Linux Audio

Check our new training course

Real-Time Linux with PREEMPT_RT training

Feb 18-20, 2025
Register
Loading...
v5.9
   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/types.h>
  50#include <linux/mm.h>
  51#include <linux/capability.h>
  52#include <linux/fcntl.h>
  53#include <linux/socket.h>
  54#include <linux/in.h>
  55#include <linux/inet.h>
  56#include <linux/netdevice.h>
  57#include <linux/if_packet.h>
  58#include <linux/wireless.h>
  59#include <linux/kernel.h>
  60#include <linux/kmod.h>
  61#include <linux/slab.h>
  62#include <linux/vmalloc.h>
  63#include <net/net_namespace.h>
  64#include <net/ip.h>
  65#include <net/protocol.h>
  66#include <linux/skbuff.h>
  67#include <net/sock.h>
  68#include <linux/errno.h>
  69#include <linux/timer.h>
  70#include <linux/uaccess.h>
  71#include <asm/ioctls.h>
  72#include <asm/page.h>
  73#include <asm/cacheflush.h>
  74#include <asm/io.h>
  75#include <linux/proc_fs.h>
  76#include <linux/seq_file.h>
  77#include <linux/poll.h>
  78#include <linux/module.h>
  79#include <linux/init.h>
  80#include <linux/mutex.h>
  81#include <linux/if_vlan.h>
  82#include <linux/virtio_net.h>
  83#include <linux/errqueue.h>
  84#include <linux/net_tstamp.h>
  85#include <linux/percpu.h>
  86#ifdef CONFIG_INET
  87#include <net/inet_common.h>
  88#endif
  89#include <linux/bpf.h>
  90#include <net/compat.h>
 
  91
  92#include "internal.h"
  93
  94/*
  95   Assumptions:
  96   - if device has no dev->hard_header routine, it adds and removes ll header
  97     inside itself. In this case ll header is invisible outside of device,
  98     but higher levels still should reserve dev->hard_header_len.
  99     Some devices are enough clever to reallocate skb, when header
 100     will not fit to reserved space (tunnel), another ones are silly
 101     (PPP).
 
 
 
 102   - packet socket receives packets with pulled ll header,
 103     so that SOCK_RAW should push it back.
 104
 105On receive:
 106-----------
 107
 108Incoming, dev->hard_header!=NULL
 109   mac_header -> ll header
 110   data       -> data
 111
 112Outgoing, dev->hard_header!=NULL
 113   mac_header -> ll header
 114   data       -> ll header
 115
 116Incoming, dev->hard_header==NULL
 117   mac_header -> UNKNOWN position. It is very likely, that it points to ll
 118		 header.  PPP makes it, that is wrong, because introduce
 119		 assymetry between rx and tx paths.
 120   data       -> data
 121
 122Outgoing, dev->hard_header==NULL
 123   mac_header -> data. ll header is still not built!
 124   data       -> data
 125
 126Resume
 127  If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 
 128
 129
 130On transmit:
 131------------
 132
 133dev->hard_header != NULL
 134   mac_header -> ll header
 135   data       -> ll header
 136
 137dev->hard_header == NULL (ll header is added by device, we cannot control it)
 138   mac_header -> data
 139   data       -> data
 140
 141   We should set nh.raw on output to correct posistion,
 142   packet classifier depends on it.
 143 */
 144
 145/* Private packet socket structures. */
 146
 147/* identical to struct packet_mreq except it has
 148 * a longer address field.
 149 */
 150struct packet_mreq_max {
 151	int		mr_ifindex;
 152	unsigned short	mr_type;
 153	unsigned short	mr_alen;
 154	unsigned char	mr_address[MAX_ADDR_LEN];
 155};
 156
 157union tpacket_uhdr {
 158	struct tpacket_hdr  *h1;
 159	struct tpacket2_hdr *h2;
 160	struct tpacket3_hdr *h3;
 161	void *raw;
 162};
 163
 164static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 165		int closing, int tx_ring);
 166
 167#define V3_ALIGNMENT	(8)
 168
 169#define BLK_HDR_LEN	(ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 170
 171#define BLK_PLUS_PRIV(sz_of_priv) \
 172	(BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 173
 174#define BLOCK_STATUS(x)	((x)->hdr.bh1.block_status)
 175#define BLOCK_NUM_PKTS(x)	((x)->hdr.bh1.num_pkts)
 176#define BLOCK_O2FP(x)		((x)->hdr.bh1.offset_to_first_pkt)
 177#define BLOCK_LEN(x)		((x)->hdr.bh1.blk_len)
 178#define BLOCK_SNUM(x)		((x)->hdr.bh1.seq_num)
 179#define BLOCK_O2PRIV(x)	((x)->offset_to_priv)
 180#define BLOCK_PRIV(x)		((void *)((char *)(x) + BLOCK_O2PRIV(x)))
 181
 182struct packet_sock;
 183static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 184		       struct packet_type *pt, struct net_device *orig_dev);
 185
 186static void *packet_previous_frame(struct packet_sock *po,
 187		struct packet_ring_buffer *rb,
 188		int status);
 189static void packet_increment_head(struct packet_ring_buffer *buff);
 190static int prb_curr_blk_in_use(struct tpacket_block_desc *);
 191static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 192			struct packet_sock *);
 193static void prb_retire_current_block(struct tpacket_kbdq_core *,
 194		struct packet_sock *, unsigned int status);
 195static int prb_queue_frozen(struct tpacket_kbdq_core *);
 196static void prb_open_block(struct tpacket_kbdq_core *,
 197		struct tpacket_block_desc *);
 198static void prb_retire_rx_blk_timer_expired(struct timer_list *);
 199static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 200static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 201static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 202		struct tpacket3_hdr *);
 203static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 204		struct tpacket3_hdr *);
 205static void packet_flush_mclist(struct sock *sk);
 206static u16 packet_pick_tx_queue(struct sk_buff *skb);
 207
 208struct packet_skb_cb {
 209	union {
 210		struct sockaddr_pkt pkt;
 211		union {
 212			/* Trick: alias skb original length with
 213			 * ll.sll_family and ll.protocol in order
 214			 * to save room.
 215			 */
 216			unsigned int origlen;
 217			struct sockaddr_ll ll;
 218		};
 219	} sa;
 220};
 221
 222#define vio_le() virtio_legacy_is_little_endian()
 223
 224#define PACKET_SKB_CB(__skb)	((struct packet_skb_cb *)((__skb)->cb))
 225
 226#define GET_PBDQC_FROM_RB(x)	((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 227#define GET_PBLOCK_DESC(x, bid)	\
 228	((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 229#define GET_CURR_PBLOCK_DESC_FROM_CORE(x)	\
 230	((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 231#define GET_NEXT_PRB_BLK_NUM(x) \
 232	(((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 233	((x)->kactive_blk_num+1) : 0)
 234
 235static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 236static void __fanout_link(struct sock *sk, struct packet_sock *po);
 237
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 238static int packet_direct_xmit(struct sk_buff *skb)
 239{
 
 
 
 
 
 
 
