net/packet/af_packet.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   4  *              operating system.  INET is implemented using the  BSD Socket
   5  *              interface as the means of communication with the user level.
   6  *
   7  *              PACKET - implements raw packet sockets.
   8  *
   9  * Authors:     Ross Biro
  10  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  11  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  12  *
  13  * Fixes:
  14  *              Alan Cox        :       verify_area() now used correctly
  15  *              Alan Cox        :       new skbuff lists, look ma no backlogs!
  16  *              Alan Cox        :       tidied skbuff lists.
  17  *              Alan Cox        :       Now uses generic datagram routines I
  18  *                                      added. Also fixed the peek/read crash
  19  *                                      from all old Linux datagram code.
  20  *              Alan Cox        :       Uses the improved datagram code.
  21  *              Alan Cox        :       Added NULL's for socket options.
  22  *              Alan Cox        :       Re-commented the code.
  23  *              Alan Cox        :       Use new kernel side addressing
  24  *              Rob Janssen     :       Correct MTU usage.
  25  *              Dave Platt      :       Counter leaks caused by incorrect
  26  *                                      interrupt locking and some slightly
  27  *                                      dubious gcc output. Can you read
  28  *                                      compiler: it said _VOLATILE_
  29  *      Richard Kooijman        :       Timestamp fixes.
  30  *              Alan Cox        :       New buffers. Use sk->mac.raw.
  31  *              Alan Cox        :       sendmsg/recvmsg support.
  32  *              Alan Cox        :       Protocol setting support
  33  *      Alexey Kuznetsov        :       Untied from IPv4 stack.
  34  *      Cyrus Durgin            :       Fixed kerneld for kmod.
  35  *      Michal Ostrowski        :       Module initialization cleanup.
  36  *         Ulises Alonso        :       Frame number limit removal and
  37  *                                      packet_set_ring memory leak.
  38  *              Eric Biederman  :       Allow for > 8 byte hardware addresses.
  39  *                                      The convention is that longer addresses
  40  *                                      will simply extend the hardware address
  41  *                                      byte arrays at the end of sockaddr_ll
  42  *                                      and packet_mreq.
  43  *              Johann Baudy    :       Added TX RING.
  44  *              Chetan Loke     :       Implemented TPACKET_V3 block abstraction
  45  *                                      layer.
  46  *                                      Copyright (C) 2011, <lokec@ccs.neu.edu>
  47  */
  48
  49 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  50
  51 #include <linux/ethtool.h>
  52 #include <linux/filter.h>
  53 #include <linux/types.h>
  54 #include <linux/mm.h>
  55 #include <linux/capability.h>
  56 #include <linux/fcntl.h>
  57 #include <linux/socket.h>
  58 #include <linux/in.h>
  59 #include <linux/inet.h>
  60 #include <linux/netdevice.h>
  61 #include <linux/if_packet.h>
  62 #include <linux/wireless.h>
  63 #include <linux/kernel.h>
  64 #include <linux/kmod.h>
  65 #include <linux/slab.h>
  66 #include <linux/vmalloc.h>
  67 #include <net/net_namespace.h>
  68 #include <net/ip.h>
  69 #include <net/protocol.h>
  70 #include <linux/skbuff.h>
  71 #include <net/sock.h>
  72 #include <linux/errno.h>
  73 #include <linux/timer.h>
  74 #include <linux/uaccess.h>
  75 #include <asm/ioctls.h>
  76 #include <asm/page.h>
  77 #include <asm/cacheflush.h>
  78 #include <asm/io.h>
  79 #include <linux/proc_fs.h>
  80 #include <linux/seq_file.h>
  81 #include <linux/poll.h>
  82 #include <linux/module.h>
  83 #include <linux/init.h>
  84 #include <linux/mutex.h>
  85 #include <linux/if_vlan.h>
  86 #include <linux/virtio_net.h>
  87 #include <linux/errqueue.h>
  88 #include <linux/net_tstamp.h>
  89 #include <linux/percpu.h>
  90 #ifdef CONFIG_INET
  91 #include <net/inet_common.h>
  92 #endif
  93 #include <linux/bpf.h>
  94 #include <net/compat.h>
  95 #include <linux/netfilter_netdev.h>
  96
  97 #include "internal.h"
  98
  99 /*
 100    Assumptions:
 101    - If the device has no dev->header_ops->create, there is no LL header
 102      visible above the device. In this case, its hard_header_len should be 0.
 103      The device may prepend its own header internally. In this case, its
 104      needed_headroom should be set to the space needed for it to add its
 105      internal header.
 106      For example, a WiFi driver pretending to be an Ethernet driver should
 107      set its hard_header_len to be the Ethernet header length, and set its
 108      needed_headroom to be (the real WiFi header length - the fake Ethernet
 109      header length).
 110    - packet socket receives packets with pulled ll header,
 111      so that SOCK_RAW should push it back.
 112
 113 On receive:
 114 -----------
 115
 116 Incoming, dev_has_header(dev) == true
 117    mac_header -> ll header
 118    data       -> data
 119
 120 Outgoing, dev_has_header(dev) == true
 121    mac_header -> ll header
 122    data       -> ll header
 123
 124 Incoming, dev_has_header(dev) == false
 125    mac_header -> data
 126      However drivers often make it point to the ll header.
 127      This is incorrect because the ll header should be invisible to us.
 128    data       -> data
 129
 130 Outgoing, dev_has_header(dev) == false
 131    mac_header -> data. ll header is invisible to us.
 132    data       -> data
 133
 134 Resume
 135   If dev_has_header(dev) == false we are unable to restore the ll header,
 136     because it is invisible to us.
 137
 138
 139 On transmit:
 140 ------------
 141
 142 dev_has_header(dev) == true
 143    mac_header -> ll header
 144    data       -> ll header
 145
 146 dev_has_header(dev) == false (ll header is invisible to us)
 147    mac_header -> data
 148    data       -> data
 149
 150    We should set network_header on output to the correct position,
 151    packet classifier depends on it.
 152  */
 153
 154 /* Private packet socket structures. */
 155
 156 /* identical to struct packet_mreq except it has
 157  * a longer address field.
 158  */
 159 struct packet_mreq_max {
 160         int             mr_ifindex;
 161         unsigned short  mr_type;
 162         unsigned short  mr_alen;
 163         unsigned char   mr_address[MAX_ADDR_LEN];
 164 };
 165
 166 union tpacket_uhdr {
 167         struct tpacket_hdr  *h1;
 168         struct tpacket2_hdr *h2;
 169         struct tpacket3_hdr *h3;
 170         void *raw;
 171 };
 172
 173 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 174                 int closing, int tx_ring);
 175
 176 #define V3_ALIGNMENT    (8)
 177
 178 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 179
 180 #define BLK_PLUS_PRIV(sz_of_priv) \
 181         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 182
 183 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
 184 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
 185 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
 186 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
 187 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
 188 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
 189
 190 struct packet_sock;
 191 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 192                        struct packet_type *pt, struct net_device *orig_dev);
 193
 194 static void *packet_previous_frame(struct packet_sock *po,
 195                 struct packet_ring_buffer *rb,
 196                 int status);
 197 static void packet_increment_head(struct packet_ring_buffer *buff);
 198 static int prb_curr_blk_in_use(struct tpacket_block_desc *);
 199 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 200                         struct packet_sock *);
 201 static void prb_retire_current_block(struct tpacket_kbdq_core *,
 202                 struct packet_sock *, unsigned int status);
 203 static int prb_queue_frozen(struct tpacket_kbdq_core *);
 204 static void prb_open_block(struct tpacket_kbdq_core *,
 205                 struct tpacket_block_desc *);
 206 static void prb_retire_rx_blk_timer_expired(struct timer_list *);
 207 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 208 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 209 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 210                 struct tpacket3_hdr *);
 211 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 212                 struct tpacket3_hdr *);
 213 static void packet_flush_mclist(struct sock *sk);
 214 static u16 packet_pick_tx_queue(struct sk_buff *skb);
 215
 216 struct packet_skb_cb {
 217         union {
 218                 struct sockaddr_pkt pkt;
 219                 union {
 220                         /* Trick: alias skb original length with
 221                          * ll.sll_family and ll.protocol in order
 222                          * to save room.
 223                          */
 224                         unsigned int origlen;
 225                         struct sockaddr_ll ll;
 226                 };
 227         } sa;
 228 };
 229
 230 #define vio_le() virtio_legacy_is_little_endian()
 231
 232 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 233
 234 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 235 #define GET_PBLOCK_DESC(x, bid) \
 236         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 237 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 238         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 239 #define GET_NEXT_PRB_BLK_NUM(x) \
 240         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 241         ((x)->kactive_blk_num+1) : 0)
 242
 243 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 244 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 245
 246 #ifdef CONFIG_NETFILTER_EGRESS
 247 static noinline struct sk_buff *nf_hook_direct_egress(struct sk_buff *skb)
 248 {
 249         struct sk_buff *next, *head = NULL, *tail;
 250         int rc;
 251
 252         rcu_read_lock();
 253         for (; skb != NULL; skb = next) {
 254                 next = skb->next;
 255                 skb_mark_not_on_list(skb);
 256
 257                 if (!nf_hook_egress(skb, &rc, skb->dev))
 258                         continue;
 259
 260                 if (!head)
 261                         head = skb;
 262                 else
 263                         tail->next = skb;
 264
 265                 tail = skb;
 266         }
 267         rcu_read_unlock();
 268
 269         return head;
 270 }
 271 #endif
 272
 273 static int packet_direct_xmit(struct sk_buff *skb)
 274 {
 275 #ifdef CONFIG_NETFILTER_EGRESS
 276         if (nf_hook_egress_active()) {
 277                 skb = nf_hook_direct_egress(skb);
 278                 if (!skb)
 279                         return NET_XMIT_DROP;
 280         }
 281 #endif
 282         return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
 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         dev_hold(dev);
 292         rcu_read_unlock();
 293
 294         return dev;
 295 }
 296
 297 static void packet_cached_dev_assign(struct packet_sock *po,
 298                                      struct net_device *dev)
 299 {
 300         rcu_assign_pointer(po->cached_dev, dev);
 301 }
 302
 303 static void packet_cached_dev_reset(struct packet_sock *po)
 304 {
 305         RCU_INIT_POINTER(po->cached_dev, NULL);
 306 }
 307
 308 static bool packet_use_direct_xmit(const struct packet_sock *po)
 309 {
 310         return po->xmit == packet_direct_xmit;
 311 }
 312
 313 static u16 packet_pick_tx_queue(struct sk_buff *skb)
 314 {
 315         struct net_device *dev = skb->dev;
 316         const struct net_device_ops *ops = dev->netdev_ops;
 317         int cpu = raw_smp_processor_id();
 318         u16 queue_index;
 319
 320 #ifdef CONFIG_XPS
 321         skb->sender_cpu = cpu + 1;
 322 #endif
 323         skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues);
 324         if (ops->ndo_select_queue) {
 325                 queue_index = ops->ndo_select_queue(dev, skb, NULL);
 326                 queue_index = netdev_cap_txqueue(dev, queue_index);
 327         } else {
 328                 queue_index = netdev_pick_tx(dev, skb, NULL);
 329         }
 330
 331         return queue_index;
 332 }
 333
 334 /* __register_prot_hook must be invoked through register_prot_hook
 335  * or from a context in which asynchronous accesses to the packet
 336  * socket is not possible (packet_create()).
 337  */
 338 static void __register_prot_hook(struct sock *sk)
 339 {
 340         struct packet_sock *po = pkt_sk(sk);
 341
 342         if (!po->running) {
 343                 if (po->fanout)
 344                         __fanout_link(sk, po);
 345                 else
 346                         dev_add_pack(&po->prot_hook);
 347
 348                 sock_hold(sk);
 349                 po->running = 1;
 350         }
 351 }
 352
 353 static void register_prot_hook(struct sock *sk)
 354 {
 355         lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
 356         __register_prot_hook(sk);
 357 }
 358
 359 /* If the sync parameter is true, we will temporarily drop
 360  * the po->bind_lock and do a synchronize_net to make sure no
 361  * asynchronous packet processing paths still refer to the elements
 362  * of po->prot_hook.  If the sync parameter is false, it is the
 363  * callers responsibility to take care of this.
 364  */
 365 static void __unregister_prot_hook(struct sock *sk, bool sync)
 366 {
 367         struct packet_sock *po = pkt_sk(sk);
 368
 369         lockdep_assert_held_once(&po->bind_lock);
 370
 371         po->running = 0;
 372
 373         if (po->fanout)
 374                 __fanout_unlink(sk, po);
 375         else
 376                 __dev_remove_pack(&po->prot_hook);
 377
 378         __sock_put(sk);
 379
 380         if (sync) {
 381                 spin_unlock(&po->bind_lock);
 382                 synchronize_net();
 383                 spin_lock(&po->bind_lock);
 384         }
 385 }
 386
 387 static void unregister_prot_hook(struct sock *sk, bool sync)
 388 {
 389         struct packet_sock *po = pkt_sk(sk);
 390
 391         if (po->running)
 392                 __unregister_prot_hook(sk, sync);
 393 }
 394
 395 static inline struct page * __pure pgv_to_page(void *addr)
 396 {
 397         if (is_vmalloc_addr(addr))
 398                 return vmalloc_to_page(addr);
 399         return virt_to_page(addr);
 400 }
 401
 402 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 403 {
 404         union tpacket_uhdr h;
 405
 406         h.raw = frame;
 407         switch (po->tp_version) {
 408         case TPACKET_V1:
 409                 h.h1->tp_status = status;
 410                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 411                 break;
 412         case TPACKET_V2:
 413                 h.h2->tp_status = status;
 414                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 415                 break;
 416         case TPACKET_V3:
 417                 h.h3->tp_status = status;
 418                 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
 419                 break;
 420         default:
 421                 WARN(1, "TPACKET version not supported.\n");
 422                 BUG();
 423         }
 424
 425         smp_wmb();
 426 }
 427
 428 static int __packet_get_status(const struct packet_sock *po, void *frame)
 429 {
 430         union tpacket_uhdr h;
 431
 432         smp_rmb();
 433
 434         h.raw = frame;
 435         switch (po->tp_version) {
 436         case TPACKET_V1:
 437                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 438                 return h.h1->tp_status;
 439         case TPACKET_V2:
 440                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 441                 return h.h2->tp_status;
 442         case TPACKET_V3:
 443                 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
 444                 return h.h3->tp_status;
 445         default:
 446                 WARN(1, "TPACKET version not supported.\n");
 447                 BUG();
 448                 return 0;
 449         }
 450 }
 451
 452 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec64 *ts,
 453                                    unsigned int flags)
 454 {
 455         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 456
 457         if (shhwtstamps &&
 458             (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
 459             ktime_to_timespec64_cond(shhwtstamps->hwtstamp, ts))
 460                 return TP_STATUS_TS_RAW_HARDWARE;
 461
 462         if ((flags & SOF_TIMESTAMPING_SOFTWARE) &&
 463             ktime_to_timespec64_cond(skb_tstamp(skb), ts))
 464                 return TP_STATUS_TS_SOFTWARE;
 465
 466         return 0;
 467 }
 468
 469 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
 470                                     struct sk_buff *skb)
 471 {
 472         union tpacket_uhdr h;
 473         struct timespec64 ts;
 474         __u32 ts_status;
 475
 476         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
 477                 return 0;
 478
 479         h.raw = frame;
 480         /*
 481          * versions 1 through 3 overflow the timestamps in y2106, since they
 482          * all store the seconds in a 32-bit unsigned integer.
 483          * If we create a version 4, that should have a 64-bit timestamp,
 484          * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit
 485          * nanoseconds.
 486          */
 487         switch (po->tp_version) {
 488         case TPACKET_V1:
 489                 h.h1->tp_sec = ts.tv_sec;
 490                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
 491                 break;
 492         case TPACKET_V2:
 493                 h.h2->tp_sec = ts.tv_sec;
 494                 h.h2->tp_nsec = ts.tv_nsec;
 495                 break;
 496         case TPACKET_V3:
 497                 h.h3->tp_sec = ts.tv_sec;
 498                 h.h3->tp_nsec = ts.tv_nsec;
 499                 break;
 500         default:
 501                 WARN(1, "TPACKET version not supported.\n");
 502                 BUG();
 503         }
 504
 505         /* one flush is safe, as both fields always lie on the same cacheline */
 506         flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
 507         smp_wmb();
 508
 509         return ts_status;
 510 }
 511
 512 static void *packet_lookup_frame(const struct packet_sock *po,
 513                                  const struct packet_ring_buffer *rb,
 514                                  unsigned int position,
 515                                  int status)
 516 {
 517         unsigned int pg_vec_pos, frame_offset;
 518         union tpacket_uhdr h;
 519
 520         pg_vec_pos = position / rb->frames_per_block;
 521         frame_offset = position % rb->frames_per_block;
 522
 523         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 524                 (frame_offset * rb->frame_size);
 525
 526         if (status != __packet_get_status(po, h.raw))
 527                 return NULL;
 528
 529         return h.raw;
 530 }
 531
 532 static void *packet_current_frame(struct packet_sock *po,
 533                 struct packet_ring_buffer *rb,
 534                 int status)
 535 {
 536         return packet_lookup_frame(po, rb, rb->head, status);
 537 }
 538
 539 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 540 {
 541         del_timer_sync(&pkc->retire_blk_timer);
 542 }
 543
 544 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 545                 struct sk_buff_head *rb_queue)
 546 {
 547         struct tpacket_kbdq_core *pkc;
 548
 549         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 550
 551         spin_lock_bh(&rb_queue->lock);
 552         pkc->delete_blk_timer = 1;
 553         spin_unlock_bh(&rb_queue->lock);
 554
 555         prb_del_retire_blk_timer(pkc);
 556 }
 557
 558 static void prb_setup_retire_blk_timer(struct packet_sock *po)
 559 {
 560         struct tpacket_kbdq_core *pkc;
 561
 562         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 563         timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
 564                     0);
 565         pkc->retire_blk_timer.expires = jiffies;
 566 }
 567
 568 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 569                                 int blk_size_in_bytes)
 570 {
 571         struct net_device *dev;
 572         unsigned int mbits, div;
 573         struct ethtool_link_ksettings ecmd;
 574         int err;
 575
 576         rtnl_lock();
 577         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 578         if (unlikely(!dev)) {
 579                 rtnl_unlock();
 580                 return DEFAULT_PRB_RETIRE_TOV;
 581         }
 582         err = __ethtool_get_link_ksettings(dev, &ecmd);
 583         rtnl_unlock();
 584         if (err)
 585                 return DEFAULT_PRB_RETIRE_TOV;
 586
 587         /* If the link speed is so slow you don't really
 588          * need to worry about perf anyways
 589          */
 590         if (ecmd.base.speed < SPEED_1000 ||
 591             ecmd.base.speed == SPEED_UNKNOWN)
 592                 return DEFAULT_PRB_RETIRE_TOV;
 593
 594         div = ecmd.base.speed / 1000;
 595         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 596
 597         if (div)
 598                 mbits /= div;
 599
 600         if (div)
 601                 return mbits + 1;
 602         return mbits;
 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         rwlock_init(&p1->blk_fill_in_prog_lock);
 639
 640         p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
 641         prb_init_ft_ops(p1, req_u);
 642         prb_setup_retire_blk_timer(po);
 643         prb_open_block(p1, pbd);
 644 }
 645
 646 /*  Do NOT update the last_blk_num first.
