linux/net/packet/af_packet.c
<<
>>
Prefs
   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#include <linux/ethtool.h>
  50#include <linux/types.h>
  51#include <linux/mm.h>
  52#include <linux/capability.h>
  53#include <linux/fcntl.h>
  54#include <linux/socket.h>
  55#include <linux/in.h>
  56#include <linux/inet.h>
  57#include <linux/netdevice.h>
  58#include <linux/if_packet.h>
  59#include <linux/wireless.h>
  60#include <linux/kernel.h>
  61#include <linux/kmod.h>
  62#include <linux/slab.h>
  63#include <linux/vmalloc.h>
  64#include <net/net_namespace.h>
  65#include <net/ip.h>
  66#include <net/protocol.h>
  67#include <linux/skbuff.h>
  68#include <net/sock.h>
  69#include <linux/errno.h>
  70#include <linux/timer.h>
  71#include <linux/uaccess.h>
  72#include <asm/ioctls.h>
  73#include <asm/page.h>
  74#include <asm/cacheflush.h>
  75#include <asm/io.h>
  76#include <linux/proc_fs.h>
  77#include <linux/seq_file.h>
  78#include <linux/poll.h>
  79#include <linux/module.h>
  80#include <linux/init.h>
  81#include <linux/mutex.h>
  82#include <linux/if_vlan.h>
  83#include <linux/virtio_net.h>
  84#include <linux/errqueue.h>
  85#include <linux/net_tstamp.h>
  86#include <linux/percpu.h>
  87#ifdef CONFIG_INET
  88#include <net/inet_common.h>
  89#endif
  90#include <linux/bpf.h>
  91#include <net/compat.h>
  92
  93#include "internal.h"
  94
  95/*
  96   Assumptions:
  97   - If the device has no dev->header_ops->create, there is no LL header
  98     visible above the device. In this case, its hard_header_len should be 0.
  99     The device may prepend its own header internally. In this case, its
 100     needed_headroom should be set to the space needed for it to add its
 101     internal header.
 102     For example, a WiFi driver pretending to be an Ethernet driver should
 103     set its hard_header_len to be the Ethernet header length, and set its
 104     needed_headroom to be (the real WiFi header length - the fake Ethernet
 105     header length).
 106   - packet socket receives packets with pulled ll header,
 107     so that SOCK_RAW should push it back.
 108
 109On receive:
 110-----------
 111
 112Incoming, dev_has_header(dev) == true
 113   mac_header -> ll header
 114   data       -> data
 115
 116Outgoing, dev_has_header(dev) == true
 117   mac_header -> ll header
 118   data       -> ll header
 119
 120Incoming, dev_has_header(dev) == false
 121   mac_header -> data
 122     However drivers often make it point to the ll header.
 123     This is incorrect because the ll header should be invisible to us.
 124   data       -> data
 125
 126Outgoing, dev_has_header(dev) == false
 127   mac_header -> data. ll header is invisible to us.
 128   data       -> data
 129
 130Resume
 131  If dev_has_header(dev) == false we are unable to restore the ll header,
 132    because it is invisible to us.
 133
 134
 135On transmit:
 136------------
 137
 138dev_has_header(dev) == true
 139   mac_header -> ll header
 140   data       -> ll header
 141
 142dev_has_header(dev) == false (ll header is invisible to us)
 143   mac_header -> data
 144   data       -> data
 145
 146   We should set network_header on output to the correct position,
 147   packet classifier depends on it.
 148 */
 149
 150/* Private packet socket structures. */
 151
 152/* identical to struct packet_mreq except it has
 153 * a longer address field.
 154 */
 155struct packet_mreq_max {
 156        int             mr_ifindex;
 157        unsigned short  mr_type;
 158        unsigned short  mr_alen;
 159        unsigned char   mr_address[MAX_ADDR_LEN];
 160};
 161
 162union tpacket_uhdr {
 163        struct tpacket_hdr  *h1;
 164        struct tpacket2_hdr *h2;
 165        struct tpacket3_hdr *h3;
 166        void *raw;
 167};
 168
 169static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 170                int closing, int tx_ring);
 171
 172#define V3_ALIGNMENT    (8)
 173
 174#define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 175
 176#define BLK_PLUS_PRIV(sz_of_priv) \
 177        (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 178
 179#define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
 180#define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
 181#define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
 182#define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
 183#define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
 184#define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
 185
 186struct packet_sock;
 187static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 188                       struct packet_type *pt, struct net_device *orig_dev);
 189
 190static void *packet_previous_frame(struct packet_sock *po,
 191                struct packet_ring_buffer *rb,
 192                int status);
 193static void packet_increment_head(struct packet_ring_buffer *buff);
 194static int prb_curr_blk_in_use(struct tpacket_block_desc *);
 195static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 196                        struct packet_sock *);
 197static void prb_retire_current_block(struct tpacket_kbdq_core *,
 198                struct packet_sock *, unsigned int status);
 199static int prb_queue_frozen(struct tpacket_kbdq_core *);
 200static void prb_open_block(struct tpacket_kbdq_core *,
 201                struct tpacket_block_desc *);
 202static void prb_retire_rx_blk_timer_expired(struct timer_list *);
 203static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 204static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 205static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 206                struct tpacket3_hdr *);
 207static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 208                struct tpacket3_hdr *);
 209static void packet_flush_mclist(struct sock *sk);
 210static u16 packet_pick_tx_queue(struct sk_buff *skb);
 211
 212struct packet_skb_cb {
 213        union {
 214                struct sockaddr_pkt pkt;
 215                union {
 216                        /* Trick: alias skb original length with
 217                         * ll.sll_family and ll.protocol in order
 218                         * to save room.
 219                         */
 220                        unsigned int origlen;
 221                        struct sockaddr_ll ll;
 222                };
 223        } sa;
 224};
 225
 226#define vio_le() virtio_legacy_is_little_endian()
 227
 228#define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 229
 230#define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 231#define GET_PBLOCK_DESC(x, bid) \
 232        ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 233#define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 234        ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 235#define GET_NEXT_PRB_BLK_NUM(x) \
 236        (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 237        ((x)->kactive_blk_num+1) : 0)
 238
 239static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 240static void __fanout_link(struct sock *sk, struct packet_sock *po);
 241
 242static int packet_direct_xmit(struct sk_buff *skb)
 243{
 244        return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
 245}
 246
 247static struct net_device *packet_cached_dev_get(struct packet_sock *po)
 248{
 249        struct net_device *dev;
 250
 251        rcu_read_lock();
 252        dev = rcu_dereference(po->cached_dev);
 253        if (likely(dev))
 254                dev_hold(dev);
 255        rcu_read_unlock();
 256
 257        return dev;
 258}
 259
 260static void packet_cached_dev_assign(struct packet_sock *po,
 261                                     struct net_device *dev)
 262{
 263        rcu_assign_pointer(po->cached_dev, dev);
 264}
 265
 266static void packet_cached_dev_reset(struct packet_sock *po)
 267{
 268        RCU_INIT_POINTER(po->cached_dev, NULL);
 269}
 270
 271static bool packet_use_direct_xmit(const struct packet_sock *po)
 272{
 273        return po->xmit == packet_direct_xmit;
 274}
 275
 276static u16 packet_pick_tx_queue(struct sk_buff *skb)
 277{
 278        struct net_device *dev = skb->dev;
 279        const struct net_device_ops *ops = dev->netdev_ops;
 280        int cpu = raw_smp_processor_id();
 281        u16 queue_index;
 282
 283#ifdef CONFIG_XPS
 284        skb->sender_cpu = cpu + 1;
 285#endif
 286        skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues);
 287        if (ops->ndo_select_queue) {
 288                queue_index = ops->ndo_select_queue(dev, skb, NULL);
 289                queue_index = netdev_cap_txqueue(dev, queue_index);
 290        } else {
 291                queue_index = netdev_pick_tx(dev, skb, NULL);
 292        }
 293
 294        return queue_index;
 295}
 296
 297/* __register_prot_hook must be invoked through register_prot_hook
 298 * or from a context in which asynchronous accesses to the packet
 299 * socket is not possible (packet_create()).
 300 */
 301static void __register_prot_hook(struct sock *sk)
 302{
 303        struct packet_sock *po = pkt_sk(sk);
 304
 305        if (!po->running) {
 306                if (po->fanout)
 307                        __fanout_link(sk, po);
 308                else
 309                        dev_add_pack(&po->prot_hook);
 310
 311                sock_hold(sk);
 312                po->running = 1;
 313        }
 314}
 315
 316static void register_prot_hook(struct sock *sk)
 317{
 318        lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
 319        __register_prot_hook(sk);
 320}
 321
 322/* If the sync parameter is true, we will temporarily drop
 323 * the po->bind_lock and do a synchronize_net to make sure no
 324 * asynchronous packet processing paths still refer to the elements
 325 * of po->prot_hook.  If the sync parameter is false, it is the
 326 * callers responsibility to take care of this.
 327 */
 328static void __unregister_prot_hook(struct sock *sk, bool sync)
 329{
 330        struct packet_sock *po = pkt_sk(sk);
 331
 332        lockdep_assert_held_once(&po->bind_lock);
 333
 334        po->running = 0;
 335
 336        if (po->fanout)
 337                __fanout_unlink(sk, po);
 338        else
 339                __dev_remove_pack(&po->prot_hook);
 340
 341        __sock_put(sk);
 342
 343        if (sync) {
 344                spin_unlock(&po->bind_lock);
 345                synchronize_net();
 346                spin_lock(&po->bind_lock);
 347        }
 348}
 349
 350static void unregister_prot_hook(struct sock *sk, bool sync)
 351{
 352        struct packet_sock *po = pkt_sk(sk);
 353
 354        if (po->running)
 355                __unregister_prot_hook(sk, sync);
 356}
 357
 358static inline struct page * __pure pgv_to_page(void *addr)
 359{
 360        if (is_vmalloc_addr(addr))
 361                return vmalloc_to_page(addr);
 362        return virt_to_page(addr);
 363}
 364
 365static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 366{
 367        union tpacket_uhdr h;
 368
 369        h.raw = frame;
 370        switch (po->tp_version) {
 371        case TPACKET_V1:
 372                h.h1->tp_status = status;
 373                flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 374                break;
 375        case TPACKET_V2:
 376                h.h2->tp_status = status;
 377                flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 378                break;
 379        case TPACKET_V3:
 380                h.h3->tp_status = status;
 381                flush_dcache_page(pgv_to_page(&h.h3->tp_status));
 382                break;
 383        default:
 384                WARN(1, "TPACKET version not supported.\n");
 385                BUG();
 386        }
 387
 388        smp_wmb();
 389}
 390
 391static int __packet_get_status(const struct packet_sock *po, void *frame)
 392{
 393        union tpacket_uhdr h;
 394
 395        smp_rmb();
 396
 397        h.raw = frame;
 398        switch (po->tp_version) {
 399        case TPACKET_V1:
 400                flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 401                return h.h1->tp_status;
 402        case TPACKET_V2:
 403                flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 404                return h.h2->tp_status;
 405        case TPACKET_V3:
 406                flush_dcache_page(pgv_to_page(&h.h3->tp_status));
 407                return h.h3->tp_status;
 408        default:
 409                WARN(1, "TPACKET version not supported.\n");
 410                BUG();
 411                return 0;
 412        }
 413}
 414
 415static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec64 *ts,
 416                                   unsigned int flags)
 417{
 418        struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 419
 420        if (shhwtstamps &&
 421            (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
 422            ktime_to_timespec64_cond(shhwtstamps->hwtstamp, ts))
 423                return TP_STATUS_TS_RAW_HARDWARE;
 424
 425        if ((flags & SOF_TIMESTAMPING_SOFTWARE) &&
 426            ktime_to_timespec64_cond(skb->tstamp, ts))
 427                return TP_STATUS_TS_SOFTWARE;
 428
 429        return 0;
 430}
 431
 432static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
 433                                    struct sk_buff *skb)
 434{
 435        union tpacket_uhdr h;
 436        struct timespec64 ts;
 437        __u32 ts_status;
 438
 439        if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
 440                return 0;
 441
 442        h.raw = frame;
 443        /*
 444         * versions 1 through 3 overflow the timestamps in y2106, since they
 445         * all store the seconds in a 32-bit unsigned integer.
 446         * If we create a version 4, that should have a 64-bit timestamp,
 447         * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit
 448         * nanoseconds.
 449         */
 450        switch (po->tp_version) {
 451        case TPACKET_V1:
 452                h.h1->tp_sec = ts.tv_sec;
 453                h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
 454                break;
 455        case TPACKET_V2:
 456                h.h2->tp_sec = ts.tv_sec;
 457                h.h2->tp_nsec = ts.tv_nsec;
 458                break;
 459        case TPACKET_V3:
 460                h.h3->tp_sec = ts.tv_sec;
 461                h.h3->tp_nsec = ts.tv_nsec;
 462                break;
 463        default:
 464                WARN(1, "TPACKET version not supported.\n");
 465                BUG();
 466        }
 467
 468        /* one flush is safe, as both fields always lie on the same cacheline */
 469        flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
 470        smp_wmb();
 471
 472        return ts_status;
 473}
 474
 475static void *packet_lookup_frame(const struct packet_sock *po,
 476                                 const struct packet_ring_buffer *rb,
 477                                 unsigned int position,
 478                                 int status)
 479{
 480        unsigned int pg_vec_pos, frame_offset;
 481        union tpacket_uhdr h;
 482
 483        pg_vec_pos = position / rb->frames_per_block;
 484        frame_offset = position % rb->frames_per_block;
 485
 486        h.raw = rb->pg_vec[pg_vec_pos].buffer +
 487                (frame_offset * rb->frame_size);
 488
 489        if (status != __packet_get_status(po, h.raw))
 490                return NULL;
 491
 492        return h.raw;
 493}
 494
 495static void *packet_current_frame(struct packet_sock *po,
 496                struct packet_ring_buffer *rb,
 497                int status)
 498{
 499        return packet_lookup_frame(po, rb, rb->head, status);
 500}
 501
 502static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 503{
 504        del_timer_sync(&pkc->retire_blk_timer);
 505}
 506
 507static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 508                struct sk_buff_head *rb_queue)
 509{
 510        struct tpacket_kbdq_core *pkc;
 511
 512        pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 513
 514        spin_lock_bh(&rb_queue->lock);
 515        pkc->delete_blk_timer = 1;
 516        spin_unlock_bh(&rb_queue->lock);
 517
 518        prb_del_retire_blk_timer(pkc);
 519}
 520
 521static void prb_setup_retire_blk_timer(struct packet_sock *po)
 522{
 523        struct tpacket_kbdq_core *pkc;
 524
 525        pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 526        timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
 527                    0);
 528        pkc->retire_blk_timer.expires = jiffies;
 529}
 530
 531static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 532                                int blk_size_in_bytes)
 533{
 534        struct net_device *dev;
 535        unsigned int mbits, div;
 536        struct ethtool_link_ksettings ecmd;
 537        int err;
 538
 539        rtnl_lock();
 540        dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 541        if (unlikely(!dev)) {
 542                rtnl_unlock();
 543                return DEFAULT_PRB_RETIRE_TOV;
 544        }
 545        err = __ethtool_get_link_ksettings(dev, &ecmd);
 546        rtnl_unlock();
 547        if (err)
 548                return DEFAULT_PRB_RETIRE_TOV;
 549
 550        /* If the link speed is so slow you don't really
 551         * need to worry about perf anyways
 552         */
 553        if (ecmd.base.speed < SPEED_1000 ||
 554            ecmd.base.speed == SPEED_UNKNOWN)
 555                return DEFAULT_PRB_RETIRE_TOV;
 556
 557        div = ecmd.base.speed / 1000;
 558        mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 559
 560        if (div)
 561                mbits /= div;
 562
 563        if (div)
 564                return mbits + 1;
 565        return mbits;
 566}
 567
 568static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 569                        union tpacket_req_u *req_u)
 570{
 571        p1->feature_req_word = req_u->req3.tp_feature_req_word;
 572}
 573
 574static void init_prb_bdqc(struct packet_sock *po,
 575                        struct packet_ring_buffer *rb,
 576                        struct pgv *pg_vec,
 577                        union tpacket_req_u *req_u)
 578{
 579        struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
 580        struct tpacket_block_desc *pbd;
 581
 582        memset(p1, 0x0, sizeof(*p1));
 583
 584        p1->knxt_seq_num = 1;
 585        p1->pkbdq = pg_vec;
 586        pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 587        p1->pkblk_start = pg_vec[0].buffer;
 588        p1->kblk_size = req_u->req3.tp_block_size;
 589        p1->knum_blocks = req_u->req3.tp_block_nr;
 590        p1->hdrlen = po->tp_hdrlen;
 591        p1->version = po->tp_version;
 592        p1->last_kactive_blk_num = 0;
 593        po->stats.stats3.tp_freeze_q_cnt = 0;
 594        if (req_u->req3.tp_retire_blk_tov)
 595                p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 596        else
 597                p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 598                                                req_u->req3.tp_block_size);
 599        p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 600        p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 601        rwlock_init(&p1->blk_fill_in_prog_lock);
 602
 603        p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
 604        prb_init_ft_ops(p1, req_u);
 605        prb_setup_retire_blk_timer(po);
 606        prb_open_block(p1, pbd);
 607}
 608
 609/*  Do NOT update the last_blk_num first.
 610 *  Assumes sk_buff_head lock is held.
 611 */
 612static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 613{
 614        mod_timer(&pkc->retire_blk_timer,
 615                        jiffies + pkc->tov_in_jiffies);
 616        pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 617}
 618
 619/*
 620 * Timer logic:
 621 * 1) We refresh the timer only when we open a block.
