biederman  :       Allow for > 8 byte hardware addresses.
  38  *                                      The convention is that longer addresses
  39  *                                      will simply extend the hardware address
  40  *                                      byte arrays at the end of sockaddr_ll
  41  *                                      and packet_mreq.
  42  *              Johann Baudy    :       Added TX RING.
  43  *              Chetan Loke     :       Implemented TPACKET_V3 block abstraction
  44  *                                      layer.
  45  *                                      Copyright (C) 2011, <lokec@ccs.neu.edu>
  46  *
  47  *
  48  *              This program is free software; you can redistribute it and/or
  49  *              modify it under the terms of the GNU General Public License
  50  *              as published by the Free Software Foundation; either version
  51  *              2 of the License, or (at your option) any later version.
  52  *
  53  */
  54
  55 #include <linux/types.h>
  56 #include <linux/mm.h>
  57 #include <linux/capability.h>
  58 #include <linux/fcntl.h>
  59 #include <linux/socket.h>
  60 #include <linux/in.h>
  61 #include <linux/inet.h>
  62 #include <linux/netdevice.h>
  63 #include <linux/if_packet.h>
  64 #include <linux/wireless.h>
  65 #include <linux/kernel.h>
  66 #include <linux/kmod.h>
  67 #include <linux/slab.h>
  68 #include <linux/vmalloc.h>
  69 #include <net/net_namespace.h>
  70 #include <net/ip.h>
  71 #include <net/protocol.h>
  72 #include <linux/skbuff.h>
  73 #include <net/sock.h>
  74 #include <linux/errno.h>
  75 #include <linux/timer.h>
  76 #include <asm/uaccess.h>
  77 #include <asm/ioctls.h>
  78 #include <asm/page.h>
  79 #include <asm/cacheflush.h>
  80 #include <asm/io.h>
  81 #include <linux/proc_fs.h>
  82 #include <linux/seq_file.h>
  83 #include <linux/poll.h>
  84 #include <linux/module.h>
  85 #include <linux/init.h>
  86 #include <linux/mutex.h>
  87 #include <linux/if_vlan.h>
  88 #include <linux/virtio_net.h>
  89 #include <linux/errqueue.h>
  90 #include <linux/net_tstamp.h>
  91 #include <linux/percpu.h>
  92 #ifdef CONFIG_INET
  93 #include <net/inet_common.h>
  94 #endif
  95 #include <linux/bpf.h>
  96
  97 #include "internal.h"
  98
  99 /*
 100    Assumptions:
 101    - if device has no dev->hard_header routine, it adds and removes ll header
 102      inside itself. In this case ll header is invisible outside of device,
 103      but higher levels still should reserve dev->hard_header_len.
 104      Some devices are enough clever to reallocate skb, when header
 105      will not fit to reserved space (tunnel), another ones are silly
 106      (PPP).
 107    - packet socket receives packets with pulled ll header,
 108      so that SOCK_RAW should push it back.
 109
 110 On receive:
 111 -----------
 112
 113 Incoming, dev->hard_header!=NULL
 114    mac_header -> ll header
 115    data       -> data
 116
 117 Outgoing, dev->hard_header!=NULL
 118    mac_header -> ll header
 119    data       -> ll header
 120
 121 Incoming, dev->hard_header==NULL
 122    mac_header -> UNKNOWN position. It is very likely, that it points to ll
 123                  header.  PPP makes it, that is wrong, because introduce
 124                  assymetry between rx and tx paths.
 125    data       -> data
 126
 127 Outgoing, dev->hard_header==NULL
 128    mac_header -> data. ll header is still not built!
 129    data       -> data
 130
 131 Resume
 132   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 133
 134
 135 On transmit:
 136 ------------
 137
 138 dev->hard_header != NULL
 139    mac_header -> ll header
 140    data       -> ll header
 141
 142 dev->hard_header == NULL (ll header is added by device, we cannot control it)
 143    mac_header -> data
 144    data       -> data
 145
 146    We should set nh.raw on output to correct posistion,
 147    packet classifier depends on it.
 148  */
 149
 150 /* Private packet socket structures. */
 151
 152 /* identical to struct packet_mreq except it has
 153  * a longer address field.
 154  */
 155 struct packet_mreq_max {
 156         int             mr_ifindex;
 157         unsigned short  mr_type;
 158         unsigned short  mr_alen;
 159         unsigned char   mr_address[MAX_ADDR_LEN];
 160 };
 161
 162 union tpacket_uhdr {
 163         struct tpacket_hdr  *h1;
 164         struct tpacket2_hdr *h2;
 165         struct tpacket3_hdr *h3;
 166         void *raw;
 167 };
 168
 169 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 170                 int closing, int tx_ring);
 171
 172 #define V3_ALIGNMENT    (8)
 173
 174 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 175
 176 #define BLK_PLUS_PRIV(sz_of_priv) \
 177         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 178
 179 #define PGV_FROM_VMALLOC 1
 180
 181 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
 182 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
 183 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
 184 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
 185 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
 186 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
 187 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
 188
 189 struct packet_sock;
 190 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
 191 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 192                        struct packet_type *pt, struct net_device *orig_dev);
 193
 194 static void *packet_previous_frame(struct packet_sock *po,
 195                 struct packet_ring_buffer *rb,
 196                 int status);
 197 static void packet_increment_head(struct packet_ring_buffer *buff);
 198 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
 199                         struct tpacket_block_desc *);
 200 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 201                         struct packet_sock *);
 202 static void prb_retire_current_block(struct tpacket_kbdq_core *,
 203                 struct packet_sock *, unsigned int status);
 204 static int prb_queue_frozen(struct tpacket_kbdq_core *);
 205 static void prb_open_block(struct tpacket_kbdq_core *,
 206                 struct tpacket_block_desc *);
 207 static void prb_retire_rx_blk_timer_expired(unsigned long);
 208 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 209 static void prb_init_blk_timer(struct packet_sock *,
 210                 struct tpacket_kbdq_core *,
 211                 void (*func) (unsigned long));
 212 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 213 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 214                 struct tpacket3_hdr *);
 215 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 216                 struct tpacket3_hdr *);
 217 static void packet_flush_mclist(struct sock *sk);
 218
 219 struct packet_skb_cb {
 220         union {
 221                 struct sockaddr_pkt pkt;
 222                 union {
 223                         /* Trick: alias skb original length with
 224                          * ll.sll_family and ll.protocol in order
 225                          * to save room.
 226                          */
 227                         unsigned int origlen;
 228                         struct sockaddr_ll ll;
 229                 };
 230         } sa;
 231 };
 232
 233 #define vio_le() virtio_legacy_is_little_endian()
 234
 235 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 236
 237 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 238 #define GET_PBLOCK_DESC(x, bid) \
 239         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 240 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 241         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 242 #define GET_NEXT_PRB_BLK_NUM(x) \
 243         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 244         ((x)->kactive_blk_num+1) : 0)
 245
 246 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 247 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 248
 249 static int packet_direct_xmit(struct sk_buff *skb)
 250 {
 251         struct net_device *dev = skb->dev;
 252         struct sk_buff *orig_skb = skb;
 253         struct netdev_queue *txq;
 254         int ret = NETDEV_TX_BUSY;
 255
 256         if (unlikely(!netif_running(dev) ||
 257                      !netif_carrier_ok(dev)))
 258                 goto drop;
 259
 260         skb = validate_xmit_skb_list(skb, dev);
 261         if (skb != orig_skb)
 262                 goto drop;
 263
 264         txq = skb_get_tx_queue(dev, skb);
 265
 266         local_bh_disable();
 267
 268         HARD_TX_LOCK(dev, txq, smp_processor_id());
 269         if (!netif_xmit_frozen_or_drv_stopped(txq))
 270                 ret = netdev_start_xmit(skb, dev, txq, false);
 271         HARD_TX_UNLOCK(dev, txq);
 272
 273         local_bh_enable();
 274
 275         if (!dev_xmit_complete(ret))
 276                 kfree_skb(skb);
 277
 278         return ret;
 279 drop:
 280         atomic_long_inc(&dev->tx_dropped);
 281         kfree_skb_list(skb);
 282         return NET_XMIT_DROP;
 283 }
 284
 285 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
 286 {
 287         struct net_device *dev;
 288
 289         rcu_read_lock();
 290         dev = rcu_dereference(po->cached_dev);
 291         if (likely(dev))
 292                 dev_hold(dev);
 293         rcu_read_unlock();
 294
 295         return dev;
 296 }
 297
 298 static void packet_cached_dev_assign(struct packet_sock *po,
 299                                      struct net_device *dev)
 300 {
 301         rcu_assign_pointer(po->cached_dev, dev);
 302 }
 303
 304 static void packet_cached_dev_reset(struct packet_sock *po)
 305 {
 306         RCU_INIT_POINTER(po->cached_dev, NULL);
 307 }
 308
 309 static bool packet_use_direct_xmit(const struct packet_sock *po)
 310 {
 311         return po->xmit == packet_direct_xmit;
 312 }
 313
 314 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
 315 {
 316         return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
 317 }
 318
 319 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
 320 {
 321         const struct net_device_ops *ops = dev->netdev_ops;
 322         u16 queue_index;
 323
 324         if (ops->ndo_select_queue) {
 325                 queue_index = ops->ndo_select_queue(dev, skb, NULL,
 326                                                     __packet_pick_tx_queue);
 327                 queue_index = netdev_cap_txqueue(dev, queue_index);
 328         } else {
 329                 queue_index = __packet_pick_tx_queue(dev, skb);
 330         }
 331
 332         skb_set_queue_mapping(skb, queue_index);
 333 }
 334
 335 /* __register_prot_hook must be invoked through register_prot_hook
 336  * or from a context in which asynchronous accesses to the packet
 337  * socket is not possible (packet_create()).
 338  */
 339 static void __register_prot_hook(struct sock *sk)
 340 {
 341         struct packet_sock *po = pkt_sk(sk);
 342
 343         if (!po->running) {
 344                 if (po->fanout)
 345                         __fanout_link(sk, po);
 346                 else
 347                         dev_add_pack(&po->prot_hook);
 348
 349                 sock_hold(sk);
 350                 po->running = 1;
 351         }
 352 }
 353
 354 static void register_prot_hook(struct sock *sk)
 355 {
 356         lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
 357         __register_prot_hook(sk);
 358 }
 359
 360 /* If the sync parameter is true, we will temporarily drop
 361  * the po->bind_lock and do a synchronize_net to make sure no
 362  * asynchronous packet processing paths still refer to the elements
 363  * of po->prot_hook.  If the sync parameter is false, it is the
 364  * callers responsibility to take care of this.
 365  */
 366 static void __unregister_prot_hook(struct sock *sk, bool sync)
 367 {
 368         struct packet_sock *po = pkt_sk(sk);
 369
 370         lockdep_assert_held_once(&po->bind_lock);
 371
 372         po->running = 0;
 373
 374         if (po->fanout)
 375                 __fanout_unlink(sk, po);
 376         else
 377                 __dev_remove_pack(&po->prot_hook);
 378
 379         __sock_put(sk);
 380
 381         if (sync) {
 382                 spin_unlock(&po->bind_lock);
 383                 synchronize_net();
 384                 spin_lock(&po->bind_lock);
 385         }
 386 }
 387
 388 static void unregister_prot_hook(struct sock *sk, bool sync)
 389 {
 390         struct packet_sock *po = pkt_sk(sk);
 391
 392         if (po->running)
 393                 __unregister_prot_hook(sk, sync);
 394 }
 395
 396 static inline struct page * __pure pgv_to_page(void *addr)
 397 {
 398         if (is_vmalloc_addr(addr))
 399                 return vmalloc_to_page(addr);
 400         return virt_to_page(addr);
 401 }
 402
 403 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 404 {
 405         union tpacket_uhdr h;
 406
 407         h.raw = frame;
 408         switch (po->tp_version) {
 409         case TPACKET_V1:
 410                 h.h1->tp_status = status;
 411                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 412                 break;
 413         case TPACKET_V2:
 414                 h.h2->tp_status = status;
 415                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 416                 break;
 417         case TPACKET_V3:
 418         default:
 419                 WARN(1, "TPACKET version not supported.\n");
 420                 BUG();
 421         }
 422
 423         smp_wmb();
 424 }
 425
 426 static int __packet_get_status(struct packet_sock *po, void *frame)
 427 {
 428         union tpacket_uhdr h;
 429
 430         smp_rmb();
 431
 432         h.raw = frame;
 433         switch (po->tp_version) {
 434         case TPACKET_V1:
 435                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 436                 return h.h1->tp_status;
 437         case TPACKET_V2:
 438                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 439                 return h.h2->tp_status;
 440         case TPACKET_V3:
 441         default:
 442                 WARN(1, "TPACKET version not supported.\n");
 443                 BUG();
 444                 return 0;
 445         }
 446 }
 447
 448 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
 449                                    unsigned int flags)
 450 {
 451         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 452
 453         if (shhwtstamps &&
 454             (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
 455             ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
 456                 return TP_STATUS_TS_RAW_HARDWARE;
 457
 458         if (ktime_to_timespec_cond(skb->tstamp, ts))
 459                 return TP_STATUS_TS_SOFTWARE;
 460
 461         return 0;
 462 }
 463
 464 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
 465                                     struct sk_buff *skb)
 466 {
 467         union tpacket_uhdr h;
 468         struct timespec ts;
 469         __u32 ts_status;
 470
 471         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
 472                 return 0;
 473
 474         h.raw = frame;
 475         switch (po->tp_version) {
 476         case TPACKET_V1:
 477                 h.h1->tp_sec = ts.tv_sec;
 478                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
 479                 break;
 480         case TPACKET_V2:
 481                 h.h2->tp_sec = ts.tv_sec;
 482                 h.h2->tp_nsec = ts.tv_nsec;
 483                 break;
 484         case TPACKET_V3:
 485         default:
 486                 WARN(1, "TPACKET version not supported.\n");
 487                 BUG();
 488         }
 489
 490         /* one flush is safe, as both fields always lie on the same cacheline */
 491         flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
 492         smp_wmb();
 493
 494         return ts_status;
 495 }
 496
 497 static void *packet_lookup_frame(struct packet_sock *po,
 498                 struct packet_ring_buffer *rb,
 499                 unsigned int position,
 500                 int status)
 501 {
 502         unsigned int pg_vec_pos, frame_offset;
 503         union tpacket_uhdr h;
 504
 505         pg_vec_pos = position / rb->frames_per_block;
 506         frame_offset = position % rb->frames_per_block;
 507
 508         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 509                 (frame_offset * rb->frame_size);
 510
 511         if (status != __packet_get_status(po, h.raw))
 512                 return NULL;
 513
 514         return h.raw;
 515 }
 516
 517 static void *packet_current_frame(struct packet_sock *po,
 518                 struct packet_ring_buffer *rb,
 519                 int status)
 520 {
 521         return packet_lookup_frame(po, rb, rb->head, status);
 522 }
 523
 524 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 525 {
 526         del_timer_sync(&pkc->retire_blk_timer);
 527 }
 528
 529 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 530                 struct sk_buff_head *rb_queue)
 531 {
 532         struct tpacket_kbdq_core *pkc;
 533
 534         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 535
 536         spin_lock_bh(&rb_queue->lock);
 537         pkc->delete_blk_timer = 1;
 538         spin_unlock_bh(&rb_queue->lock);
 539
 540         prb_del_retire_blk_timer(pkc);
 541 }
 542
 543 static void prb_init_blk_timer(struct packet_sock *po,
 544                 struct tpacket_kbdq_core *pkc,
 545                 void (*func) (unsigned long))
 546 {
 547         init_timer(&pkc->retire_blk_timer);
 548         pkc->retire_blk_timer.data = (long)po;
 549         pkc->retire_blk_timer.function = func;
 550         pkc->retire_blk_timer.expires = jiffies;
 551 }
 552
 553 static void prb_setup_retire_blk_timer(struct packet_sock *po)
 554 {
 555         struct tpacket_kbdq_core *pkc;
 556
 557         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 558         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
 559 }
 560
 561 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 562                                 int blk_size_in_bytes)
 563 {
 564         struct net_device *dev;
 565         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 566         struct ethtool_cmd ecmd;
 567         int err;
 568         u32 speed;
 569
 570         rtnl_lock();
 571         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 572         if (unlikely(!dev)) {
 573                 rtnl_unlock();
 574                 return DEFAULT_PRB_RETIRE_TOV;
 575         }
 576         err = __ethtool_get_settings(dev, &ecmd);
 577         speed = ethtool_cmd_speed(&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 (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
 585                         return DEFAULT_PRB_RETIRE_TOV;
 586                 } else {
 587                         msec = 1;
 588                         div = speed / 1000;
 589                 }
 590         } else
 591                 return DEFAULT_PRB_RETIRE_TOV;
 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(pkc, 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_kbdq_core *pkc,
 976                                       struct tpacket_block_desc *pbd)
 977 {
 978         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 979 }
 980
 981 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 982 {
 983         return pkc->reset_pending_on_curr_blk;
 984 }
 985
 986 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 987 {
 988         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 989         atomic_dec(&pkc->blk_fill_in_prog);
 990 }
 991
 992 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 993                         struct tpacket3_hdr *ppd)
 994 {
 995         ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
 996 }
 997
 998 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 999                         struct tpacket3_hdr *ppd)
1000 {
1001         ppd->hv1.tp_rxhash = 0;
1002 }
1003
1004 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1005                         struct tpacket3_hdr *ppd)
1006 {
1007         if (skb_vlan_tag_present(pkc->skb)) {
1008                 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1009                 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1010                 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1011         } else {
1012                 ppd->hv1.tp_vlan_tci = 0;
1013                 ppd->hv1.tp_vlan_tpid = 0;
1014                 ppd->tp_status = TP_STATUS_AVAILABLE;
1015         }
1016 }
1017
1018 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1019                         struct tpacket3_hdr *ppd)
1020 {
1021         ppd->hv1.tp_padding = 0;
1022         prb_fill_vlan_info(pkc, ppd);
1023
1024         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1025                 prb_fill_rxhash(pkc, ppd);
1026         else
1027                 prb_clear_rxhash(pkc, ppd);
1028 }
1029
1030 static void prb_fill_curr_block(char *curr,
1031                                 struct tpacket_kbdq_core *pkc,
1032                                 struct tpacket_block_desc *pbd,
1033                                 unsigned int len)
1034 {
1035         struct tpacket3_hdr *ppd;
1036
1037         ppd  = (struct tpacket3_hdr *)curr;
1038         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1039         pkc->prev = curr;
1040         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1041         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1042         BLOCK_NUM_PKTS(pbd) += 1;
1043         atomic_inc(&pkc->blk_fill_in_prog);
1044         prb_run_all_ft_ops(pkc, ppd);
1045 }
1046
1047 /* Assumes caller has the sk->rx_queue.lock */
1048 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1049                                             struct sk_buff *skb,
1050                                                 int status,
1051                                             unsigned int len
1052                                             )
1053 {
1054         struct tpacket_kbdq_core *pkc;
1055         struct tpacket_block_desc *pbd;
1056         char *curr, *end;
1057
1058         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1059         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1060
1061         /* Queue is frozen when user space is lagging behind */
1062         if (prb_queue_frozen(pkc)) {
1063                 /*
1064                  * Check if that last block which caused the queue to freeze,
1065                  * is still in_use by user-space.