 647  *  Assumes sk_buff_head lock is held.
 648  */
 649 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 650 {
 651         mod_timer(&pkc->retire_blk_timer,
 652                         jiffies + pkc->tov_in_jiffies);
 653         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 654 }
 655
 656 /*
 657  * Timer logic:
 658  * 1) We refresh the timer only when we open a block.
 659  *    By doing this we don't waste cycles refreshing the timer
 660  *        on packet-by-packet basis.
 661  *
 662  * With a 1MB block-size, on a 1Gbps line, it will take
 663  * i) ~8 ms to fill a block + ii) memcpy etc.
 664  * In this cut we are not accounting for the memcpy time.
 665  *
 666  * So, if the user sets the 'tmo' to 10ms then the timer
 667  * will never fire while the block is still getting filled
 668  * (which is what we want). However, the user could choose
 669  * to close a block early and that's fine.
 670  *
 671  * But when the timer does fire, we check whether or not to refresh it.
 672  * Since the tmo granularity is in msecs, it is not too expensive
 673  * to refresh the timer, lets say every '8' msecs.
 674  * Either the user can set the 'tmo' or we can derive it based on
 675  * a) line-speed and b) block-size.
 676  * prb_calc_retire_blk_tmo() calculates the tmo.
 677  *
 678  */
 679 static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
 680 {
 681         struct packet_sock *po =
 682                 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
 683         struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 684         unsigned int frozen;
 685         struct tpacket_block_desc *pbd;
 686
 687         spin_lock(&po->sk.sk_receive_queue.lock);
 688
 689         frozen = prb_queue_frozen(pkc);
 690         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 691
 692         if (unlikely(pkc->delete_blk_timer))
 693                 goto out;
 694
 695         /* We only need to plug the race when the block is partially filled.
 696          * tpacket_rcv:
 697          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 698          *              copy_bits() is in progress ...
 699          *              timer fires on other cpu:
 700          *              we can't retire the current block because copy_bits
 701          *              is in progress.
 702          *
 703          */
 704         if (BLOCK_NUM_PKTS(pbd)) {
 705                 /* Waiting for skb_copy_bits to finish... */
 706                 write_lock(&pkc->blk_fill_in_prog_lock);
 707                 write_unlock(&pkc->blk_fill_in_prog_lock);
 708         }
 709
 710         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 711                 if (!frozen) {
 712                         if (!BLOCK_NUM_PKTS(pbd)) {
 713                                 /* An empty block. Just refresh the timer. */
 714                                 goto refresh_timer;
 715                         }
 716                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 717                         if (!prb_dispatch_next_block(pkc, po))
 718                                 goto refresh_timer;
 719                         else
 720                                 goto out;
 721                 } else {
 722                         /* Case 1. Queue was frozen because user-space was
 723                          *         lagging behind.
 724                          */
 725                         if (prb_curr_blk_in_use(pbd)) {
 726                                 /*
 727                                  * Ok, user-space is still behind.
 728                                  * So just refresh the timer.
 729                                  */
 730                                 goto refresh_timer;
 731                         } else {
 732                                /* Case 2. queue was frozen,user-space caught up,
 733                                 * now the link went idle && the timer fired.
 734                                 * We don't have a block to close.So we open this
 735                                 * block and restart the timer.
 736                                 * opening a block thaws the queue,restarts timer
 737                                 * Thawing/timer-refresh is a side effect.
 738                                 */
 739                                 prb_open_block(pkc, pbd);
 740                                 goto out;
 741                         }
 742                 }
 743         }
 744
 745 refresh_timer:
 746         _prb_refresh_rx_retire_blk_timer(pkc);
 747
 748 out:
 749         spin_unlock(&po->sk.sk_receive_queue.lock);
 750 }
 751
 752 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 753                 struct tpacket_block_desc *pbd1, __u32 status)
 754 {
 755         /* Flush everything minus the block header */
 756
 757 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 758         u8 *start, *end;
 759
 760         start = (u8 *)pbd1;
 761
 762         /* Skip the block header(we know header WILL fit in 4K) */
 763         start += PAGE_SIZE;
 764
 765         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 766         for (; start < end; start += PAGE_SIZE)
 767                 flush_dcache_page(pgv_to_page(start));
 768
 769         smp_wmb();
 770 #endif
 771
 772         /* Now update the block status. */
 773
 774         BLOCK_STATUS(pbd1) = status;
 775
 776         /* Flush the block header */
 777
 778 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 779         start = (u8 *)pbd1;
 780         flush_dcache_page(pgv_to_page(start));
 781
 782         smp_wmb();
 783 #endif
 784 }
 785
 786 /*
 787  * Side effect:
 788  *
 789  * 1) flush the block
 790  * 2) Increment active_blk_num
 791  *
 792  * Note:We DONT refresh the timer on purpose.
 793  *      Because almost always the next block will be opened.
 794  */
 795 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 796                 struct tpacket_block_desc *pbd1,
 797                 struct packet_sock *po, unsigned int stat)
 798 {
 799         __u32 status = TP_STATUS_USER | stat;
 800
 801         struct tpacket3_hdr *last_pkt;
 802         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 803         struct sock *sk = &po->sk;
 804
 805         if (atomic_read(&po->tp_drops))
 806                 status |= TP_STATUS_LOSING;
 807
 808         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 809         last_pkt->tp_next_offset = 0;
 810
 811         /* Get the ts of the last pkt */
 812         if (BLOCK_NUM_PKTS(pbd1)) {
 813                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 814                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 815         } else {
 816                 /* Ok, we tmo'd - so get the current time.
 817                  *
 818                  * It shouldn't really happen as we don't close empty
 819                  * blocks. See prb_retire_rx_blk_timer_expired().
 820                  */
 821                 struct timespec64 ts;
 822                 ktime_get_real_ts64(&ts);
 823                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 824                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 825         }
 826
 827         smp_wmb();
 828
 829         /* Flush the block */
 830         prb_flush_block(pkc1, pbd1, status);
 831
 832         sk->sk_data_ready(sk);
 833
 834         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 835 }
 836
 837 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 838 {
 839         pkc->reset_pending_on_curr_blk = 0;
 840 }
 841
 842 /*
 843  * Side effect of opening a block:
 844  *
 845  * 1) prb_queue is thawed.
 846  * 2) retire_blk_timer is refreshed.
 847  *
 848  */
 849 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 850         struct tpacket_block_desc *pbd1)
 851 {
 852         struct timespec64 ts;
 853         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 854
 855         smp_rmb();
 856
 857         /* We could have just memset this but we will lose the
 858          * flexibility of making the priv area sticky
 859          */
 860
 861         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 862         BLOCK_NUM_PKTS(pbd1) = 0;
 863         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 864
 865         ktime_get_real_ts64(&ts);
 866
 867         h1->ts_first_pkt.ts_sec = ts.tv_sec;
 868         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 869
 870         pkc1->pkblk_start = (char *)pbd1;
 871         pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 872
 873         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 874         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 875
 876         pbd1->version = pkc1->version;
 877         pkc1->prev = pkc1->nxt_offset;
 878         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 879
 880         prb_thaw_queue(pkc1);
 881         _prb_refresh_rx_retire_blk_timer(pkc1);
 882
 883         smp_wmb();
 884 }
 885
 886 /*
 887  * Queue freeze logic:
 888  * 1) Assume tp_block_nr = 8 blocks.
 889  * 2) At time 't0', user opens Rx ring.
 890  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 891  * 4) user-space is either sleeping or processing block '0'.
 892  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 893  *    it will close block-7,loop around and try to fill block '0'.
 894  *    call-flow:
 895  *    __packet_lookup_frame_in_block
 896  *      prb_retire_current_block()
 897  *      prb_dispatch_next_block()
 898  *        |->(BLOCK_STATUS == USER) evaluates to true
 899  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 900  * 6) Now there are two cases:
 901  *    6.1) Link goes idle right after the queue is frozen.
 902  *         But remember, the last open_block() refreshed the timer.
 903  *         When this timer expires,it will refresh itself so that we can
 904  *         re-open block-0 in near future.
 905  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 906  *         case and __packet_lookup_frame_in_block will check if block-0
 907  *         is free and can now be re-used.
 908  */
 909 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 910                                   struct packet_sock *po)
 911 {
 912         pkc->reset_pending_on_curr_blk = 1;
 913         po->stats.stats3.tp_freeze_q_cnt++;
 914 }
 915
 916 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 917
 918 /*
 919  * If the next block is free then we will dispatch it
 920  * and return a good offset.
 921  * Else, we will freeze the queue.
 922  * So, caller must check the return value.
 923  */
 924 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 925                 struct packet_sock *po)
 926 {
 927         struct tpacket_block_desc *pbd;
 928
 929         smp_rmb();
 930
 931         /* 1. Get current block num */
 932         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 933
 934         /* 2. If this block is currently in_use then freeze the queue */
 935         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 936                 prb_freeze_queue(pkc, po);
 937                 return NULL;
 938         }
 939
 940         /*
 941          * 3.
 942          * open this block and return the offset where the first packet
 943          * needs to get stored.
 944          */
 945         prb_open_block(pkc, pbd);
 946         return (void *)pkc->nxt_offset;
 947 }
 948
 949 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 950                 struct packet_sock *po, unsigned int status)
 951 {
 952         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 953
 954         /* retire/close the current block */
 955         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 956                 /*
 957                  * Plug the case where copy_bits() is in progress on
 958                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 959                  * have space to copy the pkt in the current block and
 960                  * called prb_retire_current_block()
 961                  *
 962                  * We don't need to worry about the TMO case because
 963                  * the timer-handler already handled this case.
 964                  */
 965                 if (!(status & TP_STATUS_BLK_TMO)) {
 966                         /* Waiting for skb_copy_bits to finish... */
 967                         write_lock(&pkc->blk_fill_in_prog_lock);
 968                         write_unlock(&pkc->blk_fill_in_prog_lock);
 969                 }
 970                 prb_close_block(pkc, pbd, po, status);
 971                 return;
 972         }
 973 }
 974
 975 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
 976 {
 977         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 978 }
 979
 980 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 981 {
 982         return pkc->reset_pending_on_curr_blk;
 983 }
 984
 985 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 986         __releases(&pkc->blk_fill_in_prog_lock)
 987 {
 988         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 989
 990         read_unlock(&pkc->blk_fill_in_prog_lock);
 991 }
 992
 993 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 994                         struct tpacket3_hdr *ppd)
 995 {
 996         ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
 997 }
 998
 999 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
1000                         struct tpacket3_hdr *ppd)
1001 {
1002         ppd->hv1.tp_rxhash = 0;
1003 }
1004
1005 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1006                         struct tpacket3_hdr *ppd)
1007 {
1008         if (skb_vlan_tag_present(pkc->skb)) {
1009                 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1010                 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1011                 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1012         } else {
1013                 ppd->hv1.tp_vlan_tci = 0;
1014                 ppd->hv1.tp_vlan_tpid = 0;
1015                 ppd->tp_status = TP_STATUS_AVAILABLE;
1016         }
1017 }
1018
1019 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1020                         struct tpacket3_hdr *ppd)
1021 {
1022         ppd->hv1.tp_padding = 0;
1023         prb_fill_vlan_info(pkc, ppd);
1024
1025         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1026                 prb_fill_rxhash(pkc, ppd);
1027         else
1028                 prb_clear_rxhash(pkc, ppd);
1029 }
1030
1031 static void prb_fill_curr_block(char *curr,
1032                                 struct tpacket_kbdq_core *pkc,
1033                                 struct tpacket_block_desc *pbd,
1034                                 unsigned int len)
1035         __acquires(&pkc->blk_fill_in_prog_lock)
1036 {
1037         struct tpacket3_hdr *ppd;
1038
1039         ppd  = (struct tpacket3_hdr *)curr;
1040         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1041         pkc->prev = curr;
1042         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1043         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1044         BLOCK_NUM_PKTS(pbd) += 1;
1045         read_lock(&pkc->blk_fill_in_prog_lock);
1046         prb_run_all_ft_ops(pkc, ppd);
1047 }
1048
1049 /* Assumes caller has the sk->rx_queue.lock */
1050 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1051                                             struct sk_buff *skb,
1052                                             unsigned int len
1053                                             )
1054 {
1055         struct tpacket_kbdq_core *pkc;
1056         struct tpacket_block_desc *pbd;
1057         char *curr, *end;
1058
1059         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1060         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1061
1062         /* Queue is frozen when user space is lagging behind */
1063         if (prb_queue_frozen(pkc)) {
1064                 /*
1065                  * Check if that last block which caused the queue to freeze,
1066                  * is still in_use by user-space.
1067                  */
1068                 if (prb_curr_blk_in_use(pbd)) {
1069                         /* Can't record this packet */
1070                         return NULL;
1071                 } else {
1072                         /*
1073                          * Ok, the block was released by user-space.
1074                          * Now let's open that block.
1075                          * opening a block also thaws the queue.
1076                          * Thawing is a side effect.
1077                          */
1078                         prb_open_block(pkc, pbd);
1079                 }
1080         }
1081
1082         smp_mb();
1083         curr = pkc->nxt_offset;
1084         pkc->skb = skb;
1085         end = (char *)pbd + pkc->kblk_size;
1086
1087         /* first try the current block */
1088         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1089                 prb_fill_curr_block(curr, pkc, pbd, len);
1090                 return (void *)curr;
1091         }
1092
1093         /* Ok, close the current block */
1094         prb_retire_current_block(pkc, po, 0);
1095
1096         /* Now, try to dispatch the next block */
1097         curr = (char *)prb_dispatch_next_block(pkc, po);
1098         if (curr) {
1099                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1100                 prb_fill_curr_block(curr, pkc, pbd, len);
1101                 return (void *)curr;
1102         }
1103
1104         /*
1105          * No free blocks are available.user_space hasn't caught up yet.
1106          * Queue was just frozen and now this packet will get dropped.