 622 *    By doing this we don't waste cycles refreshing the timer
 623 *        on packet-by-packet basis.
 624 *
 625 * With a 1MB block-size, on a 1Gbps line, it will take
 626 * i) ~8 ms to fill a block + ii) memcpy etc.
 627 * In this cut we are not accounting for the memcpy time.
 628 *
 629 * So, if the user sets the 'tmo' to 10ms then the timer
 630 * will never fire while the block is still getting filled
 631 * (which is what we want). However, the user could choose
 632 * to close a block early and that's fine.
 633 *
 634 * But when the timer does fire, we check whether or not to refresh it.
 635 * Since the tmo granularity is in msecs, it is not too expensive
 636 * to refresh the timer, lets say every '8' msecs.
 637 * Either the user can set the 'tmo' or we can derive it based on
 638 * a) line-speed and b) block-size.
 639 * prb_calc_retire_blk_tmo() calculates the tmo.
 640 *
 641 */
 642static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
 643{
 644        struct packet_sock *po =
 645                from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
 646        struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 647        unsigned int frozen;
 648        struct tpacket_block_desc *pbd;
 649
 650        spin_lock(&po->sk.sk_receive_queue.lock);
 651
 652        frozen = prb_queue_frozen(pkc);
 653        pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 654
 655        if (unlikely(pkc->delete_blk_timer))
 656                goto out;
 657
 658        /* We only need to plug the race when the block is partially filled.
 659         * tpacket_rcv:
 660         *              lock(); increment BLOCK_NUM_PKTS; unlock()
 661         *              copy_bits() is in progress ...
 662         *              timer fires on other cpu:
 663         *              we can't retire the current block because copy_bits
 664         *              is in progress.
 665         *
 666         */
 667        if (BLOCK_NUM_PKTS(pbd)) {
 668                /* Waiting for skb_copy_bits to finish... */
 669                write_lock(&pkc->blk_fill_in_prog_lock);
 670                write_unlock(&pkc->blk_fill_in_prog_lock);
 671        }
 672
 673        if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 674                if (!frozen) {
 675                        if (!BLOCK_NUM_PKTS(pbd)) {
 676                                /* An empty block. Just refresh the timer. */
 677                                goto refresh_timer;
 678                        }
 679                        prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 680                        if (!prb_dispatch_next_block(pkc, po))
 681                                goto refresh_timer;
 682                        else
 683                                goto out;
 684                } else {
 685                        /* Case 1. Queue was frozen because user-space was
 686                         *         lagging behind.
 687                         */
 688                        if (prb_curr_blk_in_use(pbd)) {
 689                                /*
 690                                 * Ok, user-space is still behind.
 691                                 * So just refresh the timer.
 692                                 */
 693                                goto refresh_timer;
 694                        } else {
 695                               /* Case 2. queue was frozen,user-space caught up,
 696                                * now the link went idle && the timer fired.
 697                                * We don't have a block to close.So we open this
 698                                * block and restart the timer.
 699                                * opening a block thaws the queue,restarts timer
 700                                * Thawing/timer-refresh is a side effect.
 701                                */
 702                                prb_open_block(pkc, pbd);
 703                                goto out;
 704                        }
 705                }
 706        }
 707
 708refresh_timer:
 709        _prb_refresh_rx_retire_blk_timer(pkc);
 710
 711out:
 712        spin_unlock(&po->sk.sk_receive_queue.lock);
 713}
 714
 715static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 716                struct tpacket_block_desc *pbd1, __u32 status)
 717{
 718        /* Flush everything minus the block header */
 719
 720#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 721        u8 *start, *end;
 722
 723        start = (u8 *)pbd1;
 724
 725        /* Skip the block header(we know header WILL fit in 4K) */
 726        start += PAGE_SIZE;
 727
 728        end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 729        for (; start < end; start += PAGE_SIZE)
 730                flush_dcache_page(pgv_to_page(start));
 731
 732        smp_wmb();
 733#endif
 734
 735        /* Now update the block status. */
 736
 737        BLOCK_STATUS(pbd1) = status;
 738
 739        /* Flush the block header */
 740
 741#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 742        start = (u8 *)pbd1;
 743        flush_dcache_page(pgv_to_page(start));
 744
 745        smp_wmb();
 746#endif
 747}
 748
 749/*
 750 * Side effect:
 751 *
 752 * 1) flush the block
 753 * 2) Increment active_blk_num
 754 *
 755 * Note:We DONT refresh the timer on purpose.
 756 *      Because almost always the next block will be opened.
 757 */
 758static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 759                struct tpacket_block_desc *pbd1,
 760                struct packet_sock *po, unsigned int stat)
 761{
 762        __u32 status = TP_STATUS_USER | stat;
 763
 764        struct tpacket3_hdr *last_pkt;
 765        struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 766        struct sock *sk = &po->sk;
 767
 768        if (atomic_read(&po->tp_drops))
 769                status |= TP_STATUS_LOSING;
 770
 771        last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 772        last_pkt->tp_next_offset = 0;
 773
 774        /* Get the ts of the last pkt */
 775        if (BLOCK_NUM_PKTS(pbd1)) {
 776                h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 777                h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 778        } else {
 779                /* Ok, we tmo'd - so get the current time.
 780                 *
 781                 * It shouldn't really happen as we don't close empty
 782                 * blocks. See prb_retire_rx_blk_timer_expired().
 783                 */
 784                struct timespec64 ts;
 785                ktime_get_real_ts64(&ts);
 786                h1->ts_last_pkt.ts_sec = ts.tv_sec;
 787                h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 788        }
 789
 790        smp_wmb();
 791
 792        /* Flush the block */
 793        prb_flush_block(pkc1, pbd1, status);
 794
 795        sk->sk_data_ready(sk);
 796
 797        pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 798}
 799
 800static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 801{
 802        pkc->reset_pending_on_curr_blk = 0;
 803}
 804
 805/*
 806 * Side effect of opening a block:
 807 *
 808 * 1) prb_queue is thawed.
 809 * 2) retire_blk_timer is refreshed.
 810 *
 811 */
 812static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 813        struct tpacket_block_desc *pbd1)
 814{
 815        struct timespec64 ts;
 816        struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 817
 818        smp_rmb();
 819
 820        /* We could have just memset this but we will lose the
 821         * flexibility of making the priv area sticky
 822         */
 823
 824        BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 825        BLOCK_NUM_PKTS(pbd1) = 0;
 826        BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 827
 828        ktime_get_real_ts64(&ts);
 829
 830        h1->ts_first_pkt.ts_sec = ts.tv_sec;
 831        h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 832
 833        pkc1->pkblk_start = (char *)pbd1;
 834        pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 835
 836        BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 837        BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 838
 839        pbd1->version = pkc1->version;
 840        pkc1->prev = pkc1->nxt_offset;
 841        pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 842
 843        prb_thaw_queue(pkc1);
 844        _prb_refresh_rx_retire_blk_timer(pkc1);
 845
 846        smp_wmb();
 847}
 848
 849/*
 850 * Queue freeze logic:
 851 * 1) Assume tp_block_nr = 8 blocks.
 852 * 2) At time 't0', user opens Rx ring.
 853 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 854 * 4) user-space is either sleeping or processing block '0'.
 855 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 856 *    it will close block-7,loop around and try to fill block '0'.
 857 *    call-flow:
 858 *    __packet_lookup_frame_in_block
 859 *      prb_retire_current_block()
 860 *      prb_dispatch_next_block()
 861 *        |->(BLOCK_STATUS == USER) evaluates to true
 862 *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 863 * 6) Now there are two cases:
 864 *    6.1) Link goes idle right after the queue is frozen.
 865 *         But remember, the last open_block() refreshed the timer.
 866 *         When this timer expires,it will refresh itself so that we can
 867 *         re-open block-0 in near future.
 868 *    6.2) Link is busy and keeps on receiving packets. This is a simple
 869 *         case and __packet_lookup_frame_in_block will check if block-0
 870 *         is free and can now be re-used.
 871 */
 872static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 873                                  struct packet_sock *po)
 874{
 875        pkc->reset_pending_on_curr_blk = 1;
 876        po->stats.stats3.tp_freeze_q_cnt++;
 877}
 878
 879#define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 880
 881/*
 882 * If the next block is free then we will dispatch it
 883 * and return a good offset.
 884 * Else, we will freeze the queue.
 885 * So, caller must check the return value.
 886 */
 887static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 888                struct packet_sock *po)
 889{
 890        struct tpacket_block_desc *pbd;
 891
 892        smp_rmb();
 893
 894        /* 1. Get current block num */
 895        pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 896
 897        /* 2. If this block is currently in_use then freeze the queue */
 898        if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 899                prb_freeze_queue(pkc, po);
 900                return NULL;
 901        }
 902
 903        /*
 904         * 3.
 905         * open this block and return the offset where the first packet
 906         * needs to get stored.
 907         */
 908        prb_open_block(pkc, pbd);
 909        return (void *)pkc->nxt_offset;
 910}
 911
 912static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 913                struct packet_sock *po, unsigned int status)
 914{
 915        struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 916
 917        /* retire/close the current block */
 918        if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 919                /*
 920                 * Plug the case where copy_bits() is in progress on
 921                 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 922                 * have space to copy the pkt in the current block and
 923                 * called prb_retire_current_block()
 924                 *
 925                 * We don't need to worry about the TMO case because
 926                 * the timer-handler already handled this case.
 927                 */
 928                if (!(status & TP_STATUS_BLK_TMO)) {
 929                        /* Waiting for skb_copy_bits to finish... */
 930                        write_lock(&pkc->blk_fill_in_prog_lock);
 931                        write_unlock(&pkc->blk_fill_in_prog_lock);
 932                }
 933                prb_close_block(pkc, pbd, po, status);
 934                return;
 935        }
 936}
 937
 938static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
 939{
 940        return TP_STATUS_USER & BLOCK_STATUS(pbd);
 941}
 942
 943static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 944{
 945        return pkc->reset_pending_on_curr_blk;
 946}
 947
 948static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 949        __releases(&pkc->blk_fill_in_prog_lock)
 950{
 951        struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 952
 953        read_unlock(&pkc->blk_fill_in_prog_lock);
 954}
 955
 956static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 957                        struct tpacket3_hdr *ppd)
 958{
 959        ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
 960}
 961
 962static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 963                        struct tpacket3_hdr *ppd)
 964{
 965        ppd->hv1.tp_rxhash = 0;
 966}
 967
 968static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
 969                        struct tpacket3_hdr *ppd)
 970{
 971        if (skb_vlan_tag_present(pkc->skb)) {
 972                ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
 973                ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
 974                ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
 975        } else {
 976                ppd->hv1.tp_vlan_tci = 0;
 977                ppd->hv1.tp_vlan_tpid = 0;
 978                ppd->tp_status = TP_STATUS_AVAILABLE;
 979        }
 980}
 981
 982static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
 983                        struct tpacket3_hdr *ppd)
 984{
 985        ppd->hv1.tp_padding = 0;
 986        prb_fill_vlan_info(pkc, ppd);
 987
 988        if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
 989                prb_fill_rxhash(pkc, ppd);
 990        else
 991                prb_clear_rxhash(pkc, ppd);
 992}
 993
 994static void prb_fill_curr_block(char *curr,
 995                                struct tpacket_kbdq_core *pkc,
 996                                struct tpacket_block_desc *pbd,
 997                                unsigned int len)
 998        __acquires(&pkc->blk_fill_in_prog_lock)
 999{
1000        struct tpacket3_hdr *ppd;
1001
1002        ppd  = (struct tpacket3_hdr *)curr;
1003        ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1004        pkc->prev = curr;
1005        pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1006        BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1007        BLOCK_NUM_PKTS(pbd) += 1;
1008        read_lock(&pkc->blk_fill_in_prog_lock);
1009        prb_run_all_ft_ops(pkc, ppd);
1010}
1011
1012/* Assumes caller has the sk->rx_queue.lock */
1013static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1014                                            struct sk_buff *skb,
1015                                            unsigned int len
1016                                            )
1017{
1018        struct tpacket_kbdq_core *pkc;
1019        struct tpacket_block_desc *pbd;
1020        char *curr, *end;
1021
1022        pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1023        pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1024
1025        /* Queue is frozen when user space is lagging behind */
1026        if (prb_queue_frozen(pkc)) {
1027                /*
1028                 * Check if that last block which caused the queue to freeze,
1029                 * is still in_use by user-space.
1030                 */
1031                if (prb_curr_blk_in_use(pbd)) {
1032                        /* Can't record this packet */
1033                        return NULL;
1034                } else {
1035                        /*
1036                         * Ok, the block was released by user-space.
1037                         * Now let's open that block.
1038                         * opening a block also thaws the queue.
1039                         * Thawing is a side effect.
1040                         */
1041                        prb_open_block(pkc, pbd);
1042                }
1043        }
1044
1045        smp_mb();
1046        curr = pkc->nxt_offset;
1047        pkc->skb = skb;
1048        end = (char *)pbd + pkc->kblk_size;
1049
1050        /* first try the current block */
1051        if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1052                prb_fill_curr_block(curr, pkc, pbd, len);
1053                return (void *)curr;
1054        }
1055
1056        /* Ok, close the current block */
1057        prb_retire_current_block(pkc, po, 0);
1058
1059        /* Now, try to dispatch the next block */
1060        curr = (char *)prb_dispatch_next_block(pkc, po);
1061        if (curr) {
1062                pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1063                prb_fill_curr_block(curr, pkc, pbd, len);
1064                return (void *)curr;
1065        }
1066
1067        /*
1068         * No free blocks are available.user_space hasn't caught up yet.
1069         * Queue was just frozen and now this packet will get dropped.