1066                  */
1067                 if (prb_curr_blk_in_use(pkc, pbd)) {
1068                         /* Can't record this packet */
1069                         return NULL;
1070                 } else {
1071                         /*
1072                          * Ok, the block was released by user-space.
1073                          * Now let's open that block.
1074                          * opening a block also thaws the queue.
1075                          * Thawing is a side effect.
1076                          */
1077                         prb_open_block(pkc, pbd);
1078                 }
1079         }
1080
1081         smp_mb();
1082         curr = pkc->nxt_offset;
1083         pkc->skb = skb;
1084         end = (char *)pbd + pkc->kblk_size;
1085
1086         /* first try the current block */
1087         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1088                 prb_fill_curr_block(curr, pkc, pbd, len);
1089                 return (void *)curr;
1090         }
1091
1092         /* Ok, close the current block */
1093         prb_retire_current_block(pkc, po, 0);
1094
1095         /* Now, try to dispatch the next block */
1096         curr = (char *)prb_dispatch_next_block(pkc, po);
1097         if (curr) {
1098                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1099                 prb_fill_curr_block(curr, pkc, pbd, len);
1100                 return (void *)curr;
1101         }
1102
1103         /*
1104          * No free blocks are available.user_space hasn't caught up yet.
1105          * Queue was just frozen and now this packet will get dropped.
1106          */
1107         return NULL;
1108 }
1109
1110 static void *packet_current_rx_frame(struct packet_sock *po,
1111                                             struct sk_buff *skb,
1112                                             int status, unsigned int len)
1113 {
1114         char *curr = NULL;
1115         switch (po->tp_version) {
1116         case TPACKET_V1:
1117         case TPACKET_V2:
1118                 curr = packet_lookup_frame(po, &po->rx_ring,
1119                                         po->rx_ring.head, status);
1120                 return curr;
1121         case TPACKET_V3:
1122                 return __packet_lookup_frame_in_block(po, skb, status, len);
1123         default:
1124                 WARN(1, "TPACKET version not supported\n");
1125                 BUG();
1126                 return NULL;
1127         }
1128 }
1129
1130 static void *prb_lookup_block(struct packet_sock *po,
1131                                      struct packet_ring_buffer *rb,
1132                                      unsigned int idx,
1133                                      int status)
1134 {
1135         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1136         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1137
1138         if (status != BLOCK_STATUS(pbd))
1139                 return NULL;
1140         return pbd;
1141 }
1142
1143 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1144 {
1145         unsigned int prev;
1146         if (rb->prb_bdqc.kactive_blk_num)
1147                 prev = rb->prb_bdqc.kactive_blk_num-1;
1148         else
1149                 prev = rb->prb_bdqc.knum_blocks-1;
1150         return prev;
1151 }
1152
1153 /* Assumes caller has held the rx_queue.lock */
1154 static void *__prb_previous_block(struct packet_sock *po,
1155                                          struct packet_ring_buffer *rb,
1156                                          int status)
1157 {
1158         unsigned int previous = prb_previous_blk_num(rb);
1159         return prb_lookup_block(po, rb, previous, status);
1160 }
1161
1162 static void *packet_previous_rx_frame(struct packet_sock *po,
1163                                              struct packet_ring_buffer *rb,
1164                                              int status)
1165 {
1166         if (po->tp_version <= TPACKET_V2)
1167                 return packet_previous_frame(po, rb, status);
1168
1169         return __prb_previous_block(po, rb, status);
1170 }
1171
1172 static void packet_increment_rx_head(struct packet_sock *po,
1173                                             struct packet_ring_buffer *rb)
1174 {
1175         switch (po->tp_version) {
1176         case TPACKET_V1:
1177         case TPACKET_V2:
1178                 return packet_increment_head(rb);
1179         case TPACKET_V3:
1180         default:
1181                 WARN(1, "TPACKET version not supported.\n");
1182                 BUG();
1183                 return;
1184         }
1185 }
1186
1187 static void *packet_previous_frame(struct packet_sock *po,
1188                 struct packet_ring_buffer *rb,
1189                 int status)
1190 {
1191         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1192         return packet_lookup_frame(po, rb, previous, status);
1193 }
1194
1195 static void packet_increment_head(struct packet_ring_buffer *buff)
1196 {
1197         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1198 }
1199
1200 static void packet_inc_pending(struct packet_ring_buffer *rb)
1201 {
1202         this_cpu_inc(*rb->pending_refcnt);
1203 }
1204
1205 static void packet_dec_pending(struct packet_ring_buffer *rb)
1206 {
1207         this_cpu_dec(*rb->pending_refcnt);
1208 }
1209
1210 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1211 {
1212         unsigned int refcnt = 0;
1213         int cpu;
1214
1215         /* We don't use pending refcount in rx_ring. */
1216         if (rb->pending_refcnt == NULL)
1217                 return 0;
1218
1219         for_each_possible_cpu(cpu)
1220                 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1221
1222         return refcnt;
1223 }
1224
1225 static int packet_alloc_pending(struct packet_sock *po)
1226 {
1227         po->rx_ring.pending_refcnt = NULL;
1228
1229         po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1230         if (unlikely(po->tx_ring.pending_refcnt == NULL))
1231                 return -ENOBUFS;
1232
1233         return 0;
1234 }
1235
1236 static void packet_free_pending(struct packet_sock *po)
1237 {
1238         free_percpu(po->tx_ring.pending_refcnt);
1239 }
1240
1241 #define ROOM_POW_OFF    2
1242 #define ROOM_NONE       0x0
1243 #define ROOM_LOW        0x1
1244 #define ROOM_NORMAL     0x2
1245
1246 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1247 {
1248         int idx, len;
1249
1250         len = po->rx_ring.frame_max + 1;
1251         idx = po->rx_ring.head;
1252         if (pow_off)
1253                 idx += len >> pow_off;
1254         if (idx >= len)
1255                 idx -= len;
1256         return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1257 }
1258
1259 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1260 {
1261         int idx, len;
1262
1263         len = po->rx_ring.prb_bdqc.knum_blocks;
1264         idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1265         if (pow_off)
1266                 idx += len >> pow_off;
1267         if (idx >= len)
1268                 idx -= len;
1269         return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1270 }
1271
1272 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1273 {
1274         struct sock *sk = &po->sk;
1275         int ret = ROOM_NONE;
1276
1277         if (po->prot_hook.func != tpacket_rcv) {
1278                 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1279                                           - (skb ? skb->truesize : 0);
1280                 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1281                         return ROOM_NORMAL;
1282                 else if (avail > 0)
1283                         return ROOM_LOW;
1284                 else
1285                         return ROOM_NONE;
1286         }
1287
1288         if (po->tp_version == TPACKET_V3) {
1289                 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1290                         ret = ROOM_NORMAL;
1291                 else if (__tpacket_v3_has_room(po, 0))
1292                         ret = ROOM_LOW;
1293         } else {
1294                 if (__tpacket_has_room(po, ROOM_POW_OFF))
1295                         ret = ROOM_NORMAL;
1296                 else if (__tpacket_has_room(po, 0))
1297                         ret = ROOM_LOW;
1298         }
1299
1300         return ret;
1301 }
1302
1303 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1304 {
1305         int ret;
1306         bool has_room;
1307
1308         spin_lock_bh(&po->sk.sk_receive_queue.lock);
1309         ret = __packet_rcv_has_room(po, skb);
1310         has_room = ret == ROOM_NORMAL;
1311         if (po->pressure == has_room)
1312                 po->pressure = !has_room;
1313         spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1314
1315         return ret;
1316 }
1317
1318 static void packet_sock_destruct(struct sock *sk)
1319 {
1320         skb_queue_purge(&sk->sk_error_queue);
1321
1322         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1323         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1324
1325         if (!sock_flag(sk, SOCK_DEAD)) {
1326                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1327                 return;
1328         }
1329
1330         sk_refcnt_debug_dec(sk);
1331 }
1332
1333 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1334 {
1335         u32 *history = po->rollover->history;
1336         u32 victim, rxhash;
1337         int i, count = 0;
1338
1339         rxhash = skb_get_hash(skb);
1340         for (i = 0; i < ROLLOVER_HLEN; i++)
1341                 if (READ_ONCE(history[i]) == rxhash)
1342                         count++;
1343
1344         victim = prandom_u32() % ROLLOVER_HLEN;
1345
1346         /* Avoid dirtying the cache line if possible */
1347         if (READ_ONCE(history[victim]) != rxhash)
1348                 WRITE_ONCE(history[victim], rxhash);
1349
1350         return count > (ROLLOVER_HLEN >> 1);
1351 }
1352
1353 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1354                                       struct sk_buff *skb,
1355                                       unsigned int num)
1356 {
1357         return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1358 }
1359
1360 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1361                                     struct sk_buff *skb,
1362                                     unsigned int num)
1363 {
1364         unsigned int val = atomic_inc_return(&f->rr_cur);
1365
1366         return val % num;
1367 }
1368
1369 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1370                                      struct sk_buff *skb,
1371                                      unsigned int num)
1372 {
1373         return smp_processor_id() % num;
1374 }
1375
1376 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1377                                      struct sk_buff *skb,
1378                                      unsigned int num)
1379 {
1380         return prandom_u32_max(num);
1381 }
1382
1383 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1384                                           struct sk_buff *skb,
1385                                           unsigned int idx, bool try_self,
1386                                           unsigned int num)
1387 {
1388         struct packet_sock *po, *po_next, *po_skip = NULL;
1389         unsigned int i, j, room = ROOM_NONE;
1390
1391         po = pkt_sk(f->arr[idx]);
1392
1393         if (try_self) {
1394                 room = packet_rcv_has_room(po, skb);
1395                 if (room == ROOM_NORMAL ||
1396                     (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1397                         return idx;
1398                 po_skip = po;
1399         }
1400
1401         i = j = min_t(int, po->rollover->sock, num - 1);
1402         do {
1403                 po_next = pkt_sk(f->arr[i]);
1404                 if (po_next != po_skip && !po_next->pressure &&
1405                     packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1406                         if (i != j)
1407                                 po->rollover->sock = i;
1408                         atomic_long_inc(&po->rollover->num);
1409                         if (room == ROOM_LOW)
1410                                 atomic_long_inc(&po->rollover->num_huge);
1411                         return i;
1412                 }
1413
1414                 if (++i == num)
1415                         i = 0;
1416         } while (i != j);
1417
1418         atomic_long_inc(&po->rollover->num_failed);
1419         return idx;
1420 }
1421
1422 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1423                                     struct sk_buff *skb,
1424                                     unsigned int num)
1425 {
1426         return skb_get_queue_mapping(skb) % num;
1427 }
1428
1429 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1430                                      struct sk_buff *skb,
1431                                      unsigned int num)
1432 {
1433         struct bpf_prog *prog;
1434         unsigned int ret = 0;
1435
1436         rcu_read_lock();
1437         prog = rcu_dereference(f->bpf_prog);
1438         if (prog)
1439                 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1440         rcu_read_unlock();
1441
1442         return ret;
1443 }
1444
1445 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1446 {
1447         return f->flags & (flag >> 8);
1448 }
1449
1450 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1451                              struct packet_type *pt, struct net_device *orig_dev)
1452 {
1453         struct packet_fanout *f = pt->af_packet_priv;
1454         unsigned int num = READ_ONCE(f->num_members);
1455         struct net *net = read_pnet(&f->net);
1456         struct packet_sock *po;
1457         unsigned int idx;
1458
1459         if (!net_eq(dev_net(dev), net) || !num) {
1460                 kfree_skb(skb);
1461                 return 0;
1462         }
1463
1464         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1465                 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1466                 if (!skb)
1467                         return 0;
1468         }
1469         switch (f->type) {
1470         case PACKET_FANOUT_HASH:
1471         default:
1472                 idx = fanout_demux_hash(f, skb, num);
1473                 break;
1474         case PACKET_FANOUT_LB:
1475                 idx = fanout_demux_lb(f, skb, num);
1476                 break;
1477         case PACKET_FANOUT_CPU:
1478                 idx = fanout_demux_cpu(f, skb, num);
1479                 break;
1480         case PACKET_FANOUT_RND:
1481                 idx = fanout_demux_rnd(f, skb, num);
1482                 break;
1483         case PACKET_FANOUT_QM:
1484                 idx = fanout_demux_qm(f, skb, num);
1485                 break;
1486         case PACKET_FANOUT_ROLLOVER:
1487                 idx = fanout_demux_rollover(f, skb, 0, false, num);
1488                 break;
1489         case PACKET_FANOUT_CBPF:
1490         case PACKET_FANOUT_EBPF:
1491                 idx = fanout_demux_bpf(f, skb, num);
1492                 break;
1493         }
1494
1495         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1496                 idx = fanout_demux_rollover(f, skb, idx, true, num);
1497
1498         po = pkt_sk(f->arr[idx]);
1499         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1500 }
1501
1502 DEFINE_MUTEX(fanout_mutex);
1503 EXPORT_SYMBOL_GPL(fanout_mutex);
1504 static LIST_HEAD(fanout_list);
1505
1506 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1507 {
1508         struct packet_fanout *f = po->fanout;
1509
1510         spin_lock(&f->lock);
1511         f->arr[f->num_members] = sk;
1512         smp_wmb();
1513         f->num_members++;
1514         if (f->num_members == 1)
1515                 dev_add_pack(&f->prot_hook);
1516         spin_unlock(&f->lock);
1517 }
1518
1519 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1520 {
1521         struct packet_fanout *f = po->fanout;
1522         int i;
1523
1524         spin_lock(&f->lock);
1525         for (i = 0; i < f->num_members; i++) {
1526                 if (f->arr[i] == sk)
1527                         break;
1528         }
1529         BUG_ON(i >= f->num_members);
1530         f->arr[i] = f->arr[f->num_members - 1];
1531         f->num_members--;
1532         if (f->num_members == 0)
1533                 __dev_remove_pack(&f->prot_hook);
1534         spin_unlock(&f->lock);
1535 }
1536
1537 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1538 {
1539         if (sk->sk_family != PF_PACKET)
1540                 return false;
1541
1542         return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1543 }
1544
1545 static void fanout_init_data(struct packet_fanout *f)
1546 {
1547         switch (f->type) {
1548         case PACKET_FANOUT_LB:
1549                 atomic_set(&f->rr_cur, 0);
1550                 break;
1551         case PACKET_FANOUT_CBPF:
1552         case PACKET_FANOUT_EBPF:
1553                 RCU_INIT_POINTER(f->bpf_prog, NULL);
1554                 break;
1555         }
1556 }
1557
1558 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1559 {
1560         struct bpf_prog *old;
1561
1562         spin_lock(&f->lock);
1563         old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1564         rcu_assign_pointer(f->bpf_prog, new);
1565         spin_unlock(&f->lock);
1566
1567         if (old) {
1568                 synchronize_net();
1569                 bpf_prog_destroy(old);
1570         }
1571 }
1572
1573 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1574                                 unsigned int len)
1575 {
1576         struct bpf_prog *new;
1577         struct sock_fprog fprog;
1578         int ret;
1579
1580         if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1581                 return -EPERM;
1582         if (len != sizeof(fprog))
1583                 return -EINVAL;
1584         if (copy_from_user(&fprog, data, len))
1585                 return -EFAULT;
1586
1587         ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1588         if (ret)
1589                 return ret;
1590
1591         __fanout_set_data_bpf(po->fanout, new);
1592         return 0;
1593 }
1594
1595 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1596                                 unsigned int len)
1597 {
1598         struct bpf_prog *new;
1599         u32 fd;
1600
1601         if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1602                 return -EPERM;
1603         if (len != sizeof(fd))
1604                 return -EINVAL;
1605         if (copy_from_user(&fd, data, len))
1606                 return -EFAULT;
1607
1608         new = bpf_prog_get(fd);
1609         if (IS_ERR(new))
1610                 return PTR_ERR(new);
1611         if (new->type != BPF_PROG_TYPE_SOCKET_FILTER) {
1612                 bpf_prog_put(new);
1613                 return -EINVAL;
1614         }
1615
1616         __fanout_set_data_bpf(po->fanout, new);
1617         return 0;
1618 }
1619
1620 static int fanout_set_data(struct packet_sock *po, char __user *data,
1621                            unsigned int len)
1622 {
1623         switch (po->fanout->type) {
1624         case PACKET_FANOUT_CBPF:
1625                 return fanout_set_data_cbpf(po, data, len);
1626         case PACKET_FANOUT_EBPF:
1627                 return fanout_set_data_ebpf(po, data, len);
1628         default:
1629                 return -EINVAL;
1630         };
1631 }
1632
1633 static void fanout_release_data(struct packet_fanout *f)
1634 {
1635         switch (f->type) {
1636         case PACKET_FANOUT_CBPF:
1637         case PACKET_FANOUT_EBPF:
1638                 __fanout_set_data_bpf(f, NULL);
1639         };
1640 }
1641
1642 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1643 {
1644         struct packet_rollover *rollover = NULL;
1645         struct packet_sock *po = pkt_sk(sk);
1646         struct packet_fanout *f, *match;
1647         u8 type = type_flags & 0xff;
1648         u8 flags = type_flags >> 8;
1649         int err;
1650
1651         switch (type) {
1652         case PACKET_FANOUT_ROLLOVER:
1653                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1654                         return -EINVAL;
1655         case PACKET_FANOUT_HASH:
1656         case PACKET_FANOUT_LB:
1657         case PACKET_FANOUT_CPU:
1658         case PACKET_FANOUT_RND:
1659         case PACKET_FANOUT_QM:
1660         case PACKET_FANOUT_CBPF:
1661         case PACKET_FANOUT_EBPF:
1662                 break;
1663         default:
1664                 return -EINVAL;
1665         }
1666
1667         mutex_lock(&fanout_mutex);
1668
1669         err = -EALREADY;
1670         if (po->fanout)
1671                 goto out;
1672
1673         if (type == PACKET_FANOUT_ROLLOVER ||
1674             (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1675                 err = -ENOMEM;
1676                 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1677                 if (!rollover)
1678                         goto out;
1679                 atomic_long_set(&rollover->num, 0);
1680                 atomic_long_set(&rollover->num_huge, 0);
1681                 atomic_long_set(&rollover->num_failed, 0);
1682         }
1683
1684         match = NULL;
1685         list_for_each_entry(f, &fanout_list, list) {
1686                 if (f->id == id &&
1687                     read_pnet(&f->net) == sock_net(sk)) {
1688                         match = f;
1689                         break;
1690                 }
1691         }
1692         err = -EINVAL;
1693         if (match && match->flags != flags)
1694                 goto out;
1695         if (!match) {
1696                 err = -ENOMEM;
1697                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1698                 if (!match)
1699                         goto out;
1700                 write_pnet(&match->net, sock_net(sk));
1701                 match->id = id;
1702                 match->type = type;
1703                 match->flags = flags;
1704                 INIT_LIST_HEAD(&match->list);
1705                 spin_lock_init(&match->lock);
1706                 atomic_set(&match->sk_ref, 0);
1707                 fanout_init_data(match);
1708                 match->prot_hook.type = po->prot_hook.type;
1709                 match->prot_hook.dev = po->prot_hook.dev;
1710                 match->prot_hook.func = packet_rcv_fanout;
1711                 match->prot_hook.af_packet_priv = match;
1712                 match->prot_hook.id_match = match_fanout_group;
1713                 list_add(&match->list, &fanout_list);
1714         }
1715         err = -EINVAL;
1716
1717         spin_lock(&po->bind_lock);
1718         if (po->running &&
1719             match->type == type &&
1720             match->prot_hook.type == po->prot_hook.type &&
1721             match->prot_hook.dev == po->prot_hook.dev) {
1722                 err = -ENOSPC;
1723                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1724                         __dev_remove_pack(&po->prot_hook);
1725                         po->fanout = match;
1726                         po->rollover = rollover;
1727                         rollover = NULL;
1728                         atomic_inc(&match->sk_ref);
1729                         __fanout_link(sk, po);
1730                         err = 0;
1731                 }
1732         }
1733         spin_unlock(&po->bind_lock);
1734
1735         if (err && !atomic_read(&match->sk_ref)) {
1736                 list_del(&match->list);
1737                 kfree(match);
1738         }
1739
1740 out:
1741         kfree(rollover);
1742         mutex_unlock(&fanout_mutex);
1743         return err;
1744 }
1745
1746 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1747  * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1748  * It is the responsibility of the caller to call fanout_release_data() and
1749  * free the returned packet_fanout (after synchronize_net())
1750  */
1751 static struct packet_fanout *fanout_release(struct sock *sk)
1752 {
1753         struct packet_sock *po = pkt_sk(sk);
1754         struct packet_fanout *f;
1755
1756         mutex_lock(&fanout_mutex);
1757         f = po->fanout;
1758         if (f) {
1759                 po->fanout = NULL;
1760
1761                 if (atomic_dec_and_test(&f->sk_ref))
1762                         list_del(&f->list);
1763                 else
1764                         f = NULL;
1765         }
1766         mutex_unlock(&fanout_mutex);
1767
1768         return f;
1769 }
1770
1771 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1772                                           struct sk_buff *skb)
1773 {
1774         /* Earlier code assumed this would be a VLAN pkt, double-check
1775          * this now that we have the actual packet in hand. We can only
1776          * do this check on Ethernet devices.
1777          */
1778         if (unlikely(dev->type != ARPHRD_ETHER))
1779                 return false;
1780
1781         skb_reset_mac_header(skb);
1782         return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1783 }
1784
1785 static const struct proto_ops packet_ops;
1786
1787 static const struct proto_ops packet_ops_spkt;
1788
1789 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1790                            struct packet_type *pt, struct net_device *orig_dev)
1791 {
1792         struct sock *sk;
1793         struct sockaddr_pkt *spkt;
1794
1795         /*
1796          *      When we registered the protocol we saved the socket in the data
1797          *      field for just this event.
1798          */
1799
1800         sk = pt->af_packet_priv;
1801
1802         /*
1803          *      Yank back the headers [hope the device set this
1804          *      right or kerboom...]
1805          *
1806          *      Incoming packets have ll header pulled,
1807          *      push it back.
1808          *
1809          *      For outgoing ones skb->data == skb_mac_header(skb)
1810          *      so that this procedure is noop.
1811          */
1812
1813         if (skb->pkt_type == PACKET_LOOPBACK)
1814                 goto out;
1815
1816         if (!net_eq(dev_net(dev), sock_net(sk)))
1817                 goto out;
1818
1819         skb = skb_share_check(skb, GFP_ATOMIC);
1820         if (skb == NULL)
1821                 goto oom;
1822
1823         /* drop any routing info */
1824         skb_dst_drop(skb);
1825
1826         /* drop conntrack reference */
1827         nf_reset(skb);
1828
1829         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1830
1831         skb_push(skb, skb->data - skb_mac_header(skb));
1832
1833         /*
1834          *      The SOCK_PACKET socket receives _all_ frames.
1835          */
1836
1837         spkt->spkt_family = dev->type;
1838         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1839         spkt->spkt_protocol = skb->protocol;
1840
1841         /*
1842          *      Charge the memory to the socket. This is done specifically
1843          *      to prevent sockets using all the memory up.
1844          */
1845
1846         if (sock_queue_rcv_skb(sk, skb) == 0)
1847                 return 0;
1848
1849 out:
1850         kfree_skb(skb);
1851 oom:
1852         return 0;
1853 }
1854
1855
1856 /*
1857  *      Output a raw packet to a device layer. This bypasses all the other
1858  *      protocol layers and you must therefore supply it with a complete frame
1859  */
1860
1861 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1862                                size_t len)
1863 {
1864         struct sock *sk = sock->sk;
1865         DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1866         struct sk_buff *skb = NULL;
1867         struct net_device *dev;
1868         __be16 proto = 0;
1869         int err;
1870         int extra_len = 0;
1871
1872         /*
1873          *      Get and verify the address.
1874          */
1875
1876         if (saddr) {
1877                 if (msg->msg_namelen < sizeof(struct sockaddr))
1878                         return -EINVAL;
1879                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1880                         proto = saddr->spkt_protocol;
1881         } else
1882                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1883
1884         /*
1885          *      Find the device first to size check it
1886          */
1887
1888         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1889 retry:
1890         rcu_read_lock();
1891         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1892         err = -ENODEV;
1893         if (dev == NULL)
1894                 goto out_unlock;
1895
1896         err = -ENETDOWN;
1897         if (!(dev->flags & IFF_UP))
1898                 goto out_unlock;
1899
1900         /*
1901          * You may not queue a frame bigger than the mtu. This is the lowest level
1902          * raw protocol and you must do your own fragmentation at this level.
1903          */
1904
1905         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1906                 if (!netif_supports_nofcs(dev)) {
1907                         err = -EPROTONOSUPPORT;
1908                         goto out_unlock;
1909                 }
1910                 extra_len = 4; /* We're doing our own CRC */
1911         }
1912
1913         err = -EMSGSIZE;
1914         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1915                 goto out_unlock;
1916
1917         if (!skb) {
1918                 size_t reserved = LL_RESERVED_SPACE(dev);
1919                 int tlen = dev->needed_tailroom;
1920                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1921
1922                 rcu_read_unlock();
1923                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1924                 if (skb == NULL)
1925                         return -ENOBUFS;
1926                 /* FIXME: Save some space for broken drivers that write a hard
1927                  * header at transmission time by themselves. PPP is the notable
1928                  * one here. This should really be fixed at the driver level.
1929                  */
1930                 skb_reserve(skb, reserved);
1931                 skb_reset_network_header(skb);
1932
1933                 /* Try to align data part correctly */
1934                 if (hhlen) {
1935                         skb->data -= hhlen;
1936                         skb->tail -= hhlen;
1937                         if (len < hhlen)
1938                                 skb_reset_network_header(skb);
1939                 }
1940                 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1941                 if (err)
1942                         goto out_free;
1943                 goto retry;
1944         }
1945
1946         if (!dev_validate_header(dev, skb->data, len)) {
1947                 err = -EINVAL;
1948                 goto out_unlock;
1949         }
1950         if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1951             !packet_extra_vlan_len_allowed(dev, skb)) {
1952                 err = -EMSGSIZE;
1953                 goto out_unlock;
1954         }
1955
1956         skb->protocol = proto;
1957         skb->dev = dev;
1958         skb->priority = sk->sk_priority;
1959         skb->mark = sk->sk_mark;
1960
1961         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1962
1963         if (unlikely(extra_len == 4))
1964                 skb->no_fcs = 1;
1965
1966         skb_probe_transport_header(skb, 0);
1967
1968         dev_queue_xmit(skb);
1969         rcu_read_unlock();
1970         return len;
1971
1972 out_unlock:
1973         rcu_read_unlock();
1974 out_free:
1975         kfree_skb(skb);
1976         return err;
1977 }
1978
1979 static unsigned int run_filter(struct sk_buff *skb,
1980                                const struct sock *sk,
1981                                unsigned int res)
1982 {
1983         struct sk_filter *filter;
1984
1985         rcu_read_lock();
1986         filter = rcu_dereference(sk->sk_filter);
1987         if (filter != NULL)
1988                 res = bpf_prog_run_clear_cb(filter->prog, skb);
1989         rcu_read_unlock();
1990
1991         return res;
1992 }
1993
1994 /*
1995  * This function makes lazy skb cloning in hope that most of packets
1996  * are discarded by BPF.
1997  *
1998  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1999  * and skb->cb are mangled. It works because (and until) packets
2000  * falling here are owned by current CPU. Output packets are cloned
2001  * by dev_queue_xmit_nit(), input packets are processed by net_bh
2002  * sequencially, so that if we return skb to original state on exit,
2003  * we will not harm anyone.
2004  */
2005
2006 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2007                       struct packet_type *pt, struct net_device *orig_dev)
2008 {
2009         struct sock *sk;
2010         struct sockaddr_ll *sll;
2011         struct packet_sock *po;
2012         u8 *skb_head = skb->data;
2013         int skb_len = skb->len;
2014         unsigned int snaplen, res;
2015
2016         if (skb->pkt_type == PACKET_LOOPBACK)
2017                 goto drop;
2018
2019         sk = pt->af_packet_priv;
2020         po = pkt_sk(sk);
2021
2022         if (!net_eq(dev_net(dev), sock_net(sk)))
2023                 goto drop;
2024
2025         skb->dev = dev;
2026
2027         if (dev->header_ops) {
2028                 /* The device has an explicit notion of ll header,
2029                  * exported to higher levels.
2030                  *
2031                  * Otherwise, the device hides details of its frame
2032                  * structure, so that corresponding packet head is
2033                  * never delivered to user.