 240	return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
 241}
 242
 243static struct net_device *packet_cached_dev_get(struct packet_sock *po)
 244{
 245	struct net_device *dev;
 246
 247	rcu_read_lock();
 248	dev = rcu_dereference(po->cached_dev);
 249	if (likely(dev))
 250		dev_hold(dev);
 251	rcu_read_unlock();
 252
 253	return dev;
 254}
 255
 256static void packet_cached_dev_assign(struct packet_sock *po,
 257				     struct net_device *dev)
 258{
 259	rcu_assign_pointer(po->cached_dev, dev);
 260}
 261
 262static void packet_cached_dev_reset(struct packet_sock *po)
 263{
 264	RCU_INIT_POINTER(po->cached_dev, NULL);
 265}
 266
 267static bool packet_use_direct_xmit(const struct packet_sock *po)
 268{
 269	return po->xmit == packet_direct_xmit;
 270}
 271
 272static u16 packet_pick_tx_queue(struct sk_buff *skb)
 273{
 274	struct net_device *dev = skb->dev;
 275	const struct net_device_ops *ops = dev->netdev_ops;
 276	int cpu = raw_smp_processor_id();
 277	u16 queue_index;
 278
 279#ifdef CONFIG_XPS
 280	skb->sender_cpu = cpu + 1;
 281#endif
 282	skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues);
 283	if (ops->ndo_select_queue) {
 284		queue_index = ops->ndo_select_queue(dev, skb, NULL);
 285		queue_index = netdev_cap_txqueue(dev, queue_index);
 286	} else {
 287		queue_index = netdev_pick_tx(dev, skb, NULL);
 288	}
 289
 290	return queue_index;
 291}
 292
 293/* __register_prot_hook must be invoked through register_prot_hook
 294 * or from a context in which asynchronous accesses to the packet
 295 * socket is not possible (packet_create()).
 296 */
 297static void __register_prot_hook(struct sock *sk)
 298{
 299	struct packet_sock *po = pkt_sk(sk);
 300
 301	if (!po->running) {
 302		if (po->fanout)
 303			__fanout_link(sk, po);
 304		else
 305			dev_add_pack(&po->prot_hook);
 306
 307		sock_hold(sk);
 308		po->running = 1;
 309	}
 310}
 311
 312static void register_prot_hook(struct sock *sk)
 313{
 314	lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
 315	__register_prot_hook(sk);
 316}
 317
 318/* If the sync parameter is true, we will temporarily drop
 319 * the po->bind_lock and do a synchronize_net to make sure no
 320 * asynchronous packet processing paths still refer to the elements
 321 * of po->prot_hook.  If the sync parameter is false, it is the
 322 * callers responsibility to take care of this.
 323 */
 324static void __unregister_prot_hook(struct sock *sk, bool sync)
 325{
 326	struct packet_sock *po = pkt_sk(sk);
 327
 328	lockdep_assert_held_once(&po->bind_lock);
 329
 330	po->running = 0;
 331
 332	if (po->fanout)
 333		__fanout_unlink(sk, po);
 334	else
 335		__dev_remove_pack(&po->prot_hook);
 336
 337	__sock_put(sk);
 338
 339	if (sync) {
 340		spin_unlock(&po->bind_lock);
 341		synchronize_net();
 342		spin_lock(&po->bind_lock);
 343	}
 344}
 345
 346static void unregister_prot_hook(struct sock *sk, bool sync)
 347{
 348	struct packet_sock *po = pkt_sk(sk);
 349
 350	if (po->running)
 351		__unregister_prot_hook(sk, sync);
 352}
 353
 354static inline struct page * __pure pgv_to_page(void *addr)
 355{
 356	if (is_vmalloc_addr(addr))
 357		return vmalloc_to_page(addr);
 358	return virt_to_page(addr);
 359}
 360
 361static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 362{
 363	union tpacket_uhdr h;
 364
 365	h.raw = frame;
 366	switch (po->tp_version) {
 367	case TPACKET_V1:
 368		h.h1->tp_status = status;
 369		flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 370		break;
 371	case TPACKET_V2:
 372		h.h2->tp_status = status;
 373		flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 374		break;
 375	case TPACKET_V3:
 376		h.h3->tp_status = status;
 377		flush_dcache_page(pgv_to_page(&h.h3->tp_status));
 378		break;
 379	default:
 380		WARN(1, "TPACKET version not supported.\n");
 381		BUG();
 382	}
 383
 384	smp_wmb();
 385}
 386
 387static int __packet_get_status(const struct packet_sock *po, void *frame)
 388{
 389	union tpacket_uhdr h;
 390
 391	smp_rmb();
 392
 393	h.raw = frame;
 394	switch (po->tp_version) {
 395	case TPACKET_V1:
 396		flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 397		return h.h1->tp_status;
 398	case TPACKET_V2:
 399		flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 400		return h.h2->tp_status;
 401	case TPACKET_V3:
 402		flush_dcache_page(pgv_to_page(&h.h3->tp_status));
 403		return h.h3->tp_status;
 404	default:
 405		WARN(1, "TPACKET version not supported.\n");
 406		BUG();
 407		return 0;
 408	}
 409}
 410
 411static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec64 *ts,
 412				   unsigned int flags)
 413{
 414	struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 415
 416	if (shhwtstamps &&
 417	    (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
 418	    ktime_to_timespec64_cond(shhwtstamps->hwtstamp, ts))
 419		return TP_STATUS_TS_RAW_HARDWARE;
 420
 421	if (ktime_to_timespec64_cond(skb->tstamp, ts))
 