1107          */
1108         return NULL;
1109 }
1110
1111 static void *packet_current_rx_frame(struct packet_sock *po,
1112                                             struct sk_buff *skb,
1113                                             int status, unsigned int len)
1114 {
1115         char *curr = NULL;
1116         switch (po->tp_version) {
1117         case TPACKET_V1:
1118         case TPACKET_V2:
1119                 curr = packet_lookup_frame(po, &po->rx_ring,
1120                                         po->rx_ring.head, status);
1121                 return curr;
1122         case TPACKET_V3:
1123                 return __packet_lookup_frame_in_block(po, skb, len);
1124         default:
1125                 WARN(1, "TPACKET version not supported\n");
1126                 BUG();
1127                 return NULL;
1128         }
1129 }
1130
1131 static void *prb_lookup_block(const struct packet_sock *po,
1132                               const struct packet_ring_buffer *rb,
1133                               unsigned int idx,
1134                               int status)
1135 {
1136         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1137         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1138
1139         if (status != BLOCK_STATUS(pbd))
1140                 return NULL;
1141         return pbd;
1142 }
1143
1144 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1145 {
1146         unsigned int prev;
1147         if (rb->prb_bdqc.kactive_blk_num)
1148                 prev = rb->prb_bdqc.kactive_blk_num-1;
1149         else
1150                 prev = rb->prb_bdqc.knum_blocks-1;
1151         return prev;
1152 }
1153
1154 /* Assumes caller has held the rx_queue.lock */
1155 static void *__prb_previous_block(struct packet_sock *po,
1156                                          struct packet_ring_buffer *rb,
1157                                          int status)
1158 {
1159         unsigned int previous = prb_previous_blk_num(rb);
1160         return prb_lookup_block(po, rb, previous, status);
1161 }
1162
1163 static void *packet_previous_rx_frame(struct packet_sock *po,
1164                                              struct packet_ring_buffer *rb,
1165                                              int status)
1166 {
1167         if (po->tp_version <= TPACKET_V2)
1168                 return packet_previous_frame(po, rb, status);
1169
1170         return __prb_previous_block(po, rb, status);
1171 }
1172
1173 static void packet_increment_rx_head(struct packet_sock *po,
1174                                             struct packet_ring_buffer *rb)
1175 {
1176         switch (po->tp_version) {
1177         case TPACKET_V1:
1178         case TPACKET_V2:
1179                 return packet_increment_head(rb);
1180         case TPACKET_V3:
1181         default:
1182                 WARN(1, "TPACKET version not supported.\n");
1183                 BUG();
1184                 return;
1185         }
1186 }
1187
1188 static void *packet_previous_frame(struct packet_sock *po,
1189                 struct packet_ring_buffer *rb,
1190                 int status)
1191 {
1192         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1193         return packet_lookup_frame(po, rb, previous, status);
1194 }
1195
1196 static void packet_increment_head(struct packet_ring_buffer *buff)
1197 {
1198         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1199 }
1200
1201 static void packet_inc_pending(struct packet_ring_buffer *rb)
1202 {
1203         this_cpu_inc(*rb->pending_refcnt);
1204 }
1205
1206 static void packet_dec_pending(struct packet_ring_buffer *rb)
1207 {
1208         this_cpu_dec(*rb->pending_refcnt);
1209 }
1210
1211 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1212 {
1213         unsigned int refcnt = 0;
1214         int cpu;
1215
1216         /* We don't use pending refcount in rx_ring. */
1217         if (rb->pending_refcnt == NULL)
1218                 return 0;
1219
1220         for_each_possible_cpu(cpu)
1221                 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1222
1223         return refcnt;
1224 }
1225
1226 static int packet_alloc_pending(struct packet_sock *po)
1227 {
1228         po->rx_ring.pending_refcnt = NULL;
1229
1230         po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1231         if (unlikely(po->tx_ring.pending_refcnt == NULL))
1232                 return -ENOBUFS;
1233
1234         return 0;
1235 }
1236
1237 static void packet_free_pending(struct packet_sock *po)
1238 {
1239         free_percpu(po->tx_ring.pending_refcnt);
1240 }
1241
1242 #define ROOM_POW_OFF    2
1243 #define ROOM_NONE       0x0
1244 #define ROOM_LOW        0x1
1245 #define ROOM_NORMAL     0x2
1246
1247 static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1248 {
1249         int idx, len;
1250
1251         len = READ_ONCE(po->rx_ring.frame_max) + 1;
1252         idx = READ_ONCE(po->rx_ring.head);
1253         if (pow_off)
1254                 idx += len >> pow_off;
1255         if (idx >= len)
1256                 idx -= len;
1257         return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1258 }
1259
1260 static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1261 {
1262         int idx, len;
1263
1264         len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1265         idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1266         if (pow_off)
1267                 idx += len >> pow_off;
1268         if (idx >= len)
1269                 idx -= len;
1270         return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1271 }
1272
1273 static int __packet_rcv_has_room(const struct packet_sock *po,
1274                                  const struct sk_buff *skb)
1275 {
1276         const struct sock *sk = &po->sk;
1277         int ret = ROOM_NONE;
1278
1279         if (po->prot_hook.func != tpacket_rcv) {
1280                 int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1281                 int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1282                                    - (skb ? skb->truesize : 0);
1283
1284                 if (avail > (rcvbuf >> ROOM_POW_OFF))
1285                         return ROOM_NORMAL;
1286                 else if (avail > 0)
1287                         return ROOM_LOW;
1288                 else
1289                         return ROOM_NONE;
1290         }
1291
1292         if (po->tp_version == TPACKET_V3) {
1293                 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1294                         ret = ROOM_NORMAL;
1295                 else if (__tpacket_v3_has_room(po, 0))
1296                         ret = ROOM_LOW;
1297         } else {
1298                 if (__tpacket_has_room(po, ROOM_POW_OFF))
1299                         ret = ROOM_NORMAL;
1300                 else if (__tpacket_has_room(po, 0))
1301                         ret = ROOM_LOW;
1302         }
1303
1304         return ret;
1305 }
1306
1307 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1308 {
1309         int pressure, ret;
1310
1311         ret = __packet_rcv_has_room(po, skb);
1312         pressure = ret != ROOM_NORMAL;
1313
1314         if (READ_ONCE(po->pressure) != pressure)
1315                 WRITE_ONCE(po->pressure, pressure);
1316
1317         return ret;
1318 }
1319
1320 static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1321 {
1322         if (READ_ONCE(po->pressure) &&
1323             __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1324                 WRITE_ONCE(po->pressure,  0);
1325 }
1326
1327 static void packet_sock_destruct(struct sock *sk)
1328 {
1329         skb_queue_purge(&sk->sk_error_queue);
1330
1331         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1332         WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1333
1334         if (!sock_flag(sk, SOCK_DEAD)) {
1335                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1336                 return;
1337         }
1338
1339         sk_refcnt_debug_dec(sk);
1340 }
1341
1342 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1343 {
1344         u32 *history = po->rollover->history;
1345         u32 victim, rxhash;
1346         int i, count = 0;
1347
1348         rxhash = skb_get_hash(skb);
1349         for (i = 0; i < ROLLOVER_HLEN; i++)
1350                 if (READ_ONCE(history[i]) == rxhash)
1351                         count++;
1352
1353         victim = prandom_u32() % ROLLOVER_HLEN;
1354
1355         /* Avoid dirtying the cache line if possible */
1356         if (READ_ONCE(history[victim]) != rxhash)
1357                 WRITE_ONCE(history[victim], rxhash);
1358
1359         return count > (ROLLOVER_HLEN >> 1);
1360 }
1361
1362 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1363                                       struct sk_buff *skb,
1364                                       unsigned int num)
1365 {
1366         return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1367 }
1368
1369 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1370                                     struct sk_buff *skb,
1371                                     unsigned int num)
1372 {
1373         unsigned int val = atomic_inc_return(&f->rr_cur);
1374
1375         return val % num;
1376 }
1377
1378 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1379                                      struct sk_buff *skb,
1380                                      unsigned int num)
1381 {
1382         return smp_processor_id() % num;
1383 }
1384
1385 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1386                                      struct sk_buff *skb,
1387                                      unsigned int num)
1388 {
1389         return prandom_u32_max(num);
1390 }
1391
1392 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1393                                           struct sk_buff *skb,
1394                                           unsigned int idx, bool try_self,
1395                                           unsigned int num)
1396 {
1397         struct packet_sock *po, *po_next, *po_skip = NULL;
1398         unsigned int i, j, room = ROOM_NONE;
1399
1400         po = pkt_sk(rcu_dereference(f->arr[idx]));
1401
1402         if (try_self) {
1403                 room = packet_rcv_has_room(po, skb);
1404                 if (room == ROOM_NORMAL ||
1405                     (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1406                         return idx;
1407                 po_skip = po;
1408         }
1409
1410         i = j = min_t(int, po->rollover->sock, num - 1);
1411         do {
1412                 po_next = pkt_sk(rcu_dereference(f->arr[i]));
1413                 if (po_next != po_skip && !READ_ONCE(po_next->pressure) &&
1414                     packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1415                         if (i != j)
1416                                 po->rollover->sock = i;
1417                         atomic_long_inc(&po->rollover->num);
1418                         if (room == ROOM_LOW)
1419                                 atomic_long_inc(&po->rollover->num_huge);
1420                         return i;
1421                 }
1422
1423                 if (++i == num)
1424                         i = 0;
1425         } while (i != j);
1426
1427         atomic_long_inc(&po->rollover->num_failed);
1428         return idx;
1429 }
1430
1431 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1432                                     struct sk_buff *skb,
1433                                     unsigned int num)
1434 {
1435         return skb_get_queue_mapping(skb) % num;
1436 }
1437
1438 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1439                                      struct sk_buff *skb,
1440                                      unsigned int num)
1441 {
1442         struct bpf_prog *prog;
1443         unsigned int ret = 0;
1444
1445         rcu_read_lock();
1446         prog = rcu_dereference(f->bpf_prog);
1447         if (prog)
1448                 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1449         rcu_read_unlock();
1450
1451         return ret;
1452 }
1453
1454 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1455 {
1456         return f->flags & (flag >> 8);
1457 }
1458
1459 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1460                              struct packet_type *pt, struct net_device *orig_dev)
1461 {
1462         struct packet_fanout *f = pt->af_packet_priv;
1463         unsigned int num = READ_ONCE(f->num_members);
1464         struct net *net = read_pnet(&f->net);
1465         struct packet_sock *po;
1466         unsigned int idx;
1467
1468         if (!net_eq(dev_net(dev), net) || !num) {
1469                 kfree_skb(skb);
1470                 return 0;
1471         }
1472
1473         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1474                 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1475                 if (!skb)
1476                         return 0;
1477         }
1478         switch (f->type) {
1479         case PACKET_FANOUT_HASH:
1480         default:
1481                 idx = fanout_demux_hash(f, skb, num);
1482                 break;
1483         case PACKET_FANOUT_LB:
1484                 idx = fanout_demux_lb(f, skb, num);
1485                 break;
1486         case PACKET_FANOUT_CPU:
1487                 idx = fanout_demux_cpu(f, skb, num);
1488                 break;
1489         case PACKET_FANOUT_RND:
1490                 idx = fanout_demux_rnd(f, skb, num);
1491                 break;
1492         case PACKET_FANOUT_QM:
1493                 idx = fanout_demux_qm(f, skb, num);
1494                 break;
1495         case PACKET_FANOUT_ROLLOVER:
1496                 idx = fanout_demux_rollover(f, skb, 0, false, num);
1497                 break;
1498         case PACKET_FANOUT_CBPF:
1499         case PACKET_FANOUT_EBPF:
1500                 idx = fanout_demux_bpf(f, skb, num);
1501                 break;
1502         }
1503
1504         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1505                 idx = fanout_demux_rollover(f, skb, idx, true, num);
1506
1507         po = pkt_sk(rcu_dereference(f->arr[idx]));
1508         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1509 }
1510
1511 DEFINE_MUTEX(fanout_mutex);
1512 EXPORT_SYMBOL_GPL(fanout_mutex);
1513 static LIST_HEAD(fanout_list);
1514 static u16 fanout_next_id;
1515
1516 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1517 {
1518         struct packet_fanout *f = po->fanout;
1519
1520         spin_lock(&f->lock);
1521         rcu_assign_pointer(f->arr[f->num_members], sk);
1522         smp_wmb();
1523         f->num_members++;
1524         if (f->num_members == 1)
1525                 dev_add_pack(&f->prot_hook);
1526         spin_unlock(&f->lock);
1527 }
1528
1529 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1530 {
1531         struct packet_fanout *f = po->fanout;
1532         int i;
1533
1534         spin_lock(&f->lock);
1535         for (i = 0; i < f->num_members; i++) {
1536                 if (rcu_dereference_protected(f->arr[i],
1537                                               lockdep_is_held(&f->lock)) == sk)
1538                         break;
1539         }
1540         BUG_ON(i >= f->num_members);
1541         rcu_assign_pointer(f->arr[i],
1542                            rcu_dereference_protected(f->arr[f->num_members - 1],
1543                                                      lockdep_is_held(&f->lock)));
1544         f->num_members--;
1545         if (f->num_members == 0)
1546                 __dev_remove_pack(&f->prot_hook);
1547         spin_unlock(&f->lock);
1548 }
1549
1550 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1551 {
1552         if (sk->sk_family != PF_PACKET)
1553                 return false;
1554
1555         return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1556 }
1557
1558 static void fanout_init_data(struct packet_fanout *f)
1559 {
1560         switch (f->type) {
1561         case PACKET_FANOUT_LB:
1562                 atomic_set(&f->rr_cur, 0);
1563                 break;
1564         case PACKET_FANOUT_CBPF:
1565         case PACKET_FANOUT_EBPF:
1566                 RCU_INIT_POINTER(f->bpf_prog, NULL);
1567                 break;
1568         }
1569 }
1570
1571 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1572 {
1573         struct bpf_prog *old;
1574
1575         spin_lock(&f->lock);
1576         old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1577         rcu_assign_pointer(f->bpf_prog, new);
1578         spin_unlock(&f->lock);
1579
1580         if (old) {
1581                 synchronize_net();
1582                 bpf_prog_destroy(old);
1583         }
1584 }
1585
1586 static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data,
1587                                 unsigned int len)
1588 {
1589         struct bpf_prog *new;
1590         struct sock_fprog fprog;
1591         int ret;
1592
1593         if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1594                 return -EPERM;
1595
1596         ret = copy_bpf_fprog_from_user(&fprog, data, len);
1597         if (ret)
1598                 return ret;
1599
1600         ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1601         if (ret)
1602                 return ret;
1603
1604         __fanout_set_data_bpf(po->fanout, new);
1605         return 0;
1606 }
1607
1608 static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data,
1609                                 unsigned int len)
1610 {
1611         struct bpf_prog *new;
1612         u32 fd;
1613
1614         if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1615                 return -EPERM;
1616         if (len != sizeof(fd))
1617                 return -EINVAL;
1618         if (copy_from_sockptr(&fd, data, len))
1619                 return -EFAULT;
1620
1621         new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1622         if (IS_ERR(new))
1623                 return PTR_ERR(new);
1624
1625         __fanout_set_data_bpf(po->fanout, new);
1626         return 0;
1627 }
1628
1629 static int fanout_set_data(struct packet_sock *po, sockptr_t data,
1630                            unsigned int len)
1631 {
1632         switch (po->fanout->type) {
1633         case PACKET_FANOUT_CBPF:
1634                 return fanout_set_data_cbpf(po, data, len);
1635         case PACKET_FANOUT_EBPF:
1636                 return fanout_set_data_ebpf(po, data, len);
1637         default:
1638                 return -EINVAL;
1639         }
1640 }
1641
1642 static void fanout_release_data(struct packet_fanout *f)
1643 {
1644         switch (f->type) {
1645         case PACKET_FANOUT_CBPF:
1646         case PACKET_FANOUT_EBPF:
1647                 __fanout_set_data_bpf(f, NULL);
1648         }
1649 }
1650
1651 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1652 {
1653         struct packet_fanout *f;
1654
1655         list_for_each_entry(f, &fanout_list, list) {
1656                 if (f->id == candidate_id &&
1657                     read_pnet(&f->net) == sock_net(sk)) {
1658                         return false;
1659                 }
1660         }
1661         return true;
1662 }
1663
1664 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1665 {
1666         u16 id = fanout_next_id;
1667
1668         do {
1669                 if (__fanout_id_is_free(sk, id)) {
1670                         *new_id = id;
1671                         fanout_next_id = id + 1;
1672                         return true;
1673                 }
1674
1675                 id++;
1676         } while (id != fanout_next_id);
1677
1678         return false;
1679 }
1680
1681 static int fanout_add(struct sock *sk, struct fanout_args *args)
1682 {
1683         struct packet_rollover *rollover = NULL;
1684         struct packet_sock *po = pkt_sk(sk);
1685         u16 type_flags = args->type_flags;
1686         struct packet_fanout *f, *match;
1687         u8 type = type_flags & 0xff;
1688         u8 flags = type_flags >> 8;
1689         u16 id = args->id;
1690         int err;
1691
1692         switch (type) {
1693         case PACKET_FANOUT_ROLLOVER:
1694                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1695                         return -EINVAL;
1696                 break;
1697         case PACKET_FANOUT_HASH:
1698         case PACKET_FANOUT_LB:
1699         case PACKET_FANOUT_CPU:
1700         case PACKET_FANOUT_RND:
1701         case PACKET_FANOUT_QM:
1702         case PACKET_FANOUT_CBPF:
1703         case PACKET_FANOUT_EBPF:
1704                 break;
1705         default:
1706                 return -EINVAL;
1707         }
1708
1709         mutex_lock(&fanout_mutex);
1710
1711         err = -EALREADY;
1712         if (po->fanout)
1713                 goto out;
1714
1715         if (type == PACKET_FANOUT_ROLLOVER ||
1716             (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1717                 err = -ENOMEM;
1718                 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1719                 if (!rollover)
1720                         goto out;
1721                 atomic_long_set(&rollover->num, 0);
1722                 atomic_long_set(&rollover->num_huge, 0);
1723                 atomic_long_set(&rollover->num_failed, 0);
1724         }
1725
1726         if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1727                 if (id != 0) {
1728                         err = -EINVAL;
1729                         goto out;
1730                 }
1731                 if (!fanout_find_new_id(sk, &id)) {
1732                         err = -ENOMEM;
1733                         goto out;
1734                 }
1735                 /* ephemeral flag for the first socket in the group: drop it */
1736                 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1737         }
1738
1739         match = NULL;
1740         list_for_each_entry(f, &fanout_list, list) {
1741                 if (f->id == id &&
1742                     read_pnet(&f->net) == sock_net(sk)) {
1743                         match = f;
1744                         break;
1745                 }
1746         }
1747         err = -EINVAL;
1748         if (match) {
1749                 if (match->flags != flags)
1750                         goto out;
1751                 if (args->max_num_members &&
1752                     args->max_num_members != match->max_num_members)
1753                         goto out;
1754         } else {
1755                 if (args->max_num_members > PACKET_FANOUT_MAX)
1756                         goto out;
1757                 if (!args->max_num_members)
1758                         /* legacy PACKET_FANOUT_MAX */
1759                         args->max_num_members = 256;
1760                 err = -ENOMEM;
1761                 match = kvzalloc(struct_size(match, arr, args->max_num_members),
1762                                  GFP_KERNEL);
1763                 if (!match)
1764                         goto out;
1765                 write_pnet(&match->net, sock_net(sk));
1766                 match->id = id;
1767                 match->type = type;
1768                 match->flags = flags;
1769                 INIT_LIST_HEAD(&match->list);
1770                 spin_lock_init(&match->lock);
1771                 refcount_set(&match->sk_ref, 0);
1772                 fanout_init_data(match);
1773                 match->prot_hook.type = po->prot_hook.type;
1774                 match->prot_hook.dev = po->prot_hook.dev;
1775                 match->prot_hook.func = packet_rcv_fanout;
1776                 match->prot_hook.af_packet_priv = match;
1777                 match->prot_hook.af_packet_net = read_pnet(&match->net);
1778                 match->prot_hook.id_match = match_fanout_group;
1779                 match->max_num_members = args->max_num_members;
1780                 list_add(&match->list, &fanout_list);
1781         }
1782         err = -EINVAL;
1783
1784         spin_lock(&po->bind_lock);
1785         if (po->running &&
1786             match->type == type &&
1787             match->prot_hook.type == po->prot_hook.type &&
1788             match->prot_hook.dev == po->prot_hook.dev) {
1789                 err = -ENOSPC;
1790                 if (refcount_read(&match->sk_ref) < match->max_num_members) {
1791                         __dev_remove_pack(&po->prot_hook);
1792
1793                         /* Paired with packet_setsockopt(PACKET_FANOUT_DATA) */
1794                         WRITE_ONCE(po->fanout, match);
1795
1796                         po->rollover = rollover;
1797                         rollover = NULL;
1798                         refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1799                         __fanout_link(sk, po);
1800                         err = 0;
1801                 }
1802         }
1803         spin_unlock(&po->bind_lock);
1804
1805         if (err && !refcount_read(&match->sk_ref)) {
1806                 list_del(&match->list);
1807                 kvfree(match);
1808         }
1809
1810 out:
1811         kfree(rollover);
1812         mutex_unlock(&fanout_mutex);
1813         return err;
1814 }
1815
1816 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1817  * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1818  * It is the responsibility of the caller to call fanout_release_data() and
1819  * free the returned packet_fanout (after synchronize_net())
1820  */
1821 static struct packet_fanout *fanout_release(struct sock *sk)
1822 {
1823         struct packet_sock *po = pkt_sk(sk);
1824         struct packet_fanout *f;
1825
1826         mutex_lock(&fanout_mutex);
1827         f = po->fanout;
1828         if (f) {
1829                 po->fanout = NULL;
1830
1831                 if (refcount_dec_and_test(&f->sk_ref))
1832                         list_del(&f->list);
1833                 else
1834                         f = NULL;
1835         }
1836         mutex_unlock(&fanout_mutex);
1837
1838         return f;
1839 }
1840
1841 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1842                                           struct sk_buff *skb)
1843 {
1844         /* Earlier code assumed this would be a VLAN pkt, double-check
1845          * this now that we have the actual packet in hand. We can only
1846          * do this check on Ethernet devices.