2034                  */
2035                 if (sk->sk_type != SOCK_DGRAM)
2036                         skb_push(skb, skb->data - skb_mac_header(skb));
2037                 else if (skb->pkt_type == PACKET_OUTGOING) {
2038                         /* Special case: outgoing packets have ll header at head */
2039                         skb_pull(skb, skb_network_offset(skb));
2040                 }
2041         }
2042
2043         snaplen = skb->len;
2044
2045         res = run_filter(skb, sk, snaplen);
2046         if (!res)
2047                 goto drop_n_restore;
2048         if (snaplen > res)
2049                 snaplen = res;
2050
2051         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2052                 goto drop_n_acct;
2053
2054         if (skb_shared(skb)) {
2055                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2056                 if (nskb == NULL)
2057                         goto drop_n_acct;
2058
2059                 if (skb_head != skb->data) {
2060                         skb->data = skb_head;
2061                         skb->len = skb_len;
2062                 }
2063                 consume_skb(skb);
2064                 skb = nskb;
2065         }
2066
2067         sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2068
2069         sll = &PACKET_SKB_CB(skb)->sa.ll;
2070         sll->sll_hatype = dev->type;
2071         sll->sll_pkttype = skb->pkt_type;
2072         if (unlikely(po->origdev))
2073                 sll->sll_ifindex = orig_dev->ifindex;
2074         else
2075                 sll->sll_ifindex = dev->ifindex;
2076
2077         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2078
2079         /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2080          * Use their space for storing the original skb length.
2081          */
2082         PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2083
2084         if (pskb_trim(skb, snaplen))
2085                 goto drop_n_acct;
2086
2087         skb_set_owner_r(skb, sk);
2088         skb->dev = NULL;
2089         skb_dst_drop(skb);
2090
2091         /* drop conntrack reference */
2092         nf_reset(skb);
2093
2094         spin_lock(&sk->sk_receive_queue.lock);
2095         po->stats.stats1.tp_packets++;
2096         sock_skb_set_dropcount(sk, skb);
2097         __skb_queue_tail(&sk->sk_receive_queue, skb);
2098         spin_unlock(&sk->sk_receive_queue.lock);
2099         sk->sk_data_ready(sk);
2100         return 0;
2101
2102 drop_n_acct:
2103         spin_lock(&sk->sk_receive_queue.lock);
2104         po->stats.stats1.tp_drops++;
2105         atomic_inc(&sk->sk_drops);
2106         spin_unlock(&sk->sk_receive_queue.lock);
2107
2108 drop_n_restore:
2109         if (skb_head != skb->data && skb_shared(skb)) {
2110                 skb->data = skb_head;
2111                 skb->len = skb_len;
2112         }
2113 drop:
2114         consume_skb(skb);
2115         return 0;
2116 }
2117
2118 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2119                        struct packet_type *pt, struct net_device *orig_dev)
2120 {
2121         struct sock *sk;
2122         struct packet_sock *po;
2123         struct sockaddr_ll *sll;
2124         union tpacket_uhdr h;
2125         u8 *skb_head = skb->data;
2126         int skb_len = skb->len;
2127         unsigned int snaplen, res;
2128         unsigned long status = TP_STATUS_USER;
2129         unsigned short macoff, netoff, hdrlen;
2130         struct sk_buff *copy_skb = NULL;
2131         struct timespec ts;
2132         __u32 ts_status;
2133
2134         /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2135          * We may add members to them until current aligned size without forcing
2136          * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2137          */
2138         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2139         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2140
2141         if (skb->pkt_type == PACKET_LOOPBACK)
2142                 goto drop;
2143
2144         sk = pt->af_packet_priv;
2145         po = pkt_sk(sk);
2146
2147         if (!net_eq(dev_net(dev), sock_net(sk)))
2148                 goto drop;
2149
2150         if (dev->header_ops) {
2151                 if (sk->sk_type != SOCK_DGRAM)
2152                         skb_push(skb, skb->data - skb_mac_header(skb));
2153                 else if (skb->pkt_type == PACKET_OUTGOING) {
2154                         /* Special case: outgoing packets have ll header at head */
2155                         skb_pull(skb, skb_network_offset(skb));
2156                 }
2157         }
2158
2159         snaplen = skb->len;
2160
2161         res = run_filter(skb, sk, snaplen);
2162         if (!res)
2163                 goto drop_n_restore;
2164
2165         if (skb->ip_summed == CHECKSUM_PARTIAL)
2166                 status |= TP_STATUS_CSUMNOTREADY;
2167         else if (skb->pkt_type != PACKET_OUTGOING &&
2168                  (skb->ip_summed == CHECKSUM_COMPLETE ||
2169                   skb_csum_unnecessary(skb)))
2170                 status |= TP_STATUS_CSUM_VALID;
2171
2172         if (snaplen > res)
2173                 snaplen = res;
2174
2175         if (sk->sk_type == SOCK_DGRAM) {
2176                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2177                                   po->tp_reserve;
2178         } else {
2179                 unsigned int maclen = skb_network_offset(skb);
2180                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2181                                        (maclen < 16 ? 16 : maclen)) +
2182                         po->tp_reserve;
2183                 macoff = netoff - maclen;
2184         }
2185         if (po->tp_version <= TPACKET_V2) {
2186                 if (macoff + snaplen > po->rx_ring.frame_size) {
2187                         if (po->copy_thresh &&
2188                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2189                                 if (skb_shared(skb)) {
2190                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
2191                                 } else {
2192                                         copy_skb = skb_get(skb);
2193                                         skb_head = skb->data;
2194                                 }
2195                                 if (copy_skb)
2196                                         skb_set_owner_r(copy_skb, sk);
2197                         }
2198                         snaplen = po->rx_ring.frame_size - macoff;
2199                         if ((int)snaplen < 0)
2200                                 snaplen = 0;
2201                 }
2202         } else if (unlikely(macoff + snaplen >
2203                             GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2204                 u32 nval;
2205
2206                 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2207                 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2208                             snaplen, nval, macoff);
2209                 snaplen = nval;
2210                 if (unlikely((int)snaplen < 0)) {
2211                         snaplen = 0;
2212                         macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2213                 }
2214         }
2215         spin_lock(&sk->sk_receive_queue.lock);
2216         h.raw = packet_current_rx_frame(po, skb,
2217                                         TP_STATUS_KERNEL, (macoff+snaplen));
2218         if (!h.raw)
2219                 goto ring_is_full;
2220         if (po->tp_version <= TPACKET_V2) {
2221                 packet_increment_rx_head(po, &po->rx_ring);
2222         /*
2223          * LOSING will be reported till you read the stats,
2224          * because it's COR - Clear On Read.
2225          * Anyways, moving it for V1/V2 only as V3 doesn't need this
2226          * at packet level.
2227          */
2228                 if (po->stats.stats1.tp_drops)
2229                         status |= TP_STATUS_LOSING;
2230         }
2231         po->stats.stats1.tp_packets++;
2232         if (copy_skb) {
2233                 status |= TP_STATUS_COPY;
2234                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2235         }
2236         spin_unlock(&sk->sk_receive_queue.lock);
2237
2238         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2239
2240         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2241                 getnstimeofday(&ts);
2242
2243         status |= ts_status;
2244
2245         switch (po->tp_version) {
2246         case TPACKET_V1:
2247                 h.h1->tp_len = skb->len;
2248                 h.h1->tp_snaplen = snaplen;
2249                 h.h1->tp_mac = macoff;
2250                 h.h1->tp_net = netoff;
2251                 h.h1->tp_sec = ts.tv_sec;
2252                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2253                 hdrlen = sizeof(*h.h1);
2254                 break;
2255         case TPACKET_V2:
2256                 h.h2->tp_len = skb->len;
2257                 h.h2->tp_snaplen = snaplen;
2258                 h.h2->tp_mac = macoff;
2259                 h.h2->tp_net = netoff;
2260                 h.h2->tp_sec = ts.tv_sec;
2261                 h.h2->tp_nsec = ts.tv_nsec;
2262                 if (skb_vlan_tag_present(skb)) {
2263                         h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2264                         h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2265                         status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2266                 } else {
2267                         h.h2->tp_vlan_tci = 0;
2268                         h.h2->tp_vlan_tpid = 0;
2269                 }
2270                 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2271                 hdrlen = sizeof(*h.h2);
2272                 break;
2273         case TPACKET_V3:
2274                 /* tp_nxt_offset,vlan are already populated above.
2275                  * So DONT clear those fields here
2276                  */
2277                 h.h3->tp_status |= status;
2278                 h.h3->tp_len = skb->len;
2279                 h.h3->tp_snaplen = snaplen;
2280                 h.h3->tp_mac = macoff;
2281                 h.h3->tp_net = netoff;
2282                 h.h3->tp_sec  = ts.tv_sec;
2283                 h.h3->tp_nsec = ts.tv_nsec;
2284                 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2285                 hdrlen = sizeof(*h.h3);
2286                 break;
2287         default:
2288                 BUG();
2289         }
2290
2291         sll = h.raw + TPACKET_ALIGN(hdrlen);
2292         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2293         sll->sll_family = AF_PACKET;
2294         sll->sll_hatype = dev->type;
2295         sll->sll_protocol = skb->protocol;
2296         sll->sll_pkttype = skb->pkt_type;
2297         if (unlikely(po->origdev))
2298                 sll->sll_ifindex = orig_dev->ifindex;
2299         else
2300                 sll->sll_ifindex = dev->ifindex;
2301
2302         smp_mb();
2303
2304 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2305         if (po->tp_version <= TPACKET_V2) {
2306                 u8 *start, *end;
2307
2308                 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2309                                         macoff + snaplen);
2310
2311                 for (start = h.raw; start < end; start += PAGE_SIZE)
2312                         flush_dcache_page(pgv_to_page(start));
2313         }
2314         smp_wmb();
2315 #endif
2316
2317         if (po->tp_version <= TPACKET_V2) {
2318                 __packet_set_status(po, h.raw, status);
2319                 sk->sk_data_ready(sk);
2320         } else {
2321                 prb_clear_blk_fill_status(&po->rx_ring);
2322         }
2323
2324 drop_n_restore:
2325         if (skb_head != skb->data && skb_shared(skb)) {
2326                 skb->data = skb_head;
2327                 skb->len = skb_len;
2328         }
2329 drop:
2330         kfree_skb(skb);
2331         return 0;
2332
2333 ring_is_full:
2334         po->stats.stats1.tp_drops++;
2335         spin_unlock(&sk->sk_receive_queue.lock);
2336
2337         sk->sk_data_ready(sk);
2338         kfree_skb(copy_skb);
2339         goto drop_n_restore;
2340 }
2341
2342 static void tpacket_destruct_skb(struct sk_buff *skb)
2343 {
2344         struct packet_sock *po = pkt_sk(skb->sk);
2345
2346         if (likely(po->tx_ring.pg_vec)) {
2347                 void *ph;
2348                 __u32 ts;
2349
2350                 ph = skb_shinfo(skb)->destructor_arg;
2351                 packet_dec_pending(&po->tx_ring);
2352
2353                 ts = __packet_set_timestamp(po, ph, skb);
2354                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2355         }
2356
2357         sock_wfree(skb);
2358 }
2359
2360 static void tpacket_set_protocol(const struct net_device *dev,
2361                                  struct sk_buff *skb)
2362 {
2363         if (dev->type == ARPHRD_ETHER) {
2364                 skb_reset_mac_header(skb);
2365                 skb->protocol = eth_hdr(skb)->h_proto;
2366         }
2367 }
2368
2369 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2370                 void *frame, struct net_device *dev, int size_max,
2371                 __be16 proto, unsigned char *addr, int hlen)
2372 {
2373         union tpacket_uhdr ph;
2374         int to_write, offset, len, tp_len, nr_frags, len_max;
2375         struct socket *sock = po->sk.sk_socket;
2376         struct page *page;
2377         void *data;
2378         int err;
2379
2380         ph.raw = frame;
2381
2382         skb->protocol = proto;
2383         skb->dev = dev;
2384         skb->priority = po->sk.sk_priority;
2385         skb->mark = po->sk.sk_mark;
2386         sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2387         skb_shinfo(skb)->destructor_arg = ph.raw;
2388
2389         switch (po->tp_version) {
2390         case TPACKET_V2:
2391                 tp_len = ph.h2->tp_len;
2392                 break;
2393         default:
2394                 tp_len = ph.h1->tp_len;
2395                 break;
2396         }
2397         if (unlikely(tp_len > size_max)) {
2398                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2399                 return -EMSGSIZE;
2400         }
2401
2402         skb_reserve(skb, hlen);
2403         skb_reset_network_header(skb);
2404
2405         if (unlikely(po->tp_tx_has_off)) {
2406                 int off_min, off_max, off;
2407                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2408                 off_max = po->tx_ring.frame_size - tp_len;
2409                 if (sock->type == SOCK_DGRAM) {
2410                         switch (po->tp_version) {
2411                         case TPACKET_V2:
2412                                 off = ph.h2->tp_net;
2413                                 break;
2414                         default:
2415                                 off = ph.h1->tp_net;
2416                                 break;
2417                         }
2418                 } else {
2419                         switch (po->tp_version) {
2420                         case TPACKET_V2:
2421                                 off = ph.h2->tp_mac;
2422                                 break;
2423                         default:
2424                                 off = ph.h1->tp_mac;
2425                                 break;
2426                         }
2427                 }
2428                 if (unlikely((off < off_min) || (off_max < off)))
2429                         return -EINVAL;
2430                 data = ph.raw + off;
2431         } else {
2432                 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2433         }
2434         to_write = tp_len;
2435
2436         if (sock->type == SOCK_DGRAM) {
2437                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2438                                 NULL, tp_len);
2439                 if (unlikely(err < 0))
2440                         return -EINVAL;
2441         } else if (dev->hard_header_len) {
2442                 int hdrlen = min_t(int, dev->hard_header_len, tp_len);
2443
2444                 skb_push(skb, dev->hard_header_len);
2445                 err = skb_store_bits(skb, 0, data, hdrlen);
2446                 if (unlikely(err))
2447                         return err;
2448                 if (!dev_validate_header(dev, skb->data, hdrlen))
2449                         return -EINVAL;
2450                 if (!skb->protocol)
2451                         tpacket_set_protocol(dev, skb);
2452
2453                 data += hdrlen;
2454                 to_write -= hdrlen;
2455         }
2456
2457         offset = offset_in_page(data);
2458         len_max = PAGE_SIZE - offset;
2459         len = ((to_write > len_max) ? len_max : to_write);
2460
2461         skb->data_len = to_write;
2462         skb->len += to_write;
2463         skb->truesize += to_write;
2464         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2465
2466         while (likely(to_write)) {
2467                 nr_frags = skb_shinfo(skb)->nr_frags;
2468
2469                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2470                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2471                                MAX_SKB_FRAGS);
2472                         return -EFAULT;
2473                 }
2474
2475                 page = pgv_to_page(data);
2476                 data += len;
2477                 flush_dcache_page(page);
2478                 get_page(page);
2479                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2480                 to_write -= len;
2481                 offset = 0;
2482                 len_max = PAGE_SIZE;
2483                 len = ((to_write > len_max) ? len_max : to_write);
2484         }
2485
2486         skb_probe_transport_header(skb, 0);
2487
2488         return tp_len;
2489 }
2490
2491 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2492 {
2493         struct sk_buff *skb;
2494         struct net_device *dev;
2495         __be16 proto;
2496         int err, reserve = 0;
2497         void *ph;
2498         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2499         bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2500         unsigned char *addr = NULL;
2501         int tp_len, size_max;
2502         int len_sum = 0;
2503         int status = TP_STATUS_AVAILABLE;
2504         int hlen, tlen;
2505
2506         mutex_lock(&po->pg_vec_lock);
2507
2508         /* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2509          * we need to confirm it under protection of pg_vec_lock.