 422		return TP_STATUS_TS_SOFTWARE;
 423
 424	return 0;
 425}
 426
 427static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
 428				    struct sk_buff *skb)
 429{
 430	union tpacket_uhdr h;
 431	struct timespec64 ts;
 432	__u32 ts_status;
 433
 434	if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
 435		return 0;
 436
 437	h.raw = frame;
 438	/*
 439	 * versions 1 through 3 overflow the timestamps in y2106, since they
 440	 * all store the seconds in a 32-bit unsigned integer.
 441	 * If we create a version 4, that should have a 64-bit timestamp,
 442	 * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit
 443	 * nanoseconds.
 444	 */
 445	switch (po->tp_version) {
 446	case TPACKET_V1:
 447		h.h1->tp_sec = ts.tv_sec;
 448		h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
 449		break;
 450	case TPACKET_V2:
 451		h.h2->tp_sec = ts.tv_sec;
 452		h.h2->tp_nsec = ts.tv_nsec;
 453		break;
 454	case TPACKET_V3:
 455		h.h3->tp_sec = ts.tv_sec;
 456		h.h3->tp_nsec = ts.tv_nsec;
 457		break;
 458	default:
 459		WARN(1, "TPACKET version not supported.\n");
 460		BUG();
 461	}
 462
 463	/* one flush is safe, as both fields always lie on the same cacheline */
 464	flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
 465	smp_wmb();
 466
 467	return ts_status;
 468}
 469
 470static void *packet_lookup_frame(const struct packet_sock *po,
 471				 const struct packet_ring_buffer *rb,
 472				 unsigned int position,
 473				 int status)
 474{
 475	unsigned int pg_vec_pos, frame_offset;
 476	union tpacket_uhdr h;
 477
 478	pg_vec_pos = position / rb->frames_per_block;
 479	frame_offset = position % rb->frames_per_block;
 480
 481	h.raw = rb->pg_vec[pg_vec_pos].buffer +
 482		(frame_offset * rb->frame_size);
 483
 484	if (status != __packet_get_status(po, h.raw))
 485		return NULL;
 486
 487	return h.raw;
 488}
 489
 490static void *packet_current_frame(struct packet_sock *po,
 491		struct packet_ring_buffer *rb,
 492		int status)
 493{
 494	return packet_lookup_frame(po, rb, rb->head, status);
 495}
 496
 497static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 498{
 499	del_timer_sync(&pkc->retire_blk_timer);
 500}
 501
 502static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 503		struct sk_buff_head *rb_queue)
 504{
 505	struct tpacket_kbdq_core *pkc;
 506
 507	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 508
 509	spin_lock_bh(&rb_queue->lock);
 510	pkc->delete_blk_timer = 1;
 511	spin_unlock_bh(&rb_queue->lock);
 512
 513	prb_del_retire_blk_timer(pkc);
 514}
 515
 516static void prb_setup_retire_blk_timer(struct packet_sock *po)
 517{
 518	struct tpacket_kbdq_core *pkc;
 519
 520	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 521	timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
 522		    0);
 523	pkc->retire_blk_timer.expires = jiffies;
 524}
 525
 526static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 527				int blk_size_in_bytes)
 528{
 529	struct net_device *dev;
 530	unsigned int mbits, div;
 531	struct ethtool_link_ksettings ecmd;
 532	int err;
 533
 534	rtnl_lock();
 535	dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 536	if (unlikely(!dev)) {
 537		rtnl_unlock();
 538		return DEFAULT_PRB_RETIRE_TOV;
 539	}
 540	err = __ethtool_get_link_ksettings(dev, &ecmd);
 541	rtnl_unlock();
 542	if (err)
 543		return DEFAULT_PRB_RETIRE_TOV;
 544
 545	/* If the link speed is so slow you don't really
 546	 * need to worry about perf anyways
 547	 */
 548	if (ecmd.base.speed < SPEED_1000 ||
 549	    ecmd.base.speed == SPEED_UNKNOWN)
 550		return DEFAULT_PRB_RETIRE_TOV;
 551
 552	div = ecmd.base.speed / 1000;
 553	mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 554
 555	if (div)
 556		mbits /= div;
 557
 558	if (div)
 559		return mbits + 1;
 560	return mbits;
 561}
 562
 563static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 564			union tpacket_req_u *req_u)
 565{
 566	p1->feature_req_word = req_u->req3.tp_feature_req_word;
 567}
 568
 569static void init_prb_bdqc(struct packet_sock *po,
 570			struct packet_ring_buffer *rb,
 571			struct pgv *pg_vec,
 572			union tpacket_req_u *req_u)
 573{
 574	struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
 575	struct tpacket_block_desc *pbd;
 576
 577	memset(p1, 0x0, sizeof(*p1));
 578
 579	p1->knxt_seq_num = 1;
 580	p1->pkbdq = pg_vec;
 581	pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 582	p1->pkblk_start	= pg_vec[0].buffer;
 583	p1->kblk_size = req_u->req3.tp_block_size;
 584	p1->knum_blocks	= req_u->req3.tp_block_nr;
 585	p1->hdrlen = po->tp_hdrlen;
 586	p1->version = po->tp_version;
 587	p1->last_kactive_blk_num = 0;
 588	po->stats.stats3.tp_freeze_q_cnt = 0;
 589	if (req_u->req3.tp_retire_blk_tov)
 590		p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 591	else
 592		p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 593						req_u->req3.tp_block_size);
 594	p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 595	p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 596	rwlock_init(&p1->blk_fill_in_prog_lock);
 597
 598	p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
 599	prb_init_ft_ops(p1, req_u);
 600	prb_setup_retire_blk_timer(po);
 601	prb_open_block(p1, pbd);
 602}
 603
 604/*  Do NOT update the last_blk_num first.
 605 *  Assumes sk_buff_head lock is held.
 606 */
 607static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 608{
 609	mod_timer(&pkc->retire_blk_timer,
 610			jiffies + pkc->tov_in_jiffies);
 611	pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 612}
 613
 614/*
 615 * Timer logic:
 616 * 1) We refresh the timer only when we open a block.
 617 *    By doing this we don't waste cycles refreshing the timer
 618 *	  on packet-by-packet basis.
 619 *
 620 * With a 1MB block-size, on a 1Gbps line, it will take
 621 * i) ~8 ms to fill a block + ii) memcpy etc.
 622 * In this cut we are not accounting for the memcpy time.
 623 *
 624 * So, if the user sets the 'tmo' to 10ms then the timer
 625 * will never fire while the block is still getting filled
 626 * (which is what we want). However, the user could choose
 627 * to close a block early and that's fine.
 628 *
 629 * But when the timer does fire, we check whether or not to refresh it.
 630 * Since the tmo granularity is in msecs, it is not too expensive
 631 * to refresh the timer, lets say every '8' msecs.
 632 * Either the user can set the 'tmo' or we can derive it based on
 633 * a) line-speed and b) block-size.
 634 * prb_calc_retire_blk_tmo() calculates the tmo.
 635 *
 636 */
 637static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
 638{
 639	struct packet_sock *po =
 640		from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
 641	struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 642	unsigned int frozen;
 643	struct tpacket_block_desc *pbd;
 644
 645	spin_lock(&po->sk.sk_receive_queue.lock);
 646
 647	frozen = prb_queue_frozen(pkc);
 648	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 649
 650	if (unlikely(pkc->delete_blk_timer))
 651		goto out;
 652
 653	/* We only need to plug the race when the block is partially filled.
 654	 * tpacket_rcv:
 655	 *		lock(); increment BLOCK_NUM_PKTS; unlock()
 656	 *		copy_bits() is in progress ...
 657	 *		timer fires on other cpu:
 658	 *		we can't retire the current block because copy_bits
 659	 *		is in progress.
 660	 *
 661	 */
 662	if (BLOCK_NUM_PKTS(pbd)) {
 663		/* Waiting for skb_copy_bits to finish... */
 664		write_lock(&pkc->blk_fill_in_prog_lock);
 665		write_unlock(&pkc->blk_fill_in_prog_lock);
 666	}
 667
 668	if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 669		if (!frozen) {
 670			if (!BLOCK_NUM_PKTS(pbd)) {
 671				/* An empty block. Just refresh the timer. */
 672				goto refresh_timer;
 673			}
 674			prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 675			if (!prb_dispatch_next_block(pkc, po))
 676				goto refresh_timer;
 677			else
 678				goto out;
 679		} else {
 680			/* Case 1. Queue was frozen because user-space was
 681			 *	   lagging behind.
 682			 */
 683			if (prb_curr_blk_in_use(pbd)) {
 684				/*
 685				 * Ok, user-space is still behind.
 686				 * So just refresh the timer.
 687				 */
 688				goto refresh_timer;
 689			} else {
 690			       /* Case 2. queue was frozen,user-space caught up,
 691				* now the link went idle && the timer fired.
 692				* We don't have a block to close.So we open this
 693				* block and restart the timer.
 694				* opening a block thaws the queue,restarts timer
 695				* Thawing/timer-refresh is a side effect.
 696				*/
 697				prb_open_block(pkc, pbd);
 698				goto out;
 699			}
 700		}
 701	}
 702
 703refresh_timer:
 704	_prb_refresh_rx_retire_blk_timer(pkc);
 705
 706out:
 707	spin_unlock(&po->sk.sk_receive_queue.lock);
 708}
 709
 710static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 711		struct tpacket_block_desc *pbd1, __u32 status)
 712{
 713	/* Flush everything minus the block header */
 714
 715#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 716	u8 *start, *end;
 717
 718	start = (u8 *)pbd1;
 719
 720	/* Skip the block header(we know header WILL fit in 4K) */
 721	start += PAGE_SIZE;
 722
 723	end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 724	for (; start < end; start += PAGE_SIZE)
 725		flush_dcache_page(pgv_to_page(start));
 726
 727	smp_wmb();
 728#endif
 729
 730	/* Now update the block status. */
 731
 732	BLOCK_STATUS(pbd1) = status;
 733
 734	/* Flush the block header */
 735
 736#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 737	start = (u8 *)pbd1;
 738	flush_dcache_page(pgv_to_page(start));
 739
 740	smp_wmb();
 741#endif
 742}
 743
 744/*
 745 * Side effect:
 746 *
 747 * 1) flush the block
 748 * 2) Increment active_blk_num
 749 *
 750 * Note:We DONT refresh the timer on purpose.
 751 *	Because almost always the next block will be opened.
 752 */
 753static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 754		struct tpacket_block_desc *pbd1,
 755		struct packet_sock *po, unsigned int stat)
 756{
 757	__u32 status = TP_STATUS_USER | stat;
 758
 759	struct tpacket3_hdr *last_pkt;
 760	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 761	struct sock *sk = &po->sk;
 762
 763	if (atomic_read(&po->tp_drops))
 764		status |= TP_STATUS_LOSING;
 765
 766	last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 767	last_pkt->tp_next_offset = 0;
 768
 769	/* Get the ts of the last pkt */
 770	if (BLOCK_NUM_PKTS(pbd1)) {