1847          */
1848         if (unlikely(dev->type != ARPHRD_ETHER))
1849                 return false;
1850
1851         skb_reset_mac_header(skb);
1852         return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1853 }
1854
1855 static const struct proto_ops packet_ops;
1856
1857 static const struct proto_ops packet_ops_spkt;
1858
1859 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1860                            struct packet_type *pt, struct net_device *orig_dev)
1861 {
1862         struct sock *sk;
1863         struct sockaddr_pkt *spkt;
1864
1865         /*
1866          *      When we registered the protocol we saved the socket in the data
1867          *      field for just this event.
1868          */
1869
1870         sk = pt->af_packet_priv;
1871
1872         /*
1873          *      Yank back the headers [hope the device set this
1874          *      right or kerboom...]
1875          *
1876          *      Incoming packets have ll header pulled,
1877          *      push it back.
1878          *
1879          *      For outgoing ones skb->data == skb_mac_header(skb)
1880          *      so that this procedure is noop.
1881          */
1882
1883         if (skb->pkt_type == PACKET_LOOPBACK)
1884                 goto out;
1885
1886         if (!net_eq(dev_net(dev), sock_net(sk)))
1887                 goto out;
1888
1889         skb = skb_share_check(skb, GFP_ATOMIC);
1890         if (skb == NULL)
1891                 goto oom;
1892
1893         /* drop any routing info */
1894         skb_dst_drop(skb);
1895
1896         /* drop conntrack reference */
1897         nf_reset_ct(skb);
1898
1899         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1900
1901         skb_push(skb, skb->data - skb_mac_header(skb));
1902
1903         /*
1904          *      The SOCK_PACKET socket receives _all_ frames.
1905          */
1906
1907         spkt->spkt_family = dev->type;
1908         strscpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1909         spkt->spkt_protocol = skb->protocol;
1910
1911         /*
1912          *      Charge the memory to the socket. This is done specifically
1913          *      to prevent sockets using all the memory up.
1914          */
1915
1916         if (sock_queue_rcv_skb(sk, skb) == 0)
1917                 return 0;
1918
1919 out:
1920         kfree_skb(skb);
1921 oom:
1922         return 0;
1923 }
1924
1925 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1926 {
1927         int depth;
1928
1929         if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1930             sock->type == SOCK_RAW) {
1931                 skb_reset_mac_header(skb);
1932                 skb->protocol = dev_parse_header_protocol(skb);
1933         }
1934
1935         /* Move network header to the right position for VLAN tagged packets */
1936         if (likely(skb->dev->type == ARPHRD_ETHER) &&
1937             eth_type_vlan(skb->protocol) &&
1938             __vlan_get_protocol(skb, skb->protocol, &depth) != 0) {
1939                 if (pskb_may_pull(skb, depth))
1940                         skb_set_network_header(skb, depth);
1941         }
1942
1943         skb_probe_transport_header(skb);
1944 }
1945
1946 /*
1947  *      Output a raw packet to a device layer. This bypasses all the other
1948  *      protocol layers and you must therefore supply it with a complete frame
1949  */
1950
1951 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1952                                size_t len)
1953 {
1954         struct sock *sk = sock->sk;
1955         DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1956         struct sk_buff *skb = NULL;
1957         struct net_device *dev;
1958         struct sockcm_cookie sockc;
1959         __be16 proto = 0;
1960         int err;
1961         int extra_len = 0;
1962
1963         /*
1964          *      Get and verify the address.
1965          */
1966
1967         if (saddr) {
1968                 if (msg->msg_namelen < sizeof(struct sockaddr))
1969                         return -EINVAL;
1970                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1971                         proto = saddr->spkt_protocol;
1972         } else
1973                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1974
1975         /*
1976          *      Find the device first to size check it
1977          */
1978
1979         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1980 retry:
1981         rcu_read_lock();
1982         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1983         err = -ENODEV;
1984         if (dev == NULL)
1985                 goto out_unlock;
1986
1987         err = -ENETDOWN;
1988         if (!(dev->flags & IFF_UP))
1989                 goto out_unlock;
1990
1991         /*
1992          * You may not queue a frame bigger than the mtu. This is the lowest level
1993          * raw protocol and you must do your own fragmentation at this level.
1994          */
1995
1996         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1997                 if (!netif_supports_nofcs(dev)) {
1998                         err = -EPROTONOSUPPORT;
1999                         goto out_unlock;
2000                 }
2001                 extra_len = 4; /* We're doing our own CRC */
2002         }
2003
2004         err = -EMSGSIZE;
2005         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
2006                 goto out_unlock;
2007
2008         if (!skb) {
2009                 size_t reserved = LL_RESERVED_SPACE(dev);
2010                 int tlen = dev->needed_tailroom;
2011                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
2012
2013                 rcu_read_unlock();
2014                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
2015                 if (skb == NULL)
2016                         return -ENOBUFS;
2017                 /* FIXME: Save some space for broken drivers that write a hard
2018                  * header at transmission time by themselves. PPP is the notable
2019                  * one here. This should really be fixed at the driver level.
2020                  */
2021                 skb_reserve(skb, reserved);
2022                 skb_reset_network_header(skb);
2023
2024                 /* Try to align data part correctly */
2025                 if (hhlen) {
2026                         skb->data -= hhlen;
2027                         skb->tail -= hhlen;
2028                         if (len < hhlen)
2029                                 skb_reset_network_header(skb);
2030                 }
2031                 err = memcpy_from_msg(skb_put(skb, len), msg, len);
2032                 if (err)
2033                         goto out_free;
2034                 goto retry;
2035         }
2036
2037         if (!dev_validate_header(dev, skb->data, len)) {
2038                 err = -EINVAL;
2039                 goto out_unlock;
2040         }
2041         if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
2042             !packet_extra_vlan_len_allowed(dev, skb)) {
2043                 err = -EMSGSIZE;
2044                 goto out_unlock;
2045         }
2046
2047         sockcm_init(&sockc, sk);
2048         if (msg->msg_controllen) {
2049                 err = sock_cmsg_send(sk, msg, &sockc);
2050                 if (unlikely(err))
2051                         goto out_unlock;
2052         }
2053
2054         skb->protocol = proto;
2055         skb->dev = dev;
2056         skb->priority = sk->sk_priority;
2057         skb->mark = sk->sk_mark;
2058         skb->tstamp = sockc.transmit_time;
2059
2060         skb_setup_tx_timestamp(skb, sockc.tsflags);
2061
2062         if (unlikely(extra_len == 4))
2063                 skb->no_fcs = 1;
2064
2065         packet_parse_headers(skb, sock);
2066
2067         dev_queue_xmit(skb);
2068         rcu_read_unlock();
2069         return len;
2070
2071 out_unlock:
2072         rcu_read_unlock();
2073 out_free:
2074         kfree_skb(skb);
2075         return err;
2076 }
2077
2078 static unsigned int run_filter(struct sk_buff *skb,
2079                                const struct sock *sk,
2080                                unsigned int res)
2081 {
2082         struct sk_filter *filter;
2083
2084         rcu_read_lock();
2085         filter = rcu_dereference(sk->sk_filter);
2086         if (filter != NULL)
2087                 res = bpf_prog_run_clear_cb(filter->prog, skb);
2088         rcu_read_unlock();
2089
2090         return res;
2091 }
2092
2093 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2094                            size_t *len)
2095 {
2096         struct virtio_net_hdr vnet_hdr;
2097
2098         if (*len < sizeof(vnet_hdr))
2099                 return -EINVAL;
2100         *len -= sizeof(vnet_hdr);
2101
2102         if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2103                 return -EINVAL;
2104
2105         return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2106 }
2107
2108 /*
2109  * This function makes lazy skb cloning in hope that most of packets
2110  * are discarded by BPF.
2111  *
2112  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2113  * and skb->cb are mangled. It works because (and until) packets
2114  * falling here are owned by current CPU. Output packets are cloned
2115  * by dev_queue_xmit_nit(), input packets are processed by net_bh
2116  * sequentially, so that if we return skb to original state on exit,
2117  * we will not harm anyone.
2118  */
2119
2120 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2121                       struct packet_type *pt, struct net_device *orig_dev)
2122 {
2123         struct sock *sk;
2124         struct sockaddr_ll *sll;
2125         struct packet_sock *po;
2126         u8 *skb_head = skb->data;
2127         int skb_len = skb->len;
2128         unsigned int snaplen, res;
2129         bool is_drop_n_account = false;
2130
2131         if (skb->pkt_type == PACKET_LOOPBACK)
2132                 goto drop;
2133
2134         sk = pt->af_packet_priv;
2135         po = pkt_sk(sk);
2136
2137         if (!net_eq(dev_net(dev), sock_net(sk)))
2138                 goto drop;
2139
2140         skb->dev = dev;
2141
2142         if (dev_has_header(dev)) {
2143                 /* The device has an explicit notion of ll header,
2144                  * exported to higher levels.
2145                  *
2146                  * Otherwise, the device hides details of its frame
2147                  * structure, so that corresponding packet head is
2148                  * never delivered to user.
2149                  */
2150                 if (sk->sk_type != SOCK_DGRAM)
2151                         skb_push(skb, skb->data - skb_mac_header(skb));
2152                 else if (skb->pkt_type == PACKET_OUTGOING) {
2153                         /* Special case: outgoing packets have ll header at head */
2154                         skb_pull(skb, skb_network_offset(skb));
2155                 }
2156         }
2157
2158         snaplen = skb->len;
2159
2160         res = run_filter(skb, sk, snaplen);
2161         if (!res)
2162                 goto drop_n_restore;
2163         if (snaplen > res)
2164                 snaplen = res;
2165
2166         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2167                 goto drop_n_acct;
2168
2169         if (skb_shared(skb)) {
2170                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2171                 if (nskb == NULL)
2172                         goto drop_n_acct;
2173
2174                 if (skb_head != skb->data) {
2175                         skb->data = skb_head;
2176                         skb->len = skb_len;
2177                 }
2178                 consume_skb(skb);
2179                 skb = nskb;
2180         }
2181
2182         sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2183
2184         sll = &PACKET_SKB_CB(skb)->sa.ll;
2185         sll->sll_hatype = dev->type;
2186         sll->sll_pkttype = skb->pkt_type;
2187         if (unlikely(po->origdev))
2188                 sll->sll_ifindex = orig_dev->ifindex;
2189         else
2190                 sll->sll_ifindex = dev->ifindex;
2191
2192         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2193
2194         /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2195          * Use their space for storing the original skb length.
2196          */
2197         PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2198
2199         if (pskb_trim(skb, snaplen))
2200                 goto drop_n_acct;
2201
2202         skb_set_owner_r(skb, sk);
2203         skb->dev = NULL;
2204         skb_dst_drop(skb);
2205
2206         /* drop conntrack reference */
2207         nf_reset_ct(skb);
2208
2209         spin_lock(&sk->sk_receive_queue.lock);
2210         po->stats.stats1.tp_packets++;
2211         sock_skb_set_dropcount(sk, skb);
2212         skb_clear_delivery_time(skb);
2213         __skb_queue_tail(&sk->sk_receive_queue, skb);
2214         spin_unlock(&sk->sk_receive_queue.lock);
2215         sk->sk_data_ready(sk);
2216         return 0;
2217
2218 drop_n_acct:
2219         is_drop_n_account = true;
2220         atomic_inc(&po->tp_drops);
2221         atomic_inc(&sk->sk_drops);
2222
2223 drop_n_restore:
2224         if (skb_head != skb->data && skb_shared(skb)) {
2225                 skb->data = skb_head;
2226                 skb->len = skb_len;
2227         }
2228 drop:
2229         if (!is_drop_n_account)
2230                 consume_skb(skb);
2231         else
2232                 kfree_skb(skb);
2233         return 0;
2234 }
2235
2236 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2237                        struct packet_type *pt, struct net_device *orig_dev)
2238 {
2239         struct sock *sk;
2240         struct packet_sock *po;
2241         struct sockaddr_ll *sll;
2242         union tpacket_uhdr h;
2243         u8 *skb_head = skb->data;
2244         int skb_len = skb->len;
2245         unsigned int snaplen, res;
2246         unsigned long status = TP_STATUS_USER;
2247         unsigned short macoff, hdrlen;
2248         unsigned int netoff;
2249         struct sk_buff *copy_skb = NULL;
2250         struct timespec64 ts;
2251         __u32 ts_status;
2252         bool is_drop_n_account = false;
2253         unsigned int slot_id = 0;
2254         bool do_vnet = false;
2255
2256         /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2257          * We may add members to them until current aligned size without forcing
2258          * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2259          */
2260         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2261         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2262
2263         if (skb->pkt_type == PACKET_LOOPBACK)
2264                 goto drop;
2265
2266         sk = pt->af_packet_priv;
2267         po = pkt_sk(sk);
2268
2269         if (!net_eq(dev_net(dev), sock_net(sk)))
2270                 goto drop;
2271
2272         if (dev_has_header(dev)) {
2273                 if (sk->sk_type != SOCK_DGRAM)
2274                         skb_push(skb, skb->data - skb_mac_header(skb));
2275                 else if (skb->pkt_type == PACKET_OUTGOING) {
2276                         /* Special case: outgoing packets have ll header at head */
2277                         skb_pull(skb, skb_network_offset(skb));
2278                 }
2279         }
2280
2281         snaplen = skb->len;
2282
2283         res = run_filter(skb, sk, snaplen);
2284         if (!res)
2285                 goto drop_n_restore;
2286
2287         /* If we are flooded, just give up */
2288         if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2289                 atomic_inc(&po->tp_drops);
2290                 goto drop_n_restore;
2291         }
2292
2293         if (skb->ip_summed == CHECKSUM_PARTIAL)
2294                 status |= TP_STATUS_CSUMNOTREADY;
2295         else if (skb->pkt_type != PACKET_OUTGOING &&
2296                  (skb->ip_summed == CHECKSUM_COMPLETE ||
2297                   skb_csum_unnecessary(skb)))
2298                 status |= TP_STATUS_CSUM_VALID;
2299
2300         if (snaplen > res)
2301                 snaplen = res;
2302
2303         if (sk->sk_type == SOCK_DGRAM) {
2304                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2305                                   po->tp_reserve;
2306         } else {
2307                 unsigned int maclen = skb_network_offset(skb);
2308                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2309                                        (maclen < 16 ? 16 : maclen)) +
2310                                        po->tp_reserve;
2311                 if (po->has_vnet_hdr) {
2312                         netoff += sizeof(struct virtio_net_hdr);
2313                         do_vnet = true;
2314                 }
2315                 macoff = netoff - maclen;
2316         }
2317         if (netoff > USHRT_MAX) {
2318                 atomic_inc(&po->tp_drops);
2319                 goto drop_n_restore;
2320         }
2321         if (po->tp_version <= TPACKET_V2) {
2322                 if (macoff + snaplen > po->rx_ring.frame_size) {
2323                         if (po->copy_thresh &&
2324                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2325                                 if (skb_shared(skb)) {
2326                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
2327                                 } else {
2328                                         copy_skb = skb_get(skb);
2329                                         skb_head = skb->data;
2330                                 }
2331                                 if (copy_skb) {
2332                                         memset(&PACKET_SKB_CB(copy_skb)->sa.ll, 0,
2333                                                sizeof(PACKET_SKB_CB(copy_skb)->sa.ll));
2334                                         skb_set_owner_r(copy_skb, sk);
2335                                 }
2336                         }
2337                         snaplen = po->rx_ring.frame_size - macoff;
2338                         if ((int)snaplen < 0) {
2339                                 snaplen = 0;
2340                                 do_vnet = false;
2341                         }
2342                 }
2343         } else if (unlikely(macoff + snaplen >
2344                             GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2345                 u32 nval;
2346
2347                 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2348                 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2349                             snaplen, nval, macoff);
2350                 snaplen = nval;
2351                 if (unlikely((int)snaplen < 0)) {
2352                         snaplen = 0;
2353                         macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2354                         do_vnet = false;
2355                 }
2356         }
2357         spin_lock(&sk->sk_receive_queue.lock);
2358         h.raw = packet_current_rx_frame(po, skb,
2359                                         TP_STATUS_KERNEL, (macoff+snaplen));
2360         if (!h.raw)
2361                 goto drop_n_account;
2362
2363         if (po->tp_version <= TPACKET_V2) {
2364                 slot_id = po->rx_ring.head;
2365                 if (test_bit(slot_id, po->rx_ring.rx_owner_map))
2366                         goto drop_n_account;
2367                 __set_bit(slot_id, po->rx_ring.rx_owner_map);
2368         }
2369
2370         if (do_vnet &&
2371             virtio_net_hdr_from_skb(skb, h.raw + macoff -
2372                                     sizeof(struct virtio_net_hdr),
2373                                     vio_le(), true, 0)) {
2374                 if (po->tp_version == TPACKET_V3)
2375                         prb_clear_blk_fill_status(&po->rx_ring);
2376                 goto drop_n_account;
2377         }
2378
2379         if (po->tp_version <= TPACKET_V2) {
2380                 packet_increment_rx_head(po, &po->rx_ring);
2381         /*
2382          * LOSING will be reported till you read the stats,
2383          * because it's COR - Clear On Read.
2384          * Anyways, moving it for V1/V2 only as V3 doesn't need this
2385          * at packet level.
2386          */
2387                 if (atomic_read(&po->tp_drops))
2388                         status |= TP_STATUS_LOSING;
2389         }
2390
2391         po->stats.stats1.tp_packets++;
2392         if (copy_skb) {
2393                 status |= TP_STATUS_COPY;
2394                 skb_clear_delivery_time(copy_skb);
2395                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2396         }
2397         spin_unlock(&sk->sk_receive_queue.lock);
2398
2399         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2400
2401         /* Always timestamp; prefer an existing software timestamp taken
2402          * closer to the time of capture.
2403          */
2404         ts_status = tpacket_get_timestamp(skb, &ts,
2405                                           po->tp_tstamp | SOF_TIMESTAMPING_SOFTWARE);
2406         if (!ts_status)
2407                 ktime_get_real_ts64(&ts);
2408
2409         status |= ts_status;
2410
2411         switch (po->tp_version) {
2412         case TPACKET_V1:
2413                 h.h1->tp_len = skb->len;
2414                 h.h1->tp_snaplen = snaplen;
2415                 h.h1->tp_mac = macoff;
2416                 h.h1->tp_net = netoff;
2417                 h.h1->tp_sec = ts.tv_sec;
2418                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2419                 hdrlen = sizeof(*h.h1);
2420                 break;
2421         case TPACKET_V2:
2422                 h.h2->tp_len = skb->len;
2423                 h.h2->tp_snaplen = snaplen;
2424                 h.h2->tp_mac = macoff;
2425                 h.h2->tp_net = netoff;
2426                 h.h2->tp_sec = ts.tv_sec;
2427                 h.h2->tp_nsec = ts.tv_nsec;
2428                 if (skb_vlan_tag_present(skb)) {
2429                         h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2430                         h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2431                         status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2432                 } else {
2433                         h.h2->tp_vlan_tci = 0;
2434                         h.h2->tp_vlan_tpid = 0;
2435                 }
2436                 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2437                 hdrlen = sizeof(*h.h2);
2438                 break;
2439         case TPACKET_V3:
2440                 /* tp_nxt_offset,vlan are already populated above.