1070         */
1071        return NULL;
1072}
1073
1074static void *packet_current_rx_frame(struct packet_sock *po,
1075                                            struct sk_buff *skb,
1076                                            int status, unsigned int len)
1077{
1078        char *curr = NULL;
1079        switch (po->tp_version) {
1080        case TPACKET_V1:
1081        case TPACKET_V2:
1082                curr = packet_lookup_frame(po, &po->rx_ring,
1083                                        po->rx_ring.head, status);
1084                return curr;
1085        case TPACKET_V3:
1086                return __packet_lookup_frame_in_block(po, skb, len);
1087        default:
1088                WARN(1, "TPACKET version not supported\n");
1089                BUG();
1090                return NULL;
1091        }
1092}
1093
1094static void *prb_lookup_block(const struct packet_sock *po,
1095                              const struct packet_ring_buffer *rb,
1096                              unsigned int idx,
1097                              int status)
1098{
1099        struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1100        struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1101
1102        if (status != BLOCK_STATUS(pbd))
1103                return NULL;
1104        return pbd;
1105}
1106
1107static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1108{
1109        unsigned int prev;
1110        if (rb->prb_bdqc.kactive_blk_num)
1111                prev = rb->prb_bdqc.kactive_blk_num-1;
1112        else
1113                prev = rb->prb_bdqc.knum_blocks-1;
1114        return prev;
1115}
1116
1117/* Assumes caller has held the rx_queue.lock */
1118static void *__prb_previous_block(struct packet_sock *po,
1119                                         struct packet_ring_buffer *rb,
1120                                         int status)
1121{
1122        unsigned int previous = prb_previous_blk_num(rb);
1123        return prb_lookup_block(po, rb, previous, status);
1124}
1125
1126static void *packet_previous_rx_frame(struct packet_sock *po,
1127                                             struct packet_ring_buffer *rb,
1128                                             int status)
1129{
1130        if (po->tp_version <= TPACKET_V2)
1131                return packet_previous_frame(po, rb, status);
1132
1133        return __prb_previous_block(po, rb, status);
1134}
1135
1136static void packet_increment_rx_head(struct packet_sock *po,
1137                                            struct packet_ring_buffer *rb)
1138{
1139        switch (po->tp_version) {
1140        case TPACKET_V1:
1141        case TPACKET_V2:
1142                return packet_increment_head(rb);
1143        case TPACKET_V3:
1144        default:
1145                WARN(1, "TPACKET version not supported.\n");
1146                BUG();
1147                return;
1148        }
1149}
1150
1151static void *packet_previous_frame(struct packet_sock *po,
1152                struct packet_ring_buffer *rb,
1153                int status)
1154{
1155        unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1156        return packet_lookup_frame(po, rb, previous, status);
1157}
1158
1159static void packet_increment_head(struct packet_ring_buffer *buff)
1160{
1161        buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1162}
1163
1164static void packet_inc_pending(struct packet_ring_buffer *rb)
1165{
1166        this_cpu_inc(*rb->pending_refcnt);
1167}
1168
1169static void packet_dec_pending(struct packet_ring_buffer *rb)
1170{
1171        this_cpu_dec(*rb->pending_refcnt);
1172}
1173
1174static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1175{
1176        unsigned int refcnt = 0;
1177        int cpu;
1178
1179        /* We don't use pending refcount in rx_ring. */
1180        if (rb->pending_refcnt == NULL)
1181                return 0;
1182
1183        for_each_possible_cpu(cpu)
1184                refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1185
1186        return refcnt;
1187}
1188
1189static int packet_alloc_pending(struct packet_sock *po)
1190{
1191        po->rx_ring.pending_refcnt = NULL;
1192
1193        po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1194        if (unlikely(po->tx_ring.pending_refcnt == NULL))
1195                return -ENOBUFS;
1196
1197        return 0;
1198}
1199
1200static void packet_free_pending(struct packet_sock *po)
1201{
1202        free_percpu(po->tx_ring.pending_refcnt);
1203}
1204
1205#define ROOM_POW_OFF    2
1206#define ROOM_NONE       0x0
1207#define ROOM_LOW        0x1
1208#define ROOM_NORMAL     0x2
1209
1210static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1211{
1212        int idx, len;
1213
1214        len = READ_ONCE(po->rx_ring.frame_max) + 1;
1215        idx = READ_ONCE(po->rx_ring.head);
1216        if (pow_off)
1217                idx += len >> pow_off;
1218        if (idx >= len)
1219                idx -= len;
1220        return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1221}
1222
1223static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1224{
1225        int idx, len;
1226
1227        len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1228        idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1229        if (pow_off)
1230                idx += len >> pow_off;
1231        if (idx >= len)
1232                idx -= len;
1233        return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1234}
1235
1236static int __packet_rcv_has_room(const struct packet_sock *po,
1237                                 const struct sk_buff *skb)
1238{
1239        const struct sock *sk = &po->sk;
1240        int ret = ROOM_NONE;
1241
1242        if (po->prot_hook.func != tpacket_rcv) {
1243                int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1244                int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1245                                   - (skb ? skb->truesize : 0);
1246
1247                if (avail > (rcvbuf >> ROOM_POW_OFF))
1248                        return ROOM_NORMAL;
1249                else if (avail > 0)
1250                        return ROOM_LOW;
1251                else
1252                        return ROOM_NONE;
1253        }
1254
1255        if (po->tp_version == TPACKET_V3) {
1256                if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1257                        ret = ROOM_NORMAL;
1258                else if (__tpacket_v3_has_room(po, 0))
1259                        ret = ROOM_LOW;
1260        } else {
1261                if (__tpacket_has_room(po, ROOM_POW_OFF))
1262                        ret = ROOM_NORMAL;
1263                else if (__tpacket_has_room(po, 0))
1264                        ret = ROOM_LOW;
1265        }
1266
1267        return ret;
1268}
1269
1270static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1271{
1272        int pressure, ret;
1273
1274        ret = __packet_rcv_has_room(po, skb);
1275        pressure = ret != ROOM_NORMAL;
1276
1277        if (READ_ONCE(po->pressure) != pressure)
1278                WRITE_ONCE(po->pressure, pressure);
1279
1280        return ret;
1281}
1282
1283static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1284{
1285        if (READ_ONCE(po->pressure) &&
1286            __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1287                WRITE_ONCE(po->pressure,  0);
1288}
1289
1290static void packet_sock_destruct(struct sock *sk)
1291{
1292        skb_queue_purge(&sk->sk_error_queue);
1293
1294        WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1295        WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1296
1297        if (!sock_flag(sk, SOCK_DEAD)) {
1298                pr_err("Attempt to release alive packet socket: %p\n", sk);
1299                return;
1300        }
1301
1302        sk_refcnt_debug_dec(sk);
1303}
1304
1305static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1306{
1307        u32 *history = po->rollover->history;
1308        u32 victim, rxhash;
1309        int i, count = 0;
1310
1311        rxhash = skb_get_hash(skb);
1312        for (i = 0; i < ROLLOVER_HLEN; i++)
1313                if (READ_ONCE(history[i]) == rxhash)
1314                        count++;
1315
1316        victim = prandom_u32() % ROLLOVER_HLEN;
1317
1318        /* Avoid dirtying the cache line if possible */
1319        if (READ_ONCE(history[victim]) != rxhash)
1320                WRITE_ONCE(history[victim], rxhash);
1321
1322        return count > (ROLLOVER_HLEN >> 1);
1323}
1324
1325static unsigned int fanout_demux_hash(struct packet_fanout *f,
1326                                      struct sk_buff *skb,
1327                                      unsigned int num)
1328{
1329        return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1330}
1331
1332static unsigned int fanout_demux_lb(struct packet_fanout *f,
1333                                    struct sk_buff *skb,
1334                                    unsigned int num)
1335{
1336        unsigned int val = atomic_inc_return(&f->rr_cur);
1337
1338        return val % num;
1339}
1340
1341static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1342                                     struct sk_buff *skb,
1343                                     unsigned int num)
1344{
1345        return smp_processor_id() % num;
1346}
1347
1348static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1349                                     struct sk_buff *skb,
1350                                     unsigned int num)
1351{
1352        return prandom_u32_max(num);
1353}
1354
1355static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1356                                          struct sk_buff *skb,
1357                                          unsigned int idx, bool try_self,
1358                                          unsigned int num)
1359{
1360        struct packet_sock *po, *po_next, *po_skip = NULL;
1361        unsigned int i, j, room = ROOM_NONE;
1362
1363        po = pkt_sk(rcu_dereference(f->arr[idx]));
1364
1365        if (try_self) {
1366                room = packet_rcv_has_room(po, skb);
1367                if (room == ROOM_NORMAL ||
1368                    (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1369                        return idx;
1370                po_skip = po;
1371        }
1372
1373        i = j = min_t(int, po->rollover->sock, num - 1);
1374        do {
1375                po_next = pkt_sk(rcu_dereference(f->arr[i]));
1376                if (po_next != po_skip && !READ_ONCE(po_next->pressure) &&
1377                    packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1378                        if (i != j)
1379                                po->rollover->sock = i;
1380                        atomic_long_inc(&po->rollover->num);
1381                        if (room == ROOM_LOW)
1382                                atomic_long_inc(&po->rollover->num_huge);
1383                        return i;
1384                }
1385
1386                if (++i == num)
1387                        i = 0;
1388        } while (i != j);
1389
1390        atomic_long_inc(&po->rollover->num_failed);
1391        return idx;
1392}
1393
1394static unsigned int fanout_demux_qm(struct packet_fanout *f,
1395                                    struct sk_buff *skb,
1396                                    unsigned int num)
1397{
1398        return skb_get_queue_mapping(skb) % num;
1399}
1400
1401static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1402                                     struct sk_buff *skb,
1403                                     unsigned int num)
1404{
1405        struct bpf_prog *prog;
1406        unsigned int ret = 0;
1407
1408        rcu_read_lock();
1409        prog = rcu_dereference(f->bpf_prog);
1410        if (prog)
1411                ret = bpf_prog_run_clear_cb(prog, skb) % num;
1412        rcu_read_unlock();
1413
1414        return ret;
1415}
1416
1417static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1418{
1419        return f->flags & (flag >> 8);
1420}
1421
1422static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1423                             struct packet_type *pt, struct net_device *orig_dev)
1424{
1425        struct packet_fanout *f = pt->af_packet_priv;
1426        unsigned int num = READ_ONCE(f->num_members);
1427        struct net *net = read_pnet(&f->net);
1428        struct packet_sock *po;
1429        unsigned int idx;
1430
1431        if (!net_eq(dev_net(dev), net) || !num) {
1432                kfree_skb(skb);
1433                return 0;
1434        }
1435
1436        if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1437                skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1438                if (!skb)
1439                        return 0;
1440        }
1441        switch (f->type) {
1442        case PACKET_FANOUT_HASH:
1443        default:
1444                idx = fanout_demux_hash(f, skb, num);
1445                break;
1446        case PACKET_FANOUT_LB:
1447                idx = fanout_demux_lb(f, skb, num);
1448                break;
1449        case PACKET_FANOUT_CPU:
1450                idx = fanout_demux_cpu(f, skb, num);
1451                break;
1452        case PACKET_FANOUT_RND:
1453                idx = fanout_demux_rnd(f, skb, num);
1454                break;
1455        case PACKET_FANOUT_QM:
1456                idx = fanout_demux_qm(f, skb, num);
1457                break;
1458        case PACKET_FANOUT_ROLLOVER:
1459                idx = fanout_demux_rollover(f, skb, 0, false, num);
1460                break;
1461        case PACKET_FANOUT_CBPF:
1462        case PACKET_FANOUT_EBPF:
1463                idx = fanout_demux_bpf(f, skb, num);
1464                break;
1465        }
1466
1467        if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1468                idx = fanout_demux_rollover(f, skb, idx, true, num);
1469
1470        po = pkt_sk(rcu_dereference(f->arr[idx]));
1471        return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1472}
1473
1474DEFINE_MUTEX(fanout_mutex);
1475EXPORT_SYMBOL_GPL(fanout_mutex);
1476static LIST_HEAD(fanout_list);
1477static u16 fanout_next_id;
1478
1479static void __fanout_link(struct sock *sk, struct packet_sock *po)
1480{
1481        struct packet_fanout *f = po->fanout;
1482
1483        spin_lock(&f->lock);
1484        rcu_assign_pointer(f->arr[f->num_members], sk);
1485        smp_wmb();
1486        f->num_members++;
1487        if (f->num_members == 1)
1488                dev_add_pack(&f->prot_hook);
1489        spin_unlock(&f->lock);
1490}
1491
1492static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1493{
1494        struct packet_fanout *f = po->fanout;
1495        int i;
1496
1497        spin_lock(&f->lock);
1498        for (i = 0; i < f->num_members; i++) {
1499                if (rcu_dereference_protected(f->arr[i],
1500                                              lockdep_is_held(&f->lock)) == sk)
1501                        break;
1502        }
1503        BUG_ON(i >= f->num_members);
1504        rcu_assign_pointer(f->arr[i],
1505                           rcu_dereference_protected(f->arr[f->num_members - 1],
1506                                                     lockdep_is_held(&f->lock)));
1507        f->num_members--;
1508        if (f->num_members == 0)
1509                __dev_remove_pack(&f->prot_hook);
1510        spin_unlock(&f->lock);
1511}
1512
1513static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1514{
1515        if (sk->sk_family != PF_PACKET)
1516                return false;
1517
1518        return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1519}
1520
1521static void fanout_init_data(struct packet_fanout *f)
1522{
1523        switch (f->type) {
1524        case PACKET_FANOUT_LB:
1525                atomic_set(&f->rr_cur, 0);
1526                break;
1527        case PACKET_FANOUT_CBPF:
1528        case PACKET_FANOUT_EBPF:
1529                RCU_INIT_POINTER(f->bpf_prog, NULL);
1530                break;
1531        }
1532}
1533
1534static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1535{
1536        struct bpf_prog *old;
1537
1538        spin_lock(&f->lock);
1539        old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1540        rcu_assign_pointer(f->bpf_prog, new);
1541        spin_unlock(&f->lock);
1542
1543        if (old) {
1544                synchronize_net();
1545                bpf_prog_destroy(old);
1546        }
1547}
1548
1549static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data,
1550                                unsigned int len)
1551{
1552        struct bpf_prog *new;
1553        struct sock_fprog fprog;
1554        int ret;
1555
1556        if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1557                return -EPERM;
1558
1559        ret = copy_bpf_fprog_from_user(&fprog, data, len);
1560        if (ret)
1561                return ret;
1562
1563        ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1564        if (ret)
1565                return ret;
1566
1567        __fanout_set_data_bpf(po->fanout, new);
1568        return 0;
1569}
1570
1571static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data,
1572                                unsigned int len)
1573{
1574        struct bpf_prog *new;
1575        u32 fd;
1576
1577        if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1578                return -EPERM;
1579        if (len != sizeof(fd))
1580                return -EINVAL;
1581        if (copy_from_sockptr(&fd, data, len))
1582                return -EFAULT;
1583
1584        new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1585        if (IS_ERR(new))
1586                return PTR_ERR(new);
1587
1588        __fanout_set_data_bpf(po->fanout, new);
1589        return 0;
1590}
1591
1592static int fanout_set_data(struct packet_sock *po, sockptr_t data,
1593                           unsigned int len)
1594{
1595        switch (po->fanout->type) {
1596        case PACKET_FANOUT_CBPF:
1597                return fanout_set_data_cbpf(po, data, len);
1598        case PACKET_FANOUT_EBPF:
1599                return fanout_set_data_ebpf(po, data, len);
1600        default:
1601                return -EINVAL;
1602        }
1603}
1604
1605static void fanout_release_data(struct packet_fanout *f)
1606{
1607        switch (f->type) {
1608        case PACKET_FANOUT_CBPF:
1609        case PACKET_FANOUT_EBPF:
1610                __fanout_set_data_bpf(f, NULL);
1611        }
1612}
1613
1614static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1615{
1616        struct packet_fanout *f;
1617
1618        list_for_each_entry(f, &fanout_list, list) {
1619                if (f->id == candidate_id &&
1620                    read_pnet(&f->net) == sock_net(sk)) {
1621                        return false;
1622                }
1623        }
1624        return true;
1625}
1626
1627static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1628{
1629        u16 id = fanout_next_id;
1630
1631        do {
1632                if (__fanout_id_is_free(sk, id)) {
1633                        *new_id = id;
1634                        fanout_next_id = id + 1;
1635                        return true;
1636                }
1637
1638                id++;
1639        } while (id != fanout_next_id);
1640
1641        return false;
1642}
1643
1644static int fanout_add(struct sock *sk, struct fanout_args *args)
1645{
1646        struct packet_rollover *rollover = NULL;
1647        struct packet_sock *po = pkt_sk(sk);
1648        u16 type_flags = args->type_flags;
1649        struct packet_fanout *f, *match;
1650        u8 type = type_flags & 0xff;
1651        u8 flags = type_flags >> 8;
1652        u16 id = args->id;
1653        int err;
1654
1655        switch (type) {
1656        case PACKET_FANOUT_ROLLOVER:
1657                if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1658                        return -EINVAL;
1659                break;
1660        case PACKET_FANOUT_HASH:
1661        case PACKET_FANOUT_LB:
1662        case PACKET_FANOUT_CPU:
1663        case PACKET_FANOUT_RND:
1664        case PACKET_FANOUT_QM:
1665        case PACKET_FANOUT_CBPF:
1666        case PACKET_FANOUT_EBPF:
1667                break;
1668        default:
1669                return -EINVAL;
1670        }
1671
1672        mutex_lock(&fanout_mutex);
1673
1674        err = -EALREADY;
1675        if (po->fanout)
1676                goto out;
1677
1678        if (type == PACKET_FANOUT_ROLLOVER ||
1679            (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1680                err = -ENOMEM;
1681                rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1682                if (!rollover)
1683                        goto out;
1684                atomic_long_set(&rollover->num, 0);
1685                atomic_long_set(&rollover->num_huge, 0);
1686                atomic_long_set(&rollover->num_failed, 0);
1687        }
1688
1689        if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1690                if (id != 0) {
1691                        err = -EINVAL;
1692                        goto out;
1693                }
1694                if (!fanout_find_new_id(sk, &id)) {
1695                        err = -ENOMEM;
1696                        goto out;
1697                }
1698                /* ephemeral flag for the first socket in the group: drop it */
1699                flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1700        }
1701
1702        match = NULL;
1703        list_for_each_entry(f, &fanout_list, list) {
1704                if (f->id == id &&
1705                    read_pnet(&f->net) == sock_net(sk)) {
1706                        match = f;
1707                        break;
1708                }
1709        }
1710        err = -EINVAL;
1711        if (match) {
1712                if (match->flags != flags)
1713                        goto out;
1714                if (args->max_num_members &&
1715                    args->max_num_members != match->max_num_members)
1716                        goto out;
1717        } else {
1718                if (args->max_num_members > PACKET_FANOUT_MAX)
1719                        goto out;
1720                if (!args->max_num_members)
1721                        /* legacy PACKET_FANOUT_MAX */
1722                        args->max_num_members = 256;
1723                err = -ENOMEM;
1724                match = kvzalloc(struct_size(match, arr, args->max_num_members),
1725                                 GFP_KERNEL);
1726                if (!match)
1727                        goto out;
1728                write_pnet(&match->net, sock_net(sk));
1729                match->id = id;
1730                match->type = type;
1731                match->flags = flags;
1732                INIT_LIST_HEAD(&match->list);
1733                spin_lock_init(&match->lock);
1734                refcount_set(&match->sk_ref, 0);
1735                fanout_init_data(match);
1736                match->prot_hook.type = po->prot_hook.type;
1737                match->prot_hook.dev = po->prot_hook.dev;
1738                match->prot_hook.func = packet_rcv_fanout;
1739                match->prot_hook.af_packet_priv = match;
1740                match->prot_hook.id_match = match_fanout_group;
1741                match->max_num_members = args->max_num_members;
1742                list_add(&match->list, &fanout_list);
1743        }
1744        err = -EINVAL;
1745
1746        spin_lock(&po->bind_lock);
1747        if (po->running &&
1748            match->type == type &&
1749            match->prot_hook.type == po->prot_hook.type &&
1750            match->prot_hook.dev == po->prot_hook.dev) {
1751                err = -ENOSPC;
1752                if (refcount_read(&match->sk_ref) < match->max_num_members) {
1753                        __dev_remove_pack(&po->prot_hook);
1754                        po->fanout = match;
1755                        po->rollover = rollover;
1756                        rollover = NULL;
1757                        refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1758                        __fanout_link(sk, po);
1759                        err = 0;
1760                }
1761        }
1762        spin_unlock(&po->bind_lock);
1763
1764        if (err && !refcount_read(&match->sk_ref)) {
1765                list_del(&match->list);
1766                kvfree(match);
1767        }
1768
1769out:
1770        kfree(rollover);
1771        mutex_unlock(&fanout_mutex);
1772        return err;
1773}
1774
1775/* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1776 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1777 * It is the responsibility of the caller to call fanout_release_data() and
1778 * free the returned packet_fanout (after synchronize_net())
1779 */
1780static struct packet_fanout *fanout_release(struct sock *sk)
1781{
1782        struct packet_sock *po = pkt_sk(sk);
1783        struct packet_fanout *f;
1784
1785        mutex_lock(&fanout_mutex);
1786        f = po->fanout;
1787        if (f) {
1788                po->fanout = NULL;
1789
1790                if (refcount_dec_and_test(&f->sk_ref))
1791                        list_del(&f->list);
1792                else
1793                        f = NULL;
1794        }
1795        mutex_unlock(&fanout_mutex);
1796
1797        return f;
1798}
1799
1800static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1801                                          struct sk_buff *skb)
1802{
1803        /* Earlier code assumed this would be a VLAN pkt, double-check
1804         * this now that we have the actual packet in hand. We can only
1805         * do this check on Ethernet devices.