 771		h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 772		h1->ts_last_pkt.ts_nsec	= last_pkt->tp_nsec;
 773	} else {
 774		/* Ok, we tmo'd - so get the current time.
 775		 *
 776		 * It shouldn't really happen as we don't close empty
 777		 * blocks. See prb_retire_rx_blk_timer_expired().
 778		 */
 779		struct timespec64 ts;
 780		ktime_get_real_ts64(&ts);
 781		h1->ts_last_pkt.ts_sec = ts.tv_sec;
 782		h1->ts_last_pkt.ts_nsec	= ts.tv_nsec;
 783	}
 784
 785	smp_wmb();
 786
 787	/* Flush the block */
 788	prb_flush_block(pkc1, pbd1, status);
 789
 790	sk->sk_data_ready(sk);
 791
 792	pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 793}
 794
 795static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 796{
 797	pkc->reset_pending_on_curr_blk = 0;
 798}
 799
 800/*
 801 * Side effect of opening a block:
 802 *
 803 * 1) prb_queue is thawed.
 804 * 2) retire_blk_timer is refreshed.
 805 *
 806 */
 807static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 808	struct tpacket_block_desc *pbd1)
 809{
 810	struct timespec64 ts;
 811	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 812
 813	smp_rmb();
 814
 815	/* We could have just memset this but we will lose the
 816	 * flexibility of making the priv area sticky
 817	 */
 818
 819	BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 820	BLOCK_NUM_PKTS(pbd1) = 0;
 821	BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 822
 823	ktime_get_real_ts64(&ts);
 824
 825	h1->ts_first_pkt.ts_sec = ts.tv_sec;
 826	h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 827
 828	pkc1->pkblk_start = (char *)pbd1;
 829	pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 830
 831	BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 832	BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 833
 834	pbd1->version = pkc1->version;
 835	pkc1->prev = pkc1->nxt_offset;
 836	pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 837
 838	prb_thaw_queue(pkc1);
 839	_prb_refresh_rx_retire_blk_timer(pkc1);
 840
 841	smp_wmb();
 842}
 843
 844/*
 845 * Queue freeze logic:
 846 * 1) Assume tp_block_nr = 8 blocks.
 847 * 2) At time 't0', user opens Rx ring.
 848 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 849 * 4) user-space is either sleeping or processing block '0'.
 850 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 851 *    it will close block-7,loop around and try to fill block '0'.
 852 *    call-flow:
 853 *    __packet_lookup_frame_in_block
 854 *      prb_retire_current_block()
 855 *      prb_dispatch_next_block()
 856 *        |->(BLOCK_STATUS == USER) evaluates to true
 857 *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 858 * 6) Now there are two cases:
 859 *    6.1) Link goes idle right after the queue is frozen.
 860 *         But remember, the last open_block() refreshed the timer.
 861 *         When this timer expires,it will refresh itself so that we can
 862 *         re-open block-0 in near future.
 863 *    6.2) Link is busy and keeps on receiving packets. This is a simple
 864 *         case and __packet_lookup_frame_in_block will check if block-0
 865 *         is free and can now be re-used.
 866 */
 867static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 868				  struct packet_sock *po)
 869{
 870	pkc->reset_pending_on_curr_blk = 1;
 871	po->stats.stats3.tp_freeze_q_cnt++;
 872}
 873
 874#define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 875
 876/*
 877 * If the next block is free then we will dispatch it
 878 * and return a good offset.
 879 * Else, we will freeze the queue.
 880 * So, caller must check the return value.
 881 */
 882static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 883		struct packet_sock *po)
 884{
 885	struct tpacket_block_desc *pbd;
 886
 887	smp_rmb();
 888
 889	/* 1. Get current block num */
 890	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 891
 892	/* 2. If this block is currently in_use then freeze the queue */
 893	if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 894		prb_freeze_queue(pkc, po);
 895		return NULL;
 896	}
 897
 898	/*
 899	 * 3.
 900	 * open this block and return the offset where the first packet
 901	 * needs to get stored.
 902	 */
 903	prb_open_block(pkc, pbd);
 904	return (void *)pkc->nxt_offset;
 905}
 906
 907static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 908		struct packet_sock *po, unsigned int status)
 909{
 910	struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 911
 912	/* retire/close the current block */
 913	if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 914		/*
 915		 * Plug the case where copy_bits() is in progress on
 916		 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 917		 * have space to copy the pkt in the current block and
 918		 * called prb_retire_current_block()
 919		 *
 920		 * We don't need to worry about the TMO case because
 921		 * the timer-handler already handled this case.
 922		 */
 923		if (!(status & TP_STATUS_BLK_TMO)) {
 924			/* Waiting for skb_copy_bits to finish... */
 925			write_lock(&pkc->blk_fill_in_prog_lock);
 926			write_unlock(&pkc->blk_fill_in_prog_lock);
 927		}
 928		prb_close_block(pkc, pbd, po, status);
 929		return;
 930	}
 931}
 932
 933static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
 934{
 935	return TP_STATUS_USER & BLOCK_STATUS(pbd);
 936}
 937
 938static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 939{
 940	return pkc->reset_pending_on_curr_blk;
 941}
 942
 943static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 944	__releases(&pkc->blk_fill_in_prog_lock)
 945{
 946	struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 947
 948	read_unlock(&pkc->blk_fill_in_prog_lock);
 949}
 950
 951static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 952			struct tpacket3_hdr *ppd)
 953{
 954	ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
 955}
 956
 957static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 958			struct tpacket3_hdr *ppd)
 959{
 960	ppd->hv1.tp_rxhash = 0;
 961}
 962
 963static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
 964			struct tpacket3_hdr *ppd)
 965{
 966	if (skb_vlan_tag_present(pkc->skb)) {
 967		ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
 968		ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
 969		ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
 970	} else {
 971		ppd->hv1.tp_vlan_tci = 0;
 972		ppd->hv1.tp_vlan_tpid = 0;
 973		ppd->tp_status = TP_STATUS_AVAILABLE;
 974	}
 975}
 976
 977static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
 978			struct tpacket3_hdr *ppd)
 979{
 980	ppd->hv1.tp_padding = 0;
 981	prb_fill_vlan_info(pkc, ppd);
 982
 983	if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
 984		prb_fill_rxhash(pkc, ppd);
 985	else
 986		prb_clear_rxhash(pkc, ppd);
 987}
 988
 989static void prb_fill_curr_block(char *curr,
 990				struct tpacket_kbdq_core *pkc,
 991				struct tpacket_block_desc *pbd,
 992				unsigned int len)
 993	__acquires(&pkc->blk_fill_in_prog_lock)
 994{
 995	struct tpacket3_hdr *ppd;
 996
 997	ppd  = (struct tpacket3_hdr *)curr;
 998	ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
 999	pkc->prev = curr;
1000	pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1001	BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1002	BLOCK_NUM_PKTS(pbd) += 1;
1003	read_lock(&pkc->blk_fill_in_prog_lock);
1004	prb_run_all_ft_ops(pkc, ppd);
1005}
1006
1007/* Assumes caller has the sk->rx_queue.lock */
1008static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1009					    struct sk_buff *skb,
1010					    unsigned int len
1011					    )
1012{
1013	struct tpacket_kbdq_core *pkc;
1014	struct tpacket_block_desc *pbd;
1015	char *curr, *end;
1016
1017	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1018	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1019
1020	/* Queue is frozen when user space is lagging behind */
1021	if (prb_queue_frozen(pkc)) {
1022		/*
1023		 * Check if that last block which caused the queue to freeze,
1024		 * is still in_use by user-space.
1025		 */
1026		if (prb_curr_blk_in_use(pbd)) {
1027			/* Can't record this packet */
1028			return NULL;
1029		} else {
1030			/*
1031			 * Ok, the block was released by user-space.
1032			 * Now let's open that block.
1033			 * opening a block also thaws the queue.
1034			 * Thawing is a side effect.
1035			 */
1036			prb_open_block(pkc, pbd);
1037		}
1038	}
1039
1040	smp_mb();
1041	curr = pkc->nxt_offset;
1042	pkc->skb = skb;
1043	end = (char *)pbd + pkc->kblk_size;
1044
1045	/* first try the current block */
1046	if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1047		prb_fill_curr_block(curr, pkc, pbd, len);
1048		return (void *)curr;
1049	}
1050
1051	/* Ok, close the current block */
1052	prb_retire_current_block(pkc, po, 0);
1053
1054	/* Now, try to dispatch the next block */
1055	curr = (char *)prb_dispatch_next_block(pkc, po);
1056	if (curr) {
1057		pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1058		prb_fill_curr_block(curr, pkc, pbd, len);
1059		return (void *)curr;
1060	}
1061
1062	/*
1063	 * No free blocks are available.user_space hasn't caught up yet.
1064	 * Queue was just frozen and now this packet will get dropped.
1065	 */
1066	return NULL;
1067}
1068
1069static void *packet_current_rx_frame(struct packet_sock *po,
1070					    struct sk_buff *skb,
1071					    int status, unsigned int len)
1072{
1073	char *curr = NULL;
1074	switch (po->tp_version) {
1075	case TPACKET_V1:
1076	case TPACKET_V2:
1077		curr = packet_lookup_frame(po, &po->rx_ring,
1078					po->rx_ring.head, status);
1079		return curr;
1080	case TPACKET_V3:
1081		return __packet_lookup_frame_in_block(po, skb, len);
1082	default:
1083		WARN(1, "TPACKET version not supported\n");
1084		BUG();
1085		return NULL;
1086	}
1087}
1088
1089static void *prb_lookup_block(const struct packet_sock *po,
1090			      const struct packet_ring_buffer *rb,
1091			      unsigned int idx,
1092			      int status)
1093{
1094	struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1095	struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1096
1097	if (status != BLOCK_STATUS(pbd))
1098		return NULL;
1099	return pbd;
1100}
1101
1102static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1103{
1104	unsigned int prev;
1105	if (rb->prb_bdqc.kactive_blk_num)
1106		prev = rb->prb_bdqc.kactive_blk_num-1;
1107	else
1108		prev = rb->prb_bdqc.knum_blocks-1;
1109	return prev;
1110}
1111