2441                  * So DONT clear those fields here
2442                  */
2443                 h.h3->tp_status |= status;
2444                 h.h3->tp_len = skb->len;
2445                 h.h3->tp_snaplen = snaplen;
2446                 h.h3->tp_mac = macoff;
2447                 h.h3->tp_net = netoff;
2448                 h.h3->tp_sec  = ts.tv_sec;
2449                 h.h3->tp_nsec = ts.tv_nsec;
2450                 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2451                 hdrlen = sizeof(*h.h3);
2452                 break;
2453         default:
2454                 BUG();
2455         }
2456
2457         sll = h.raw + TPACKET_ALIGN(hdrlen);
2458         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2459         sll->sll_family = AF_PACKET;
2460         sll->sll_hatype = dev->type;
2461         sll->sll_protocol = skb->protocol;
2462         sll->sll_pkttype = skb->pkt_type;
2463         if (unlikely(po->origdev))
2464                 sll->sll_ifindex = orig_dev->ifindex;
2465         else
2466                 sll->sll_ifindex = dev->ifindex;
2467
2468         smp_mb();
2469
2470 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2471         if (po->tp_version <= TPACKET_V2) {
2472                 u8 *start, *end;
2473
2474                 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2475                                         macoff + snaplen);
2476
2477                 for (start = h.raw; start < end; start += PAGE_SIZE)
2478                         flush_dcache_page(pgv_to_page(start));
2479         }
2480         smp_wmb();
2481 #endif
2482
2483         if (po->tp_version <= TPACKET_V2) {
2484                 spin_lock(&sk->sk_receive_queue.lock);
2485                 __packet_set_status(po, h.raw, status);
2486                 __clear_bit(slot_id, po->rx_ring.rx_owner_map);
2487                 spin_unlock(&sk->sk_receive_queue.lock);
2488                 sk->sk_data_ready(sk);
2489         } else if (po->tp_version == TPACKET_V3) {
2490                 prb_clear_blk_fill_status(&po->rx_ring);
2491         }
2492
2493 drop_n_restore:
2494         if (skb_head != skb->data && skb_shared(skb)) {
2495                 skb->data = skb_head;
2496                 skb->len = skb_len;
2497         }
2498 drop:
2499         if (!is_drop_n_account)
2500                 consume_skb(skb);
2501         else
2502                 kfree_skb(skb);
2503         return 0;
2504
2505 drop_n_account:
2506         spin_unlock(&sk->sk_receive_queue.lock);
2507         atomic_inc(&po->tp_drops);
2508         is_drop_n_account = true;
2509
2510         sk->sk_data_ready(sk);
2511         kfree_skb(copy_skb);
2512         goto drop_n_restore;
2513 }
2514
2515 static void tpacket_destruct_skb(struct sk_buff *skb)
2516 {
2517         struct packet_sock *po = pkt_sk(skb->sk);
2518
2519         if (likely(po->tx_ring.pg_vec)) {
2520                 void *ph;
2521                 __u32 ts;
2522
2523                 ph = skb_zcopy_get_nouarg(skb);
2524                 packet_dec_pending(&po->tx_ring);
2525
2526                 ts = __packet_set_timestamp(po, ph, skb);
2527                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2528
2529                 if (!packet_read_pending(&po->tx_ring))
2530                         complete(&po->skb_completion);
2531         }
2532
2533         sock_wfree(skb);
2534 }
2535
2536 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2537 {
2538         if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2539             (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2540              __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2541               __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2542                 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2543                          __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2544                         __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2545
2546         if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2547                 return -EINVAL;
2548
2549         return 0;
2550 }
2551
2552 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2553                                  struct virtio_net_hdr *vnet_hdr)
2554 {
2555         if (*len < sizeof(*vnet_hdr))
2556                 return -EINVAL;
2557         *len -= sizeof(*vnet_hdr);
2558
2559         if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2560                 return -EFAULT;
2561
2562         return __packet_snd_vnet_parse(vnet_hdr, *len);
2563 }
2564
2565 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2566                 void *frame, struct net_device *dev, void *data, int tp_len,
2567                 __be16 proto, unsigned char *addr, int hlen, int copylen,
2568                 const struct sockcm_cookie *sockc)
2569 {
2570         union tpacket_uhdr ph;
2571         int to_write, offset, len, nr_frags, len_max;
2572         struct socket *sock = po->sk.sk_socket;
2573         struct page *page;
2574         int err;
2575
2576         ph.raw = frame;
2577
2578         skb->protocol = proto;
2579         skb->dev = dev;
2580         skb->priority = po->sk.sk_priority;
2581         skb->mark = po->sk.sk_mark;
2582         skb->tstamp = sockc->transmit_time;
2583         skb_setup_tx_timestamp(skb, sockc->tsflags);
2584         skb_zcopy_set_nouarg(skb, ph.raw);
2585
2586         skb_reserve(skb, hlen);
2587         skb_reset_network_header(skb);
2588
2589         to_write = tp_len;
2590
2591         if (sock->type == SOCK_DGRAM) {
2592                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2593                                 NULL, tp_len);
2594                 if (unlikely(err < 0))
2595                         return -EINVAL;
2596         } else if (copylen) {
2597                 int hdrlen = min_t(int, copylen, tp_len);
2598
2599                 skb_push(skb, dev->hard_header_len);
2600                 skb_put(skb, copylen - dev->hard_header_len);
2601                 err = skb_store_bits(skb, 0, data, hdrlen);
2602                 if (unlikely(err))
2603                         return err;
2604                 if (!dev_validate_header(dev, skb->data, hdrlen))
2605                         return -EINVAL;
2606
2607                 data += hdrlen;
2608                 to_write -= hdrlen;
2609         }
2610
2611         offset = offset_in_page(data);
2612         len_max = PAGE_SIZE - offset;
2613         len = ((to_write > len_max) ? len_max : to_write);
2614
2615         skb->data_len = to_write;
2616         skb->len += to_write;
2617         skb->truesize += to_write;
2618         refcount_add(to_write, &po->sk.sk_wmem_alloc);
2619
2620         while (likely(to_write)) {
2621                 nr_frags = skb_shinfo(skb)->nr_frags;
2622
2623                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2624                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2625                                MAX_SKB_FRAGS);
2626                         return -EFAULT;
2627                 }
2628
2629                 page = pgv_to_page(data);
2630                 data += len;
2631                 flush_dcache_page(page);
2632                 get_page(page);
2633                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2634                 to_write -= len;
2635                 offset = 0;
2636                 len_max = PAGE_SIZE;
2637                 len = ((to_write > len_max) ? len_max : to_write);
2638         }
2639
2640         packet_parse_headers(skb, sock);
2641
2642         return tp_len;
2643 }
2644
2645 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2646                                 int size_max, void **data)
2647 {
2648         union tpacket_uhdr ph;
2649         int tp_len, off;
2650
2651         ph.raw = frame;
2652
2653         switch (po->tp_version) {
2654         case TPACKET_V3:
2655                 if (ph.h3->tp_next_offset != 0) {
2656                         pr_warn_once("variable sized slot not supported");
2657                         return -EINVAL;
2658                 }
2659                 tp_len = ph.h3->tp_len;
2660                 break;
2661         case TPACKET_V2:
2662                 tp_len = ph.h2->tp_len;
2663                 break;
2664         default:
2665                 tp_len = ph.h1->tp_len;
2666                 break;
2667         }
2668         if (unlikely(tp_len > size_max)) {
2669                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2670                 return -EMSGSIZE;
2671         }
2672
2673         if (unlikely(po->tp_tx_has_off)) {
2674                 int off_min, off_max;
2675
2676                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2677                 off_max = po->tx_ring.frame_size - tp_len;
2678                 if (po->sk.sk_type == SOCK_DGRAM) {
2679                         switch (po->tp_version) {
2680                         case TPACKET_V3:
2681                                 off = ph.h3->tp_net;
2682                                 break;
2683                         case TPACKET_V2:
2684                                 off = ph.h2->tp_net;
2685                                 break;
2686                         default:
2687                                 off = ph.h1->tp_net;
2688                                 break;
2689                         }
2690                 } else {
2691                         switch (po->tp_version) {
2692                         case TPACKET_V3:
2693                                 off = ph.h3->tp_mac;
2694                                 break;
2695                         case TPACKET_V2:
2696                                 off = ph.h2->tp_mac;
2697                                 break;
2698                         default:
2699                                 off = ph.h1->tp_mac;
2700                                 break;
2701                         }
2702                 }
2703                 if (unlikely((off < off_min) || (off_max < off)))
2704                         return -EINVAL;
2705         } else {
2706                 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2707         }
2708
2709         *data = frame + off;
2710         return tp_len;
2711 }
2712
2713 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2714 {
2715         struct sk_buff *skb = NULL;
2716         struct net_device *dev;
2717         struct virtio_net_hdr *vnet_hdr = NULL;
2718         struct sockcm_cookie sockc;
2719         __be16 proto;
2720         int err, reserve = 0;
2721         void *ph;
2722         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2723         bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2724         unsigned char *addr = NULL;
2725         int tp_len, size_max;
2726         void *data;
2727         int len_sum = 0;
2728         int status = TP_STATUS_AVAILABLE;
2729         int hlen, tlen, copylen = 0;
2730         long timeo = 0;
2731
2732         mutex_lock(&po->pg_vec_lock);
2733
2734         /* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2735          * we need to confirm it under protection of pg_vec_lock.
2736          */
2737         if (unlikely(!po->tx_ring.pg_vec)) {
2738                 err = -EBUSY;
2739                 goto out;
2740         }
2741         if (likely(saddr == NULL)) {
2742                 dev     = packet_cached_dev_get(po);
2743                 proto   = READ_ONCE(po->num);
2744         } else {
2745                 err = -EINVAL;
2746                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2747                         goto out;
2748                 if (msg->msg_namelen < (saddr->sll_halen
2749                                         + offsetof(struct sockaddr_ll,
2750                                                 sll_addr)))
2751                         goto out;
2752                 proto   = saddr->sll_protocol;
2753                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2754                 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2755                         if (dev && msg->msg_namelen < dev->addr_len +
2756                                    offsetof(struct sockaddr_ll, sll_addr))
2757                                 goto out_put;
2758                         addr = saddr->sll_addr;
2759                 }
2760         }
2761
2762         err = -ENXIO;
2763         if (unlikely(dev == NULL))
2764                 goto out;
2765         err = -ENETDOWN;
2766         if (unlikely(!(dev->flags & IFF_UP)))
2767                 goto out_put;
2768
2769         sockcm_init(&sockc, &po->sk);
2770         if (msg->msg_controllen) {
2771                 err = sock_cmsg_send(&po->sk, msg, &sockc);
2772                 if (unlikely(err))
2773                         goto out_put;
2774         }
2775
2776         if (po->sk.sk_socket->type == SOCK_RAW)
2777                 reserve = dev->hard_header_len;
2778         size_max = po->tx_ring.frame_size
2779                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2780
2781         if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2782                 size_max = dev->mtu + reserve + VLAN_HLEN;
2783
2784         reinit_completion(&po->skb_completion);
2785
2786         do {
2787                 ph = packet_current_frame(po, &po->tx_ring,
2788                                           TP_STATUS_SEND_REQUEST);
2789                 if (unlikely(ph == NULL)) {
2790                         if (need_wait && skb) {
2791                                 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2792                                 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2793                                 if (timeo <= 0) {
2794                                         err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2795                                         goto out_put;
2796                                 }
2797                         }
2798                         /* check for additional frames */
2799                         continue;
2800                 }
2801
2802                 skb = NULL;
2803                 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2804                 if (tp_len < 0)
2805                         goto tpacket_error;
2806
2807                 status = TP_STATUS_SEND_REQUEST;
2808                 hlen = LL_RESERVED_SPACE(dev);
2809                 tlen = dev->needed_tailroom;
2810                 if (po->has_vnet_hdr) {
2811                         vnet_hdr = data;
2812                         data += sizeof(*vnet_hdr);
2813                         tp_len -= sizeof(*vnet_hdr);
2814                         if (tp_len < 0 ||
2815                             __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2816                                 tp_len = -EINVAL;
2817                                 goto tpacket_error;
2818                         }
2819                         copylen = __virtio16_to_cpu(vio_le(),
2820                                                     vnet_hdr->hdr_len);
2821                 }
2822                 copylen = max_t(int, copylen, dev->hard_header_len);
2823                 skb = sock_alloc_send_skb(&po->sk,
2824                                 hlen + tlen + sizeof(struct sockaddr_ll) +
2825                                 (copylen - dev->hard_header_len),
2826                                 !need_wait, &err);
2827
2828                 if (unlikely(skb == NULL)) {
2829                         /* we assume the socket was initially writeable ... */
2830                         if (likely(len_sum > 0))
2831                                 err = len_sum;
2832                         goto out_status;
2833                 }
2834                 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2835                                           addr, hlen, copylen, &sockc);
2836                 if (likely(tp_len >= 0) &&
2837                     tp_len > dev->mtu + reserve &&
2838                     !po->has_vnet_hdr &&
2839                     !packet_extra_vlan_len_allowed(dev, skb))
2840                         tp_len = -EMSGSIZE;
2841
2842                 if (unlikely(tp_len < 0)) {
2843 tpacket_error:
2844                         if (po->tp_loss) {
2845                                 __packet_set_status(po, ph,
2846                                                 TP_STATUS_AVAILABLE);
2847                                 packet_increment_head(&po->tx_ring);
2848                                 kfree_skb(skb);
2849                                 continue;
2850                         } else {
2851                                 status = TP_STATUS_WRONG_FORMAT;
2852                                 err = tp_len;
2853                                 goto out_status;
2854                         }
2855                 }
2856
2857                 if (po->has_vnet_hdr) {
2858                         if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2859                                 tp_len = -EINVAL;
2860                                 goto tpacket_error;
2861                         }
2862                         virtio_net_hdr_set_proto(skb, vnet_hdr);
2863                 }
2864
2865                 skb->destructor = tpacket_destruct_skb;
2866                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2867                 packet_inc_pending(&po->tx_ring);
2868
2869                 status = TP_STATUS_SEND_REQUEST;
2870                 err = po->xmit(skb);
2871                 if (unlikely(err != 0)) {
2872                         if (err > 0)
2873                                 err = net_xmit_errno(err);
2874                         if (err && __packet_get_status(po, ph) ==
2875                                    TP_STATUS_AVAILABLE) {
2876                                 /* skb was destructed already */
2877                                 skb = NULL;
2878                                 goto out_status;
2879                         }
2880                         /*
2881                          * skb was dropped but not destructed yet;
2882                          * let's treat it like congestion or err < 0
2883                          */
2884                         err = 0;
2885                 }
2886                 packet_increment_head(&po->tx_ring);
2887                 len_sum += tp_len;
2888         } while (likely((ph != NULL) ||
2889                 /* Note: packet_read_pending() might be slow if we have
2890                  * to call it as it's per_cpu variable, but in fast-path
2891                  * we already short-circuit the loop with the first
2892                  * condition, and luckily don't have to go that path
2893                  * anyway.
2894                  */
2895                  (need_wait && packet_read_pending(&po->tx_ring))));
2896
2897         err = len_sum;
2898         goto out_put;
2899
2900 out_status:
2901         __packet_set_status(po, ph, status);
2902         kfree_skb(skb);
2903 out_put:
2904         dev_put(dev);
2905 out:
2906         mutex_unlock(&po->pg_vec_lock);
2907         return err;
2908 }
2909
2910 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2911                                         size_t reserve, size_t len,
2912                                         size_t linear, int noblock,
2913                                         int *err)
2914 {
2915         struct sk_buff *skb;
2916
2917         /* Under a page?  Don't bother with paged skb. */
2918         if (prepad + len < PAGE_SIZE || !linear)
2919                 linear = len;
2920
2921         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2922                                    err, 0);
2923         if (!skb)
2924                 return NULL;
2925
2926         skb_reserve(skb, reserve);
2927         skb_put(skb, linear);
2928         skb->data_len = len - linear;
2929         skb->len += len - linear;
2930
2931         return skb;
2932 }
2933
2934 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2935 {
2936         struct sock *sk = sock->sk;
2937         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2938         struct sk_buff *skb;
2939         struct net_device *dev;
2940         __be16 proto;
2941         unsigned char *addr = NULL;
2942         int err, reserve = 0;
2943         struct sockcm_cookie sockc;
2944         struct virtio_net_hdr vnet_hdr = { 0 };
2945         int offset = 0;
2946         struct packet_sock *po = pkt_sk(sk);
2947         bool has_vnet_hdr = false;
2948         int hlen, tlen, linear;
2949         int extra_len = 0;
2950
2951         /*
2952          *      Get and verify the address.