1806         */
1807        if (unlikely(dev->type != ARPHRD_ETHER))
1808                return false;
1809
1810        skb_reset_mac_header(skb);
1811        return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1812}
1813
1814static const struct proto_ops packet_ops;
1815
1816static const struct proto_ops packet_ops_spkt;
1817
1818static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1819                           struct packet_type *pt, struct net_device *orig_dev)
1820{
1821        struct sock *sk;
1822        struct sockaddr_pkt *spkt;
1823
1824        /*
1825         *      When we registered the protocol we saved the socket in the data
1826         *      field for just this event.
1827         */
1828
1829        sk = pt->af_packet_priv;
1830
1831        /*
1832         *      Yank back the headers [hope the device set this
1833         *      right or kerboom...]
1834         *
1835         *      Incoming packets have ll header pulled,
1836         *      push it back.
1837         *
1838         *      For outgoing ones skb->data == skb_mac_header(skb)
1839         *      so that this procedure is noop.
1840         */
1841
1842        if (skb->pkt_type == PACKET_LOOPBACK)
1843                goto out;
1844
1845        if (!net_eq(dev_net(dev), sock_net(sk)))
1846                goto out;
1847
1848        skb = skb_share_check(skb, GFP_ATOMIC);
1849        if (skb == NULL)
1850                goto oom;
1851
1852        /* drop any routing info */
1853        skb_dst_drop(skb);
1854
1855        /* drop conntrack reference */
1856        nf_reset_ct(skb);
1857
1858        spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1859
1860        skb_push(skb, skb->data - skb_mac_header(skb));
1861
1862        /*
1863         *      The SOCK_PACKET socket receives _all_ frames.
1864         */
1865
1866        spkt->spkt_family = dev->type;
1867        strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1868        spkt->spkt_protocol = skb->protocol;
1869
1870        /*
1871         *      Charge the memory to the socket. This is done specifically
1872         *      to prevent sockets using all the memory up.
1873         */
1874
1875        if (sock_queue_rcv_skb(sk, skb) == 0)
1876                return 0;
1877
1878out:
1879        kfree_skb(skb);
1880oom:
1881        return 0;
1882}
1883
1884static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1885{
1886        if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1887            sock->type == SOCK_RAW) {
1888                skb_reset_mac_header(skb);
1889                skb->protocol = dev_parse_header_protocol(skb);
1890        }
1891
1892        skb_probe_transport_header(skb);
1893}
1894
1895/*
1896 *      Output a raw packet to a device layer. This bypasses all the other
1897 *      protocol layers and you must therefore supply it with a complete frame
1898 */
1899
1900static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1901                               size_t len)
1902{
1903        struct sock *sk = sock->sk;
1904        DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1905        struct sk_buff *skb = NULL;
1906        struct net_device *dev;
1907        struct sockcm_cookie sockc;
1908        __be16 proto = 0;
1909        int err;
1910        int extra_len = 0;
1911
1912        /*
1913         *      Get and verify the address.
1914         */
1915
1916        if (saddr) {
1917                if (msg->msg_namelen < sizeof(struct sockaddr))
1918                        return -EINVAL;
1919                if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1920                        proto = saddr->spkt_protocol;
1921        } else
1922                return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1923
1924        /*
1925         *      Find the device first to size check it
1926         */
1927
1928        saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1929retry:
1930        rcu_read_lock();
1931        dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1932        err = -ENODEV;
1933        if (dev == NULL)
1934                goto out_unlock;
1935
1936        err = -ENETDOWN;
1937        if (!(dev->flags & IFF_UP))
1938                goto out_unlock;
1939
1940        /*
1941         * You may not queue a frame bigger than the mtu. This is the lowest level
1942         * raw protocol and you must do your own fragmentation at this level.
1943         */
1944
1945        if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1946                if (!netif_supports_nofcs(dev)) {
1947                        err = -EPROTONOSUPPORT;
1948                        goto out_unlock;
1949                }
1950                extra_len = 4; /* We're doing our own CRC */
1951        }
1952
1953        err = -EMSGSIZE;
1954        if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1955                goto out_unlock;
1956
1957        if (!skb) {
1958                size_t reserved = LL_RESERVED_SPACE(dev);
1959                int tlen = dev->needed_tailroom;
1960                unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1961
1962                rcu_read_unlock();
1963                skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1964                if (skb == NULL)
1965                        return -ENOBUFS;
1966                /* FIXME: Save some space for broken drivers that write a hard
1967                 * header at transmission time by themselves. PPP is the notable
1968                 * one here. This should really be fixed at the driver level.
1969                 */
1970                skb_reserve(skb, reserved);
1971                skb_reset_network_header(skb);
1972
1973                /* Try to align data part correctly */
1974                if (hhlen) {
1975                        skb->data -= hhlen;
1976                        skb->tail -= hhlen;
1977                        if (len < hhlen)
1978                                skb_reset_network_header(skb);
1979                }
1980                err = memcpy_from_msg(skb_put(skb, len), msg, len);
1981                if (err)
1982                        goto out_free;
1983                goto retry;
1984        }
1985
1986        if (!dev_validate_header(dev, skb->data, len)) {
1987                err = -EINVAL;
1988                goto out_unlock;
1989        }
1990        if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1991            !packet_extra_vlan_len_allowed(dev, skb)) {
1992                err = -EMSGSIZE;
1993                goto out_unlock;
1994        }
1995
1996        sockcm_init(&sockc, sk);
1997        if (msg->msg_controllen) {
1998                err = sock_cmsg_send(sk, msg, &sockc);
1999                if (unlikely(err))
2000                        goto out_unlock;
2001        }
2002
2003        skb->protocol = proto;
2004        skb->dev = dev;
2005        skb->priority = sk->sk_priority;
2006        skb->mark = sk->sk_mark;
2007        skb->tstamp = sockc.transmit_time;
2008
2009        skb_setup_tx_timestamp(skb, sockc.tsflags);
2010
2011        if (unlikely(extra_len == 4))
2012                skb->no_fcs = 1;
2013
2014        packet_parse_headers(skb, sock);
2015
2016        dev_queue_xmit(skb);
2017        rcu_read_unlock();
2018        return len;
2019
2020out_unlock:
2021        rcu_read_unlock();
2022out_free:
2023        kfree_skb(skb);
2024        return err;
2025}
2026
2027static unsigned int run_filter(struct sk_buff *skb,
2028                               const struct sock *sk,
2029                               unsigned int res)
2030{
2031        struct sk_filter *filter;
2032
2033        rcu_read_lock();
2034        filter = rcu_dereference(sk->sk_filter);
2035        if (filter != NULL)
2036                res = bpf_prog_run_clear_cb(filter->prog, skb);
2037        rcu_read_unlock();
2038
2039        return res;
2040}
2041
2042static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2043                           size_t *len)
2044{
2045        struct virtio_net_hdr vnet_hdr;
2046
2047        if (*len < sizeof(vnet_hdr))
2048                return -EINVAL;
2049        *len -= sizeof(vnet_hdr);
2050
2051        if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2052                return -EINVAL;
2053
2054        return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2055}
2056
2057/*
2058 * This function makes lazy skb cloning in hope that most of packets
2059 * are discarded by BPF.
2060 *
2061 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2062 * and skb->cb are mangled. It works because (and until) packets
2063 * falling here are owned by current CPU. Output packets are cloned
2064 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2065 * sequentially, so that if we return skb to original state on exit,
2066 * we will not harm anyone.
2067 */
2068
2069static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2070                      struct packet_type *pt, struct net_device *orig_dev)
2071{
2072        struct sock *sk;
2073        struct sockaddr_ll *sll;
2074        struct packet_sock *po;
2075        u8 *skb_head = skb->data;
2076        int skb_len = skb->len;
2077        unsigned int snaplen, res;
2078        bool is_drop_n_account = false;
2079
2080        if (skb->pkt_type == PACKET_LOOPBACK)
2081                goto drop;
2082
2083        sk = pt->af_packet_priv;
2084        po = pkt_sk(sk);
2085
2086        if (!net_eq(dev_net(dev), sock_net(sk)))
2087                goto drop;
2088
2089        skb->dev = dev;
2090
2091        if (dev_has_header(dev)) {
2092                /* The device has an explicit notion of ll header,
2093                 * exported to higher levels.
2094                 *
2095                 * Otherwise, the device hides details of its frame
2096                 * structure, so that corresponding packet head is
2097                 * never delivered to user.
2098                 */
2099                if (sk->sk_type != SOCK_DGRAM)
2100                        skb_push(skb, skb->data - skb_mac_header(skb));
2101                else if (skb->pkt_type == PACKET_OUTGOING) {
2102                        /* Special case: outgoing packets have ll header at head */
2103                        skb_pull(skb, skb_network_offset(skb));
2104                }
2105        }
2106
2107        snaplen = skb->len;
2108
2109        res = run_filter(skb, sk, snaplen);
2110        if (!res)
2111                goto drop_n_restore;
2112        if (snaplen > res)
2113                snaplen = res;
2114
2115        if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2116                goto drop_n_acct;
2117
2118        if (skb_shared(skb)) {
2119                struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2120                if (nskb == NULL)
2121                        goto drop_n_acct;
2122
2123                if (skb_head != skb->data) {
2124                        skb->data = skb_head;
2125                        skb->len = skb_len;
2126                }
2127                consume_skb(skb);
2128                skb = nskb;
2129        }
2130
2131        sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2132
2133        sll = &PACKET_SKB_CB(skb)->sa.ll;
2134        sll->sll_hatype = dev->type;
2135        sll->sll_pkttype = skb->pkt_type;
2136        if (unlikely(po->origdev))
2137                sll->sll_ifindex = orig_dev->ifindex;
2138        else
2139                sll->sll_ifindex = dev->ifindex;
2140
2141        sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2142
2143        /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2144         * Use their space for storing the original skb length.
2145         */
2146        PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2147
2148        if (pskb_trim(skb, snaplen))
2149                goto drop_n_acct;
2150
2151        skb_set_owner_r(skb, sk);
2152        skb->dev = NULL;
2153        skb_dst_drop(skb);
2154
2155        /* drop conntrack reference */
2156        nf_reset_ct(skb);
2157
2158        spin_lock(&sk->sk_receive_queue.lock);
2159        po->stats.stats1.tp_packets++;
2160        sock_skb_set_dropcount(sk, skb);
2161        __skb_queue_tail(&sk->sk_receive_queue, skb);
2162        spin_unlock(&sk->sk_receive_queue.lock);
2163        sk->sk_data_ready(sk);
2164        return 0;
2165
2166drop_n_acct:
2167        is_drop_n_account = true;
2168        atomic_inc(&po->tp_drops);
2169        atomic_inc(&sk->sk_drops);
2170
2171drop_n_restore:
2172        if (skb_head != skb->data && skb_shared(skb)) {
2173                skb->data = skb_head;
2174                skb->len = skb_len;
2175        }
2176drop:
2177        if (!is_drop_n_account)
2178                consume_skb(skb);
2179        else
2180                kfree_skb(skb);
2181        return 0;
2182}
2183
2184static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2185                       struct packet_type *pt, struct net_device *orig_dev)
2186{
2187        struct sock *sk;
2188        struct packet_sock *po;
2189        struct sockaddr_ll *sll;
2190        union tpacket_uhdr h;
2191        u8 *skb_head = skb->data;
2192        int skb_len = skb->len;
2193        unsigned int snaplen, res;
2194        unsigned long status = TP_STATUS_USER;
2195        unsigned short macoff, hdrlen;
2196        unsigned int netoff;
2197        struct sk_buff *copy_skb = NULL;
2198        struct timespec64 ts;
2199        __u32 ts_status;
2200        bool is_drop_n_account = false;
2201        unsigned int slot_id = 0;
2202        bool do_vnet = false;
2203
2204        /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2205         * We may add members to them until current aligned size without forcing
2206         * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2207         */
2208        BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2209        BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2210
2211        if (skb->pkt_type == PACKET_LOOPBACK)
2212                goto drop;
2213
2214        sk = pt->af_packet_priv;
2215        po = pkt_sk(sk);
2216
2217        if (!net_eq(dev_net(dev), sock_net(sk)))
2218                goto drop;
2219
2220        if (dev_has_header(dev)) {
2221                if (sk->sk_type != SOCK_DGRAM)
2222                        skb_push(skb, skb->data - skb_mac_header(skb));
2223                else if (skb->pkt_type == PACKET_OUTGOING) {
2224                        /* Special case: outgoing packets have ll header at head */
2225                        skb_pull(skb, skb_network_offset(skb));
2226                }
2227        }
2228
2229        snaplen = skb->len;
2230
2231        res = run_filter(skb, sk, snaplen);
2232        if (!res)
2233                goto drop_n_restore;
2234
2235        /* If we are flooded, just give up */
2236        if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2237                atomic_inc(&po->tp_drops);
2238                goto drop_n_restore;
2239        }
2240
2241        if (skb->ip_summed == CHECKSUM_PARTIAL)
2242                status |= TP_STATUS_CSUMNOTREADY;
2243        else if (skb->pkt_type != PACKET_OUTGOING &&
2244                 (skb->ip_summed == CHECKSUM_COMPLETE ||
2245                  skb_csum_unnecessary(skb)))
2246                status |= TP_STATUS_CSUM_VALID;
2247
2248        if (snaplen > res)
2249                snaplen = res;
2250
2251        if (sk->sk_type == SOCK_DGRAM) {
2252                macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2253                                  po->tp_reserve;
2254        } else {
2255                unsigned int maclen = skb_network_offset(skb);
2256                netoff = TPACKET_ALIGN(po->tp_hdrlen +
2257                                       (maclen < 16 ? 16 : maclen)) +
2258                                       po->tp_reserve;
2259                if (po->has_vnet_hdr) {
2260                        netoff += sizeof(struct virtio_net_hdr);
2261                        do_vnet = true;
2262                }
2263                macoff = netoff - maclen;
2264        }
2265        if (netoff > USHRT_MAX) {
2266                atomic_inc(&po->tp_drops);
2267                goto drop_n_restore;
2268        }
2269        if (po->tp_version <= TPACKET_V2) {
2270                if (macoff + snaplen > po->rx_ring.frame_size) {
2271                        if (po->copy_thresh &&
2272                            atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2273                                if (skb_shared(skb)) {
2274                                        copy_skb = skb_clone(skb, GFP_ATOMIC);
2275                                } else {
2276                                        copy_skb = skb_get(skb);
2277                                        skb_head = skb->data;
2278                                }
2279                                if (copy_skb)
2280                                        skb_set_owner_r(copy_skb, sk);
2281                        }
2282                        snaplen = po->rx_ring.frame_size - macoff;
2283                        if ((int)snaplen < 0) {
2284                                snaplen = 0;
2285                                do_vnet = false;
2286                        }
2287                }
2288        } else if (unlikely(macoff + snaplen >
2289                            GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2290                u32 nval;
2291
2292                nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2293                pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2294                            snaplen, nval, macoff);
2295                snaplen = nval;
2296                if (unlikely((int)snaplen < 0)) {
2297                        snaplen = 0;
2298                        macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2299                        do_vnet = false;
2300                }
2301        }
2302        spin_lock(&sk->sk_receive_queue.lock);
2303        h.raw = packet_current_rx_frame(po, skb,
2304                                        TP_STATUS_KERNEL, (macoff+snaplen));
2305        if (!h.raw)
2306                goto drop_n_account;
2307
2308        if (po->tp_version <= TPACKET_V2) {
2309                slot_id = po->rx_ring.head;
2310                if (test_bit(slot_id, po->rx_ring.rx_owner_map))
2311                        goto drop_n_account;
2312                __set_bit(slot_id, po->rx_ring.rx_owner_map);
2313        }
2314
2315        if (do_vnet &&
2316            virtio_net_hdr_from_skb(skb, h.raw + macoff -
2317                                    sizeof(struct virtio_net_hdr),
2318                                    vio_le(), true, 0)) {
2319                if (po->tp_version == TPACKET_V3)
2320                        prb_clear_blk_fill_status(&po->rx_ring);
2321                goto drop_n_account;
2322        }
2323
2324        if (po->tp_version <= TPACKET_V2) {
2325                packet_increment_rx_head(po, &po->rx_ring);
2326        /*
2327         * LOSING will be reported till you read the stats,
2328         * because it's COR - Clear On Read.
2329         * Anyways, moving it for V1/V2 only as V3 doesn't need this
2330         * at packet level.
2331         */
2332                if (atomic_read(&po->tp_drops))
2333                        status |= TP_STATUS_LOSING;
2334        }
2335
2336        po->stats.stats1.tp_packets++;
2337        if (copy_skb) {
2338                status |= TP_STATUS_COPY;
2339                __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2340        }
2341        spin_unlock(&sk->sk_receive_queue.lock);
2342
2343        skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2344
2345        /* Always timestamp; prefer an existing software timestamp taken
2346         * closer to the time of capture.