1112/* Assumes caller has held the rx_queue.lock */
1113static void *__prb_previous_block(struct packet_sock *po,
1114					 struct packet_ring_buffer *rb,
1115					 int status)
1116{
1117	unsigned int previous = prb_previous_blk_num(rb);
1118	return prb_lookup_block(po, rb, previous, status);
1119}
1120
1121static void *packet_previous_rx_frame(struct packet_sock *po,
1122					     struct packet_ring_buffer *rb,
1123					     int status)
1124{
1125	if (po->tp_version <= TPACKET_V2)
1126		return packet_previous_frame(po, rb, status);
1127
1128	return __prb_previous_block(po, rb, status);
1129}
1130
1131static void packet_increment_rx_head(struct packet_sock *po,
1132					    struct packet_ring_buffer *rb)
1133{
1134	switch (po->tp_version) {
1135	case TPACKET_V1:
1136	case TPACKET_V2:
1137		return packet_increment_head(rb);
1138	case TPACKET_V3:
1139	default:
1140		WARN(1, "TPACKET version not supported.\n");
1141		BUG();
1142		return;
1143	}
1144}
1145
1146static void *packet_previous_frame(struct packet_sock *po,
1147		struct packet_ring_buffer *rb,
1148		int status)
1149{
1150	unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1151	return packet_lookup_frame(po, rb, previous, status);
1152}
1153
1154static void packet_increment_head(struct packet_ring_buffer *buff)
1155{
1156	buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1157}
1158
1159static void packet_inc_pending(struct packet_ring_buffer *rb)
1160{
1161	this_cpu_inc(*rb->pending_refcnt);
1162}
1163
1164static void packet_dec_pending(struct packet_ring_buffer *rb)
1165{
1166	this_cpu_dec(*rb->pending_refcnt);
1167}
1168
1169static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1170{
1171	unsigned int refcnt = 0;
1172	int cpu;
1173
1174	/* We don't use pending refcount in rx_ring. */
1175	if (rb->pending_refcnt == NULL)
1176		return 0;
1177
1178	for_each_possible_cpu(cpu)
1179		refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1180
1181	return refcnt;
1182}
1183
1184static int packet_alloc_pending(struct packet_sock *po)
1185{
1186	po->rx_ring.pending_refcnt = NULL;
1187
1188	po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1189	if (unlikely(po->tx_ring.pending_refcnt == NULL))
1190		return -ENOBUFS;
1191
1192	return 0;
1193}
1194
1195static void packet_free_pending(struct packet_sock *po)
1196{
1197	free_percpu(po->tx_ring.pending_refcnt);
1198}
1199
1200#define ROOM_POW_OFF	2
1201#define ROOM_NONE	0x0
1202#define ROOM_LOW	0x1
1203#define ROOM_NORMAL	0x2
1204
1205static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1206{
1207	int idx, len;
1208
1209	len = READ_ONCE(po->rx_ring.frame_max) + 1;
1210	idx = READ_ONCE(po->rx_ring.head);
1211	if (pow_off)
1212		idx += len >> pow_off;
1213	if (idx >= len)
1214		idx -= len;
1215	return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1216}
1217
1218static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1219{
1220	int idx, len;
1221
1222	len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1223	idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1224	if (pow_off)
1225		idx += len >> pow_off;
1226	if (idx >= len)
1227		idx -= len;
1228	return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1229}
1230
1231static int __packet_rcv_has_room(const struct packet_sock *po,
1232				 const struct sk_buff *skb)
1233{
1234	const struct sock *sk = &po->sk;
1235	int ret = ROOM_NONE;
1236
1237	if (po->prot_hook.func != tpacket_rcv) {
1238		int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1239		int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1240				   - (skb ? skb->truesize : 0);
1241
1242		if (avail > (rcvbuf >> ROOM_POW_OFF))
1243			return ROOM_NORMAL;
1244		else if (avail > 0)
1245			return ROOM_LOW;
1246		else
1247			return ROOM_NONE;
1248	}
1249
1250	if (po->tp_version == TPACKET_V3) {
1251		if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1252			ret = ROOM_NORMAL;
1253		else if (__tpacket_v3_has_room(po, 0))
1254			ret = ROOM_LOW;
1255	} else {
1256		if (__tpacket_has_room(po, ROOM_POW_OFF))
1257			ret = ROOM_NORMAL;
1258		else if (__tpacket_has_room(po, 0))
1259			ret = ROOM_LOW;
1260	}
1261
1262	return ret;
1263}
1264
1265static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1266{
1267	int pressure, ret;
1268
1269	ret = __packet_rcv_has_room(po, skb);
1270	pressure = ret != ROOM_NORMAL;
1271
1272	if (READ_ONCE(po->pressure) != pressure)
1273		WRITE_ONCE(po->pressure, pressure);
1274
1275	return ret;
1276}
1277
1278static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1279{
1280	if (READ_ONCE(po->pressure) &&
1281	    __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1282		WRITE_ONCE(po->pressure,  0);
1283}
1284
1285static void packet_sock_destruct(struct sock *sk)
1286{
1287	skb_queue_purge(&sk->sk_error_queue);
1288
1289	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1290	WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1291
1292	if (!sock_flag(sk, SOCK_DEAD)) {
1293		pr_err("Attempt to release alive packet socket: %p\n", sk);
1294		return;
1295	}
1296
1297	sk_refcnt_debug_dec(sk);
1298}
1299
1300static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1301{
1302	u32 *history = po->rollover->history;
1303	u32 victim, rxhash;
1304	int i, count = 0;
1305
1306	rxhash = skb_get_hash(skb);
1307	for (i = 0; i < ROLLOVER_HLEN; i++)
1308		if (READ_ONCE(history[i]) == rxhash)
1309			count++;
1310
1311	victim = prandom_u32() % ROLLOVER_HLEN;
1312
1313	/* Avoid dirtying the cache line if possible */
1314	if (READ_ONCE(history[victim]) != rxhash)
1315		WRITE_ONCE(history[victim], rxhash);
1316
1317	return count > (ROLLOVER_HLEN >> 1);
1318}
1319
1320static unsigned int fanout_demux_hash(struct packet_fanout *f,
1321				      struct sk_buff *skb,
1322				      unsigned int num)
1323{
1324	return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1325}
1326
1327static unsigned int fanout_demux_lb(struct packet_fanout *f,
1328				    struct sk_buff *skb,
1329				    unsigned int num)
1330{
1331	unsigned int val = atomic_inc_return(&f->rr_cur);
1332
1333	return val % num;
1334}
1335
1336static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1337				     struct sk_buff *skb,
1338				     unsigned int num)
1339{
1340	return smp_processor_id() % num;
1341}
1342
1343static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1344				     struct sk_buff *skb,
1345				     unsigned int num)
1346{
1347	return prandom_u32_max(num);
1348}
1349
1350static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1351					  struct sk_buff *skb,
1352					  unsigned int idx, bool try_self,
1353					  unsigned int num)
1354{
1355	struct packet_sock *po, *po_next, *po_skip = NULL;
1356	unsigned int i, j, room = ROOM_NONE;
1357
1358	po = pkt_sk(f->arr[idx]);
1359
1360	if (try_self) {
1361		room = packet_rcv_has_room(po, skb);
1362		if (room == ROOM_NORMAL ||
1363		    (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1364			return idx;
1365		po_skip = po;
1366	}
1367
1368	i = j = min_t(int, po->rollover->sock, num - 1);
1369	do {
1370		po_next = pkt_sk(f->arr[i]);
1371		if (po_next != po_skip && !READ_ONCE(po_next->pressure) &&
1372		    packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1373			if (i != j)
1374				po->rollover->sock = i;
1375			atomic_long_inc(&po->rollover->num);
1376			if (room == ROOM_LOW)
1377				atomic_long_inc(&po->rollover->num_huge);
1378			return i;
1379		}
1380
1381		if (++i == num)
1382			i = 0;
1383	} while (i != j);
1384
1385	atomic_long_inc(&po->rollover->num_failed);
1386	return idx;
1387}
1388
1389static unsigned int fanout_demux_qm(struct packet_fanout *f,
1390				    struct sk_buff *skb,
1391				    unsigned int num)
1392{
1393	return skb_get_queue_mapping(skb) % num;
1394}
1395
1396static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1397				     struct sk_buff *skb,
1398				     unsigned int num)
1399{
1400	struct bpf_prog *prog;
1401	unsigned int ret = 0;
1402
1403	rcu_read_lock();
1404	prog = rcu_dereference(f->bpf_prog);
1405	if (prog)
1406		ret = bpf_prog_run_clear_cb(prog, skb) % num;
1407	rcu_read_unlock();
1408
1409	return ret;
1410}
1411
1412static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1413{
1414	return f->flags & (flag >> 8);
1415}
1416
1417static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1418			     struct packet_type *pt, struct net_device *orig_dev)
1419{
1420	struct packet_fanout *f = pt->af_packet_priv;
1421	unsigned int num = READ_ONCE(f->num_members);
1422	struct net *net = read_pnet(&f->net);
1423	struct packet_sock *po;
1424	unsigned int idx;
1425
1426	if (!net_eq(dev_net(dev), net) || !num) {
1427		kfree_skb(skb);
1428		return 0;
1429	}
1430
1431	if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1432		skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1433		if (!skb)
1434			return 0;
1435	}
1436	switch (f->type) {
1437	case PACKET_FANOUT_HASH:
1438	default:
1439		idx = fanout_demux_hash(f, skb, num);
1440		break;
1441	case PACKET_FANOUT_LB:
1442		idx = fanout_demux_lb(f, skb, num);
1443		break;
1444	case PACKET_FANOUT_CPU:
1445		idx = fanout_demux_cpu(f, skb, num);
1446		break;
1447	case PACKET_FANOUT_RND:
1448		idx = fanout_demux_rnd(f, skb, num);
1449		break;
1450	case PACKET_FANOUT_QM:
1451		idx = fanout_demux_qm(f, skb, num);
1452		break;
1453	case PACKET_FANOUT_ROLLOVER:
1454		idx = fanout_demux_rollover(f, skb, 0, false, num);
1455		break;
1456	case PACKET_FANOUT_CBPF:
1457	case PACKET_FANOUT_EBPF:
1458		idx = fanout_demux_bpf(f, skb, num);
1459		break;
1460	}
1461
1462	if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1463		idx = fanout_demux_rollover(f, skb, idx, true, num);
1464
1465	po = pkt_sk(f->arr[idx]);
1466	return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1467}
1468
1469DEFINE_MUTEX(fanout_mutex);
1470EXPORT_SYMBOL_GPL(fanout_mutex);
1471static LIST_HEAD(fanout_list);
1472static u16 fanout_next_id;
1473
1474static void __fanout_link(struct sock *sk, struct packet_sock *po)
1475{
1476	struct packet_fanout *f = po->fanout;
1477
1478	spin_lock(&f->lock);
1479	f->arr[f->num_members] = sk;
1480	smp_wmb();
1481	f->num_members++;
1482	if (f->num_members == 1)
1483		dev_add_pack(&f->prot_hook);
1484	spin_unlock(&f->lock);
1485}
1486
1487static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1488{
1489	struct packet_fanout *f = po->fanout;
1490	int i;
1491
1492	spin_lock(&f->lock);
1493	for (i = 0; i < f->num_members; i++) {
1494		if (f->arr[i] == sk)
 