2953          */
2954
2955         if (likely(saddr == NULL)) {
2956                 dev     = packet_cached_dev_get(po);
2957                 proto   = READ_ONCE(po->num);
2958         } else {
2959                 err = -EINVAL;
2960                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2961                         goto out;
2962                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2963                         goto out;
2964                 proto   = saddr->sll_protocol;
2965                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2966                 if (sock->type == SOCK_DGRAM) {
2967                         if (dev && msg->msg_namelen < dev->addr_len +
2968                                    offsetof(struct sockaddr_ll, sll_addr))
2969                                 goto out_unlock;
2970                         addr = saddr->sll_addr;
2971                 }
2972         }
2973
2974         err = -ENXIO;
2975         if (unlikely(dev == NULL))
2976                 goto out_unlock;
2977         err = -ENETDOWN;
2978         if (unlikely(!(dev->flags & IFF_UP)))
2979                 goto out_unlock;
2980
2981         sockcm_init(&sockc, sk);
2982         sockc.mark = sk->sk_mark;
2983         if (msg->msg_controllen) {
2984                 err = sock_cmsg_send(sk, msg, &sockc);
2985                 if (unlikely(err))
2986                         goto out_unlock;
2987         }
2988
2989         if (sock->type == SOCK_RAW)
2990                 reserve = dev->hard_header_len;
2991         if (po->has_vnet_hdr) {
2992                 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2993                 if (err)
2994                         goto out_unlock;
2995                 has_vnet_hdr = true;
2996         }
2997
2998         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2999                 if (!netif_supports_nofcs(dev)) {
3000                         err = -EPROTONOSUPPORT;
3001                         goto out_unlock;
3002                 }
3003                 extra_len = 4; /* We're doing our own CRC */
3004         }
3005
3006         err = -EMSGSIZE;
3007         if (!vnet_hdr.gso_type &&
3008             (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
3009                 goto out_unlock;
3010
3011         err = -ENOBUFS;
3012         hlen = LL_RESERVED_SPACE(dev);
3013         tlen = dev->needed_tailroom;
3014         linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
3015         linear = max(linear, min_t(int, len, dev->hard_header_len));
3016         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
3017                                msg->msg_flags & MSG_DONTWAIT, &err);
3018         if (skb == NULL)
3019                 goto out_unlock;
3020
3021         skb_reset_network_header(skb);
3022
3023         err = -EINVAL;
3024         if (sock->type == SOCK_DGRAM) {
3025                 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
3026                 if (unlikely(offset < 0))
3027                         goto out_free;
3028         } else if (reserve) {
3029                 skb_reserve(skb, -reserve);
3030                 if (len < reserve + sizeof(struct ipv6hdr) &&
3031                     dev->min_header_len != dev->hard_header_len)
3032                         skb_reset_network_header(skb);
3033         }
3034
3035         /* Returns -EFAULT on error */
3036         err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
3037         if (err)
3038                 goto out_free;
3039
3040         if ((sock->type == SOCK_RAW &&
3041              !dev_validate_header(dev, skb->data, len)) || !skb->len) {
3042                 err = -EINVAL;
3043                 goto out_free;
3044         }
3045
3046         skb_setup_tx_timestamp(skb, sockc.tsflags);
3047
3048         if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
3049             !packet_extra_vlan_len_allowed(dev, skb)) {
3050                 err = -EMSGSIZE;
3051                 goto out_free;
3052         }
3053
3054         skb->protocol = proto;
3055         skb->dev = dev;
3056         skb->priority = sk->sk_priority;
3057         skb->mark = sockc.mark;
3058         skb->tstamp = sockc.transmit_time;
3059
3060         if (unlikely(extra_len == 4))
3061                 skb->no_fcs = 1;
3062
3063         packet_parse_headers(skb, sock);
3064
3065         if (has_vnet_hdr) {
3066                 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
3067                 if (err)
3068                         goto out_free;
3069                 len += sizeof(vnet_hdr);
3070                 virtio_net_hdr_set_proto(skb, &vnet_hdr);
3071         }
3072
3073         err = po->xmit(skb);
3074         if (unlikely(err != 0)) {
3075                 if (err > 0)
3076                         err = net_xmit_errno(err);
3077                 if (err)
3078                         goto out_unlock;
3079         }
3080
3081         dev_put(dev);
3082
3083         return len;
3084
3085 out_free:
3086         kfree_skb(skb);
3087 out_unlock:
3088         dev_put(dev);
3089 out:
3090         return err;
3091 }
3092
3093 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
3094 {
3095         struct sock *sk = sock->sk;
3096         struct packet_sock *po = pkt_sk(sk);
3097
3098         /* Reading tx_ring.pg_vec without holding pg_vec_lock is racy.
3099          * tpacket_snd() will redo the check safely.
3100          */
3101         if (data_race(po->tx_ring.pg_vec))
3102                 return tpacket_snd(po, msg);
3103
3104         return packet_snd(sock, msg, len);
3105 }
3106
3107 /*
3108  *      Close a PACKET socket. This is fairly simple. We immediately go
3109  *      to 'closed' state and remove our protocol entry in the device list.
3110  */
3111
3112 static int packet_release(struct socket *sock)
3113 {
3114         struct sock *sk = sock->sk;
3115         struct packet_sock *po;
3116         struct packet_fanout *f;
3117         struct net *net;
3118         union tpacket_req_u req_u;
3119
3120         if (!sk)
3121                 return 0;
3122
3123         net = sock_net(sk);
3124         po = pkt_sk(sk);
3125
3126         mutex_lock(&net->packet.sklist_lock);
3127         sk_del_node_init_rcu(sk);
3128         mutex_unlock(&net->packet.sklist_lock);
3129
3130         sock_prot_inuse_add(net, sk->sk_prot, -1);
3131
3132         spin_lock(&po->bind_lock);
3133         unregister_prot_hook(sk, false);
3134         packet_cached_dev_reset(po);
3135
3136         if (po->prot_hook.dev) {
3137                 netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3138                 po->prot_hook.dev = NULL;
3139         }
3140         spin_unlock(&po->bind_lock);
3141
3142         packet_flush_mclist(sk);
3143
3144         lock_sock(sk);
3145         if (po->rx_ring.pg_vec) {
3146                 memset(&req_u, 0, sizeof(req_u));
3147                 packet_set_ring(sk, &req_u, 1, 0);
3148         }
3149
3150         if (po->tx_ring.pg_vec) {
3151                 memset(&req_u, 0, sizeof(req_u));
3152                 packet_set_ring(sk, &req_u, 1, 1);
3153         }
3154         release_sock(sk);
3155
3156         f = fanout_release(sk);
3157
3158         synchronize_net();
3159
3160         kfree(po->rollover);
3161         if (f) {
3162                 fanout_release_data(f);
3163                 kvfree(f);
3164         }
3165         /*
3166          *      Now the socket is dead. No more input will appear.
3167          */
3168         sock_orphan(sk);
3169         sock->sk = NULL;
3170
3171         /* Purge queues */
3172
3173         skb_queue_purge(&sk->sk_receive_queue);
3174         packet_free_pending(po);
3175         sk_refcnt_debug_release(sk);
3176
3177         sock_put(sk);
3178         return 0;
3179 }
3180
3181 /*
3182  *      Attach a packet hook.
3183  */
3184
3185 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3186                           __be16 proto)
3187 {
3188         struct packet_sock *po = pkt_sk(sk);
3189         struct net_device *dev = NULL;
3190         bool unlisted = false;
3191         bool need_rehook;
3192         int ret = 0;
3193
3194         lock_sock(sk);
3195         spin_lock(&po->bind_lock);
3196         rcu_read_lock();
3197
3198         if (po->fanout) {
3199                 ret = -EINVAL;
3200                 goto out_unlock;
3201         }
3202
3203         if (name) {
3204                 dev = dev_get_by_name_rcu(sock_net(sk), name);
3205                 if (!dev) {
3206                         ret = -ENODEV;
3207                         goto out_unlock;
3208                 }
3209         } else if (ifindex) {
3210                 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3211                 if (!dev) {
3212                         ret = -ENODEV;
3213                         goto out_unlock;
3214                 }
3215         }
3216
3217         need_rehook = po->prot_hook.type != proto || po->prot_hook.dev != dev;
3218
3219         if (need_rehook) {
3220                 dev_hold(dev);
3221                 if (po->running) {
3222                         rcu_read_unlock();
3223                         /* prevents packet_notifier() from calling
3224                          * register_prot_hook()
3225                          */
3226                         WRITE_ONCE(po->num, 0);
3227                         __unregister_prot_hook(sk, true);
3228                         rcu_read_lock();
3229                         if (dev)
3230                                 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3231                                                                  dev->ifindex);
3232                 }
3233
3234                 BUG_ON(po->running);
3235                 WRITE_ONCE(po->num, proto);
3236                 po->prot_hook.type = proto;
3237
3238                 netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3239
3240                 if (unlikely(unlisted)) {
3241                         po->prot_hook.dev = NULL;
3242                         WRITE_ONCE(po->ifindex, -1);
3243                         packet_cached_dev_reset(po);
3244                 } else {
3245                         netdev_hold(dev, &po->prot_hook.dev_tracker,
3246                                     GFP_ATOMIC);
3247                         po->prot_hook.dev = dev;
3248                         WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0);
3249                         packet_cached_dev_assign(po, dev);
3250                 }
3251                 dev_put(dev);
3252         }
3253
3254         if (proto == 0 || !need_rehook)
3255                 goto out_unlock;
3256
3257         if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3258                 register_prot_hook(sk);
3259         } else {
3260                 sk->sk_err = ENETDOWN;
3261                 if (!sock_flag(sk, SOCK_DEAD))
3262                         sk_error_report(sk);
3263         }
3264
3265 out_unlock:
3266         rcu_read_unlock();
3267         spin_unlock(&po->bind_lock);
3268         release_sock(sk);
3269         return ret;
3270 }
3271
3272 /*
3273  *      Bind a packet socket to a device
3274  */
3275
3276 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3277                             int addr_len)
3278 {
3279         struct sock *sk = sock->sk;
3280         char name[sizeof(uaddr->sa_data) + 1];
3281
3282         /*
3283          *      Check legality
3284          */
3285
3286         if (addr_len != sizeof(struct sockaddr))
3287                 return -EINVAL;
3288         /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3289          * zero-terminated.
3290          */
3291         memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3292         name[sizeof(uaddr->sa_data)] = 0;
3293
3294         return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3295 }
3296
3297 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3298 {
3299         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3300         struct sock *sk = sock->sk;
3301
3302         /*
3303          *      Check legality
3304          */
3305
3306         if (addr_len < sizeof(struct sockaddr_ll))
3307                 return -EINVAL;
3308         if (sll->sll_family != AF_PACKET)
3309                 return -EINVAL;
3310
3311         return packet_do_bind(sk, NULL, sll->sll_ifindex,
3312                               sll->sll_protocol ? : pkt_sk(sk)->num);
3313 }
3314
3315 static struct proto packet_proto = {
3316         .name     = "PACKET",
3317         .owner    = THIS_MODULE,
3318         .obj_size = sizeof(struct packet_sock),
3319 };
3320
3321 /*
3322  *      Create a packet of type SOCK_PACKET.
3323  */
3324
3325 static int packet_create(struct net *net, struct socket *sock, int protocol,
3326                          int kern)
3327 {
3328         struct sock *sk;
3329         struct packet_sock *po;
3330         __be16 proto = (__force __be16)protocol; /* weird, but documented */
3331         int err;
3332
3333         if (!ns_capable(net->user_ns, CAP_NET_RAW))
3334                 return -EPERM;
3335         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3336             sock->type != SOCK_PACKET)
3337                 return -ESOCKTNOSUPPORT;
3338
3339         sock->state = SS_UNCONNECTED;
3340
3341         err = -ENOBUFS;
3342         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3343         if (sk == NULL)
3344                 goto out;
3345
3346         sock->ops = &packet_ops;
3347         if (sock->type == SOCK_PACKET)
3348                 sock->ops = &packet_ops_spkt;
3349
3350         sock_init_data(sock, sk);
3351
3352         po = pkt_sk(sk);
3353         init_completion(&po->skb_completion);
3354         sk->sk_family = PF_PACKET;
3355         po->num = proto;
3356         po->xmit = dev_queue_xmit;
3357
3358         err = packet_alloc_pending(po);
3359         if (err)
3360                 goto out2;
3361
3362         packet_cached_dev_reset(po);
3363
3364         sk->sk_destruct = packet_sock_destruct;
3365         sk_refcnt_debug_inc(sk);
3366
3367         /*
3368          *      Attach a protocol block
3369          */
3370
3371         spin_lock_init(&po->bind_lock);
3372         mutex_init(&po->pg_vec_lock);
3373         po->rollover = NULL;
3374         po->prot_hook.func = packet_rcv;
3375
3376         if (sock->type == SOCK_PACKET)
3377                 po->prot_hook.func = packet_rcv_spkt;
3378
3379         po->prot_hook.af_packet_priv = sk;
3380         po->prot_hook.af_packet_net = sock_net(sk);
3381
3382         if (proto) {
3383                 po->prot_hook.type = proto;
3384                 __register_prot_hook(sk);
3385         }
3386
3387         mutex_lock(&net->packet.sklist_lock);
3388         sk_add_node_tail_rcu(sk, &net->packet.sklist);
3389         mutex_unlock(&net->packet.sklist_lock);
3390
3391         sock_prot_inuse_add(net, &packet_proto, 1);
3392
3393         return 0;
3394 out2:
3395         sk_free(sk);
3396 out:
3397         return err;
3398 }
3399
3400 /*
3401  *      Pull a packet from our receive queue and hand it to the user.
3402  *      If necessary we block.
3403  */
3404
3405 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3406                           int flags)
3407 {
3408         struct sock *sk = sock->sk;
3409         struct sk_buff *skb;
3410         int copied, err;
3411         int vnet_hdr_len = 0;
3412         unsigned int origlen = 0;
3413
3414         err = -EINVAL;
3415         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3416                 goto out;
3417
3418 #if 0
3419         /* What error should we return now? EUNATTACH? */
3420         if (pkt_sk(sk)->ifindex < 0)
3421                 return -ENODEV;
3422 #endif
3423
3424         if (flags & MSG_ERRQUEUE) {
3425                 err = sock_recv_errqueue(sk, msg, len,
3426                                          SOL_PACKET, PACKET_TX_TIMESTAMP);
3427                 goto out;
3428         }
3429
3430         /*
3431          *      Call the generic datagram receiver. This handles all sorts
3432          *      of horrible races and re-entrancy so we can forget about it
3433          *      in the protocol layers.
3434          *
3435          *      Now it will return ENETDOWN, if device have just gone down,
3436          *      but then it will block.
3437          */
3438
3439         skb = skb_recv_datagram(sk, flags, &err);
3440
3441         /*
3442          *      An error occurred so return it. Because skb_recv_datagram()
3443          *      handles the blocking we don't see and worry about blocking
3444          *      retries.
3445          */
3446
3447         if (skb == NULL)
3448                 goto out;
3449
3450         packet_rcv_try_clear_pressure(pkt_sk(sk));
3451
3452         if (pkt_sk(sk)->has_vnet_hdr) {
3453                 err = packet_rcv_vnet(msg, skb, &len);
3454                 if (err)
3455                         goto out_free;
3456                 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3457         }
3458
3459         /* You lose any data beyond the buffer you gave. If it worries
3460          * a user program they can ask the device for its MTU
3461          * anyway.
3462          */
3463         copied = skb->len;
3464         if (copied > len) {
3465                 copied = len;
3466                 msg->msg_flags |= MSG_TRUNC;
3467         }
3468
3469         err = skb_copy_datagram_msg(skb, 0, msg, copied);
3470         if (err)
3471                 goto out_free;
3472
3473         if (sock->type != SOCK_PACKET) {
3474                 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3475
3476                 /* Original length was stored in sockaddr_ll fields */
3477                 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3478                 sll->sll_family = AF_PACKET;
3479                 sll->sll_protocol = skb->protocol;
3480         }
3481
3482         sock_recv_cmsgs(msg, sk, skb);
3483
3484         if (msg->msg_name) {
3485                 const size_t max_len = min(sizeof(skb->cb),
3486                                            sizeof(struct sockaddr_storage));
3487                 int copy_len;
3488
3489                 /* If the address length field is there to be filled
3490                  * in, we fill it in now.
3491                  */
3492                 if (sock->type == SOCK_PACKET) {
3493                         __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3494                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
3495                         copy_len = msg->msg_namelen;
3496                 } else {
3497                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3498
3499                         msg->msg_namelen = sll->sll_halen +
3500                                 offsetof(struct sockaddr_ll, sll_addr);
3501                         copy_len = msg->msg_namelen;
3502                         if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3503                                 memset(msg->msg_name +
3504                                        offsetof(struct sockaddr_ll, sll_addr),
3505                                        0, sizeof(sll->sll_addr));
3506                                 msg->msg_namelen = sizeof(struct sockaddr_ll);
3507                         }
3508                 }
3509                 if (WARN_ON_ONCE(copy_len > max_len)) {
3510                         copy_len = max_len;
3511                         msg->msg_namelen = copy_len;
3512                 }
3513                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3514         }
3515
3516         if (pkt_sk(sk)->auxdata) {
3517                 struct tpacket_auxdata aux;
3518
3519                 aux.tp_status = TP_STATUS_USER;
3520                 if (skb->ip_summed == CHECKSUM_PARTIAL)
3521                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3522                 else if (skb->pkt_type != PACKET_OUTGOING &&
3523                          (skb->ip_summed == CHECKSUM_COMPLETE ||
3524                           skb_csum_unnecessary(skb)))
3525                         aux.tp_status |= TP_STATUS_CSUM_VALID;
3526
3527                 aux.tp_len = origlen;
3528                 aux.tp_snaplen = skb->len;
3529                 aux.tp_mac = 0;
3530                 aux.tp_net = skb_network_offset(skb);
3531                 if (skb_vlan_tag_present(skb)) {
3532                         aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3533                         aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3534                         aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3535                 } else {
3536                         aux.tp_vlan_tci = 0;
3537                         aux.tp_vlan_tpid = 0;
3538                 }
3539                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3540         }
3541
3542         /*
3543          *      Free or return the buffer as appropriate. Again this
3544          *      hides all the races and re-entrancy issues from us.