2510          */
2511         if (unlikely(!po->tx_ring.pg_vec)) {
2512                 err = -EBUSY;
2513                 goto out;
2514         }
2515         if (likely(saddr == NULL)) {
2516                 dev     = packet_cached_dev_get(po);
2517                 proto   = po->num;
2518                 addr    = NULL;
2519         } else {
2520                 err = -EINVAL;
2521                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2522                         goto out;
2523                 if (msg->msg_namelen < (saddr->sll_halen
2524                                         + offsetof(struct sockaddr_ll,
2525                                                 sll_addr)))
2526                         goto out;
2527                 proto   = saddr->sll_protocol;
2528                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2529                 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2530                         if (dev && msg->msg_namelen < dev->addr_len +
2531                                    offsetof(struct sockaddr_ll, sll_addr))
2532                                 goto out_put;
2533                         addr = saddr->sll_addr;
2534                 }
2535         }
2536
2537         err = -ENXIO;
2538         if (unlikely(dev == NULL))
2539                 goto out;
2540         err = -ENETDOWN;
2541         if (unlikely(!(dev->flags & IFF_UP)))
2542                 goto out_put;
2543
2544         if (po->sk.sk_socket->type == SOCK_RAW)
2545                 reserve = dev->hard_header_len;
2546         size_max = po->tx_ring.frame_size
2547                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2548
2549         if (size_max > dev->mtu + reserve + VLAN_HLEN)
2550                 size_max = dev->mtu + reserve + VLAN_HLEN;
2551
2552         do {
2553                 ph = packet_current_frame(po, &po->tx_ring,
2554                                           TP_STATUS_SEND_REQUEST);
2555                 if (unlikely(ph == NULL)) {
2556                         if (need_wait && need_resched())
2557                                 schedule();
2558                         continue;
2559                 }
2560
2561                 status = TP_STATUS_SEND_REQUEST;
2562                 hlen = LL_RESERVED_SPACE(dev);
2563                 tlen = dev->needed_tailroom;
2564                 skb = sock_alloc_send_skb(&po->sk,
2565                                 hlen + tlen + sizeof(struct sockaddr_ll),
2566                                 !need_wait, &err);
2567
2568                 if (unlikely(skb == NULL)) {
2569                         /* we assume the socket was initially writeable ... */
2570                         if (likely(len_sum > 0))
2571                                 err = len_sum;
2572                         goto out_status;
2573                 }
2574                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2575                                           addr, hlen);
2576                 if (likely(tp_len >= 0) &&
2577                     tp_len > dev->mtu + reserve &&
2578                     !packet_extra_vlan_len_allowed(dev, skb))
2579                         tp_len = -EMSGSIZE;
2580
2581                 if (unlikely(tp_len < 0)) {
2582                         if (po->tp_loss) {
2583                                 __packet_set_status(po, ph,
2584                                                 TP_STATUS_AVAILABLE);
2585                                 packet_increment_head(&po->tx_ring);
2586                                 kfree_skb(skb);
2587                                 continue;
2588                         } else {
2589                                 status = TP_STATUS_WRONG_FORMAT;
2590                                 err = tp_len;
2591                                 goto out_status;
2592                         }
2593                 }
2594
2595                 packet_pick_tx_queue(dev, skb);
2596
2597                 skb->destructor = tpacket_destruct_skb;
2598                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2599                 packet_inc_pending(&po->tx_ring);
2600
2601                 status = TP_STATUS_SEND_REQUEST;
2602                 err = po->xmit(skb);
2603                 if (unlikely(err > 0)) {
2604                         err = net_xmit_errno(err);
2605                         if (err && __packet_get_status(po, ph) ==
2606                                    TP_STATUS_AVAILABLE) {
2607                                 /* skb was destructed already */
2608                                 skb = NULL;
2609                                 goto out_status;
2610                         }
2611                         /*
2612                          * skb was dropped but not destructed yet;
2613                          * let's treat it like congestion or err < 0
2614                          */
2615                         err = 0;
2616                 }
2617                 packet_increment_head(&po->tx_ring);
2618                 len_sum += tp_len;
2619         } while (likely((ph != NULL) ||
2620                 /* Note: packet_read_pending() might be slow if we have
2621                  * to call it as it's per_cpu variable, but in fast-path
2622                  * we already short-circuit the loop with the first
2623                  * condition, and luckily don't have to go that path
2624                  * anyway.
2625                  */
2626                  (need_wait && packet_read_pending(&po->tx_ring))));
2627
2628         err = len_sum;
2629         goto out_put;
2630
2631 out_status:
2632         __packet_set_status(po, ph, status);
2633         kfree_skb(skb);
2634 out_put:
2635         dev_put(dev);
2636 out:
2637         mutex_unlock(&po->pg_vec_lock);
2638         return err;
2639 }
2640
2641 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2642                                         size_t reserve, size_t len,
2643                                         size_t linear, int noblock,
2644                                         int *err)
2645 {
2646         struct sk_buff *skb;
2647
2648         /* Under a page?  Don't bother with paged skb. */
2649         if (prepad + len < PAGE_SIZE || !linear)
2650                 linear = len;
2651
2652         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2653                                    err, 0);
2654         if (!skb)
2655                 return NULL;
2656
2657         skb_reserve(skb, reserve);
2658         skb_put(skb, linear);
2659         skb->data_len = len - linear;
2660         skb->len += len - linear;
2661
2662         return skb;
2663 }
2664
2665 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2666 {
2667         struct sock *sk = sock->sk;
2668         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2669         struct sk_buff *skb;
2670         struct net_device *dev;
2671         __be16 proto;
2672         unsigned char *addr = NULL;
2673         int err, reserve = 0;
2674         struct sockcm_cookie sockc;
2675         struct virtio_net_hdr vnet_hdr = { 0 };
2676         int offset = 0;
2677         int vnet_hdr_len;
2678         struct packet_sock *po = pkt_sk(sk);
2679         unsigned short gso_type = 0;
2680         bool has_vnet_hdr = false;
2681         int hlen, tlen, linear;
2682         int extra_len = 0;
2683         ssize_t n;
2684
2685         /*
2686          *      Get and verify the address.
2687          */
2688
2689         if (likely(saddr == NULL)) {
2690                 dev     = packet_cached_dev_get(po);
2691                 proto   = po->num;
2692         } else {
2693                 err = -EINVAL;
2694                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2695                         goto out;
2696                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2697                         goto out;
2698                 proto   = saddr->sll_protocol;
2699                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2700                 if (sock->type == SOCK_DGRAM) {
2701                         if (dev && msg->msg_namelen < dev->addr_len +
2702                                    offsetof(struct sockaddr_ll, sll_addr))
2703                                 goto out_unlock;
2704                         addr = saddr->sll_addr;
2705                 }
2706         }
2707
2708         err = -ENXIO;
2709         if (unlikely(dev == NULL))
2710                 goto out_unlock;
2711         err = -ENETDOWN;
2712         if (unlikely(!(dev->flags & IFF_UP)))
2713                 goto out_unlock;
2714
2715         sockc.mark = sk->sk_mark;
2716         if (msg->msg_controllen) {
2717                 err = sock_cmsg_send(sk, msg, &sockc);
2718                 if (unlikely(err))
2719                         goto out_unlock;
2720         }
2721
2722         if (sock->type == SOCK_RAW)
2723                 reserve = dev->hard_header_len;
2724         if (po->has_vnet_hdr) {
2725                 vnet_hdr_len = sizeof(vnet_hdr);
2726
2727                 err = -EINVAL;
2728                 if (len < vnet_hdr_len)
2729                         goto out_unlock;
2730
2731                 len -= vnet_hdr_len;
2732
2733                 err = -EFAULT;
2734                 n = copy_from_iter(&vnet_hdr, vnet_hdr_len, &msg->msg_iter);
2735                 if (n != vnet_hdr_len)
2736                         goto out_unlock;
2737
2738                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2739                     (__virtio16_to_cpu(vio_le(), vnet_hdr.csum_start) +
2740                      __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset) + 2 >
2741                       __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len)))
2742                         vnet_hdr.hdr_len = __cpu_to_virtio16(vio_le(),
2743                                  __virtio16_to_cpu(vio_le(), vnet_hdr.csum_start) +
2744                                 __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset) + 2);
2745
2746                 err = -EINVAL;
2747                 if (__virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len) > len)
2748                         goto out_unlock;
2749
2750                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2751                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2752                         case VIRTIO_NET_HDR_GSO_TCPV4:
2753                                 gso_type = SKB_GSO_TCPV4;
2754                                 break;
2755                         case VIRTIO_NET_HDR_GSO_TCPV6:
2756                                 gso_type = SKB_GSO_TCPV6;
2757                                 break;
2758                         case VIRTIO_NET_HDR_GSO_UDP:
2759                                 gso_type = SKB_GSO_UDP;
2760                                 break;
2761                         default:
2762                                 goto out_unlock;
2763                         }
2764
2765                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2766                                 gso_type |= SKB_GSO_TCP_ECN;
2767
2768                         if (vnet_hdr.gso_size == 0)
2769                                 goto out_unlock;
2770
2771                 }
2772                 has_vnet_hdr = true;
2773         }
2774
2775         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2776                 if (!netif_supports_nofcs(dev)) {
2777                         err = -EPROTONOSUPPORT;
2778                         goto out_unlock;
2779                 }
2780                 extra_len = 4; /* We're doing our own CRC */
2781         }
2782
2783         err = -EMSGSIZE;
2784         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2785                 goto out_unlock;
2786
2787         err = -ENOBUFS;
2788         hlen = LL_RESERVED_SPACE(dev);
2789         tlen = dev->needed_tailroom;
2790         linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2791         linear = max(linear, min_t(int, len, dev->hard_header_len));
2792         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2793                                msg->msg_flags & MSG_DONTWAIT, &err);
2794         if (skb == NULL)
2795                 goto out_unlock;
2796
2797         skb_reset_network_header(skb);
2798
2799         err = -EINVAL;
2800         if (sock->type == SOCK_DGRAM) {
2801                 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2802                 if (unlikely(offset < 0))
2803                         goto out_free;
2804         } else if (reserve) {
2805                 skb_reserve(skb, -reserve);
2806                 if (len < reserve)
2807                         skb_reset_network_header(skb);
2808         }
2809
2810         /* Returns -EFAULT on error */
2811         err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2812         if (err)
2813                 goto out_free;
2814
2815         if (sock->type == SOCK_RAW &&
2816             !dev_validate_header(dev, skb->data, len)) {
2817                 err = -EINVAL;
2818                 goto out_free;
2819         }
2820
2821         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2822
2823         if (!gso_type && (len > dev->mtu + reserve + extra_len) &&
2824             !packet_extra_vlan_len_allowed(dev, skb)) {
2825                 err = -EMSGSIZE;
2826                 goto out_free;
2827         }
2828
2829         skb->protocol = proto;
2830         skb->dev = dev;
2831         skb->priority = sk->sk_priority;
2832         skb->mark = sockc.mark;
2833
2834         packet_pick_tx_queue(dev, skb);
2835
2836         if (has_vnet_hdr) {
2837                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2838                         u16 s = __virtio16_to_cpu(vio_le(), vnet_hdr.csum_start);
2839                         u16 o = __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset);
2840                         if (!skb_partial_csum_set(skb, s, o)) {
2841                                 err = -EINVAL;
2842                                 goto out_free;
2843                         }
2844                 }
2845
2846                 skb_shinfo(skb)->gso_size =
2847                         __virtio16_to_cpu(vio_le(), vnet_hdr.gso_size);
2848                 skb_shinfo(skb)->gso_type = gso_type;
2849
2850                 /* Header must be checked, and gso_segs computed. */
2851                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2852                 skb_shinfo(skb)->gso_segs = 0;
2853
2854                 len += vnet_hdr_len;
2855         }
2856
2857         skb_probe_transport_header(skb, reserve);
2858
2859         if (unlikely(extra_len == 4))
2860                 skb->no_fcs = 1;
2861
2862         err = po->xmit(skb);
2863         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2864                 goto out_unlock;
2865
2866         dev_put(dev);
2867
2868         return len;
2869
2870 out_free:
2871         kfree_skb(skb);
2872 out_unlock:
2873         if (dev)
2874                 dev_put(dev);
2875 out:
2876         return err;
2877 }
2878
2879 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2880 {
2881         struct sock *sk = sock->sk;
2882         struct packet_sock *po = pkt_sk(sk);
2883
2884         if (po->tx_ring.pg_vec)
2885                 return tpacket_snd(po, msg);
2886         else
2887                 return packet_snd(sock, msg, len);
2888 }
2889
2890 /*
2891  *      Close a PACKET socket. This is fairly simple. We immediately go
2892  *      to 'closed' state and remove our protocol entry in the device list.
2893  */
2894
2895 static int packet_release(struct socket *sock)
2896 {
2897         struct sock *sk = sock->sk;
2898         struct packet_sock *po;
2899         struct packet_fanout *f;
2900         struct net *net;
2901         union tpacket_req_u req_u;
2902
2903         if (!sk)
2904                 return 0;
2905
2906         net = sock_net(sk);
2907         po = pkt_sk(sk);
2908
2909         mutex_lock(&net->packet.sklist_lock);
2910         sk_del_node_init_rcu(sk);
2911         mutex_unlock(&net->packet.sklist_lock);
2912
2913         preempt_disable();
2914         sock_prot_inuse_add(net, sk->sk_prot, -1);
2915         preempt_enable();
2916
2917         spin_lock(&po->bind_lock);
2918         unregister_prot_hook(sk, false);
2919         packet_cached_dev_reset(po);
2920
2921         if (po->prot_hook.dev) {
2922                 dev_put(po->prot_hook.dev);
2923                 po->prot_hook.dev = NULL;
2924         }
2925         spin_unlock(&po->bind_lock);
2926
2927         packet_flush_mclist(sk);
2928
2929         lock_sock(sk);
2930         if (po->rx_ring.pg_vec) {
2931                 memset(&req_u, 0, sizeof(req_u));
2932                 packet_set_ring(sk, &req_u, 1, 0);
2933         }
2934
2935         if (po->tx_ring.pg_vec) {
2936                 memset(&req_u, 0, sizeof(req_u));
2937                 packet_set_ring(sk, &req_u, 1, 1);
2938         }
2939         release_sock(sk);
2940
2941         f = fanout_release(sk);
2942
2943         synchronize_net();
2944
2945         if (f) {
2946                 kfree(po->rollover);
2947                 fanout_release_data(f);
2948                 kfree(f);
2949         }
2950         /*
2951          *      Now the socket is dead. No more input will appear.
2952          */
2953         sock_orphan(sk);
2954         sock->sk = NULL;
2955
2956         /* Purge queues */
2957
2958         skb_queue_purge(&sk->sk_receive_queue);
2959         packet_free_pending(po);
2960         sk_refcnt_debug_release(sk);
2961
2962         sock_put(sk);
2963         return 0;
2964 }
2965
2966 /*
2967  *      Attach a packet hook.