1495			break;
1496	}
1497	BUG_ON(i >= f->num_members);
1498	f->arr[i] = f->arr[f->num_members - 1];
 
 
1499	f->num_members--;
1500	if (f->num_members == 0)
1501		__dev_remove_pack(&f->prot_hook);
1502	spin_unlock(&f->lock);
1503}
1504
1505static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1506{
1507	if (sk->sk_family != PF_PACKET)
1508		return false;
1509
1510	return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1511}
1512
1513static void fanout_init_data(struct packet_fanout *f)
1514{
1515	switch (f->type) {
1516	case PACKET_FANOUT_LB:
1517		atomic_set(&f->rr_cur, 0);
1518		break;
1519	case PACKET_FANOUT_CBPF:
1520	case PACKET_FANOUT_EBPF:
1521		RCU_INIT_POINTER(f->bpf_prog, NULL);
1522		break;
1523	}
1524}
1525
1526static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1527{
1528	struct bpf_prog *old;
1529
1530	spin_lock(&f->lock);
1531	old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1532	rcu_assign_pointer(f->bpf_prog, new);
1533	spin_unlock(&f->lock);
1534
1535	if (old) {
1536		synchronize_net();
1537		bpf_prog_destroy(old);
1538	}
1539}
1540
1541static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data,
1542				unsigned int len)
1543{
1544	struct bpf_prog *new;
1545	struct sock_fprog fprog;
1546	int ret;
1547
1548	if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1549		return -EPERM;
1550
1551	ret = copy_bpf_fprog_from_user(&fprog, data, len);
1552	if (ret)
1553		return ret;
1554
1555	ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1556	if (ret)
1557		return ret;
1558
1559	__fanout_set_data_bpf(po->fanout, new);
1560	return 0;
1561}
1562
1563static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data,
1564				unsigned int len)
1565{
1566	struct bpf_prog *new;
1567	u32 fd;
1568
1569	if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1570		return -EPERM;
1571	if (len != sizeof(fd))
1572		return -EINVAL;
1573	if (copy_from_sockptr(&fd, data, len))
1574		return -EFAULT;
1575
1576	new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1577	if (IS_ERR(new))
1578		return PTR_ERR(new);
1579
1580	__fanout_set_data_bpf(po->fanout, new);
1581	return 0;
1582}
1583
1584static int fanout_set_data(struct packet_sock *po, sockptr_t data,
1585			   unsigned int len)
1586{
1587	switch (po->fanout->type) {
1588	case PACKET_FANOUT_CBPF:
1589		return fanout_set_data_cbpf(po, data, len);
1590	case PACKET_FANOUT_EBPF:
1591		return fanout_set_data_ebpf(po, data, len);
1592	default:
1593		return -EINVAL;
1594	}
1595}
1596
1597static void fanout_release_data(struct packet_fanout *f)
1598{
1599	switch (f->type) {
1600	case PACKET_FANOUT_CBPF:
1601	case PACKET_FANOUT_EBPF:
1602		__fanout_set_data_bpf(f, NULL);
1603	}
1604}
1605
1606static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1607{
1608	struct packet_fanout *f;
1609
1610	list_for_each_entry(f, &fanout_list, list) {
1611		if (f->id == candidate_id &&
1612		    read_pnet(&f->net) == sock_net(sk)) {
1613			return false;
1614		}
1615	}
1616	return true;
1617}
1618
1619static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1620{
1621	u16 id = fanout_next_id;
1622
1623	do {
1624		if (__fanout_id_is_free(sk, id)) {
1625			*new_id = id;
1626			fanout_next_id = id + 1;
1627			return true;
1628		}
1629
1630		id++;
1631	} while (id != fanout_next_id);
1632
1633	return false;
1634}
1635
1636static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1637{
1638	struct packet_rollover *rollover = NULL;
1639	struct packet_sock *po = pkt_sk(sk);
 