3545          */
3546         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3547
3548 out_free:
3549         skb_free_datagram(sk, skb);
3550 out:
3551         return err;
3552 }
3553
3554 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3555                                int peer)
3556 {
3557         struct net_device *dev;
3558         struct sock *sk = sock->sk;
3559
3560         if (peer)
3561                 return -EOPNOTSUPP;
3562
3563         uaddr->sa_family = AF_PACKET;
3564         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3565         rcu_read_lock();
3566         dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex));
3567         if (dev)
3568                 strscpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3569         rcu_read_unlock();
3570
3571         return sizeof(*uaddr);
3572 }
3573
3574 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3575                           int peer)
3576 {
3577         struct net_device *dev;
3578         struct sock *sk = sock->sk;
3579         struct packet_sock *po = pkt_sk(sk);
3580         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3581         int ifindex;
3582
3583         if (peer)
3584                 return -EOPNOTSUPP;
3585
3586         ifindex = READ_ONCE(po->ifindex);
3587         sll->sll_family = AF_PACKET;
3588         sll->sll_ifindex = ifindex;
3589         sll->sll_protocol = READ_ONCE(po->num);
3590         sll->sll_pkttype = 0;
3591         rcu_read_lock();
3592         dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3593         if (dev) {
3594                 sll->sll_hatype = dev->type;
3595                 sll->sll_halen = dev->addr_len;
3596                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3597         } else {
3598                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
3599                 sll->sll_halen = 0;
3600         }
3601         rcu_read_unlock();
3602
3603         return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3604 }
3605
3606 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3607                          int what)
3608 {
3609         switch (i->type) {
3610         case PACKET_MR_MULTICAST:
3611                 if (i->alen != dev->addr_len)
3612                         return -EINVAL;
3613                 if (what > 0)
3614                         return dev_mc_add(dev, i->addr);
3615                 else
3616                         return dev_mc_del(dev, i->addr);
3617                 break;
3618         case PACKET_MR_PROMISC:
3619                 return dev_set_promiscuity(dev, what);
3620         case PACKET_MR_ALLMULTI:
3621                 return dev_set_allmulti(dev, what);
3622         case PACKET_MR_UNICAST:
3623                 if (i->alen != dev->addr_len)
3624                         return -EINVAL;
3625                 if (what > 0)
3626                         return dev_uc_add(dev, i->addr);
3627                 else
3628                         return dev_uc_del(dev, i->addr);
3629                 break;
3630         default:
3631                 break;
3632         }
3633         return 0;
3634 }
3635
3636 static void packet_dev_mclist_delete(struct net_device *dev,
3637                                      struct packet_mclist **mlp)
3638 {
3639         struct packet_mclist *ml;
3640
3641         while ((ml = *mlp) != NULL) {
3642                 if (ml->ifindex == dev->ifindex) {
3643                         packet_dev_mc(dev, ml, -1);
3644                         *mlp = ml->next;
3645                         kfree(ml);
3646                 } else
3647                         mlp = &ml->next;
3648         }
3649 }
3650
3651 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3652 {
3653         struct packet_sock *po = pkt_sk(sk);
3654         struct packet_mclist *ml, *i;
3655         struct net_device *dev;
3656         int err;
3657
3658         rtnl_lock();
3659
3660         err = -ENODEV;
3661         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3662         if (!dev)
3663                 goto done;
3664
3665         err = -EINVAL;
3666         if (mreq->mr_alen > dev->addr_len)
3667                 goto done;
3668
3669         err = -ENOBUFS;
3670         i = kmalloc(sizeof(*i), GFP_KERNEL);
3671         if (i == NULL)
3672                 goto done;
3673
3674         err = 0;
3675         for (ml = po->mclist; ml; ml = ml->next) {
3676                 if (ml->ifindex == mreq->mr_ifindex &&
3677                     ml->type == mreq->mr_type &&
3678                     ml->alen == mreq->mr_alen &&
3679                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3680                         ml->count++;
3681                         /* Free the new element ... */
3682                         kfree(i);
3683                         goto done;
3684                 }
3685         }
3686
3687         i->type = mreq->mr_type;
3688         i->ifindex = mreq->mr_ifindex;
3689         i->alen = mreq->mr_alen;
3690         memcpy(i->addr, mreq->mr_address, i->alen);
3691         memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3692         i->count = 1;
3693         i->next = po->mclist;
3694         po->mclist = i;
3695         err = packet_dev_mc(dev, i, 1);
3696         if (err) {
3697                 po->mclist = i->next;
3698                 kfree(i);
3699         }
3700
3701 done:
3702         rtnl_unlock();
3703         return err;
3704 }
3705
3706 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3707 {
3708         struct packet_mclist *ml, **mlp;
3709
3710         rtnl_lock();
3711
3712         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3713                 if (ml->ifindex == mreq->mr_ifindex &&
3714                     ml->type == mreq->mr_type &&
3715                     ml->alen == mreq->mr_alen &&
3716                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3717                         if (--ml->count == 0) {
3718                                 struct net_device *dev;
3719                                 *mlp = ml->next;
3720                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3721                                 if (dev)
3722                                         packet_dev_mc(dev, ml, -1);
3723                                 kfree(ml);
3724                         }
3725                         break;
3726                 }
3727         }
3728         rtnl_unlock();
3729         return 0;
3730 }
3731
3732 static void packet_flush_mclist(struct sock *sk)
3733 {
3734         struct packet_sock *po = pkt_sk(sk);
3735         struct packet_mclist *ml;
3736
3737         if (!po->mclist)
3738                 return;
3739
3740         rtnl_lock();
3741         while ((ml = po->mclist) != NULL) {
3742                 struct net_device *dev;
3743
3744                 po->mclist = ml->next;
3745                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3746                 if (dev != NULL)
3747                         packet_dev_mc(dev, ml, -1);
3748                 kfree(ml);
3749         }
3750         rtnl_unlock();
3751 }
3752
3753 static int
3754 packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval,
3755                   unsigned int optlen)
3756 {
3757         struct sock *sk = sock->sk;
3758         struct packet_sock *po = pkt_sk(sk);
3759         int ret;
3760
3761         if (level != SOL_PACKET)
3762                 return -ENOPROTOOPT;
3763
3764         switch (optname) {
3765         case PACKET_ADD_MEMBERSHIP:
3766         case PACKET_DROP_MEMBERSHIP:
3767         {
3768                 struct packet_mreq_max mreq;
3769                 int len = optlen;
3770                 memset(&mreq, 0, sizeof(mreq));
3771                 if (len < sizeof(struct packet_mreq))
3772                         return -EINVAL;
3773                 if (len > sizeof(mreq))
3774                         len = sizeof(mreq);
3775                 if (copy_from_sockptr(&mreq, optval, len))
3776                         return -EFAULT;
3777                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3778                         return -EINVAL;
3779                 if (optname == PACKET_ADD_MEMBERSHIP)
3780                         ret = packet_mc_add(sk, &mreq);
3781                 else
3782                         ret = packet_mc_drop(sk, &mreq);
3783                 return ret;
3784         }
3785
3786         case PACKET_RX_RING:
3787         case PACKET_TX_RING:
3788         {
3789                 union tpacket_req_u req_u;
3790                 int len;
3791
3792                 lock_sock(sk);
3793                 switch (po->tp_version) {
3794                 case TPACKET_V1:
3795                 case TPACKET_V2:
3796                         len = sizeof(req_u.req);
3797                         break;
3798                 case TPACKET_V3:
3799                 default:
3800                         len = sizeof(req_u.req3);
3801                         break;
3802                 }
3803                 if (optlen < len) {
3804                         ret = -EINVAL;
3805                 } else {
3806                         if (copy_from_sockptr(&req_u.req, optval, len))
3807                                 ret = -EFAULT;
3808                         else
3809                                 ret = packet_set_ring(sk, &req_u, 0,
3810                                                     optname == PACKET_TX_RING);
3811                 }
3812                 release_sock(sk);
3813                 return ret;
3814         }
3815         case PACKET_COPY_THRESH:
3816         {
3817                 int val;
3818
3819                 if (optlen != sizeof(val))
3820                         return -EINVAL;
3821                 if (copy_from_sockptr(&val, optval, sizeof(val)))
3822                         return -EFAULT;
3823
3824                 pkt_sk(sk)->copy_thresh = val;
3825                 return 0;
3826         }
3827         case PACKET_VERSION:
3828         {
3829                 int val;
3830
3831                 if (optlen != sizeof(val))
3832                         return -EINVAL;
3833                 if (copy_from_sockptr(&val, optval, sizeof(val)))
3834                         return -EFAULT;
3835                 switch (val) {
3836                 case TPACKET_V1:
3837                 case TPACKET_V2:
3838                 case TPACKET_V3:
3839                         break;
3840                 default:
3841                         return -EINVAL;
3842                 }
3843                 lock_sock(sk);
3844                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3845                         ret = -EBUSY;
3846                 } else {
3847                         po->tp_version = val;
3848                         ret = 0;
3849                 }
3850                 release_sock(sk);
3851                 return ret;
3852         }
3853         case PACKET_RESERVE:
3854         {
3855                 unsigned int val;
3856
3857                 if (optlen != sizeof(val))
3858                         return -EINVAL;
3859                 if (copy_from_sockptr(&val, optval, sizeof(val)))
3860                         return -EFAULT;
3861                 if (val > INT_MAX)
3862                         return -EINVAL;
3863                 lock_sock(sk);
3864                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3865                         ret = -EBUSY;
3866                 } else {
3867                         po->tp_reserve = val;
3868                         ret = 0;
3869                 }
3870                 release_sock(sk);
3871                 return ret;
3872         }
3873         case PACKET_LOSS:
3874         {
3875                 unsigned int val;
3876
3877                 if (optlen != sizeof(val))
3878                         return -EINVAL;
3879                 if (copy_from_sockptr(&val, optval, sizeof(val)))
3880                         return -EFAULT;
3881
3882                 lock_sock(sk);
3883                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3884                         ret = -EBUSY;
3885                 } else {
3886                         po->tp_loss = !!val;
3887                         ret = 0;
3888                 }
3889                 release_sock(sk);
3890                 return ret;
3891         }
3892         case PACKET_AUXDATA:
3893         {
3894                 int val;
3895
3896                 if (optlen < sizeof(val))
3897                         return -EINVAL;
3898                 if (copy_from_sockptr(&val, optval, sizeof(val)))
3899                         return -EFAULT;
3900
3901                 lock_sock(sk);
3902                 po->auxdata = !!val;
3903                 release_sock(sk);
3904                 return 0;
3905         }
3906         case PACKET_ORIGDEV:
3907         {
3908                 int val;
3909
3910                 if (optlen < sizeof(val))
3911                         return -EINVAL;
3912                 if (copy_from_sockptr(&val, optval, sizeof(val)))
3913                         return -EFAULT;
3914
3915                 lock_sock(sk);
3916                 po->origdev = !!val;
3917                 release_sock(sk);
3918                 return 0;
3919         }
3920         case PACKET_VNET_HDR:
3921         {
3922                 int val;
3923
3924                 if (sock->type != SOCK_RAW)
3925                         return -EINVAL;
3926                 if (optlen < sizeof(val))
3927                         return -EINVAL;
3928                 if (copy_from_sockptr(&val, optval, sizeof(val)))
3929                         return -EFAULT;
3930
3931                 lock_sock(sk);
3932                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3933                         ret = -EBUSY;
3934                 } else {
3935                         po->has_vnet_hdr = !!val;
3936                         ret = 0;
3937                 }
3938                 release_sock(sk);
3939                 return ret;
3940         }
3941         case PACKET_TIMESTAMP:
3942         {
3943                 int val;
3944
3945                 if (optlen != sizeof(val))
3946                         return -EINVAL;
3947                 if (copy_from_sockptr(&val, optval, sizeof(val)))
3948                         return -EFAULT;
3949
3950                 po->tp_tstamp = val;
3951                 return 0;
3952         }
3953         case PACKET_FANOUT:
3954         {
3955                 struct fanout_args args = { 0 };
3956
3957                 if (optlen != sizeof(int) && optlen != sizeof(args))
3958                         return -EINVAL;
3959                 if (copy_from_sockptr(&args, optval, optlen))
3960                         return -EFAULT;
3961
3962                 return fanout_add(sk, &args);
3963         }
3964         case PACKET_FANOUT_DATA:
3965         {
3966                 /* Paired with the WRITE_ONCE() in fanout_add() */
3967                 if (!READ_ONCE(po->fanout))
3968                         return -EINVAL;
3969
3970                 return fanout_set_data(po, optval, optlen);
3971         }
3972         case PACKET_IGNORE_OUTGOING:
3973         {
3974                 int val;
3975
3976                 if (optlen != sizeof(val))
3977                         return -EINVAL;
3978                 if (copy_from_sockptr(&val, optval, sizeof(val)))
3979                         return -EFAULT;
3980                 if (val < 0 || val > 1)
3981                         return -EINVAL;
3982
3983                 po->prot_hook.ignore_outgoing = !!val;
3984                 return 0;
3985         }
3986         case PACKET_TX_HAS_OFF:
3987         {
3988                 unsigned int val;
3989
3990                 if (optlen != sizeof(val))
3991                         return -EINVAL;
3992                 if (copy_from_sockptr(&val, optval, sizeof(val)))
3993                         return -EFAULT;
3994
3995                 lock_sock(sk);
3996                 if (!po->rx_ring.pg_vec && !po->tx_ring.pg_vec)
3997                         po->tp_tx_has_off = !!val;
3998
3999                 release_sock(sk);
4000                 return 0;
4001         }
4002         case PACKET_QDISC_BYPASS:
4003         {
4004                 int val;
4005
4006                 if (optlen != sizeof(val))
4007                         return -EINVAL;
4008                 if (copy_from_sockptr(&val, optval, sizeof(val)))
4009                         return -EFAULT;
4010
4011                 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
4012                 return 0;
4013         }
4014         default:
4015                 return -ENOPROTOOPT;
4016         }
4017 }
4018
4019 static int packet_getsockopt(struct socket *sock, int level, int optname,
4020                              char __user *optval, int __user *optlen)
4021 {
4022         int len;
4023         int val, lv = sizeof(val);
4024         struct sock *sk = sock->sk;
4025         struct packet_sock *po = pkt_sk(sk);
4026         void *data = &val;
4027         union tpacket_stats_u st;
4028         struct tpacket_rollover_stats rstats;
4029         int drops;
4030
4031         if (level != SOL_PACKET)
4032                 return -ENOPROTOOPT;
4033
4034         if (get_user(len, optlen))
4035                 return -EFAULT;
4036
4037         if (len < 0)
4038                 return -EINVAL;
4039
4040         switch (optname) {
4041         case PACKET_STATISTICS:
4042                 spin_lock_bh(&sk->sk_receive_queue.lock);
4043                 memcpy(&st, &po->stats, sizeof(st));
4044                 memset(&po->stats, 0, sizeof(po->stats));
4045                 spin_unlock_bh(&sk->sk_receive_queue.lock);
4046                 drops = atomic_xchg(&po->tp_drops, 0);
4047
4048                 if (po->tp_version == TPACKET_V3) {
4049                         lv = sizeof(struct tpacket_stats_v3);
4050                         st.stats3.tp_drops = drops;
4051                         st.stats3.tp_packets += drops;
4052                         data = &st.stats3;
4053                 } else {
4054                         lv = sizeof(struct tpacket_stats);
4055                         st.stats1.tp_drops = drops;
4056                         st.stats1.tp_packets += drops;
4057                         data = &st.stats1;
4058                 }
4059
4060                 break;
4061         case PACKET_AUXDATA:
4062                 val = po->auxdata;
4063                 break;
4064         case PACKET_ORIGDEV:
4065                 val = po->origdev;
4066                 break;
4067         case PACKET_VNET_HDR:
4068                 val = po->has_vnet_hdr;
4069                 break;
4070         case PACKET_VERSION:
4071                 val = po->tp_version;
4072                 break;
4073         case PACKET_HDRLEN:
4074                 if (len > sizeof(int))
4075                         len = sizeof(int);
4076                 if (len < sizeof(int))
4077                         return -EINVAL;
4078                 if (copy_from_user(&val, optval, len))
4079                         return -EFAULT;
4080                 switch (val) {
4081                 case TPACKET_V1:
4082                         val = sizeof(struct tpacket_hdr);
4083                         break;
4084                 case TPACKET_V2:
4085                         val = sizeof(struct tpacket2_hdr);
4086                         break;
4087                 case TPACKET_V3:
4088                         val = sizeof(struct tpacket3_hdr);
4089                         break;
4090                 default:
4091                         return -EINVAL;
4092                 }
4093                 break;
4094         case PACKET_RESERVE:
4095                 val = po->tp_reserve;
4096                 break;
4097         case PACKET_LOSS:
4098                 val = po->tp_loss;
4099                 break;
4100         case PACKET_TIMESTAMP:
4101                 val = po->tp_tstamp;
4102                 break;
4103         case PACKET_FANOUT:
4104                 val = (po->fanout ?
4105                        ((u32)po->fanout->id |
4106                         ((u32)po->fanout->type << 16) |
4107                         ((u32)po->fanout->flags << 24)) :
4108                        0);
4109                 break;
4110         case PACKET_IGNORE_OUTGOING:
4111                 val = po->prot_hook.ignore_outgoing;
4112                 break;
4113         case PACKET_ROLLOVER_STATS:
4114                 if (!po->rollover)
4115                         return -EINVAL;
4116                 rstats.tp_all = atomic_long_read(&po->rollover->num);
4117                 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4118                 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4119                 data = &rstats;
4120                 lv = sizeof(rstats);
4121                 break;
4122         case PACKET_TX_HAS_OFF:
4123                 val = po->tp_tx_has_off;
4124                 break;
4125         case PACKET_QDISC_BYPASS:
4126                 val = packet_use_direct_xmit(po);
4127                 break;
4128         default:
4129                 return -ENOPROTOOPT;
4130         }
4131
4132         if (len > lv)
4133                 len = lv;
4134         if (put_user(len, optlen))
4135                 return -EFAULT;
4136         if (copy_to_user(optval, data, len))
4137                 return -EFAULT;
4138         return 0;
4139 }
4140
4141 static int packet_notifier(struct notifier_block *this,
4142                            unsigned long msg, void *ptr)
4143 {
4144         struct sock *sk;
4145         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4146         struct net *net = dev_net(dev);
4147
4148         rcu_read_lock();
4149         sk_for_each_rcu(sk, &net->packet.sklist) {
4150                 struct packet_sock *po = pkt_sk(sk);
4151
4152                 switch (msg) {
4153                 case NETDEV_UNREGISTER:
4154                         if (po->mclist)
4155                                 packet_dev_mclist_delete(dev, &po->mclist);
4156                         fallthrough;
4157
4158                 case NETDEV_DOWN:
4159                         if (dev->ifindex == po->ifindex) {
4160                                 spin_lock(&po->bind_lock);
4161                                 if (po->running) {
4162                                         __unregister_prot_hook(sk, false);
4163                                         sk->sk_err = ENETDOWN;
4164                                         if (!sock_flag(sk, SOCK_DEAD))
4165                                                 sk_error_report(sk);
4166                                 }
4167                                 if (msg == NETDEV_UNREGISTER) {
4168                                         packet_cached_dev_reset(po);
4169                                         WRITE_ONCE(po->ifindex, -1);
4170                                         netdev_put(po->prot_hook.dev,
4171                                                    &po->prot_hook.dev_tracker);
4172                                         po->prot_hook.dev = NULL;
4173                                 }
4174                                 spin_unlock(&po->bind_lock);
4175                         }
4176                         break;
4177                 case NETDEV_UP:
4178                         if (dev->ifindex == po->ifindex) {
4179                                 spin_lock(&po->bind_lock);
4180                                 if (po->num)
4181                                         register_prot_hook(sk);
4182                                 spin_unlock(&po->bind_lock);
4183                         }
4184                         break;
4185                 }
4186         }
4187         rcu_read_unlock();
4188         return NOTIFY_DONE;
4189 }
4190
4191
4192 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4193                         unsigned long arg)
4194 {
4195         struct sock *sk = sock->sk;
4196
4197         switch (cmd) {
4198         case SIOCOUTQ:
4199         {
4200                 int amount = sk_wmem_alloc_get(sk);
4201
4202                 return put_user(amount, (int __user *)arg);
4203         }
4204         case SIOCINQ:
4205         {
4206                 struct sk_buff *skb;
4207                 int amount = 0;
4208
4209                 spin_lock_bh(&sk->sk_receive_queue.lock);
4210                 skb = skb_peek(&sk->sk_receive_queue);
4211                 if (skb)
4212                         amount = skb->len;
4213                 spin_unlock_bh(&sk->sk_receive_queue.lock);
4214                 return put_user(amount, (int __user *)arg);
4215         }
4216 #ifdef CONFIG_INET
4217         case SIOCADDRT:
4218         case SIOCDELRT:
4219         case SIOCDARP:
4220         case SIOCGARP:
4221         case SIOCSARP:
4222         case SIOCGIFADDR:
4223         case SIOCSIFADDR:
4224         case SIOCGIFBRDADDR:
4225         case SIOCSIFBRDADDR:
4226         case SIOCGIFNETMASK:
4227         case SIOCSIFNETMASK:
4228         case SIOCGIFDSTADDR:
4229         case SIOCSIFDSTADDR:
4230         case SIOCSIFFLAGS:
4231                 return inet_dgram_ops.ioctl(sock, cmd, arg);
4232 #endif
4233
4234         default:
4235                 return -ENOIOCTLCMD;
4236         }
4237         return 0;
4238 }
4239
4240 static __poll_t packet_poll(struct file *file, struct socket *sock,
4241                                 poll_table *wait)
4242 {
4243         struct sock *sk = sock->sk;
4244         struct packet_sock *po = pkt_sk(sk);
4245         __poll_t mask = datagram_poll(file, sock, wait);
4246
4247         spin_lock_bh(&sk->sk_receive_queue.lock);
4248         if (po->rx_ring.pg_vec) {
4249                 if (!packet_previous_rx_frame(po, &po->rx_ring,
4250                         TP_STATUS_KERNEL))
4251                         mask |= EPOLLIN | EPOLLRDNORM;
4252         }
4253         packet_rcv_try_clear_pressure(po);
4254         spin_unlock_bh(&sk->sk_receive_queue.lock);
4255         spin_lock_bh(&sk->sk_write_queue.lock);
4256         if (po->tx_ring.pg_vec) {
4257                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4258                         mask |= EPOLLOUT | EPOLLWRNORM;
4259         }
4260         spin_unlock_bh(&sk->sk_write_queue.lock);
4261         return mask;
4262 }
4263
4264
4265 /* Dirty? Well, I still did not learn better way to account
4266  * for user mmaps.