2968  */
2969
2970 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
2971                           __be16 proto)
2972 {
2973         struct packet_sock *po = pkt_sk(sk);
2974         struct net_device *dev_curr;
2975         __be16 proto_curr;
2976         bool need_rehook;
2977         struct net_device *dev = NULL;
2978         int ret = 0;
2979         bool unlisted = false;
2980
2981         lock_sock(sk);
2982         spin_lock(&po->bind_lock);
2983         rcu_read_lock();
2984
2985         if (po->fanout) {
2986                 ret = -EINVAL;
2987                 goto out_unlock;
2988         }
2989
2990         if (name) {
2991                 dev = dev_get_by_name_rcu(sock_net(sk), name);
2992                 if (!dev) {
2993                         ret = -ENODEV;
2994                         goto out_unlock;
2995                 }
2996         } else if (ifindex) {
2997                 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
2998                 if (!dev) {
2999                         ret = -ENODEV;
3000                         goto out_unlock;
3001                 }
3002         }
3003
3004         if (dev)
3005                 dev_hold(dev);
3006
3007         proto_curr = po->prot_hook.type;
3008         dev_curr = po->prot_hook.dev;
3009
3010         need_rehook = proto_curr != proto || dev_curr != dev;
3011
3012         if (need_rehook) {
3013                 if (po->running) {
3014                         rcu_read_unlock();
3015                         /* prevents packet_notifier() from calling
3016                          * register_prot_hook()
3017                          */
3018                         po->num = 0;
3019                         __unregister_prot_hook(sk, true);
3020                         rcu_read_lock();
3021                         dev_curr = po->prot_hook.dev;
3022                         if (dev)
3023                                 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3024                                                                  dev->ifindex);
3025                 }
3026
3027                 BUG_ON(po->running);
3028                 po->num = proto;
3029                 po->prot_hook.type = proto;
3030
3031                 if (unlikely(unlisted)) {
3032                         dev_put(dev);
3033                         po->prot_hook.dev = NULL;
3034                         po->ifindex = -1;
3035                         packet_cached_dev_reset(po);
3036                 } else {
3037                         po->prot_hook.dev = dev;
3038                         po->ifindex = dev ? dev->ifindex : 0;
3039                         packet_cached_dev_assign(po, dev);
3040                 }
3041         }
3042         if (dev_curr)
3043                 dev_put(dev_curr);
3044
3045         if (proto == 0 || !need_rehook)
3046                 goto out_unlock;
3047
3048         if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3049                 register_prot_hook(sk);
3050         } else {
3051                 sk->sk_err = ENETDOWN;
3052                 if (!sock_flag(sk, SOCK_DEAD))
3053                         sk->sk_error_report(sk);
3054         }
3055
3056 out_unlock:
3057         rcu_read_unlock();
3058         spin_unlock(&po->bind_lock);
3059         release_sock(sk);
3060         return ret;
3061 }
3062
3063 /*
3064  *      Bind a packet socket to a device
3065  */
3066
3067 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3068                             int addr_len)
3069 {
3070         struct sock *sk = sock->sk;
3071         char name[sizeof(uaddr->sa_data) + 1];
3072
3073         /*
3074          *      Check legality
3075          */
3076
3077         if (addr_len != sizeof(struct sockaddr))
3078                 return -EINVAL;
3079         /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3080          * zero-terminated.
3081          */
3082         memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3083         name[sizeof(uaddr->sa_data)] = 0;
3084
3085         return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3086 }
3087
3088 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3089 {
3090         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3091         struct sock *sk = sock->sk;
3092
3093         /*
3094          *      Check legality
3095          */
3096
3097         if (addr_len < sizeof(struct sockaddr_ll))
3098                 return -EINVAL;
3099         if (sll->sll_family != AF_PACKET)
3100                 return -EINVAL;
3101
3102         return packet_do_bind(sk, NULL, sll->sll_ifindex,
3103                               sll->sll_protocol ? : pkt_sk(sk)->num);
3104 }
3105
3106 static struct proto packet_proto = {
3107         .name     = "PACKET",
3108         .owner    = THIS_MODULE,
3109         .obj_size = sizeof(struct packet_sock),
3110 };
3111
3112 /*
3113  *      Create a packet of type SOCK_PACKET.
3114  */
3115
3116 static int packet_create(struct net *net, struct socket *sock, int protocol,
3117                          int kern)
3118 {
3119         struct sock *sk;
3120         struct packet_sock *po;
3121         __be16 proto = (__force __be16)protocol; /* weird, but documented */
3122         int err;
3123
3124         if (!ns_capable(net->user_ns, CAP_NET_RAW))
3125                 return -EPERM;
3126         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3127             sock->type != SOCK_PACKET)
3128                 return -ESOCKTNOSUPPORT;
3129
3130         sock->state = SS_UNCONNECTED;
3131
3132         err = -ENOBUFS;
3133         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3134         if (sk == NULL)
3135                 goto out;
3136
3137         sock->ops = &packet_ops;
3138         if (sock->type == SOCK_PACKET)
3139                 sock->ops = &packet_ops_spkt;
3140
3141         sock_init_data(sock, sk);
3142
3143         po = pkt_sk(sk);
3144         sk->sk_family = PF_PACKET;
3145         po->num = proto;
3146         po->xmit = dev_queue_xmit;
3147
3148         err = packet_alloc_pending(po);
3149         if (err)
3150                 goto out2;
3151
3152         packet_cached_dev_reset(po);
3153
3154         sk->sk_destruct = packet_sock_destruct;
3155         sk_refcnt_debug_inc(sk);
3156
3157         /*
3158          *      Attach a protocol block
3159          */
3160
3161         spin_lock_init(&po->bind_lock);
3162         mutex_init(&po->pg_vec_lock);
3163         po->rollover = NULL;
3164         po->prot_hook.func = packet_rcv;
3165
3166         if (sock->type == SOCK_PACKET)
3167                 po->prot_hook.func = packet_rcv_spkt;
3168
3169         po->prot_hook.af_packet_priv = sk;
3170
3171         if (proto) {
3172                 po->prot_hook.type = proto;
3173                 __register_prot_hook(sk);
3174         }
3175
3176         mutex_lock(&net->packet.sklist_lock);
3177         sk_add_node_tail_rcu(sk, &net->packet.sklist);
3178         mutex_unlock(&net->packet.sklist_lock);
3179
3180         preempt_disable();
3181         sock_prot_inuse_add(net, &packet_proto, 1);
3182         preempt_enable();
3183
3184         return 0;
3185 out2:
3186         sk_free(sk);
3187 out:
3188         return err;
3189 }
3190
3191 /*
3192  *      Pull a packet from our receive queue and hand it to the user.
3193  *      If necessary we block.
3194  */
3195
3196 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3197                           int flags)
3198 {
3199         struct sock *sk = sock->sk;
3200         struct sk_buff *skb;
3201         int copied, err;
3202         int vnet_hdr_len = 0;
3203         unsigned int origlen = 0;
3204
3205         err = -EINVAL;
3206         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3207                 goto out;
3208
3209 #if 0
3210         /* What error should we return now? EUNATTACH? */
3211         if (pkt_sk(sk)->ifindex < 0)
3212                 return -ENODEV;
3213 #endif
3214
3215         if (flags & MSG_ERRQUEUE) {
3216                 err = sock_recv_errqueue(sk, msg, len,
3217                                          SOL_PACKET, PACKET_TX_TIMESTAMP);
3218                 goto out;
3219         }
3220
3221         /*
3222          *      Call the generic datagram receiver. This handles all sorts
3223          *      of horrible races and re-entrancy so we can forget about it
3224          *      in the protocol layers.
3225          *
3226          *      Now it will return ENETDOWN, if device have just gone down,
3227          *      but then it will block.
3228          */
3229
3230         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3231
3232         /*
3233          *      An error occurred so return it. Because skb_recv_datagram()
3234          *      handles the blocking we don't see and worry about blocking
3235          *      retries.
3236          */
3237
3238         if (skb == NULL)
3239                 goto out;
3240
3241         if (pkt_sk(sk)->pressure)
3242                 packet_rcv_has_room(pkt_sk(sk), NULL);
3243
3244         if (pkt_sk(sk)->has_vnet_hdr) {
3245                 struct virtio_net_hdr vnet_hdr = { 0 };
3246
3247                 err = -EINVAL;
3248                 vnet_hdr_len = sizeof(vnet_hdr);
3249                 if (len < vnet_hdr_len)
3250                         goto out_free;
3251
3252                 len -= vnet_hdr_len;
3253
3254                 if (skb_is_gso(skb)) {
3255                         struct skb_shared_info *sinfo = skb_shinfo(skb);
3256
3257                         /* This is a hint as to how much should be linear. */
3258                         vnet_hdr.hdr_len =
3259                                 __cpu_to_virtio16(vio_le(), skb_headlen(skb));
3260                         vnet_hdr.gso_size =
3261                                 __cpu_to_virtio16(vio_le(), sinfo->gso_size);
3262                         if (sinfo->gso_type & SKB_GSO_TCPV4)
3263                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
3264                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
3265                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
3266                         else if (sinfo->gso_type & SKB_GSO_UDP)
3267                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
3268                         else if (sinfo->gso_type & SKB_GSO_FCOE)
3269                                 goto out_free;
3270                         else
3271                                 BUG();
3272                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
3273                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
3274                 } else
3275                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
3276
3277                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3278                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
3279                         vnet_hdr.csum_start = __cpu_to_virtio16(vio_le(),
3280                                           skb_checksum_start_offset(skb));
3281                         vnet_hdr.csum_offset = __cpu_to_virtio16(vio_le(),
3282                                                          skb->csum_offset);
3283                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3284                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
3285                 } /* else everything is zero */
3286
3287                 err = memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_len);
3288                 if (err < 0)
3289                         goto out_free;
3290         }
3291
3292         /* You lose any data beyond the buffer you gave. If it worries
3293          * a user program they can ask the device for its MTU
3294          * anyway.
3295          */
3296         copied = skb->len;
3297         if (copied > len) {
3298                 copied = len;
3299                 msg->msg_flags |= MSG_TRUNC;
3300         }
3301
3302         err = skb_copy_datagram_msg(skb, 0, msg, copied);
3303         if (err)
3304                 goto out_free;
3305
3306         if (sock->type != SOCK_PACKET) {
3307                 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3308
3309                 /* Original length was stored in sockaddr_ll fields */
3310                 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3311                 sll->sll_family = AF_PACKET;
3312                 sll->sll_protocol = skb->protocol;
3313         }
3314
3315         sock_recv_ts_and_drops(msg, sk, skb);
3316
3317         if (msg->msg_name) {
3318                 int copy_len;
3319
3320                 /* If the address length field is there to be filled
3321                  * in, we fill it in now.
3322                  */
3323                 if (sock->type == SOCK_PACKET) {
3324                         __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3325                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
3326                         copy_len = msg->msg_namelen;
3327                 } else {
3328                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3329
3330                         msg->msg_namelen = sll->sll_halen +
3331                                 offsetof(struct sockaddr_ll, sll_addr);
3332                         copy_len = msg->msg_namelen;
3333                         if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3334                                 memset(msg->msg_name +
3335                                        offsetof(struct sockaddr_ll, sll_addr),
3336                                        0, sizeof(sll->sll_addr));
3337                                 msg->msg_namelen = sizeof(struct sockaddr_ll);
3338                         }
3339                 }
3340                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3341         }
3342
3343         if (pkt_sk(sk)->auxdata) {
3344                 struct tpacket_auxdata aux;
3345
3346                 aux.tp_status = TP_STATUS_USER;
3347                 if (skb->ip_summed == CHECKSUM_PARTIAL)
3348                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3349                 else if (skb->pkt_type != PACKET_OUTGOING &&
3350                          (skb->ip_summed == CHECKSUM_COMPLETE ||
3351                           skb_csum_unnecessary(skb)))
3352                         aux.tp_status |= TP_STATUS_CSUM_VALID;
3353
3354                 aux.tp_len = origlen;
3355                 aux.tp_snaplen = skb->len;
3356                 aux.tp_mac = 0;
3357                 aux.tp_net = skb_network_offset(skb);
3358                 if (skb_vlan_tag_present(skb)) {
3359                         aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3360                         aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3361                         aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3362                 } else {
3363                         aux.tp_vlan_tci = 0;
3364                         aux.tp_vlan_tpid = 0;
3365                 }
3366                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3367         }
3368
3369         /*
3370          *      Free or return the buffer as appropriate. Again this
3371          *      hides all the races and re-entrancy issues from us.
3372          */
3373         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3374
3375 out_free:
3376         skb_free_datagram(sk, skb);
3377 out:
3378         return err;
3379 }
3380
3381 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3382                                int *uaddr_len, int peer)
3383 {
3384         struct net_device *dev;
3385         struct sock *sk = sock->sk;
3386
3387         if (peer)
3388                 return -EOPNOTSUPP;
3389
3390         uaddr->sa_family = AF_PACKET;
3391         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3392         rcu_read_lock();
3393         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3394         if (dev)
3395                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3396         rcu_read_unlock();
3397         *uaddr_len = sizeof(*uaddr);
3398
3399         return 0;
3400 }
3401
3402 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3403                           int *uaddr_len, int peer)
3404 {
3405         struct net_device *dev;
3406         struct sock *sk = sock->sk;
3407         struct packet_sock *po = pkt_sk(sk);
3408         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3409
3410         if (peer)
3411                 return -EOPNOTSUPP;
3412
3413         sll->sll_family = AF_PACKET;
3414         sll->sll_ifindex = po->ifindex;
3415         sll->sll_protocol = po->num;
3416         sll->sll_pkttype = 0;
3417         rcu_read_lock();
3418         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3419         if (dev) {
3420                 sll->sll_hatype = dev->type;
3421                 sll->sll_halen = dev->addr_len;
3422                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3423         } else {
3424                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
3425                 sll->sll_halen = 0;
3426         }
3427         rcu_read_unlock();
3428         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3429
3430         return 0;
3431 }
3432
3433 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3434                          int what)
3435 {
3436         switch (i->type) {
3437         case PACKET_MR_MULTICAST:
3438                 if (i->alen != dev->addr_len)
3439                         return -EINVAL;
3440                 if (what > 0)
3441                         return dev_mc_add(dev, i->addr);
3442                 else
3443                         return dev_mc_del(dev, i->addr);
3444                 break;
3445         case PACKET_MR_PROMISC:
3446                 return dev_set_promiscuity(dev, what);
3447         case PACKET_MR_ALLMULTI:
3448                 return dev_set_allmulti(dev, what);
3449         case PACKET_MR_UNICAST:
3450                 if (i->alen != dev->addr_len)
3451                         return -EINVAL;
3452                 if (what > 0)
3453                         return dev_uc_add(dev, i->addr);
3454                 else
3455                         return dev_uc_del(dev, i->addr);
3456                 break;
3457         default:
3458                 break;
3459         }
3460         return 0;
3461 }
3462
3463 static void packet_dev_mclist_delete(struct net_device *dev,
3464                                      struct packet_mclist **mlp)
3465 {
3466         struct packet_mclist *ml;
3467
3468         while ((ml = *mlp) != NULL) {
3469                 if (ml->ifindex == dev->ifindex) {
3470                         packet_dev_mc(dev, ml, -1);
3471                         *mlp = ml->next;
3472                         kfree(ml);
3473                 } else
3474                         mlp = &ml->next;
3475         }
3476 }
3477
3478 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3479 {
3480         struct packet_sock *po = pkt_sk(sk);
3481         struct packet_mclist *ml, *i;
3482         struct net_device *dev;
3483         int err;
3484
3485         rtnl_lock();
3486
3487         err = -ENODEV;
3488         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3489         if (!dev)
3490                 goto done;
3491
3492         err = -EINVAL;
3493         if (mreq->mr_alen > dev->addr_len)
3494                 goto done;
3495
3496         err = -ENOBUFS;
3497         i = kmalloc(sizeof(*i), GFP_KERNEL);
3498         if (i == NULL)
3499                 goto done;
3500
3501         err = 0;
3502         for (ml = po->mclist; ml; ml = ml->next) {
3503                 if (ml->ifindex == mreq->mr_ifindex &&
3504                     ml->type == mreq->mr_type &&
3505                     ml->alen == mreq->mr_alen &&
3506                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3507                         ml->count++;
3508                         /* Free the new element ... */
3509                         kfree(i);
3510                         goto done;
3511                 }
3512         }
3513
3514         i->type = mreq->mr_type;
3515         i->ifindex = mreq->mr_ifindex;
3516         i->alen = mreq->mr_alen;
3517         memcpy(i->addr, mreq->mr_address, i->alen);
3518         memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3519         i->count = 1;
3520         i->next = po->mclist;