1640	struct packet_fanout *f, *match;
1641	u8 type = type_flags & 0xff;
1642	u8 flags = type_flags >> 8;
 
1643	int err;
1644
1645	switch (type) {
1646	case PACKET_FANOUT_ROLLOVER:
1647		if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1648			return -EINVAL;
 
1649	case PACKET_FANOUT_HASH:
1650	case PACKET_FANOUT_LB:
1651	case PACKET_FANOUT_CPU:
1652	case PACKET_FANOUT_RND:
1653	case PACKET_FANOUT_QM:
1654	case PACKET_FANOUT_CBPF:
1655	case PACKET_FANOUT_EBPF:
1656		break;
1657	default:
1658		return -EINVAL;
1659	}
1660
1661	mutex_lock(&fanout_mutex);
1662
1663	err = -EALREADY;
1664	if (po->fanout)
1665		goto out;
1666
1667	if (type == PACKET_FANOUT_ROLLOVER ||
1668	    (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1669		err = -ENOMEM;
1670		rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1671		if (!rollover)
1672			goto out;
1673		atomic_long_set(&rollover->num, 0);
1674		atomic_long_set(&rollover->num_huge, 0);
1675		atomic_long_set(&rollover->num_failed, 0);
1676	}
1677
1678	if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1679		if (id != 0) {
1680			err = -EINVAL;
1681			goto out;
1682		}
1683		if (!fanout_find_new_id(sk, &id)) {
1684			err = -ENOMEM;
1685			goto out;
1686		}
1687		/* ephemeral flag for the first socket in the group: drop it */
1688		flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1689	}
1690
1691	match = NULL;
1692	list_for_each_entry(f, &fanout_list, list) {
1693		if (f->id == id &&
1694		    read_pnet(&f->net) == sock_net(sk)) {
1695			match = f;
1696			break;
1697		}
1698	}
1699	err = -EINVAL;
1700	if (match && match->flags != flags)
1701		goto out;
1702	if (!match) {
 
 
 
 
 
 
 
 
 
1703		err = -ENOMEM;
1704		match = kzalloc(sizeof(*match), GFP_KERNEL);
 
1705		if (!match)
1706			goto out;
1707		write_pnet(&match->net, sock_net(sk));
1708		match->id = id;
1709		match->type = type;
1710		match->flags = flags;
1711		INIT_LIST_HEAD(&match->list);
1712		spin_lock_init(&match->lock);
1713		refcount_set(&match->sk_ref, 0);
1714		fanout_init_data(match);
1715		match->prot_hook.type = po->prot_hook.type;
1716		match->prot_hook.dev = po->prot_hook.dev;
1717		match->prot_hook.func = packet_rcv_fanout;
1718		match->prot_hook.af_packet_priv = match;
 
1719		match->prot_hook.id_match = match_fanout_group;
 
 
1720		list_add(&match->list, &fanout_list);
1721	}
1722	err = -EINVAL;
1723
1724	spin_lock(&po->bind_lock);
1725	if (po->running &&
1726	    match->type == type &&
1727	    match->prot_hook.type == po->prot_hook.type &&
1728	    match->prot_hook.dev == po->prot_hook.dev) {
1729		err = -ENOSPC;
1730		if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1731			__dev_remove_pack(&po->prot_hook);
1732			po->fanout = match;
 
 
 