4267  */
4268
4269 static void packet_mm_open(struct vm_area_struct *vma)
4270 {
4271         struct file *file = vma->vm_file;
4272         struct socket *sock = file->private_data;
4273         struct sock *sk = sock->sk;
4274
4275         if (sk)
4276                 atomic_inc(&pkt_sk(sk)->mapped);
4277 }
4278
4279 static void packet_mm_close(struct vm_area_struct *vma)
4280 {
4281         struct file *file = vma->vm_file;
4282         struct socket *sock = file->private_data;
4283         struct sock *sk = sock->sk;
4284
4285         if (sk)
4286                 atomic_dec(&pkt_sk(sk)->mapped);
4287 }
4288
4289 static const struct vm_operations_struct packet_mmap_ops = {
4290         .open   =       packet_mm_open,
4291         .close  =       packet_mm_close,
4292 };
4293
4294 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4295                         unsigned int len)
4296 {
4297         int i;
4298
4299         for (i = 0; i < len; i++) {
4300                 if (likely(pg_vec[i].buffer)) {
4301                         if (is_vmalloc_addr(pg_vec[i].buffer))
4302                                 vfree(pg_vec[i].buffer);
4303                         else
4304                                 free_pages((unsigned long)pg_vec[i].buffer,
4305                                            order);
4306                         pg_vec[i].buffer = NULL;
4307                 }
4308         }
4309         kfree(pg_vec);
4310 }
4311
4312 static char *alloc_one_pg_vec_page(unsigned long order)
4313 {
4314         char *buffer;
4315         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4316                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4317
4318         buffer = (char *) __get_free_pages(gfp_flags, order);
4319         if (buffer)
4320                 return buffer;
4321
4322         /* __get_free_pages failed, fall back to vmalloc */
4323         buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4324         if (buffer)
4325                 return buffer;
4326
4327         /* vmalloc failed, lets dig into swap here */
4328         gfp_flags &= ~__GFP_NORETRY;
4329         buffer = (char *) __get_free_pages(gfp_flags, order);
4330         if (buffer)
4331                 return buffer;
4332
4333         /* complete and utter failure */
4334         return NULL;
4335 }
4336
4337 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4338 {
4339         unsigned int block_nr = req->tp_block_nr;
4340         struct pgv *pg_vec;
4341         int i;
4342
4343         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4344         if (unlikely(!pg_vec))
4345                 goto out;
4346
4347         for (i = 0; i < block_nr; i++) {
4348                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4349                 if (unlikely(!pg_vec[i].buffer))
4350                         goto out_free_pgvec;
4351         }
4352
4353 out:
4354         return pg_vec;
4355
4356 out_free_pgvec:
4357         free_pg_vec(pg_vec, order, block_nr);
4358         pg_vec = NULL;
4359         goto out;
4360 }
4361
4362 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4363                 int closing, int tx_ring)
4364 {
4365         struct pgv *pg_vec = NULL;
4366         struct packet_sock *po = pkt_sk(sk);
4367         unsigned long *rx_owner_map = NULL;
4368         int was_running, order = 0;
4369         struct packet_ring_buffer *rb;
4370         struct sk_buff_head *rb_queue;
4371         __be16 num;
4372         int err;
4373         /* Added to avoid minimal code churn */
4374         struct tpacket_req *req = &req_u->req;
4375
4376         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4377         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4378
4379         err = -EBUSY;
4380         if (!closing) {
4381                 if (atomic_read(&po->mapped))
4382                         goto out;
4383                 if (packet_read_pending(rb))
4384                         goto out;
4385         }
4386
4387         if (req->tp_block_nr) {
4388                 unsigned int min_frame_size;
4389
4390                 /* Sanity tests and some calculations */
4391                 err = -EBUSY;
4392                 if (unlikely(rb->pg_vec))
4393                         goto out;
4394
4395                 switch (po->tp_version) {
4396                 case TPACKET_V1:
4397                         po->tp_hdrlen = TPACKET_HDRLEN;
4398                         break;
4399                 case TPACKET_V2:
4400                         po->tp_hdrlen = TPACKET2_HDRLEN;
4401                         break;
4402                 case TPACKET_V3:
4403                         po->tp_hdrlen = TPACKET3_HDRLEN;
4404                         break;
4405                 }
4406
4407                 err = -EINVAL;
4408                 if (unlikely((int)req->tp_block_size <= 0))
4409                         goto out;
4410                 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4411                         goto out;
4412                 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4413                 if (po->tp_version >= TPACKET_V3 &&
4414                     req->tp_block_size <
4415                     BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4416                         goto out;
4417                 if (unlikely(req->tp_frame_size < min_frame_size))
4418                         goto out;
4419                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4420                         goto out;
4421
4422                 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4423                 if (unlikely(rb->frames_per_block == 0))
4424                         goto out;
4425                 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4426                         goto out;
4427                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4428                                         req->tp_frame_nr))
4429                         goto out;
4430
4431                 err = -ENOMEM;
4432                 order = get_order(req->tp_block_size);
4433                 pg_vec = alloc_pg_vec(req, order);
4434                 if (unlikely(!pg_vec))
4435                         goto out;
4436                 switch (po->tp_version) {
4437                 case TPACKET_V3:
4438                         /* Block transmit is not supported yet */
4439                         if (!tx_ring) {
4440                                 init_prb_bdqc(po, rb, pg_vec, req_u);
4441                         } else {
4442                                 struct tpacket_req3 *req3 = &req_u->req3;
4443
4444                                 if (req3->tp_retire_blk_tov ||
4445                                     req3->tp_sizeof_priv ||
4446                                     req3->tp_feature_req_word) {
4447                                         err = -EINVAL;
4448                                         goto out_free_pg_vec;
4449                                 }
4450                         }
4451                         break;
4452                 default:
4453                         if (!tx_ring) {
4454                                 rx_owner_map = bitmap_alloc(req->tp_frame_nr,
4455                                         GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
4456                                 if (!rx_owner_map)
4457                                         goto out_free_pg_vec;
4458                         }
4459                         break;
4460                 }
4461         }
4462         /* Done */
4463         else {
4464                 err = -EINVAL;
4465                 if (unlikely(req->tp_frame_nr))
4466                         goto out;
4467         }
4468
4469
4470         /* Detach socket from network */
4471         spin_lock(&po->bind_lock);
4472         was_running = po->running;
4473         num = po->num;
4474         if (was_running) {
4475                 WRITE_ONCE(po->num, 0);
4476                 __unregister_prot_hook(sk, false);
4477         }
4478         spin_unlock(&po->bind_lock);
4479
4480         synchronize_net();
4481
4482         err = -EBUSY;
4483         mutex_lock(&po->pg_vec_lock);
4484         if (closing || atomic_read(&po->mapped) == 0) {
4485                 err = 0;
4486                 spin_lock_bh(&rb_queue->lock);
4487                 swap(rb->pg_vec, pg_vec);
4488                 if (po->tp_version <= TPACKET_V2)
4489                         swap(rb->rx_owner_map, rx_owner_map);
4490                 rb->frame_max = (req->tp_frame_nr - 1);
4491                 rb->head = 0;
4492                 rb->frame_size = req->tp_frame_size;
4493                 spin_unlock_bh(&rb_queue->lock);
4494
4495                 swap(rb->pg_vec_order, order);
4496                 swap(rb->pg_vec_len, req->tp_block_nr);
4497
4498                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4499                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4500                                                 tpacket_rcv : packet_rcv;
4501                 skb_queue_purge(rb_queue);
4502                 if (atomic_read(&po->mapped))
4503                         pr_err("packet_mmap: vma is busy: %d\n",
4504                                atomic_read(&po->mapped));
4505         }
4506         mutex_unlock(&po->pg_vec_lock);
4507
4508         spin_lock(&po->bind_lock);
4509         if (was_running) {
4510                 WRITE_ONCE(po->num, num);
4511                 register_prot_hook(sk);
4512         }
4513         spin_unlock(&po->bind_lock);
4514         if (pg_vec && (po->tp_version > TPACKET_V2)) {
4515                 /* Because we don't support block-based V3 on tx-ring */
4516                 if (!tx_ring)
4517                         prb_shutdown_retire_blk_timer(po, rb_queue);
4518         }
4519
4520 out_free_pg_vec:
4521         if (pg_vec) {
4522                 bitmap_free(rx_owner_map);
4523                 free_pg_vec(pg_vec, order, req->tp_block_nr);
4524         }
4525 out:
4526         return err;
4527 }
4528
4529 static int packet_mmap(struct file *file, struct socket *sock,
4530                 struct vm_area_struct *vma)
4531 {
4532         struct sock *sk = sock->sk;
4533         struct packet_sock *po = pkt_sk(sk);
4534         unsigned long size, expected_size;
4535         struct packet_ring_buffer *rb;
4536         unsigned long start;
4537         int err = -EINVAL;
4538         int i;
4539
4540         if (vma->vm_pgoff)
4541                 return -EINVAL;
4542
4543         mutex_lock(&po->pg_vec_lock);
4544
4545         expected_size = 0;
4546         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4547                 if (rb->pg_vec) {
4548                         expected_size += rb->pg_vec_len
4549                                                 * rb->pg_vec_pages
4550                                                 * PAGE_SIZE;
4551                 }
4552         }
4553
4554         if (expected_size == 0)
4555                 goto out;
4556
4557         size = vma->vm_end - vma->vm_start;
4558         if (size != expected_size)
4559                 goto out;
4560
4561         start = vma->vm_start;
4562         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4563                 if (rb->pg_vec == NULL)
4564                         continue;
4565
4566                 for (i = 0; i < rb->pg_vec_len; i++) {
4567                         struct page *page;
4568                         void *kaddr = rb->pg_vec[i].buffer;
4569                         int pg_num;
4570
4571                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4572                                 page = pgv_to_page(kaddr);
4573                                 err = vm_insert_page(vma, start, page);
4574                                 if (unlikely(err))
4575                                         goto out;
4576                                 start += PAGE_SIZE;
4577                                 kaddr += PAGE_SIZE;
4578                         }
4579                 }
4580         }
4581
4582         atomic_inc(&po->mapped);
4583         vma->vm_ops = &packet_mmap_ops;
4584         err = 0;
4585
4586 out:
4587         mutex_unlock(&po->pg_vec_lock);
4588         return err;
4589 }
4590
4591 static const struct proto_ops packet_ops_spkt = {
4592         .family =       PF_PACKET,
4593         .owner =        THIS_MODULE,
4594         .release =      packet_release,
4595         .bind =         packet_bind_spkt,
4596         .connect =      sock_no_connect,
4597         .socketpair =   sock_no_socketpair,
4598         .accept =       sock_no_accept,
4599         .getname =      packet_getname_spkt,
4600         .poll =         datagram_poll,
4601         .ioctl =        packet_ioctl,
4602         .gettstamp =    sock_gettstamp,
4603         .listen =       sock_no_listen,
4604         .shutdown =     sock_no_shutdown,
4605         .sendmsg =      packet_sendmsg_spkt,
4606         .recvmsg =      packet_recvmsg,
4607         .mmap =         sock_no_mmap,
4608         .sendpage =     sock_no_sendpage,
4609 };
4610
4611 static const struct proto_ops packet_ops = {
4612         .family =       PF_PACKET,
4613         .owner =        THIS_MODULE,
4614         .release =      packet_release,
4615         .bind =         packet_bind,
4616         .connect =      sock_no_connect,
4617         .socketpair =   sock_no_socketpair,
4618         .accept =       sock_no_accept,
4619         .getname =      packet_getname,
4620         .poll =         packet_poll,
4621         .ioctl =        packet_ioctl,
4622         .gettstamp =    sock_gettstamp,
4623         .listen =       sock_no_listen,
4624         .shutdown =     sock_no_shutdown,
4625         .setsockopt =   packet_setsockopt,
4626         .getsockopt =   packet_getsockopt,
4627         .sendmsg =      packet_sendmsg,
4628         .recvmsg =      packet_recvmsg,
4629         .mmap =         packet_mmap,
4630         .sendpage =     sock_no_sendpage,
4631 };
4632
4633 static const struct net_proto_family packet_family_ops = {
4634         .family =       PF_PACKET,
4635         .create =       packet_create,
4636         .owner  =       THIS_MODULE,
4637 };
4638
4639 static struct notifier_block packet_netdev_notifier = {
4640         .notifier_call =        packet_notifier,
4641 };
4642
4643 #ifdef CONFIG_PROC_FS
4644
4645 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4646         __acquires(RCU)
4647 {
4648         struct net *net = seq_file_net(seq);
4649
4650         rcu_read_lock();
4651         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4652 }
4653
4654 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4655 {
4656         struct net *net = seq_file_net(seq);
4657         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4658 }
4659
4660 static void packet_seq_stop(struct seq_file *seq, void *v)
4661         __releases(RCU)
4662 {
4663         rcu_read_unlock();
4664 }
4665
4666 static int packet_seq_show(struct seq_file *seq, void *v)
4667 {
4668         if (v == SEQ_START_TOKEN)
4669                 seq_printf(seq,
4670                            "%*sRefCnt Type Proto  Iface R Rmem   User   Inode\n",
4671                            IS_ENABLED(CONFIG_64BIT) ? -17 : -9, "sk");
4672         else {
4673                 struct sock *s = sk_entry(v);
4674                 const struct packet_sock *po = pkt_sk(s);
4675
4676                 seq_printf(seq,
4677                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4678                            s,
4679                            refcount_read(&s->sk_refcnt),
4680                            s->sk_type,
4681                            ntohs(READ_ONCE(po->num)),
4682                            READ_ONCE(po->ifindex),
4683                            po->running,
4684                            atomic_read(&s->sk_rmem_alloc),
4685                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4686                            sock_i_ino(s));
4687         }
4688
4689         return 0;
4690 }
4691
4692 static const struct seq_operations packet_seq_ops = {
4693         .start  = packet_seq_start,
4694         .next   = packet_seq_next,
4695         .stop   = packet_seq_stop,
4696         .show   = packet_seq_show,
4697 };
4698 #endif
4699
4700 static int __net_init packet_net_init(struct net *net)
4701 {
4702         mutex_init(&net->packet.sklist_lock);
4703         INIT_HLIST_HEAD(&net->packet.sklist);
4704
4705 #ifdef CONFIG_PROC_FS
4706         if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4707                         sizeof(struct seq_net_private)))
4708                 return -ENOMEM;
4709 #endif /* CONFIG_PROC_FS */
4710
4711         return 0;
4712 }
4713
4714 static void __net_exit packet_net_exit(struct net *net)
4715 {
4716         remove_proc_entry("packet", net->proc_net);
4717         WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4718 }
4719
4720 static struct pernet_operations packet_net_ops = {
4721         .init = packet_net_init,
4722         .exit = packet_net_exit,
4723 };
4724
4725
4726 static void __exit packet_exit(void)
4727 {
4728         sock_unregister(PF_PACKET);
4729         proto_unregister(&packet_proto);
4730         unregister_netdevice_notifier(&packet_netdev_notifier);
4731         unregister_pernet_subsys(&packet_net_ops);
4732 }
4733
4734 static int __init packet_init(void)
4735 {
4736         int rc;
4737
4738         rc = register_pernet_subsys(&packet_net_ops);
4739         if (rc)
4740                 goto out;
4741         rc = register_netdevice_notifier(&packet_netdev_notifier);
4742         if (rc)
4743                 goto out_pernet;
4744         rc = proto_register(&packet_proto, 0);
4745         if (rc)
4746                 goto out_notifier;
4747         rc = sock_register(&packet_family_ops);
4748         if (rc)
4749                 goto out_proto;
4750
4751         return 0;
4752
4753 out_proto:
4754         proto_unregister(&packet_proto);
4755 out_notifier:
4756         unregister_netdevice_notifier(&packet_netdev_notifier);
4757 out_pernet:
4758         unregister_pernet_subsys(&packet_net_ops);
4759 out:
4760         return rc;
4761 }
4762
4763 module_init(packet_init);
4764 module_exit(packet_exit);
4765 MODULE_LICENSE("GPL");
4766 MODULE_ALIAS_NETPROTO(PF_PACKET);