3521         po->mclist = i;
3522         err = packet_dev_mc(dev, i, 1);
3523         if (err) {
3524                 po->mclist = i->next;
3525                 kfree(i);
3526         }
3527
3528 done:
3529         rtnl_unlock();
3530         return err;
3531 }
3532
3533 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3534 {
3535         struct packet_mclist *ml, **mlp;
3536
3537         rtnl_lock();
3538
3539         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3540                 if (ml->ifindex == mreq->mr_ifindex &&
3541                     ml->type == mreq->mr_type &&
3542                     ml->alen == mreq->mr_alen &&
3543                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3544                         if (--ml->count == 0) {
3545                                 struct net_device *dev;
3546                                 *mlp = ml->next;
3547                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3548                                 if (dev)
3549                                         packet_dev_mc(dev, ml, -1);
3550                                 kfree(ml);
3551                         }
3552                         break;
3553                 }
3554         }
3555         rtnl_unlock();
3556         return 0;
3557 }
3558
3559 static void packet_flush_mclist(struct sock *sk)
3560 {
3561         struct packet_sock *po = pkt_sk(sk);
3562         struct packet_mclist *ml;
3563
3564         if (!po->mclist)
3565                 return;
3566
3567         rtnl_lock();
3568         while ((ml = po->mclist) != NULL) {
3569                 struct net_device *dev;
3570
3571                 po->mclist = ml->next;
3572                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3573                 if (dev != NULL)
3574                         packet_dev_mc(dev, ml, -1);
3575                 kfree(ml);
3576         }
3577         rtnl_unlock();
3578 }
3579
3580 static int
3581 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3582 {
3583         struct sock *sk = sock->sk;
3584         struct packet_sock *po = pkt_sk(sk);
3585         int ret;
3586
3587         if (level != SOL_PACKET)
3588                 return -ENOPROTOOPT;
3589
3590         switch (optname) {
3591         case PACKET_ADD_MEMBERSHIP:
3592         case PACKET_DROP_MEMBERSHIP:
3593         {
3594                 struct packet_mreq_max mreq;
3595                 int len = optlen;
3596                 memset(&mreq, 0, sizeof(mreq));
3597                 if (len < sizeof(struct packet_mreq))
3598                         return -EINVAL;
3599                 if (len > sizeof(mreq))
3600                         len = sizeof(mreq);
3601                 if (copy_from_user(&mreq, optval, len))
3602                         return -EFAULT;
3603                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3604                         return -EINVAL;
3605                 if (optname == PACKET_ADD_MEMBERSHIP)
3606                         ret = packet_mc_add(sk, &mreq);
3607                 else
3608                         ret = packet_mc_drop(sk, &mreq);
3609                 return ret;
3610         }
3611
3612         case PACKET_RX_RING:
3613         case PACKET_TX_RING:
3614         {
3615                 union tpacket_req_u req_u;
3616                 int len;
3617
3618                 lock_sock(sk);
3619                 switch (po->tp_version) {
3620                 case TPACKET_V1:
3621                 case TPACKET_V2:
3622                         len = sizeof(req_u.req);
3623                         break;
3624                 case TPACKET_V3:
3625                 default:
3626                         len = sizeof(req_u.req3);
3627                         break;
3628                 }
3629                 if (optlen < len) {
3630                         ret = -EINVAL;
3631                 } else {
3632                         if (pkt_sk(sk)->has_vnet_hdr) {
3633                                 ret = -EINVAL;
3634                         } else {
3635                                 if (copy_from_user(&req_u.req, optval, len))
3636                                         ret = -EFAULT;
3637                                 else
3638                                         ret = packet_set_ring(sk, &req_u, 0,
3639                                                               optname == PACKET_TX_RING);
3640                         }
3641                 }
3642                 release_sock(sk);
3643                 return ret;
3644         }
3645         case PACKET_COPY_THRESH:
3646         {
3647                 int val;
3648
3649                 if (optlen != sizeof(val))
3650                         return -EINVAL;
3651                 if (copy_from_user(&val, optval, sizeof(val)))
3652                         return -EFAULT;
3653
3654                 pkt_sk(sk)->copy_thresh = val;
3655                 return 0;
3656         }
3657         case PACKET_VERSION:
3658         {
3659                 int val;
3660
3661                 if (optlen != sizeof(val))
3662                         return -EINVAL;
3663                 if (copy_from_user(&val, optval, sizeof(val)))
3664                         return -EFAULT;
3665                 switch (val) {
3666                 case TPACKET_V1:
3667                 case TPACKET_V2:
3668                 case TPACKET_V3:
3669                         break;
3670                 default:
3671                         return -EINVAL;
3672                 }
3673                 lock_sock(sk);
3674                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3675                         ret = -EBUSY;
3676                 } else {
3677                         po->tp_version = val;
3678                         ret = 0;
3679                 }
3680                 release_sock(sk);
3681                 return ret;
3682         }
3683         case PACKET_RESERVE:
3684         {
3685                 unsigned int val;
3686
3687                 if (optlen != sizeof(val))
3688                         return -EINVAL;
3689                 if (copy_from_user(&val, optval, sizeof(val)))
3690                         return -EFAULT;
3691                 if (val > INT_MAX)
3692                         return -EINVAL;
3693                 lock_sock(sk);
3694                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3695                         ret = -EBUSY;
3696                 } else {
3697                         po->tp_reserve = val;
3698                         ret = 0;
3699                 }
3700                 release_sock(sk);
3701                 return ret;
3702         }
3703         case PACKET_LOSS:
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
3712                 lock_sock(sk);
3713                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3714                         ret = -EBUSY;
3715                 } else {
3716                         po->tp_loss = !!val;
3717                         ret = 0;
3718                 }
3719                 release_sock(sk);
3720                 return ret;
3721         }
3722         case PACKET_AUXDATA:
3723         {
3724                 int val;
3725
3726                 if (optlen < sizeof(val))
3727                         return -EINVAL;
3728                 if (copy_from_user(&val, optval, sizeof(val)))
3729                         return -EFAULT;
3730
3731                 lock_sock(sk);
3732                 po->auxdata = !!val;
3733                 release_sock(sk);
3734                 return 0;
3735         }
3736         case PACKET_ORIGDEV:
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                 lock_sock(sk);
3746                 po->origdev = !!val;
3747                 release_sock(sk);
3748                 return 0;
3749         }
3750         case PACKET_VNET_HDR:
3751         {
3752                 int val;
3753
3754                 if (sock->type != SOCK_RAW)
3755                         return -EINVAL;
3756                 if (optlen < sizeof(val))
3757                         return -EINVAL;
3758                 if (copy_from_user(&val, optval, sizeof(val)))
3759                         return -EFAULT;
3760
3761                 lock_sock(sk);
3762                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3763                         ret = -EBUSY;
3764                 } else {
3765                         po->has_vnet_hdr = !!val;
3766                         ret = 0;
3767                 }
3768                 release_sock(sk);
3769                 return ret;
3770         }
3771         case PACKET_TIMESTAMP:
3772         {
3773                 int val;
3774
3775                 if (optlen != sizeof(val))
3776                         return -EINVAL;
3777                 if (copy_from_user(&val, optval, sizeof(val)))
3778                         return -EFAULT;
3779
3780                 po->tp_tstamp = val;
3781                 return 0;
3782         }
3783         case PACKET_FANOUT:
3784         {
3785                 int val;
3786
3787                 if (optlen != sizeof(val))
3788                         return -EINVAL;
3789                 if (copy_from_user(&val, optval, sizeof(val)))
3790                         return -EFAULT;
3791
3792                 return fanout_add(sk, val & 0xffff, val >> 16);
3793         }
3794         case PACKET_FANOUT_DATA:
3795         {
3796                 if (!po->fanout)
3797                         return -EINVAL;
3798
3799                 return fanout_set_data(po, optval, optlen);
3800         }
3801         case PACKET_TX_HAS_OFF:
3802         {
3803                 unsigned int val;
3804
3805                 if (optlen != sizeof(val))
3806                         return -EINVAL;
3807                 if (copy_from_user(&val, optval, sizeof(val)))
3808                         return -EFAULT;
3809
3810                 lock_sock(sk);
3811                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3812                         ret = -EBUSY;
3813                 } else {
3814                         po->tp_tx_has_off = !!val;
3815                         ret = 0;
3816                 }
3817                 release_sock(sk);
3818                 return 0;
3819         }
3820         case PACKET_QDISC_BYPASS:
3821         {
3822                 int val;
3823
3824                 if (optlen != sizeof(val))
3825                         return -EINVAL;
3826                 if (copy_from_user(&val, optval, sizeof(val)))
3827                         return -EFAULT;
3828
3829                 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3830                 return 0;
3831         }
3832         default:
3833                 return -ENOPROTOOPT;
3834         }
3835 }
3836
3837 static int packet_getsockopt(struct socket *sock, int level, int optname,
3838                              char __user *optval, int __user *optlen)
3839 {
3840         int len;
3841         int val, lv = sizeof(val);
3842         struct sock *sk = sock->sk;
3843         struct packet_sock *po = pkt_sk(sk);
3844         void *data = &val;
3845         union tpacket_stats_u st;
3846         struct tpacket_rollover_stats rstats;
3847
3848         if (level != SOL_PACKET)
3849                 return -ENOPROTOOPT;
3850
3851         if (get_user(len, optlen))
3852                 return -EFAULT;
3853
3854         if (len < 0)
3855                 return -EINVAL;
3856
3857         switch (optname) {
3858         case PACKET_STATISTICS:
3859                 spin_lock_bh(&sk->sk_receive_queue.lock);
3860                 memcpy(&st, &po->stats, sizeof(st));
3861                 memset(&po->stats, 0, sizeof(po->stats));
3862                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3863
3864                 if (po->tp_version == TPACKET_V3) {
3865                         lv = sizeof(struct tpacket_stats_v3);
3866                         st.stats3.tp_packets += st.stats3.tp_drops;
3867                         data = &st.stats3;
3868                 } else {
3869                         lv = sizeof(struct tpacket_stats);
3870                         st.stats1.tp_packets += st.stats1.tp_drops;
3871                         data = &st.stats1;
3872                 }
3873
3874                 break;
3875         case PACKET_AUXDATA:
3876                 val = po->auxdata;
3877                 break;
3878         case PACKET_ORIGDEV:
3879                 val = po->origdev;
3880                 break;
3881         case PACKET_VNET_HDR:
3882                 val = po->has_vnet_hdr;
3883                 break;
3884         case PACKET_VERSION:
3885                 val = po->tp_version;
3886                 break;
3887         case PACKET_HDRLEN:
3888                 if (len > sizeof(int))
3889                         len = sizeof(int);
3890                 if (len < sizeof(int))
3891                         return -EINVAL;
3892                 if (copy_from_user(&val, optval, len))
3893                         return -EFAULT;
3894                 switch (val) {
3895                 case TPACKET_V1:
3896                         val = sizeof(struct tpacket_hdr);
3897                         break;
3898                 case TPACKET_V2:
3899                         val = sizeof(struct tpacket2_hdr);
3900                         break;
3901                 case TPACKET_V3:
3902                         val = sizeof(struct tpacket3_hdr);
3903                         break;
3904                 default:
3905                         return -EINVAL;
3906                 }
3907                 break;
3908         case PACKET_RESERVE:
3909                 val = po->tp_reserve;
3910                 break;
3911         case PACKET_LOSS:
3912                 val = po->tp_loss;
3913                 break;
3914         case PACKET_TIMESTAMP:
3915                 val = po->tp_tstamp;
3916                 break;
3917         case PACKET_FANOUT:
3918                 val = (po->fanout ?
3919                        ((u32)po->fanout->id |
3920                         ((u32)po->fanout->type << 16) |
3921                         ((u32)po->fanout->flags << 24)) :
3922                        0);
3923                 break;
3924         case PACKET_ROLLOVER_STATS:
3925                 if (!po->rollover)
3926                         return -EINVAL;
3927                 rstats.tp_all = atomic_long_read(&po->rollover->num);
3928                 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3929                 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3930                 data = &rstats;
3931                 lv = sizeof(rstats);
3932                 break;
3933         case PACKET_TX_HAS_OFF:
3934                 val = po->tp_tx_has_off;
3935                 break;
3936         case PACKET_QDISC_BYPASS:
3937                 val = packet_use_direct_xmit(po);
3938                 break;
3939         default:
3940                 return -ENOPROTOOPT;
3941         }
3942
3943         if (len > lv)
3944                 len = lv;
3945         if (put_user(len, optlen))
3946                 return -EFAULT;
3947         if (copy_to_user(optval, data, len))
3948                 return -EFAULT;
3949         return 0;
3950 }
3951
3952
3953 static int packet_notifier(struct notifier_block *this,
3954                            unsigned long msg, void *ptr)
3955 {
3956         struct sock *sk;
3957         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3958         struct net *net = dev_net(dev);
3959
3960         rcu_read_lock();
3961         sk_for_each_rcu(sk, &net->packet.sklist) {
3962                 struct packet_sock *po = pkt_sk(sk);
3963
3964                 switch (msg) {
3965                 case NETDEV_UNREGISTER:
3966                         if (po->mclist)
3967                                 packet_dev_mclist_delete(dev, &po->mclist);
3968                         /* fallthrough */
3969
3970                 case NETDEV_DOWN:
3971                         if (dev->ifindex == po->ifindex) {
3972                                 spin_lock(&po->bind_lock);
3973                                 if (po->running) {
3974                                         __unregister_prot_hook(sk, false);
3975                                         sk->sk_err = ENETDOWN;
3976                                         if (!sock_flag(sk, SOCK_DEAD))
3977                                                 sk->sk_error_report(sk);
3978                                 }
3979                                 if (msg == NETDEV_UNREGISTER) {
3980                                         packet_cached_dev_reset(po);
3981                                         po->ifindex = -1;
3982                                         if (po->prot_hook.dev)
3983                                                 dev_put(po->prot_hook.dev);
3984                                         po->prot_hook.dev = NULL;
3985                                 }
3986                                 spin_unlock(&po->bind_lock);
3987                         }
3988                         break;
3989                 case NETDEV_UP:
3990                         if (dev->ifindex == po->ifindex) {
3991                                 spin_lock(&po->bind_lock);
3992                                 if (po->num)
3993                                         register_prot_hook(sk);
3994                                 spin_unlock(&po->bind_lock);
3995                         }
3996                         break;
3997                 }
3998         }
3999         rcu_read_unlock();
4000         return NOTIFY_DONE;
4001 }
4002
4003
4004 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4005                         unsigned long arg)
4006 {
4007         struct sock *sk = sock->sk;
4008
4009         switch (cmd) {
4010         case SIOCOUTQ:
4011         {
4012                 int amount = sk_wmem_alloc_get(sk);
4013
4014                 return put_user(amount, (int __user *)arg);
4015         }
4016         case SIOCINQ:
4017         {
4018                 struct sk_buff *skb;
4019                 int amount = 0;
4020
4021                 spin_lock_bh(&sk->sk_receive_queue.lock);
4022                 skb = skb_peek(&sk->sk_receive_queue);
4023                 if (skb)
4024                         amount = skb->len;
4025                 spin_unlock_bh(&sk->sk_receive_queue.lock);
4026                 return put_user(amount, (int __user *)arg);
4027         }
4028         case SIOCGSTAMP:
4029                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
4030         case SIOCGSTAMPNS:
4031                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
4032
4033 #ifdef CONFIG_INET
4034         case SIOCADDRT:
4035         case SIOCDELRT:
4036         case SIOCDARP:
4037         case SIOCGARP:
4038         case SIOCSARP:
4039         case SIOCGIFADDR:
4040         case SIOCSIFADDR:
4041         case SIOCGIFBRDADDR:
4042         case SIOCSIFBRDADDR:
4043         case SIOCGIFNETMASK:
4044         case SIOCSIFNETMASK:
4045         case SIOCGIFDSTADDR:
4046         case SIOCSIFDSTADDR:
4047         case SIOCSIFFLAGS:
4048                 return inet_dgram_ops.ioctl(sock, cmd, arg);
4049 #endif
4050
4051         default:
4052                 return -ENOIOCTLCMD;
4053         }
4054         return 0;
4055 }
4056
4057 static unsigned int packet_poll(struct file *file, struct socket *sock,
4058                                 poll_table *wait)
4059 {
4060         struct sock *sk = sock->sk;
4061         struct packet_sock *po = pkt_sk(sk);
4062         unsigned int mask = datagram_poll(file, sock, wait);
4063
4064         spin_lock_bh(&sk->sk_receive_queue.lock);
4065         if (po->rx_ring.pg_vec) {
4066                 if (!packet_previous_rx_frame(po, &po->rx_ring,
4067                         TP_STATUS_KERNEL))
4068                         mask |= POLLIN | POLLRDNORM;
4069         }
4070         if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
4071                 po->pressure = 0;
4072         spin_unlock_bh(&sk->sk_receive_queue.lock);
4073         spin_lock_bh(&sk->sk_write_queue.lock);
4074         if (po->tx_ring.pg_vec) {
4075                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4076                         mask |= POLLOUT | POLLWRNORM;
4077         }
4078         spin_unlock_bh(&sk->sk_write_queue.lock);
4079         return mask;
4080 }
4081
4082
4083 /* Dirty? Well, I still did not learn better way to account
4084  * for user mmaps.