2347         */
2348        ts_status = tpacket_get_timestamp(skb, &ts,
2349                                          po->tp_tstamp | SOF_TIMESTAMPING_SOFTWARE);
2350        if (!ts_status)
2351                ktime_get_real_ts64(&ts);
2352
2353        status |= ts_status;
2354
2355        switch (po->tp_version) {
2356        case TPACKET_V1:
2357                h.h1->tp_len = skb->len;
2358                h.h1->tp_snaplen = snaplen;
2359                h.h1->tp_mac = macoff;
2360                h.h1->tp_net = netoff;
2361                h.h1->tp_sec = ts.tv_sec;
2362                h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2363                hdrlen = sizeof(*h.h1);
2364                break;
2365        case TPACKET_V2:
2366                h.h2->tp_len = skb->len;
2367                h.h2->tp_snaplen = snaplen;
2368                h.h2->tp_mac = macoff;
2369                h.h2->tp_net = netoff;
2370                h.h2->tp_sec = ts.tv_sec;
2371                h.h2->tp_nsec = ts.tv_nsec;
2372                if (skb_vlan_tag_present(skb)) {
2373                        h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2374                        h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2375                        status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2376                } else {
2377                        h.h2->tp_vlan_tci = 0;
2378                        h.h2->tp_vlan_tpid = 0;
2379                }
2380                memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2381                hdrlen = sizeof(*h.h2);
2382                break;
2383        case TPACKET_V3:
2384                /* tp_nxt_offset,vlan are already populated above.
2385                 * So DONT clear those fields here
2386                 */
2387                h.h3->tp_status |= status;
2388                h.h3->tp_len = skb->len;
2389                h.h3->tp_snaplen = snaplen;
2390                h.h3->tp_mac = macoff;
2391                h.h3->tp_net = netoff;
2392                h.h3->tp_sec  = ts.tv_sec;
2393                h.h3->tp_nsec = ts.tv_nsec;
2394                memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2395                hdrlen = sizeof(*h.h3);
2396                break;
2397        default:
2398                BUG();
2399        }
2400
2401        sll = h.raw + TPACKET_ALIGN(hdrlen);
2402        sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2403        sll->sll_family = AF_PACKET;
2404        sll->sll_hatype = dev->type;
2405        sll->sll_protocol = skb->protocol;
2406        sll->sll_pkttype = skb->pkt_type;
2407        if (unlikely(po->origdev))
2408                sll->sll_ifindex = orig_dev->ifindex;
2409        else
2410                sll->sll_ifindex = dev->ifindex;
2411
2412        smp_mb();
2413
2414#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2415        if (po->tp_version <= TPACKET_V2) {
2416                u8 *start, *end;
2417
2418                end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2419                                        macoff + snaplen);
2420
2421                for (start = h.raw; start < end; start += PAGE_SIZE)
2422                        flush_dcache_page(pgv_to_page(start));
2423        }
2424        smp_wmb();
2425#endif
2426
2427        if (po->tp_version <= TPACKET_V2) {
2428                spin_lock(&sk->sk_receive_queue.lock);
2429                __packet_set_status(po, h.raw, status);
2430                __clear_bit(slot_id, po->rx_ring.rx_owner_map);
2431                spin_unlock(&sk->sk_receive_queue.lock);
2432                sk->sk_data_ready(sk);
2433        } else if (po->tp_version == TPACKET_V3) {
2434                prb_clear_blk_fill_status(&po->rx_ring);
2435        }
2436
2437drop_n_restore:
2438        if (skb_head != skb->data && skb_shared(skb)) {
2439                skb->data = skb_head;
2440                skb->len = skb_len;
2441        }
2442drop:
2443        if (!is_drop_n_account)
2444                consume_skb(skb);
2445        else
2446                kfree_skb(skb);
2447        return 0;
2448
2449drop_n_account:
2450        spin_unlock(&sk->sk_receive_queue.lock);
2451        atomic_inc(&po->tp_drops);
2452        is_drop_n_account = true;
2453
2454        sk->sk_data_ready(sk);
2455        kfree_skb(copy_skb);
2456        goto drop_n_restore;
2457}
2458
2459static void tpacket_destruct_skb(struct sk_buff *skb)
2460{
2461        struct packet_sock *po = pkt_sk(skb->sk);
2462
2463        if (likely(po->tx_ring.pg_vec)) {
2464                void *ph;
2465                __u32 ts;
2466
2467                ph = skb_zcopy_get_nouarg(skb);
2468                packet_dec_pending(&po->tx_ring);
2469
2470                ts = __packet_set_timestamp(po, ph, skb);
2471                __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2472
2473                if (!packet_read_pending(&po->tx_ring))
2474                        complete(&po->skb_completion);
2475        }
2476
2477        sock_wfree(skb);
2478}
2479
2480static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2481{
2482        if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2483            (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2484             __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2485              __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2486                vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2487                         __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2488                        __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2489
2490        if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2491                return -EINVAL;
2492
2493        return 0;
2494}
2495
2496static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2497                                 struct virtio_net_hdr *vnet_hdr)
2498{
2499        if (*len < sizeof(*vnet_hdr))
2500                return -EINVAL;
2501        *len -= sizeof(*vnet_hdr);
2502
2503        if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2504                return -EFAULT;
2505
2506        return __packet_snd_vnet_parse(vnet_hdr, *len);
2507}
2508
2509static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2510                void *frame, struct net_device *dev, void *data, int tp_len,
2511                __be16 proto, unsigned char *addr, int hlen, int copylen,
2512                const struct sockcm_cookie *sockc)
2513{
2514        union tpacket_uhdr ph;
2515        int to_write, offset, len, nr_frags, len_max;
2516        struct socket *sock = po->sk.sk_socket;
2517        struct page *page;
2518        int err;
2519
2520        ph.raw = frame;
2521
2522        skb->protocol = proto;
2523        skb->dev = dev;
2524        skb->priority = po->sk.sk_priority;
2525        skb->mark = po->sk.sk_mark;
2526        skb->tstamp = sockc->transmit_time;
2527        skb_setup_tx_timestamp(skb, sockc->tsflags);
2528        skb_zcopy_set_nouarg(skb, ph.raw);
2529
2530        skb_reserve(skb, hlen);
2531        skb_reset_network_header(skb);
2532
2533        to_write = tp_len;
2534
2535        if (sock->type == SOCK_DGRAM) {
2536                err = dev_hard_header(skb, dev, ntohs(proto), addr,
2537                                NULL, tp_len);
2538                if (unlikely(err < 0))
2539                        return -EINVAL;
2540        } else if (copylen) {
2541                int hdrlen = min_t(int, copylen, tp_len);
2542
2543                skb_push(skb, dev->hard_header_len);
2544                skb_put(skb, copylen - dev->hard_header_len);
2545                err = skb_store_bits(skb, 0, data, hdrlen);
2546                if (unlikely(err))
2547                        return err;
2548                if (!dev_validate_header(dev, skb->data, hdrlen))
2549                        return -EINVAL;
2550
2551                data += hdrlen;
2552                to_write -= hdrlen;
2553        }
2554
2555        offset = offset_in_page(data);
2556        len_max = PAGE_SIZE - offset;
2557        len = ((to_write > len_max) ? len_max : to_write);
2558
2559        skb->data_len = to_write;
2560        skb->len += to_write;
2561        skb->truesize += to_write;
2562        refcount_add(to_write, &po->sk.sk_wmem_alloc);
2563
2564        while (likely(to_write)) {
2565                nr_frags = skb_shinfo(skb)->nr_frags;
2566
2567                if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2568                        pr_err("Packet exceed the number of skb frags(%lu)\n",
2569                               MAX_SKB_FRAGS);
2570                        return -EFAULT;
2571                }
2572
2573                page = pgv_to_page(data);
2574                data += len;
2575                flush_dcache_page(page);
2576                get_page(page);
2577                skb_fill_page_desc(skb, nr_frags, page, offset, len);
2578                to_write -= len;
2579                offset = 0;
2580                len_max = PAGE_SIZE;
2581                len = ((to_write > len_max) ? len_max : to_write);
2582        }
2583
2584        packet_parse_headers(skb, sock);
2585
2586        return tp_len;
2587}
2588
2589static int tpacket_parse_header(struct packet_sock *po, void *frame,
2590                                int size_max, void **data)
2591{
2592        union tpacket_uhdr ph;
2593        int tp_len, off;
2594
2595        ph.raw = frame;
2596
2597        switch (po->tp_version) {
2598        case TPACKET_V3:
2599                if (ph.h3->tp_next_offset != 0) {
2600                        pr_warn_once("variable sized slot not supported");
2601                        return -EINVAL;
2602                }
2603                tp_len = ph.h3->tp_len;
2604                break;
2605        case TPACKET_V2:
2606                tp_len = ph.h2->tp_len;
2607                break;
2608        default:
2609                tp_len = ph.h1->tp_len;
2610                break;
2611        }
2612        if (unlikely(tp_len > size_max)) {
2613                pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2614                return -EMSGSIZE;
2615        }
2616
2617        if (unlikely(po->tp_tx_has_off)) {
2618                int off_min, off_max;
2619
2620                off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2621                off_max = po->tx_ring.frame_size - tp_len;
2622                if (po->sk.sk_type == SOCK_DGRAM) {
2623                        switch (po->tp_version) {
2624                        case TPACKET_V3:
2625                                off = ph.h3->tp_net;
2626                                break;
2627                        case TPACKET_V2:
2628                                off = ph.h2->tp_net;
2629                                break;
2630                        default:
2631                                off = ph.h1->tp_net;
2632                                break;
2633                        }
2634                } else {
2635                        switch (po->tp_version) {
2636                        case TPACKET_V3:
2637                                off = ph.h3->tp_mac;
2638                                break;
2639                        case TPACKET_V2:
2640                                off = ph.h2->tp_mac;
2641                                break;
2642                        default:
2643                                off = ph.h1->tp_mac;
2644                                break;
2645                        }
2646                }
2647                if (unlikely((off < off_min) || (off_max < off)))
2648                        return -EINVAL;
2649        } else {
2650                off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2651        }
2652
2653        *data = frame + off;
2654        return tp_len;
2655}
2656
2657static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2658{
2659        struct sk_buff *skb = NULL;
2660        struct net_device *dev;
2661        struct virtio_net_hdr *vnet_hdr = NULL;
2662        struct sockcm_cookie sockc;
2663        __be16 proto;
2664        int err, reserve = 0;
2665        void *ph;
2666        DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2667        bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2668        unsigned char *addr = NULL;
2669        int tp_len, size_max;
2670        void *data;
2671        int len_sum = 0;
2672        int status = TP_STATUS_AVAILABLE;
2673        int hlen, tlen, copylen = 0;
2674        long timeo = 0;
2675
2676        mutex_lock(&po->pg_vec_lock);
2677
2678        /* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2679         * we need to confirm it under protection of pg_vec_lock.
2680         */
2681        if (unlikely(!po->tx_ring.pg_vec)) {
2682                err = -EBUSY;
2683                goto out;
2684        }
2685        if (likely(saddr == NULL)) {
2686                dev     = packet_cached_dev_get(po);
2687                proto   = READ_ONCE(po->num);
2688        } else {
2689                err = -EINVAL;
2690                if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2691                        goto out;
2692                if (msg->msg_namelen < (saddr->sll_halen
2693                                        + offsetof(struct sockaddr_ll,
2694                                                sll_addr)))
2695                        goto out;
2696                proto   = saddr->sll_protocol;
2697                dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2698                if (po->sk.sk_socket->type == SOCK_DGRAM) {
2699                        if (dev && msg->msg_namelen < dev->addr_len +
2700                                   offsetof(struct sockaddr_ll, sll_addr))
2701                                goto out_put;
2702                        addr = saddr->sll_addr;
2703                }
2704        }
2705
2706        err = -ENXIO;
2707        if (unlikely(dev == NULL))
2708                goto out;
2709        err = -ENETDOWN;
2710        if (unlikely(!(dev->flags & IFF_UP)))
2711                goto out_put;
2712
2713        sockcm_init(&sockc, &po->sk);
2714        if (msg->msg_controllen) {
2715                err = sock_cmsg_send(&po->sk, msg, &sockc);
2716                if (unlikely(err))
2717                        goto out_put;
2718        }
2719
2720        if (po->sk.sk_socket->type == SOCK_RAW)
2721                reserve = dev->hard_header_len;
2722        size_max = po->tx_ring.frame_size
2723                - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2724
2725        if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2726                size_max = dev->mtu + reserve + VLAN_HLEN;
2727
2728        reinit_completion(&po->skb_completion);
2729
2730        do {
2731                ph = packet_current_frame(po, &po->tx_ring,
2732                                          TP_STATUS_SEND_REQUEST);
2733                if (unlikely(ph == NULL)) {
2734                        if (need_wait && skb) {
2735                                timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2736                                timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2737                                if (timeo <= 0) {
2738                                        err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2739                                        goto out_put;
2740                                }
2741                        }
2742                        /* check for additional frames */
2743                        continue;
2744                }
2745
2746                skb = NULL;
2747                tp_len = tpacket_parse_header(po, ph, size_max, &data);
2748                if (tp_len < 0)
2749                        goto tpacket_error;
2750
2751                status = TP_STATUS_SEND_REQUEST;
2752                hlen = LL_RESERVED_SPACE(dev);
2753                tlen = dev->needed_tailroom;
2754                if (po->has_vnet_hdr) {
2755                        vnet_hdr = data;
2756                        data += sizeof(*vnet_hdr);
2757                        tp_len -= sizeof(*vnet_hdr);
2758                        if (tp_len < 0 ||
2759                            __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2760                                tp_len = -EINVAL;
2761                                goto tpacket_error;
2762                        }
2763                        copylen = __virtio16_to_cpu(vio_le(),
2764                                                    vnet_hdr->hdr_len);
2765                }
2766                copylen = max_t(int, copylen, dev->hard_header_len);
2767                skb = sock_alloc_send_skb(&po->sk,
2768                                hlen + tlen + sizeof(struct sockaddr_ll) +
2769                                (copylen - dev->hard_header_len),
2770                                !need_wait, &err);
2771
2772                if (unlikely(skb == NULL)) {
2773                        /* we assume the socket was initially writeable ... */
2774                        if (likely(len_sum > 0))
2775                                err = len_sum;
2776                        goto out_status;
2777                }
2778                tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2779                                          addr, hlen, copylen, &sockc);
2780                if (likely(tp_len >= 0) &&
2781                    tp_len > dev->mtu + reserve &&
2782                    !po->has_vnet_hdr &&
2783                    !packet_extra_vlan_len_allowed(dev, skb))
2784                        tp_len = -EMSGSIZE;
2785
2786                if (unlikely(tp_len < 0)) {
2787tpacket_error:
2788                        if (po->tp_loss) {
2789                                __packet_set_status(po, ph,
2790                                                TP_STATUS_AVAILABLE);
2791                                packet_increment_head(&po->tx_ring);
2792                                kfree_skb(skb);
2793                                continue;
2794                        } else {
2795                                status = TP_STATUS_WRONG_FORMAT;
2796                                err = tp_len;
2797                                goto out_status;
2798                        }
2799                }
2800
2801                if (po->has_vnet_hdr) {
2802                        if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2803                                tp_len = -EINVAL;
2804                                goto tpacket_error;
2805                        }
2806                        virtio_net_hdr_set_proto(skb, vnet_hdr);
2807                }
2808
2809                skb->destructor = tpacket_destruct_skb;
2810                __packet_set_status(po, ph, TP_STATUS_SENDING);
2811                packet_inc_pending(&po->tx_ring);
2812
2813                status = TP_STATUS_SEND_REQUEST;
2814                err = po->xmit(skb);
2815                if (unlikely(err > 0)) {
2816                        err = net_xmit_errno(err);
2817                        if (err && __packet_get_status(po, ph) ==
2818                                   TP_STATUS_AVAILABLE) {
2819                                /* skb was destructed already */
2820                                skb = NULL;
2821                                goto out_status;
2822                        }
2823                        /*
2824                         * skb was dropped but not destructed yet;
2825                         * let's treat it like congestion or err < 0
2826                         */
2827                        err = 0;
2828                }
2829                packet_increment_head(&po->tx_ring);
2830                len_sum += tp_len;
2831        } while (likely((ph != NULL) ||
2832                /* Note: packet_read_pending() might be slow if we have
2833                 * to call it as it's per_cpu variable, but in fast-path
2834                 * we already short-circuit the loop with the first
2835                 * condition, and luckily don't have to go that path
2836                 * anyway.
2837                 */
2838                 (need_wait && packet_read_pending(&po->tx_ring))));
2839
2840        err = len_sum;
2841        goto out_put;
2842
2843out_status:
2844        __packet_set_status(po, ph, status);
2845        kfree_skb(skb);
2846out_put:
2847        dev_put(dev);
2848out:
2849        mutex_unlock(&po->pg_vec_lock);
2850        return err;
2851}
2852
2853static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2854                                        size_t reserve, size_t len,
2855                                        size_t linear, int noblock,
2856                                        int *err)
2857{
2858        struct sk_buff *skb;
2859
2860        /* Under a page?  Don't bother with paged skb. */
2861        if (prepad + len < PAGE_SIZE || !linear)
2862                linear = len;
2863
2864        skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2865                                   err, 0);
2866        if (!skb)
2867                return NULL;
2868
2869        skb_reserve(skb, reserve);
2870        skb_put(skb, linear);
2871        skb->data_len = len - linear;
2872        skb->len += len - linear;
2873
2874        return skb;
2875}
2876
2877static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2878{
2879        struct sock *sk = sock->sk;
2880        DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2881        struct sk_buff *skb;
2882        struct net_device *dev;
2883        __be16 proto;
2884        unsigned char *addr = NULL;
2885        int err, reserve = 0;
2886        struct sockcm_cookie sockc;
2887        struct virtio_net_hdr vnet_hdr = { 0 };
2888        int offset = 0;
2889        struct packet_sock *po = pkt_sk(sk);
2890        bool has_vnet_hdr = false;
2891        int hlen, tlen, linear;
2892        int extra_len = 0;
2893
2894        /*
2895         *      Get and verify the address.