4085  */
4086
4087 static void packet_mm_open(struct vm_area_struct *vma)
4088 {
4089         struct file *file = vma->vm_file;
4090         struct socket *sock = file->private_data;
4091         struct sock *sk = sock->sk;
4092
4093         if (sk)
4094                 atomic_inc(&pkt_sk(sk)->mapped);
4095 }
4096
4097 static void packet_mm_close(struct vm_area_struct *vma)
4098 {
4099         struct file *file = vma->vm_file;
4100         struct socket *sock = file->private_data;
4101         struct sock *sk = sock->sk;
4102
4103         if (sk)
4104                 atomic_dec(&pkt_sk(sk)->mapped);
4105 }
4106
4107 static const struct vm_operations_struct packet_mmap_ops = {
4108         .open   =       packet_mm_open,
4109         .close  =       packet_mm_close,
4110 };
4111
4112 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4113                         unsigned int len)
4114 {
4115         int i;
4116
4117         for (i = 0; i < len; i++) {
4118                 if (likely(pg_vec[i].buffer)) {
4119                         if (is_vmalloc_addr(pg_vec[i].buffer))
4120                                 vfree(pg_vec[i].buffer);
4121                         else
4122                                 free_pages((unsigned long)pg_vec[i].buffer,
4123                                            order);
4124                         pg_vec[i].buffer = NULL;
4125                 }
4126         }
4127         kfree(pg_vec);
4128 }
4129
4130 static char *alloc_one_pg_vec_page(unsigned long order)
4131 {
4132         char *buffer;
4133         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4134                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4135
4136         buffer = (char *) __get_free_pages(gfp_flags, order);
4137         if (buffer)
4138                 return buffer;
4139
4140         /* __get_free_pages failed, fall back to vmalloc */
4141         buffer = vzalloc((1 << order) * PAGE_SIZE);
4142         if (buffer)
4143                 return buffer;
4144
4145         /* vmalloc failed, lets dig into swap here */
4146         gfp_flags &= ~__GFP_NORETRY;
4147         buffer = (char *) __get_free_pages(gfp_flags, order);
4148         if (buffer)
4149                 return buffer;
4150
4151         /* complete and utter failure */
4152         return NULL;
4153 }
4154
4155 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4156 {
4157         unsigned int block_nr = req->tp_block_nr;
4158         struct pgv *pg_vec;
4159         int i;
4160
4161         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4162         if (unlikely(!pg_vec))
4163                 goto out;
4164
4165         for (i = 0; i < block_nr; i++) {
4166                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4167                 if (unlikely(!pg_vec[i].buffer))
4168                         goto out_free_pgvec;
4169         }
4170
4171 out:
4172         return pg_vec;
4173
4174 out_free_pgvec:
4175         free_pg_vec(pg_vec, order, block_nr);
4176         pg_vec = NULL;
4177         goto out;
4178 }
4179
4180 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4181                 int closing, int tx_ring)
4182 {
4183         struct pgv *pg_vec = NULL;
4184         struct packet_sock *po = pkt_sk(sk);
4185         int was_running, order = 0;
4186         struct packet_ring_buffer *rb;
4187         struct sk_buff_head *rb_queue;
4188         __be16 num;
4189         int err = -EINVAL;
4190         /* Added to avoid minimal code churn */
4191         struct tpacket_req *req = &req_u->req;
4192
4193         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
4194         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
4195                 net_warn_ratelimited("Tx-ring is not supported.\n");
4196                 goto out;
4197         }
4198
4199         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4200         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4201
4202         err = -EBUSY;
4203         if (!closing) {
4204                 if (atomic_read(&po->mapped))
4205                         goto out;
4206                 if (packet_read_pending(rb))
4207                         goto out;
4208         }
4209
4210         if (req->tp_block_nr) {
4211                 unsigned int min_frame_size;
4212
4213                 /* Sanity tests and some calculations */
4214                 err = -EBUSY;
4215                 if (unlikely(rb->pg_vec))
4216                         goto out;
4217
4218                 switch (po->tp_version) {
4219                 case TPACKET_V1:
4220                         po->tp_hdrlen = TPACKET_HDRLEN;
4221                         break;
4222                 case TPACKET_V2:
4223                         po->tp_hdrlen = TPACKET2_HDRLEN;
4224                         break;
4225                 case TPACKET_V3:
4226                         po->tp_hdrlen = TPACKET3_HDRLEN;
4227                         break;
4228                 }
4229
4230                 err = -EINVAL;
4231                 if (unlikely((int)req->tp_block_size <= 0))
4232                         goto out;
4233                 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4234                         goto out;
4235                 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4236                 if (po->tp_version >= TPACKET_V3 &&
4237                     req->tp_block_size <
4238                     BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4239                         goto out;
4240                 if (unlikely(req->tp_frame_size < min_frame_size))
4241                         goto out;
4242                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4243                         goto out;
4244
4245                 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4246                 if (unlikely(rb->frames_per_block == 0))
4247                         goto out;
4248                 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4249                         goto out;
4250                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4251                                         req->tp_frame_nr))
4252                         goto out;
4253
4254                 err = -ENOMEM;
4255                 order = get_order(req->tp_block_size);
4256                 pg_vec = alloc_pg_vec(req, order);
4257                 if (unlikely(!pg_vec))
4258                         goto out;
4259                 switch (po->tp_version) {
4260                 case TPACKET_V3:
4261                 /* Transmit path is not supported. We checked
4262                  * it above but just being paranoid
4263                  */
4264                         if (!tx_ring)
4265                                 init_prb_bdqc(po, rb, pg_vec, req_u);
4266                         break;
4267                 default:
4268                         break;
4269                 }
4270         }
4271         /* Done */
4272         else {
4273                 err = -EINVAL;
4274                 if (unlikely(req->tp_frame_nr))
4275                         goto out;
4276         }
4277
4278
4279         /* Detach socket from network */
4280         spin_lock(&po->bind_lock);
4281         was_running = po->running;
4282         num = po->num;
4283         if (was_running) {
4284                 po->num = 0;
4285                 __unregister_prot_hook(sk, false);
4286         }
4287         spin_unlock(&po->bind_lock);
4288
4289         synchronize_net();
4290
4291         err = -EBUSY;
4292         mutex_lock(&po->pg_vec_lock);
4293         if (closing || atomic_read(&po->mapped) == 0) {
4294                 err = 0;
4295                 spin_lock_bh(&rb_queue->lock);
4296                 swap(rb->pg_vec, pg_vec);
4297                 rb->frame_max = (req->tp_frame_nr - 1);
4298                 rb->head = 0;
4299                 rb->frame_size = req->tp_frame_size;
4300                 spin_unlock_bh(&rb_queue->lock);
4301
4302                 swap(rb->pg_vec_order, order);
4303                 swap(rb->pg_vec_len, req->tp_block_nr);
4304
4305                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4306                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4307                                                 tpacket_rcv : packet_rcv;
4308                 skb_queue_purge(rb_queue);
4309                 if (atomic_read(&po->mapped))
4310                         pr_err("packet_mmap: vma is busy: %d\n",
4311                                atomic_read(&po->mapped));
4312         }
4313         mutex_unlock(&po->pg_vec_lock);
4314
4315         spin_lock(&po->bind_lock);
4316         if (was_running) {
4317                 po->num = num;
4318                 register_prot_hook(sk);
4319         }
4320         spin_unlock(&po->bind_lock);
4321         if (pg_vec && (po->tp_version > TPACKET_V2)) {
4322                 /* Because we don't support block-based V3 on tx-ring */
4323                 if (!tx_ring)
4324                         prb_shutdown_retire_blk_timer(po, rb_queue);
4325         }
4326
4327         if (pg_vec)
4328                 free_pg_vec(pg_vec, order, req->tp_block_nr);
4329 out:
4330         return err;
4331 }
4332
4333 static int packet_mmap(struct file *file, struct socket *sock,
4334                 struct vm_area_struct *vma)
4335 {
4336         struct sock *sk = sock->sk;
4337         struct packet_sock *po = pkt_sk(sk);
4338         unsigned long size, expected_size;
4339         struct packet_ring_buffer *rb;
4340         unsigned long start;
4341         int err = -EINVAL;
4342         int i;
4343
4344         if (vma->vm_pgoff)
4345                 return -EINVAL;
4346
4347         mutex_lock(&po->pg_vec_lock);
4348
4349         expected_size = 0;
4350         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4351                 if (rb->pg_vec) {
4352                         expected_size += rb->pg_vec_len
4353                                                 * rb->pg_vec_pages
4354                                                 * PAGE_SIZE;
4355                 }
4356         }
4357
4358         if (expected_size == 0)
4359                 goto out;
4360
4361         size = vma->vm_end - vma->vm_start;
4362         if (size != expected_size)
4363                 goto out;
4364
4365         start = vma->vm_start;
4366         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4367                 if (rb->pg_vec == NULL)
4368                         continue;
4369
4370                 for (i = 0; i < rb->pg_vec_len; i++) {
4371                         struct page *page;
4372                         void *kaddr = rb->pg_vec[i].buffer;
4373                         int pg_num;
4374
4375                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4376                                 page = pgv_to_page(kaddr);
4377                                 err = vm_insert_page(vma, start, page);
4378                                 if (unlikely(err))
4379                                         goto out;
4380                                 start += PAGE_SIZE;
4381                                 kaddr += PAGE_SIZE;
4382                         }
4383                 }
4384         }
4385
4386         atomic_inc(&po->mapped);
4387         vma->vm_ops = &packet_mmap_ops;
4388         err = 0;
4389
4390 out:
4391         mutex_unlock(&po->pg_vec_lock);
4392         return err;
4393 }
4394
4395 static const struct proto_ops packet_ops_spkt = {
4396         .family =       PF_PACKET,
4397         .owner =        THIS_MODULE,
4398         .release =      packet_release,
4399         .bind =         packet_bind_spkt,
4400         .connect =      sock_no_connect,
4401         .socketpair =   sock_no_socketpair,
4402         .accept =       sock_no_accept,
4403         .getname =      packet_getname_spkt,
4404         .poll =         datagram_poll,
4405         .ioctl =        packet_ioctl,
4406         .listen =       sock_no_listen,
4407         .shutdown =     sock_no_shutdown,
4408         .setsockopt =   sock_no_setsockopt,
4409         .getsockopt =   sock_no_getsockopt,
4410         .sendmsg =      packet_sendmsg_spkt,
4411         .recvmsg =      packet_recvmsg,
4412         .mmap =         sock_no_mmap,
4413         .sendpage =     sock_no_sendpage,
4414 };
4415
4416 static const struct proto_ops packet_ops = {
4417         .family =       PF_PACKET,
4418         .owner =        THIS_MODULE,
4419         .release =      packet_release,
4420         .bind =         packet_bind,
4421         .connect =      sock_no_connect,
4422         .socketpair =   sock_no_socketpair,
4423         .accept =       sock_no_accept,
4424         .getname =      packet_getname,
4425         .poll =         packet_poll,
4426         .ioctl =        packet_ioctl,
4427         .listen =       sock_no_listen,
4428         .shutdown =     sock_no_shutdown,
4429         .setsockopt =   packet_setsockopt,
4430         .getsockopt =   packet_getsockopt,
4431         .sendmsg =      packet_sendmsg,
4432         .recvmsg =      packet_recvmsg,
4433         .mmap =         packet_mmap,
4434         .sendpage =     sock_no_sendpage,
4435 };
4436
4437 static const struct net_proto_family packet_family_ops = {
4438         .family =       PF_PACKET,
4439         .create =       packet_create,
4440         .owner  =       THIS_MODULE,
4441 };
4442
4443 static struct notifier_block packet_netdev_notifier = {
4444         .notifier_call =        packet_notifier,
4445 };
4446
4447 #ifdef CONFIG_PROC_FS
4448
4449 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4450         __acquires(RCU)
4451 {
4452         struct net *net = seq_file_net(seq);
4453
4454         rcu_read_lock();
4455         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4456 }
4457
4458 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4459 {
4460         struct net *net = seq_file_net(seq);
4461         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4462 }
4463
4464 static void packet_seq_stop(struct seq_file *seq, void *v)
4465         __releases(RCU)
4466 {
4467         rcu_read_unlock();
4468 }
4469
4470 static int packet_seq_show(struct seq_file *seq, void *v)
4471 {
4472         if (v == SEQ_START_TOKEN)
4473                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
4474         else {
4475                 struct sock *s = sk_entry(v);
4476                 const struct packet_sock *po = pkt_sk(s);
4477
4478                 seq_printf(seq,
4479                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4480                            s,
4481                            atomic_read(&s->sk_refcnt),
4482                            s->sk_type,
4483                            ntohs(po->num),
4484                            po->ifindex,
4485                            po->running,
4486                            atomic_read(&s->sk_rmem_alloc),
4487                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4488                            sock_i_ino(s));
4489         }
4490
4491         return 0;
4492 }
4493
4494 static const struct seq_operations packet_seq_ops = {
4495         .start  = packet_seq_start,
4496         .next   = packet_seq_next,
4497         .stop   = packet_seq_stop,
4498         .show   = packet_seq_show,
4499 };
4500
4501 static int packet_seq_open(struct inode *inode, struct file *file)
4502 {
4503         return seq_open_net(inode, file, &packet_seq_ops,
4504                             sizeof(struct seq_net_private));
4505 }
4506
4507 static const struct file_operations packet_seq_fops = {
4508         .owner          = THIS_MODULE,
4509         .open           = packet_seq_open,
4510         .read           = seq_read,
4511         .llseek         = seq_lseek,
4512         .release        = seq_release_net,
4513 };
4514
4515 #endif
4516
4517 static int __net_init packet_net_init(struct net *net)
4518 {
4519         mutex_init(&net->packet.sklist_lock);
4520         INIT_HLIST_HEAD(&net->packet.sklist);
4521
4522         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4523                 return -ENOMEM;
4524
4525         return 0;
4526 }
4527
4528 static void __net_exit packet_net_exit(struct net *net)
4529 {
4530         remove_proc_entry("packet", net->proc_net);
4531 }
4532
4533 static struct pernet_operations packet_net_ops = {
4534         .init = packet_net_init,
4535         .exit = packet_net_exit,
4536 };
4537
4538
4539 static void __exit packet_exit(void)
4540 {
4541         unregister_netdevice_notifier(&packet_netdev_notifier);
4542         unregister_pernet_subsys(&packet_net_ops);
4543         sock_unregister(PF_PACKET);
4544         proto_unregister(&packet_proto);
4545 }
4546
4547 static int __init packet_init(void)
4548 {
4549         int rc;
4550
4551         rc = proto_register(&packet_proto, 0);
4552         if (rc)
4553                 goto out;
4554         rc = sock_register(&packet_family_ops);
4555         if (rc)
4556                 goto out_proto;
4557         rc = register_pernet_subsys(&packet_net_ops);
4558         if (rc)
4559                 goto out_sock;
4560         rc = register_netdevice_notifier(&packet_netdev_notifier);
4561         if (rc)
4562                 goto out_pernet;
4563
4564         return 0;
4565
4566 out_pernet:
4567         unregister_pernet_subsys(&packet_net_ops);
4568 out_sock:
4569         sock_unregister(PF_PACKET);
4570 out_proto:
4571         proto_unregister(&packet_proto);
4572 out:
4573         return rc;
4574 }
4575
4576 module_init(packet_init);
4577 module_exit(packet_exit);
4578 MODULE_LICENSE("GPL");
4579 MODULE_ALIAS_NETPROTO(PF_PACKET);