2896         */
2897
2898        if (likely(saddr == NULL)) {
2899                dev     = packet_cached_dev_get(po);
2900                proto   = READ_ONCE(po->num);
2901        } else {
2902                err = -EINVAL;
2903                if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2904                        goto out;
2905                if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2906                        goto out;
2907                proto   = saddr->sll_protocol;
2908                dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2909                if (sock->type == SOCK_DGRAM) {
2910                        if (dev && msg->msg_namelen < dev->addr_len +
2911                                   offsetof(struct sockaddr_ll, sll_addr))
2912                                goto out_unlock;
2913                        addr = saddr->sll_addr;
2914                }
2915        }
2916
2917        err = -ENXIO;
2918        if (unlikely(dev == NULL))
2919                goto out_unlock;
2920        err = -ENETDOWN;
2921        if (unlikely(!(dev->flags & IFF_UP)))
2922                goto out_unlock;
2923
2924        sockcm_init(&sockc, sk);
2925        sockc.mark = sk->sk_mark;
2926        if (msg->msg_controllen) {
2927                err = sock_cmsg_send(sk, msg, &sockc);
2928                if (unlikely(err))
2929                        goto out_unlock;
2930        }
2931
2932        if (sock->type == SOCK_RAW)
2933                reserve = dev->hard_header_len;
2934        if (po->has_vnet_hdr) {
2935                err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2936                if (err)
2937                        goto out_unlock;
2938                has_vnet_hdr = true;
2939        }
2940
2941        if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2942                if (!netif_supports_nofcs(dev)) {
2943                        err = -EPROTONOSUPPORT;
2944                        goto out_unlock;
2945                }
2946                extra_len = 4; /* We're doing our own CRC */
2947        }
2948
2949        err = -EMSGSIZE;
2950        if (!vnet_hdr.gso_type &&
2951            (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2952                goto out_unlock;
2953
2954        err = -ENOBUFS;
2955        hlen = LL_RESERVED_SPACE(dev);
2956        tlen = dev->needed_tailroom;
2957        linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2958        linear = max(linear, min_t(int, len, dev->hard_header_len));
2959        skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2960                               msg->msg_flags & MSG_DONTWAIT, &err);
2961        if (skb == NULL)
2962                goto out_unlock;
2963
2964        skb_reset_network_header(skb);
2965
2966        err = -EINVAL;
2967        if (sock->type == SOCK_DGRAM) {
2968                offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2969                if (unlikely(offset < 0))
2970                        goto out_free;
2971        } else if (reserve) {
2972                skb_reserve(skb, -reserve);
2973                if (len < reserve + sizeof(struct ipv6hdr) &&
2974                    dev->min_header_len != dev->hard_header_len)
2975                        skb_reset_network_header(skb);
2976        }
2977
2978        /* Returns -EFAULT on error */
2979        err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2980        if (err)
2981                goto out_free;
2982
2983        if (sock->type == SOCK_RAW &&
2984            !dev_validate_header(dev, skb->data, len)) {
2985                err = -EINVAL;
2986                goto out_free;
2987        }
2988
2989        skb_setup_tx_timestamp(skb, sockc.tsflags);
2990
2991        if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2992            !packet_extra_vlan_len_allowed(dev, skb)) {
2993                err = -EMSGSIZE;
2994                goto out_free;
2995        }
2996
2997        skb->protocol = proto;
2998        skb->dev = dev;
2999        skb->priority = sk->sk_priority;
3000        skb->mark = sockc.mark;
3001        skb->tstamp = sockc.transmit_time;
3002
3003        if (has_vnet_hdr) {
3004                err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
3005                if (err)
3006                        goto out_free;
3007                len += sizeof(vnet_hdr);
3008                virtio_net_hdr_set_proto(skb, &vnet_hdr);
3009        }
3010
3011        packet_parse_headers(skb, sock);
3012
3013        if (unlikely(extra_len == 4))
3014                skb->no_fcs = 1;
3015
3016        err = po->xmit(skb);
3017        if (err > 0 && (err = net_xmit_errno(err)) != 0)
3018                goto out_unlock;
3019
3020        dev_put(dev);
3021
3022        return len;
3023
3024out_free:
3025        kfree_skb(skb);
3026out_unlock:
3027        if (dev)
3028                dev_put(dev);
3029out:
3030        return err;
3031}
3032
3033static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
3034{
3035        struct sock *sk = sock->sk;
3036        struct packet_sock *po = pkt_sk(sk);
3037
3038        /* Reading tx_ring.pg_vec without holding pg_vec_lock is racy.
3039         * tpacket_snd() will redo the check safely.
3040         */
3041        if (data_race(po->tx_ring.pg_vec))
3042                return tpacket_snd(po, msg);
3043
3044        return packet_snd(sock, msg, len);
3045}
3046
3047/*
3048 *      Close a PACKET socket. This is fairly simple. We immediately go
3049 *      to 'closed' state and remove our protocol entry in the device list.
3050 */
3051
3052static int packet_release(struct socket *sock)
3053{
3054        struct sock *sk = sock->sk;
3055        struct packet_sock *po;
3056        struct packet_fanout *f;
3057        struct net *net;
3058        union tpacket_req_u req_u;
3059
3060        if (!sk)
3061                return 0;
3062
3063        net = sock_net(sk);
3064        po = pkt_sk(sk);
3065
3066        mutex_lock(&net->packet.sklist_lock);
3067        sk_del_node_init_rcu(sk);
3068        mutex_unlock(&net->packet.sklist_lock);
3069
3070        preempt_disable();
3071        sock_prot_inuse_add(net, sk->sk_prot, -1);
3072        preempt_enable();
3073
3074        spin_lock(&po->bind_lock);
3075        unregister_prot_hook(sk, false);
3076        packet_cached_dev_reset(po);
3077
3078        if (po->prot_hook.dev) {
3079                dev_put(po->prot_hook.dev);
3080                po->prot_hook.dev = NULL;
3081        }
3082        spin_unlock(&po->bind_lock);
3083
3084        packet_flush_mclist(sk);
3085
3086        lock_sock(sk);
3087        if (po->rx_ring.pg_vec) {
3088                memset(&req_u, 0, sizeof(req_u));
3089                packet_set_ring(sk, &req_u, 1, 0);
3090        }
3091
3092        if (po->tx_ring.pg_vec) {
3093                memset(&req_u, 0, sizeof(req_u));
3094                packet_set_ring(sk, &req_u, 1, 1);
3095        }
3096        release_sock(sk);
3097
3098        f = fanout_release(sk);
3099
3100        synchronize_net();
3101
3102        kfree(po->rollover);
3103        if (f) {
3104                fanout_release_data(f);
3105                kvfree(f);
3106        }
3107        /*
3108         *      Now the socket is dead. No more input will appear.
3109         */
3110        sock_orphan(sk);
3111        sock->sk = NULL;
3112
3113        /* Purge queues */
3114
3115        skb_queue_purge(&sk->sk_receive_queue);
3116        packet_free_pending(po);
3117        sk_refcnt_debug_release(sk);
3118
3119        sock_put(sk);
3120        return 0;
3121}
3122
3123/*
3124 *      Attach a packet hook.
3125 */
3126
3127static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3128                          __be16 proto)
3129{
3130        struct packet_sock *po = pkt_sk(sk);
3131        struct net_device *dev_curr;
3132        __be16 proto_curr;
3133        bool need_rehook;
3134        struct net_device *dev = NULL;
3135        int ret = 0;
3136        bool unlisted = false;
3137
3138        lock_sock(sk);
3139        spin_lock(&po->bind_lock);
3140        rcu_read_lock();
3141
3142        if (po->fanout) {
3143                ret = -EINVAL;
3144                goto out_unlock;
3145        }
3146
3147        if (name) {
3148                dev = dev_get_by_name_rcu(sock_net(sk), name);
3149                if (!dev) {
3150                        ret = -ENODEV;
3151                        goto out_unlock;
3152                }
3153        } else if (ifindex) {
3154                dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3155                if (!dev) {
3156                        ret = -ENODEV;
3157                        goto out_unlock;
3158                }
3159        }
3160
3161        if (dev)
3162                dev_hold(dev);
3163
3164        proto_curr = po->prot_hook.type;
3165        dev_curr = po->prot_hook.dev;
3166
3167        need_rehook = proto_curr != proto || dev_curr != dev;
3168
3169        if (need_rehook) {
3170                if (po->running) {
3171                        rcu_read_unlock();
3172                        /* prevents packet_notifier() from calling
3173                         * register_prot_hook()
3174                         */
3175                        WRITE_ONCE(po->num, 0);
3176                        __unregister_prot_hook(sk, true);
3177                        rcu_read_lock();
3178                        dev_curr = po->prot_hook.dev;
3179                        if (dev)
3180                                unlisted = !dev_get_by_index_rcu(sock_net(sk),
3181                                                                 dev->ifindex);
3182                }
3183
3184                BUG_ON(po->running);
3185                WRITE_ONCE(po->num, proto);
3186                po->prot_hook.type = proto;
3187
3188                if (unlikely(unlisted)) {
3189                        dev_put(dev);
3190                        po->prot_hook.dev = NULL;
3191                        WRITE_ONCE(po->ifindex, -1);
3192                        packet_cached_dev_reset(po);
3193                } else {
3194                        po->prot_hook.dev = dev;
3195                        WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0);
3196                        packet_cached_dev_assign(po, dev);
3197                }
3198        }
3199        if (dev_curr)
3200                dev_put(dev_curr);
3201
3202        if (proto == 0 || !need_rehook)
3203                goto out_unlock;
3204
3205        if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3206                register_prot_hook(sk);
3207        } else {
3208                sk->sk_err = ENETDOWN;
3209                if (!sock_flag(sk, SOCK_DEAD))
3210                        sk_error_report(sk);
3211        }
3212
3213out_unlock:
3214        rcu_read_unlock();
3215        spin_unlock(&po->bind_lock);
3216        release_sock(sk);
3217        return ret;
3218}
3219
3220/*
3221 *      Bind a packet socket to a device
3222 */
3223
3224static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3225                            int addr_len)
3226{
3227        struct sock *sk = sock->sk;
3228        char name[sizeof(uaddr->sa_data) + 1];
3229
3230        /*
3231         *      Check legality
3232         */
3233
3234        if (addr_len != sizeof(struct sockaddr))
3235                return -EINVAL;
3236        /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3237         * zero-terminated.
3238         */
3239        memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3240        name[sizeof(uaddr->sa_data)] = 0;
3241
3242        return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3243}
3244
3245static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3246{
3247        struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3248        struct sock *sk = sock->sk;
3249
3250        /*
3251         *      Check legality
3252         */
3253
3254        if (addr_len < sizeof(struct sockaddr_ll))
3255                return -EINVAL;
3256        if (sll->sll_family != AF_PACKET)
3257                return -EINVAL;
3258
3259        return packet_do_bind(sk, NULL, sll->sll_ifindex,
3260                              sll->sll_protocol ? : pkt_sk(sk)->num);
3261}
3262
3263static struct proto packet_proto = {
3264        .name     = "PACKET",
3265        .owner    = THIS_MODULE,
3266        .obj_size = sizeof(struct packet_sock),
3267};
3268
3269/*
3270 *      Create a packet of type SOCK_PACKET.
3271 */
3272
3273static int packet_create(struct net *net, struct socket *sock, int protocol,
3274                         int kern)
3275{
3276        struct sock *sk;
3277        struct packet_sock *po;
3278        __be16 proto = (__force __be16)protocol; /* weird, but documented */
3279        int err;
3280
3281        if (!ns_capable(net->user_ns, CAP_NET_RAW))
3282                return -EPERM;
3283        if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3284            sock->type != SOCK_PACKET)
3285                return -ESOCKTNOSUPPORT;
3286
3287        sock->state = SS_UNCONNECTED;
3288
3289        err = -ENOBUFS;
3290        sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3291        if (sk == NULL)
3292                goto out;
3293
3294        sock->ops = &packet_ops;
3295        if (sock->type == SOCK_PACKET)
3296                sock->ops = &packet_ops_spkt;
3297
3298        sock_init_data(sock, sk);
3299
3300        po = pkt_sk(sk);
3301        init_completion(&po->skb_completion);
3302        sk->sk_family = PF_PACKET;
3303        po->num = proto;
3304        po->xmit = dev_queue_xmit;
3305
3306        err = packet_alloc_pending(po);
3307        if (err)
3308                goto out2;
3309
3310        packet_cached_dev_reset(po);
3311
3312        sk->sk_destruct = packet_sock_destruct;
3313        sk_refcnt_debug_inc(sk);
3314
3315        /*
3316         *      Attach a protocol block
3317         */
3318
3319        spin_lock_init(&po->bind_lock);
3320        mutex_init(&po->pg_vec_lock);
3321        po->rollover = NULL;
3322        po->prot_hook.func = packet_rcv;
3323
3324        if (sock->type == SOCK_PACKET)
3325                po->prot_hook.func = packet_rcv_spkt;
3326
3327        po->prot_hook.af_packet_priv = sk;
3328
3329        if (proto) {
3330                po->prot_hook.type = proto;
3331                __register_prot_hook(sk);
3332        }
3333
3334        mutex_lock(&net->packet.sklist_lock);
3335        sk_add_node_tail_rcu(sk, &net->packet.sklist);
3336        mutex_unlock(&net->packet.sklist_lock);
3337
3338        preempt_disable();
3339        sock_prot_inuse_add(net, &packet_proto, 1);
3340        preempt_enable();
3341
3342        return 0;
3343out2:
3344        sk_free(sk);
3345out:
3346        return err;
3347}
3348
3349/*
3350 *      Pull a packet from our receive queue and hand it to the user.
3351 *      If necessary we block.
3352 */
3353
3354static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3355                          int flags)
3356{
3357        struct sock *sk = sock->sk;
3358        struct sk_buff *skb;
3359        int copied, err;
3360        int vnet_hdr_len = 0;
3361        unsigned int origlen = 0;
3362
3363        err = -EINVAL;
3364        if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3365                goto out;
3366
3367#if 0
3368        /* What error should we return now? EUNATTACH? */
3369        if (pkt_sk(sk)->ifindex < 0)
3370                return -ENODEV;
3371#endif
3372
3373        if (flags & MSG_ERRQUEUE) {
3374                err = sock_recv_errqueue(sk, msg, len,
3375                                         SOL_PACKET, PACKET_TX_TIMESTAMP);
3376                goto out;
3377        }
3378
3379        /*
3380         *      Call the generic datagram receiver. This handles all sorts
3381         *      of horrible races and re-entrancy so we can forget about it
3382         *      in the protocol layers.
3383         *
3384         *      Now it will return ENETDOWN, if device have just gone down,
3385         *      but then it will block.
3386         */
3387
3388        skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3389
3390        /*
3391         *      An error occurred so return it. Because skb_recv_datagram()
3392         *      handles the blocking we don't see and worry about blocking
3393         *      retries.
3394         */
3395
3396        if (skb == NULL)
3397                goto out;
3398
3399        packet_rcv_try_clear_pressure(pkt_sk(sk));
3400
3401        if (pkt_sk(sk)->has_vnet_hdr) {
3402                err = packet_rcv_vnet(msg, skb, &len);
3403                if (err)
3404                        goto out_free;
3405                vnet_hdr_len = sizeof(struct virtio_net_hdr);
3406        }
3407
3408        /* You lose any data beyond the buffer you gave. If it worries
3409         * a user program they can ask the device for its MTU
3410         * anyway.
3411         */
3412        copied = skb->len;
3413        if (copied > len) {
3414                copied = len;
3415                msg->msg_flags |= MSG_TRUNC;
3416        }
3417
3418        err = skb_copy_datagram_msg(skb, 0, msg, copied);
3419        if (err)
3420                goto out_free;
3421
3422        if (sock->type != SOCK_PACKET) {
3423                struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3424
3425                /* Original length was stored in sockaddr_ll fields */
3426                origlen = PACKET_SKB_CB(skb)->sa.origlen;
3427                sll->sll_family = AF_PACKET;
3428                sll->sll_protocol = skb->protocol;
3429        }
3430
3431        sock_recv_ts_and_drops(msg, sk, skb);
3432
3433        if (msg->msg_name) {
3434                int copy_len;
3435
3436                /* If the address length field is there to be filled
3437                 * in, we fill it in now.
3438                 */
3439                if (sock->type == SOCK_PACKET) {
3440                        __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3441                        msg->msg_namelen = sizeof(struct sockaddr_pkt);
3442                        copy_len = msg->msg_namelen;
3443                } else {
3444                        struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3445
3446                        msg->msg_namelen = sll->sll_halen +
3447                                offsetof(struct sockaddr_ll, sll_addr);
3448                        copy_len = msg->msg_namelen;
3449                        if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3450                                memset(msg->msg_name +
3451                                       offsetof(struct sockaddr_ll, sll_addr),
3452                                       0, sizeof(sll->sll_addr));
3453                                msg->msg_namelen = sizeof(struct sockaddr_ll);
3454                        }
3455                }
3456                memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3457        }
3458
3459        if (pkt_sk(sk)->auxdata) {
3460                struct tpacket_auxdata aux;
3461
3462                aux.tp_status = TP_STATUS_USER;
3463                if (skb->ip_summed == CHECKSUM_PARTIAL)
3464                        aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3465                else if (skb->pkt_type != PACKET_OUTGOING &&
3466                         (skb->ip_summed == CHECKSUM_COMPLETE ||
3467                          skb_csum_unnecessary(skb)))
3468                        aux.tp_status |= TP_STATUS_CSUM_VALID;
3469
3470                aux.tp_len = origlen;
3471                aux.tp_snaplen = skb->len;
3472                aux.tp_mac = 0;
3473                aux.tp_net = skb_network_offset(skb);
3474                if (skb_vlan_tag_present(skb)) {
3475                        aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3476                        aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3477                        aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3478                } else {
3479                        aux.tp_vlan_tci = 0;
3480                        aux.tp_vlan_tpid = 0;
3481                }
3482                put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3483        }
3484
3485        /*
3486         *      Free or return the buffer as appropriate. Again this
3487         *      hides all the races and re-entrancy issues from us.
3488         */
3489        err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3490
3491out_free:
3492        skb_free_datagram(sk, skb);
3493out:
3494        return err;
3495}
3496
3497static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3498                               int peer)
3499{
3500        struct net_device *dev;
3501        struct sock *sk = sock->sk;
3502
3503        if (peer)
3504                return -EOPNOTSUPP;
3505
3506        uaddr->sa_family = AF_PACKET;
3507        memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3508        rcu_read_lock();
3509        dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex));
3510        if (dev)
3511                strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3512        rcu_read_unlock();
3513
3514        return sizeof(*uaddr);
3515}
3516
3517static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3518                          int peer)
3519{
3520        struct net_device *dev;
3521        struct sock *sk = sock->sk;
3522        struct packet_sock *po = pkt_sk(sk);
3523        DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3524        int ifindex;
3525
3526        if (peer)
3527                return -EOPNOTSUPP;
3528
3529        ifindex = READ_ONCE(po->ifindex);
3530        sll->sll_family = AF_PACKET;
3531        sll->sll_ifindex = ifindex;
3532        sll->sll_protocol = READ_ONCE(po->num);
3533        sll->sll_pkttype = 0;
3534        rcu_read_lock();
3535        dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3536        if (dev) {
3537                sll->sll_hatype = dev->type;
3538                sll->sll_halen = dev->addr_len;
3539                memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3540        } else {
3541                sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
3542                sll->sll_halen = 0;
3543        }
3544        rcu_read_unlock();
3545
3546        return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3547}
3548
3549static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3550                         int what)
3551{
3552        switch (i->type) {
3553        case PACKET_MR_MULTICAST:
3554                if (i->alen != dev->addr_len)
3555                        return -EINVAL;
3556                if (what > 0)
3557                        return dev_mc_add(dev, i->addr);
3558                else
3559                        return dev_mc_del(dev, i->addr);
3560                break;
3561        case PACKET_MR_PROMISC:
3562                return dev_set_promiscuity(dev, what);
3563        case PACKET_MR_ALLMULTI:
3564                return dev_set_allmulti(dev, what);
3565        case PACKET_MR_UNICAST:
3566                if (i->alen != dev->addr_len)
3567                        return -EINVAL;
3568                if (what > 0)
3569                        return dev_uc_add(dev, i->addr);
3570                else
3571                        return dev_uc_del(dev, i->addr);
3572                break;
3573        default:
3574                break;
3575        }
3576        return 0;
3577}
3578
3579static void packet_dev_mclist_delete(struct net_device *dev,
3580                                     struct packet_mclist **mlp)
3581{
3582        struct packet_mclist *ml;
3583
3584        while ((ml = *mlp) != NULL) {
3585                if (ml->ifindex == dev->ifindex) {
3586                        packet_dev_mc(dev, ml, -1);
3587                        *mlp = ml->next;
3588                        kfree(ml);
3589                } else
3590                        mlp = &ml->next;
3591        }
3592}
3593
3594static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3595{
3596        struct packet_sock *po = pkt_sk(sk);
3597        struct packet_mclist *ml, *i;
3598        struct net_device *dev;
3599        int err;
3600
3601        rtnl_lock();
3602
3603        err = -ENODEV;
3604        dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3605        if (!dev)
3606                goto done;
3607
3608        err = -EINVAL;
3609        if (mreq->mr_alen > dev->addr_len)
3610                goto done;
3611
3612        err = -ENOBUFS;
3613        i = kmalloc(sizeof(*i), GFP_KERNEL);
3614        if (i == NULL)
3615                goto done;
3616
3617        err = 0;
3618        for (ml = po->mclist; ml; ml = ml->next) {
3619                if (ml->ifindex == mreq->mr_ifindex &&
3620                    ml->type == mreq->mr_type &&
3621                    ml->alen == mreq->mr_alen &&
3622                    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3623                        ml->count++;
3624                        /* Free the new element ... */
3625                        kfree(i);
3626                        goto done;
3627                }
3628        }
3629
3630        i->type = mreq->mr_type;
3631        i->ifindex = mreq->mr_ifindex;
3632        i->alen = mreq->mr_alen;
3633        memcpy(i->addr, mreq->mr_address, i->alen);
3634        memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3635        i->count = 1;
3636        i->next = po->mclist;
3637        po->mclist = i;
3638        err = packet_dev_mc(dev, i, 1);
3639        if (err) {
3640                po->mclist = i->next;
3641                kfree(i);
3642        }
3643
3644done:
3645        rtnl_unlock();
3646        return err;
3647}
3648
3649static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3650{
3651        struct packet_mclist *ml, **mlp;
3652
3653        rtnl_lock();
3654
3655        for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3656                if (ml->ifindex == mreq->mr_ifindex &&
3657                    ml->type == mreq->mr_type &&
3658                    ml->alen == mreq->mr_alen &&
3659                    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3660                        if (--ml->count == 0) {
3661                                struct net_device *dev;
3662                                *mlp = ml->next;
3663                                dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3664                                if (dev)
3665                                        packet_dev_mc(dev, ml, -1);
3666                                kfree(ml);
3667                        }
3668                        break;
3669                }
3670        }
3671        rtnl_unlock();
3672        return 0;
3673}
3674
3675static void packet_flush_mclist(struct sock *sk)
3676{
3677        struct packet_sock *po = pkt_sk(sk);
3678        struct packet_mclist *ml;
3679
3680        if (!po->mclist)
3681                return;
3682
3683        rtnl_lock();
3684        while ((ml = po->mclist) != NULL) {
3685                struct net_device *dev;
3686
3687                po->mclist = ml->next;
3688                dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3689                if (dev != NULL)
3690                        packet_dev_mc(dev, ml, -1);
3691                kfree(ml);
3692        }
3693        rtnl_unlock();
3694}
3695
3696static int
3697packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval,
3698                  unsigned int optlen)
3699{
3700        struct sock *sk = sock->sk;
3701        struct packet_sock *po = pkt_sk(sk);
3702        int ret;
3703
3704        if (level != SOL_PACKET)
3705                return -ENOPROTOOPT;
3706
3707        switch (optname) {
3708        case PACKET_ADD_MEMBERSHIP:
3709        case PACKET_DROP_MEMBERSHIP:
3710        {
3711                struct packet_mreq_max mreq;
3712                int len = optlen;
3713                memset(&mreq, 0, sizeof(mreq));
3714                if (len < sizeof(struct packet_mreq))
3715                        return -EINVAL;
3716                if (len > sizeof(mreq))
3717                        len = sizeof(mreq);
3718                if (copy_from_sockptr(&mreq, optval, len))
3719                        return -EFAULT;
3720                if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3721                        return -EINVAL;
3722                if (optname == PACKET_ADD_MEMBERSHIP)
3723                        ret = packet_mc_add(sk, &mreq);
3724                else
3725                        ret = packet_mc_drop(sk, &mreq);
3726                return ret;
3727        }
3728
3729        case PACKET_RX_RING:
3730        case PACKET_TX_RING:
3731        {
3732                union tpacket_req_u req_u;
3733                int len;
3734
3735                lock_sock(sk);
3736                switch (po->tp_version) {
3737                case TPACKET_V1:
3738                case TPACKET_V2:
3739                        len = sizeof(req_u.req);
3740                        break;
3741                case TPACKET_V3:
3742                default:
3743                        len = sizeof(req_u.req3);
3744                        break;
3745                }
3746                if (optlen < len) {
3747                        ret = -EINVAL;
3748                } else {
3749                        if (copy_from_sockptr(&req_u.req, optval, len))
3750                                ret = -EFAULT;
3751                        else
3752                                ret = packet_set_ring(sk, &req_u, 0,
3753                                                    optname == PACKET_TX_RING);
3754                }
3755                release_sock(sk);
3756                return ret;
3757        }
3758        case PACKET_COPY_THRESH:
3759        {
3760                int val;
3761
3762                if (optlen != sizeof(val))
3763                        return -EINVAL;
3764                if (copy_from_sockptr(&val, optval, sizeof(val)))
3765                        return -EFAULT;
3766
3767                pkt_sk(sk)->copy_thresh = val;
3768                return 0;
3769        }
3770        case PACKET_VERSION:
3771        {
3772                int val;
3773
3774                if (optlen != sizeof(val))
3775                        return -EINVAL;
3776                if (copy_from_sockptr(&val, optval, sizeof(val)))
3777                        return -EFAULT;
3778                switch (val) {
3779                case TPACKET_V1:
3780                case TPACKET_V2:
3781                case TPACKET_V3:
3782                        break;
3783                default:
3784                        return -EINVAL;
3785                }
3786                lock_sock(sk);
3787                if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3788                        ret = -EBUSY;
3789                } else {
3790                        po->tp_version = val;
3791                        ret = 0;
3792                }
3793                release_sock(sk);
3794                return ret;
3795        }
3796        case PACKET_RESERVE:
3797        {
3798                unsigned int val;
3799
3800                if (optlen != sizeof(val))
3801                        return -EINVAL;
3802                if (copy_from_sockptr(&val, optval, sizeof(val)))
3803                        return -EFAULT;
3804                if (val > INT_MAX)
3805                        return -EINVAL;
3806                lock_sock(sk);
3807                if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3808                        ret = -EBUSY;
3809                } else {
3810                        po->tp_reserve = val;
3811                        ret = 0;
3812                }
3813                release_sock(sk);
3814                return ret;
3815        }
3816        case PACKET_LOSS:
3817        {
3818                unsigned int val;
3819
3820                if (optlen != sizeof(val))
3821                        return -EINVAL;
3822                if (copy_from_sockptr(&val, optval, sizeof(val)))
3823                        return -EFAULT;
3824
3825                lock_sock(sk);
3826                if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3827                        ret = -EBUSY;
3828                } else {
3829                        po->tp_loss = !!val;
3830                        ret = 0;
3831                }
3832                release_sock(sk);
3833                return ret;
3834        }
3835        case PACKET_AUXDATA:
3836        {
3837                int val;
3838
3839                if (optlen < sizeof(val))
3840                        return -EINVAL;
3841                if (copy_from_sockptr(&val, optval, sizeof(val)))
3842                        return -EFAULT;
3843
3844                lock_sock(sk);
3845                po->auxdata = !!val;
3846                release_sock(sk);
3847                return 0;
3848        }
3849        case PACKET_ORIGDEV:
3850        {
3851                int val;
3852
3853                if (optlen < sizeof(val))
3854                        return -EINVAL;
3855                if (copy_from_sockptr(&val, optval, sizeof(val)))
3856                        return -EFAULT;
3857
3858                lock_sock(sk);
3859                po->origdev = !!val;
3860                release_sock(sk);
3861                return 0;
3862        }
3863        case PACKET_VNET_HDR:
3864        {
3865                int val;
3866
3867                if (sock->type != SOCK_RAW)
3868                        return -EINVAL;
3869                if (optlen < sizeof(val))
3870                        return -EINVAL;
3871                if (copy_from_sockptr(&val, optval, sizeof(val)))
3872                        return -EFAULT;
3873
3874                lock_sock(sk);
3875                if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3876                        ret = -EBUSY;
3877                } else {
3878                        po->has_vnet_hdr = !!val;
3879                        ret = 0;
3880                }
3881                release_sock(sk);
3882                return ret;
3883        }
3884        case PACKET_TIMESTAMP:
3885        {
3886                int val;
3887
3888                if (optlen != sizeof(val))
3889                        return -EINVAL;
3890                if (copy_from_sockptr(&val, optval, sizeof(val)))
3891                        return -EFAULT;
3892
3893                po->tp_tstamp = val;
3894                return 0;
3895        }
3896        case PACKET_FANOUT:
3897        {
3898                struct fanout_args args = { 0 };
3899
3900                if (optlen != sizeof(int) && optlen != sizeof(args))
3901                        return -EINVAL;
3902                if (copy_from_sockptr(&args, optval, optlen))
3903                        return -EFAULT;
3904
3905                return fanout_add(sk, &args);
3906        }
3907        case PACKET_FANOUT_DATA:
3908        {
3909                if (!po->fanout)
3910                        return -EINVAL;
3911
3912                return fanout_set_data(po, optval, optlen);
3913        }
3914        case PACKET_IGNORE_OUTGOING:
3915        {
3916                int val;
3917
3918                if (optlen != sizeof(val))
3919                        return -EINVAL;
3920                if (copy_from_sockptr(&val, optval, sizeof(val)))
3921                        return -EFAULT;
3922                if (val < 0 || val > 1)
3923                        return -EINVAL;
3924
3925                po->prot_hook.ignore_outgoing = !!val;
3926                return 0;
3927        }
3928        case PACKET_TX_HAS_OFF:
3929        {
3930                unsigned int val;
3931
3932                if (optlen != sizeof(val))
3933                        return -EINVAL;
3934                if (copy_from_sockptr(&val, optval, sizeof(val)))
3935                        return -EFAULT;
3936
3937                lock_sock(sk);
3938                if (!po->rx_ring.pg_vec && !po->tx_ring.pg_vec)
3939                        po->tp_tx_has_off = !!val;
3940
3941                release_sock(sk);
3942                return 0;
3943        }
3944        case PACKET_QDISC_BYPASS:
3945        {
3946                int val;
3947
3948                if (optlen != sizeof(val))
3949                        return -EINVAL;
3950                if (copy_from_sockptr(&val, optval, sizeof(val)))
3951                        return -EFAULT;
3952
3953                po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3954                return 0;
3955        }
3956        default:
3957                return -ENOPROTOOPT;
3958        }
3959}
3960
3961static int packet_getsockopt(struct socket *sock, int level, int optname,
3962                             char __user *optval, int __user *optlen)
3963{
3964        int len;
3965        int val, lv = sizeof(val);
3966        struct sock *sk = sock->sk;
3967        struct packet_sock *po = pkt_sk(sk);
3968        void *data = &val;
3969        union tpacket_stats_u st;
3970        struct tpacket_rollover_stats rstats;
3971        int drops;
3972
3973        if (level != SOL_PACKET)
3974                return -ENOPROTOOPT;
3975
3976        if (get_user(len, optlen))
3977                return -EFAULT;
3978
3979        if (len < 0)
3980                return -EINVAL;
3981
3982        switch (optname) {
3983        case PACKET_STATISTICS:
3984                spin_lock_bh(&sk->sk_receive_queue.lock);
3985                memcpy(&st, &po->stats, sizeof(st));
3986                memset(&po->stats, 0, sizeof(po->stats));
3987                spin_unlock_bh(&sk->sk_receive_queue.lock);
3988                drops = atomic_xchg(&po->tp_drops, 0);
3989
3990                if (po->tp_version == TPACKET_V3) {
3991                        lv = sizeof(struct tpacket_stats_v3);
3992                        st.stats3.tp_drops = drops;
3993                        st.stats3.tp_packets += drops;
3994                        data = &st.stats3;
3995                } else {
3996                        lv = sizeof(struct tpacket_stats);
3997                        st.stats1.tp_drops = drops;
3998                        st.stats1.tp_packets += drops;
3999                        data = &st.stats1;
4000                }
4001
4002                break;
4003        case PACKET_AUXDATA:
4004                val = po->auxdata;
4005                break;
4006        case PACKET_ORIGDEV:
4007                val = po->origdev;
4008                break;
4009        case PACKET_VNET_HDR:
4010                val = po->has_vnet_hdr;
4011                break;
4012        case PACKET_VERSION:
4013                val = po->tp_version;
4014                break;
4015        case PACKET_HDRLEN:
4016                if (len > sizeof(int))
4017                        len = sizeof(int);
4018                if (len < sizeof(int))
4019                        return -EINVAL;
4020                if (copy_from_user(&val, optval, len))
4021                        return -EFAULT;
4022                switch (val) {
4023                case TPACKET_V1:
4024                        val = sizeof(struct tpacket_hdr);
4025                        break;
4026                case TPACKET_V2:
4027                        val = sizeof(struct tpacket2_hdr);
4028                        break;
4029                case TPACKET_V3:
4030                        val = sizeof(struct tpacket3_hdr);
4031                        break;
4032                default:
4033                        return -EINVAL;
4034                }
4035                break;
4036        case PACKET_RESERVE:
4037                val = po->tp_reserve;
4038                break;
4039        case PACKET_LOSS:
4040                val = po->tp_loss;
4041                break;
4042        case PACKET_TIMESTAMP:
4043                val = po->tp_tstamp;
4044                break;
4045        case PACKET_FANOUT:
4046                val = (po->fanout ?
4047                       ((u32)po->fanout->id |
4048                        ((u32)po->fanout->type << 16) |
4049                        ((u32)po->fanout->flags << 24)) :
4050                       0);
4051                break;
4052        case PACKET_IGNORE_OUTGOING:
4053                val = po->prot_hook.ignore_outgoing;
4054                break;
4055        case PACKET_ROLLOVER_STATS:
4056                if (!po->rollover)
4057                        return -EINVAL;
4058                rstats.tp_all = atomic_long_read(&po->rollover->num);
4059                rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4060                rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4061                data = &rstats;
4062                lv = sizeof(rstats);
4063                break;
4064        case PACKET_TX_HAS_OFF:
4065                val = po->tp_tx_has_off;
4066                break;
4067        case PACKET_QDISC_BYPASS:
4068                val = packet_use_direct_xmit(po);
4069                break;
4070        default:
4071                return -ENOPROTOOPT;
4072        }
4073
4074        if (len > lv)
4075                len = lv;
4076        if (put_user(len, optlen))
4077                return -EFAULT;
4078        if (copy_to_user(optval, data, len))
4079                return -EFAULT;
4080        return 0;
4081}
4082
4083static int packet_notifier(struct notifier_block *this,
4084                           unsigned long msg, void *ptr)
4085{
4086        struct sock *sk;
4087        struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4088        struct net *net = dev_net(dev);
4089
4090        rcu_read_lock();
4091        sk_for_each_rcu(sk, &net->packet.sklist) {
4092                struct packet_sock *po = pkt_sk(sk);
4093
4094                switch (msg) {
4095                case NETDEV_UNREGISTER:
4096                        if (po->mclist)
4097                                packet_dev_mclist_delete(dev, &po->mclist);
4098                        fallthrough;
4099
4100                case NETDEV_DOWN:
4101                        if (dev->ifindex == po->ifindex) {
4102                                spin_lock(&po->bind_lock);
4103                                if (po->running) {
4104                                        __unregister_prot_hook(sk, false);
4105                                        sk->sk_err = ENETDOWN;
4106                                        if (!sock_flag(sk, SOCK_DEAD))
4107                                                sk_error_report(sk);
4108                                }
4109                                if (msg == NETDEV_UNREGISTER) {
4110                                        packet_cached_dev_reset(po);
4111                                        WRITE_ONCE(po->ifindex, -1);
4112                                        if (po->prot_hook.dev)
4113                                                dev_put(po->prot_hook.dev);
4114                                        po->prot_hook.dev = NULL;
4115                                }
4116                                spin_unlock(&po->bind_lock);
4117                        }
4118                        break;
4119                case NETDEV_UP:
4120                        if (dev->ifindex == po->ifindex) {
4121                                spin_lock(&po->bind_lock);
4122                                if (po->num)
4123                                        register_prot_hook(sk);
4124                                spin_unlock(&po->bind_lock);
4125                        }
4126                        break;
4127                }
4128        }
4129        rcu_read_unlock();
4130        return NOTIFY_DONE;
4131}
4132
4133
4134static int packet_ioctl(struct socket *sock, unsigned int cmd,
4135                        unsigned long arg)
4136{
4137        struct sock *sk = sock->sk;
4138
4139        switch (cmd) {
4140        case SIOCOUTQ:
4141        {
4142                int amount = sk_wmem_alloc_get(sk);
4143
4144                return put_user(amount, (int __user *)arg);
4145        }
4146        case SIOCINQ:
4147        {
4148                struct sk_buff *skb;
4149                int amount = 0;
4150
4151                spin_lock_bh(&sk->sk_receive_queue.lock);
4152                skb = skb_peek(&sk->sk_receive_queue);
4153                if (skb)
4154                        amount = skb->len;
4155                spin_unlock_bh(&sk->sk_receive_queue.lock);
4156                return put_user(amount, (int __user *)arg);
4157        }
4158#ifdef CONFIG_INET
4159        case SIOCADDRT:
4160        case SIOCDELRT:
4161        case SIOCDARP:
4162        case SIOCGARP:
4163        case SIOCSARP:
4164        case SIOCGIFADDR:
4165        case SIOCSIFADDR:
4166        case SIOCGIFBRDADDR:
4167        case SIOCSIFBRDADDR:
4168        case SIOCGIFNETMASK:
4169        case SIOCSIFNETMASK:
4170        case SIOCGIFDSTADDR:
4171        case SIOCSIFDSTADDR:
4172        case SIOCSIFFLAGS:
4173                return inet_dgram_ops.ioctl(sock, cmd, arg);
4174#endif
4175
4176        default:
4177                return -ENOIOCTLCMD;
4178        }
4179        return 0;
4180}
4181
4182static __poll_t packet_poll(struct file *file, struct socket *sock,
4183                                poll_table *wait)
4184{
4185        struct sock *sk = sock->sk;
4186        struct packet_sock *po = pkt_sk(sk);
4187        __poll_t mask = datagram_poll(file, sock, wait);
4188
4189        spin_lock_bh(&sk->sk_receive_queue.lock);
4190        if (po->rx_ring.pg_vec) {
4191                if (!packet_previous_rx_frame(po, &po->rx_ring,
4192                        TP_STATUS_KERNEL))
4193                        mask |= EPOLLIN | EPOLLRDNORM;
4194        }
4195        packet_rcv_try_clear_pressure(po);
4196        spin_unlock_bh(&sk->sk_receive_queue.lock);
4197        spin_lock_bh(&sk->sk_write_queue.lock);
4198        if (po->tx_ring.pg_vec) {
4199                if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4200                        mask |= EPOLLOUT | EPOLLWRNORM;
4201        }
4202        spin_unlock_bh(&sk->sk_write_queue.lock);
4203        return mask;
4204}
4205
4206
4207/* Dirty? Well, I still did not learn better way to account
4208 * for user mmaps.
4209 */
4210
4211static void packet_mm_open(struct vm_area_struct *vma)
4212{
4213        struct file *file = vma->vm_file;
4214        struct socket *sock = file->private_data;
4215        struct sock *sk = sock->sk;
4216
4217        if (sk)
4218                atomic_inc(&pkt_sk(sk)->mapped);
4219}
4220
4221static void packet_mm_close(struct vm_area_struct *vma)
4222{
4223        struct file *file = vma->vm_file;
4224        struct socket *sock = file->private_data;
4225        struct sock *sk = sock->sk;
4226
4227        if (sk)
4228                atomic_dec(&pkt_sk(sk)->mapped);
4229}
4230
4231static const struct vm_operations_struct packet_mmap_ops = {
4232        .open   =       packet_mm_open,
4233        .close  =       packet_mm_close,
4234};
4235
4236static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4237                        unsigned int len)
4238{
4239        int i;
4240
4241        for (i = 0; i < len; i++) {
4242                if (likely(pg_vec[i].buffer)) {
4243                        if (is_vmalloc_addr(pg_vec[i].buffer))
4244                                vfree(pg_vec[i].buffer);
4245                        else
4246                                free_pages((unsigned long)pg_vec[i].buffer,
4247                                           order);
4248                        pg_vec[i].buffer = NULL;
4249                }
4250        }
4251        kfree(pg_vec);
4252}
4253
4254static char *alloc_one_pg_vec_page(unsigned long order)
4255{
4256        char *buffer;
4257        gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4258                          __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4259
4260        buffer = (char *) __get_free_pages(gfp_flags, order);
4261        if (buffer)
4262                return buffer;
4263
4264        /* __get_free_pages failed, fall back to vmalloc */
4265        buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4266        if (buffer)
4267                return buffer;
4268
4269        /* vmalloc failed, lets dig into swap here */
4270        gfp_flags &= ~__GFP_NORETRY;
4271        buffer = (char *) __get_free_pages(gfp_flags, order);
4272        if (buffer)
4273                return buffer;
4274
4275        /* complete and utter failure */
4276        return NULL;
4277}
4278
4279static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4280{
4281        unsigned int block_nr = req->tp_block_nr;
4282        struct pgv *pg_vec;
4283        int i;
4284
4285        pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4286        if (unlikely(!pg_vec))
4287                goto out;
4288
4289        for (i = 0; i < block_nr; i++) {
4290                pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4291                if (unlikely(!pg_vec[i].buffer))
4292                        goto out_free_pgvec;
4293        }
4294
4295out:
4296        return pg_vec;
4297
4298out_free_pgvec:
4299        free_pg_vec(pg_vec, order, block_nr);
4300        pg_vec = NULL;
4301        goto out;
4302}
4303
4304static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4305                int closing, int tx_ring)
4306{
4307        struct pgv *pg_vec = NULL;
4308        struct packet_sock *po = pkt_sk(sk);
4309        unsigned long *rx_owner_map = NULL;
4310        int was_running, order = 0;
4311        struct packet_ring_buffer *rb;
4312        struct sk_buff_head *rb_queue;
4313        __be16 num;
4314        int err;
4315        /* Added to avoid minimal code churn */
4316        struct tpacket_req *req = &req_u->req;
4317
4318        rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4319        rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4320
4321        err = -EBUSY;
4322        if (!closing) {
4323                if (atomic_read(&po->mapped))
4324                        goto out;
4325                if (packet_read_pending(rb))
4326                        goto out;
4327        }
4328
4329        if (req->tp_block_nr) {
4330                unsigned int min_frame_size;
4331
4332                /* Sanity tests and some calculations */
4333                err = -EBUSY;
4334                if (unlikely(rb->pg_vec))
4335                        goto out;
4336
4337                switch (po->tp_version) {
4338                case TPACKET_V1:
4339                        po->tp_hdrlen = TPACKET_HDRLEN;
4340                        break;
4341                case TPACKET_V2:
4342                        po->tp_hdrlen = TPACKET2_HDRLEN;
4343                        break;
4344                case TPACKET_V3:
4345                        po->tp_hdrlen = TPACKET3_HDRLEN;
4346                        break;
4347                }
4348
4349                err = -EINVAL;
4350                if (unlikely((int)req->tp_block_size <= 0))
4351                        goto out;
4352                if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4353                        goto out;
4354                min_frame_size = po->tp_hdrlen + po->tp_reserve;
4355                if (po->tp_version >= TPACKET_V3 &&
4356                    req->tp_block_size <
4357                    BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4358                        goto out;
4359                if (unlikely(req->tp_frame_size < min_frame_size))
4360                        goto out;
4361                if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4362                        goto out;
4363
4364                rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4365                if (unlikely(rb->frames_per_block == 0))
4366                        goto out;
4367                if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4368                        goto out;
4369                if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4370                                        req->tp_frame_nr))
4371                        goto out;
4372
4373                err = -ENOMEM;
4374                order = get_order(req->tp_block_size);
4375                pg_vec = alloc_pg_vec(req, order);
4376                if (unlikely(!pg_vec))
4377                        goto out;
4378                switch (po->tp_version) {
4379                case TPACKET_V3:
4380                        /* Block transmit is not supported yet */
4381                        if (!tx_ring) {
4382                                init_prb_bdqc(po, rb, pg_vec, req_u);
4383                        } else {
4384                                struct tpacket_req3 *req3 = &req_u->req3;
4385
4386                                if (req3->tp_retire_blk_tov ||
4387                                    req3->tp_sizeof_priv ||
4388                                    req3->tp_feature_req_word) {
4389                                        err = -EINVAL;
4390                                        goto out_free_pg_vec;
4391                                }
4392                        }
4393                        break;
4394                default:
4395                        if (!tx_ring) {
4396                                rx_owner_map = bitmap_alloc(req->tp_frame_nr,
4397                                        GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
4398                                if (!rx_owner_map)
4399                                        goto out_free_pg_vec;
4400                        }
4401                        break;
4402                }
4403        }
4404        /* Done */
4405        else {
4406                err = -EINVAL;
4407                if (unlikely(req->tp_frame_nr))
4408                        goto out;
4409        }
4410
4411
4412        /* Detach socket from network */
4413        spin_lock(&po->bind_lock);
4414        was_running = po->running;
4415        num = po->num;
4416        if (was_running) {
4417                WRITE_ONCE(po->num, 0);
4418                __unregister_prot_hook(sk, false);
4419        }
4420        spin_unlock(&po->bind_lock);
4421
4422        synchronize_net();
4423
4424        err = -EBUSY;
4425        mutex_lock(&po->pg_vec_lock);
4426        if (closing || atomic_read(&po->mapped) == 0) {
4427                err = 0;
4428                spin_lock_bh(&rb_queue->lock);
4429                swap(rb->pg_vec, pg_vec);
4430                if (po->tp_version <= TPACKET_V2)
4431                        swap(rb->rx_owner_map, rx_owner_map);
4432                rb->frame_max = (req->tp_frame_nr - 1);
4433                rb->head = 0;
4434                rb->frame_size = req->tp_frame_size;
4435                spin_unlock_bh(&rb_queue->lock);
4436
4437                swap(rb->pg_vec_order, order);
4438                swap(rb->pg_vec_len, req->tp_block_nr);
4439
4440                rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4441                po->prot_hook.func = (po->rx_ring.pg_vec) ?
4442                                                tpacket_rcv : packet_rcv;
4443                skb_queue_purge(rb_queue);
4444                if (atomic_read(&po->mapped))
4445                        pr_err("packet_mmap: vma is busy: %d\n",
4446                               atomic_read(&po->mapped));
4447        }
4448        mutex_unlock(&po->pg_vec_lock);
4449
4450        spin_lock(&po->bind_lock);
4451        if (was_running) {
4452                WRITE_ONCE(po->num, num);
4453                register_prot_hook(sk);
4454        }
4455        spin_unlock(&po->bind_lock);
4456        if (pg_vec && (po->tp_version > TPACKET_V2)) {
4457                /* Because we don't support block-based V3 on tx-ring */
4458                if (!tx_ring)
4459                        prb_shutdown_retire_blk_timer(po, rb_queue);
4460        }
4461
4462out_free_pg_vec:
4463        bitmap_free(rx_owner_map);
4464        if (pg_vec)
4465                free_pg_vec(pg_vec, order, req->tp_block_nr);
4466out:
4467        return err;
4468}
4469
4470static int packet_mmap(struct file *file, struct socket *sock,
4471                struct vm_area_struct *vma)
4472{
4473        struct sock *sk = sock->sk;
4474        struct packet_sock *po = pkt_sk(sk);
4475        unsigned long size, expected_size;
4476        struct packet_ring_buffer *rb;
4477        unsigned long start;
4478        int err = -EINVAL;
4479        int i;
4480
4481        if (vma->vm_pgoff)
4482                return -EINVAL;
4483
4484        mutex_lock(&po->pg_vec_lock);
4485
4486        expected_size = 0;
4487        for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4488                if (rb->pg_vec) {
4489                        expected_size += rb->pg_vec_len
4490                                                * rb->pg_vec_pages
4491                                                * PAGE_SIZE;
4492                }
4493        }
4494
4495        if (expected_size == 0)
4496                goto out;
4497
4498        size = vma->vm_end - vma->vm_start;
4499        if (size != expected_size)
4500                goto out;
4501
4502        start = vma->vm_start;
4503        for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4504                if (rb->pg_vec == NULL)
4505                        continue;
4506
4507                for (i = 0; i < rb->pg_vec_len; i++) {
4508                        struct page *page;
4509                        void *kaddr = rb->pg_vec[i].buffer;
4510                        int pg_num;
4511
4512                        for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4513                                page = pgv_to_page(kaddr);
4514                                err = vm_insert_page(vma, start, page);
4515                                if (unlikely(err))
4516                                        goto out;
4517                                start += PAGE_SIZE;
4518                                kaddr += PAGE_SIZE;
4519                        }
4520                }
4521        }
4522
4523        atomic_inc(&po->mapped);
4524        vma->vm_ops = &packet_mmap_ops;
4525        err = 0;
4526
4527out:
4528        mutex_unlock(&po->pg_vec_lock);
4529        return err;
4530}
4531
4532static const struct proto_ops packet_ops_spkt = {
4533        .family =       PF_PACKET,
4534        .owner =        THIS_MODULE,
4535        .release =      packet_release,
4536        .bind =         packet_bind_spkt,
4537        .connect =      sock_no_connect,
4538        .socketpair =   sock_no_socketpair,
4539        .accept =       sock_no_accept,
4540        .getname =      packet_getname_spkt,
4541        .poll =         datagram_poll,
4542        .ioctl =        packet_ioctl,
4543        .gettstamp =    sock_gettstamp,
4544        .listen =       sock_no_listen,
4545        .shutdown =     sock_no_shutdown,
4546        .sendmsg =      packet_sendmsg_spkt,
4547        .recvmsg =      packet_recvmsg,
4548        .mmap =         sock_no_mmap,
4549        .sendpage =     sock_no_sendpage,
4550};
4551
4552static const struct proto_ops packet_ops = {
4553        .family =       PF_PACKET,
4554        .owner =        THIS_MODULE,
4555        .release =      packet_release,
4556        .bind =         packet_bind,
4557        .connect =      sock_no_connect,
4558        .socketpair =   sock_no_socketpair,
4559        .accept =       sock_no_accept,
4560        .getname =      packet_getname,
4561        .poll =         packet_poll,
4562        .ioctl =        packet_ioctl,
4563        .gettstamp =    sock_gettstamp,
4564        .listen =       sock_no_listen,
4565        .shutdown =     sock_no_shutdown,
4566        .setsockopt =   packet_setsockopt,
4567        .getsockopt =   packet_getsockopt,
4568        .sendmsg =      packet_sendmsg,
4569        .recvmsg =      packet_recvmsg,
4570        .mmap =         packet_mmap,
4571        .sendpage =     sock_no_sendpage,
4572};
4573
4574static const struct net_proto_family packet_family_ops = {
4575        .family =       PF_PACKET,
4576        .create =       packet_create,
4577        .owner  =       THIS_MODULE,
4578};
4579
4580static struct notifier_block packet_netdev_notifier = {
4581        .notifier_call =        packet_notifier,
4582};
4583
4584#ifdef CONFIG_PROC_FS
4585
4586static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4587        __acquires(RCU)
4588{
4589        struct net *net = seq_file_net(seq);
4590
4591        rcu_read_lock();
4592        return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4593}
4594
4595static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)