1733			po->rollover = rollover;
1734			rollover = NULL;
1735			refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1736			__fanout_link(sk, po);
1737			err = 0;
1738		}
1739	}
1740	spin_unlock(&po->bind_lock);
1741
1742	if (err && !refcount_read(&match->sk_ref)) {
1743		list_del(&match->list);
1744		kfree(match);
1745	}
1746
1747out:
1748	kfree(rollover);
1749	mutex_unlock(&fanout_mutex);
1750	return err;
1751}
1752
1753/* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1754 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1755 * It is the responsibility of the caller to call fanout_release_data() and
1756 * free the returned packet_fanout (after synchronize_net())
1757 */
1758static struct packet_fanout *fanout_release(struct sock *sk)
1759{
1760	struct packet_sock *po = pkt_sk(sk);
1761	struct packet_fanout *f;
1762
1763	mutex_lock(&fanout_mutex);
1764	f = po->fanout;
1765	if (f) {
1766		po->fanout = NULL;
1767
1768		if (refcount_dec_and_test(&f->sk_ref))
1769			list_del(&f->list);
1770		else
1771			f = NULL;
1772	}
1773	mutex_unlock(&fanout_mutex);
1774
1775	return f;
1776}
1777
1778static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1779					  struct sk_buff *skb)
1780{
1781	/* Earlier code assumed this would be a VLAN pkt, double-check
1782	 * this now that we have the actual packet in hand. We can only
1783	 * do this check on Ethernet devices.
1784	 */
1785	if (unlikely(dev->type != ARPHRD_ETHER))
1786		return false;
1787
1788	skb_reset_mac_header(skb);
1789	return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1790}
1791
1792static const struct proto_ops packet_ops;
1793
1794static const struct proto_ops packet_ops_spkt;
1795
1796static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1797			   struct packet_type *pt, struct net_device *orig_dev)
1798{
1799	struct sock *sk;
1800	struct sockaddr_pkt *spkt;
1801
1802	/*
1803	 *	When we registered the protocol we saved the socket in the data
1804	 *	field for just this event.
1805	 */
1806
1807	sk = pt->af_packet_priv;
1808
1809	/*
1810	 *	Yank back the headers [hope the device set this
1811	 *	right or kerboom...]
1812	 *
1813	 *	Incoming packets have ll header pulled,
1814	 *	push it back.
1815	 *
1816	 *	For outgoing ones skb->data == skb_mac_header(skb)
1817	 *	so that this procedure is noop.
1818	 */
1819
1820	if (skb->pkt_type == PACKET_LOOPBACK)
1821		goto out;
1822
1823	if (!net_eq(dev_net(dev), sock_net(sk)))
1824		goto out;
1825
1826	skb = skb_share_check(skb, GFP_ATOMIC);
1827	if (skb == NULL)
1828		goto oom;
1829
1830	/* drop any routing info */
1831	skb_dst_drop(skb);
1832
1833	/* drop conntrack reference */
1834	nf_reset_ct(skb);
1835
1836	spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1837
1838	skb_push(skb, skb->data - skb_mac_header(skb));
1839
1840	/*
1841	 *	The SOCK_PACKET socket receives _all_ frames.
1842	 */
1843
1844	spkt->spkt_family = dev->type;
1845	strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1846	spkt->spkt_protocol = skb->protocol;
1847
1848	/*
1849	 *	Charge the memory to the socket. This is done specifically
1850	 *	to prevent sockets using all the memory up.
1851	 */
1852
1853	if (sock_queue_rcv_skb(sk, skb) == 0)
1854		return 0;
1855
1856out:
1857	kfree_skb(skb);
1858oom:
1859	return 0;
1860}
1861
1862static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1863{
 
 
1864	if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1865	    sock->type == SOCK_RAW) {
1866		skb_reset_mac_header(skb);
1867		skb->protocol = dev_parse_header_protocol(skb);
1868	}
1869
 
 
 
 
 
 
 
 
1870	skb_probe_transport_header(skb);
1871}
1872
1873/*
1874 *	Output a raw packet to a device layer. This bypasses all the other
1875 *	protocol layers and you must therefore supply it with a complete frame
1876 */
1877
1878static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1879			       size_t len)
1880{
1881	struct sock *sk = sock->sk;
1882	DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1883	struct sk_buff *skb = NULL;
1884	struct net_device *dev;
1885	struct sockcm_cookie sockc;
1886	__be16 proto = 0;
1887	int err;
1888	int extra_len = 0;
1889
1890	/*
1891	 *	Get and verify the address.
1892	 */
1893
1894	if (saddr) {
1895		if (msg->msg_namelen < sizeof(struct sockaddr))
1896			return -EINVAL;
1897		if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1898			proto = saddr->spkt_protocol;
1899	} else
1900		return -ENOTCONN;	/* SOCK_PACKET must be sent giving an address */
1901
1902	/*
1903	 *	Find the device first to size check it
1904	 */
1905
1906	saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1907retry:
1908	rcu_read_lock();
1909	dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1910	err = -ENODEV;
1911	if (dev == NULL)
1912		goto out_unlock;
1913
1914	err = -ENETDOWN;
1915	if (!(dev->flags & IFF_UP))
1916		goto out_unlock;
1917
1918	/*
1919	 * You may not queue a frame bigger than the mtu. This is the lowest level
1920	 * raw protocol and you must do your own fragmentation at this level.
1921	 */
1922
1923	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1924		if (!netif_supports_nofcs(dev)) {
1925			err = -EPROTONOSUPPORT;
1926			goto out_unlock;
1927		}
1928		extra_len = 4; /* We're doing our own CRC */
1929	}
1930
1931	err = -EMSGSIZE;
1932	if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1933		goto out_unlock;
1934
1935	if (!skb) {
1936		size_t reserved = LL_RESERVED_SPACE(dev);
1937		int tlen = dev->needed_tailroom;
1938		unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1939
1940		rcu_read_unlock();
1941		skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1942		if (skb == NULL)
1943			return -ENOBUFS;
1944		/* FIXME: Save some space for broken drivers that write a hard
1945		 * header at transmission time by themselves. PPP is the notable
1946		 * one here. This should really be fixed at the driver level.
1947		 */
1948		skb_reserve(skb, reserved);
1949		skb_reset_network_header(skb);
1950
1951		/* Try to align data part correctly */
1952		if (hhlen) {
1953			skb->data -= hhlen;
1954			skb->tail -= hhlen;
1955			if (len < hhlen)
1956				skb_reset_network_header(skb);
1957		}
1958		err = memcpy_from_msg(skb_put(skb, len), msg, len);
1959		if (err)
1960			goto out_free;
1961		goto retry;
1962	}
1963
1964	if (!dev_validate_header(dev, skb->data, len)) {
1965		err = -EINVAL;
1966		goto out_unlock;
1967	}
1968	if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1969	    !packet_extra_vlan_len_allowed(dev, skb)) {
1970		err = -EMSGSIZE;
1971		goto out_unlock;
1972	}
1973
1974	sockcm_init(&sockc, sk);
1975	if (msg->msg_controllen) {
1976		err = sock_cmsg_send(sk, msg, &sockc);
1977		if (unlikely(err))
1978			goto out_unlock;
1979	}
1980
1981	skb->protocol = proto;
1982	skb->dev = dev;
1983	skb->priority = sk->sk_priority;
1984	skb->mark = sk->sk_mark;
1985	skb->tstamp = sockc.transmit_time;
1986
1987	skb_setup_tx_timestamp(skb, sockc.tsflags);
1988
1989	if (unlikely(extra_len == 4))
1990		skb->no_fcs = 1;
1991
1992	packet_parse_headers(skb, sock);
1993
1994	dev_queue_xmit(skb);
1995	rcu_read_unlock();
1996	return len;
1997
1998out_unlock:
1999	rcu_read_unlock();
2000out_free:
2001	kfree_skb(skb);
2002	return err;
2003}
2004
2005static unsigned int run_filter(struct sk_buff *skb,
2006			       const struct sock *sk,
2007			       unsigned int res)
2008{
2009	struct sk_filter *filter;
2010
2011	rcu_read_lock();
2012	filter = rcu_dereference(sk->sk_filter);
2013	if (filter != NULL)
2014		res = bpf_prog_run_clear_cb(filter->prog, skb);
2015	rcu_read_unlock();
2016
2017	return res;
2018}
2019
2020static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2021			   size_t *len)
2022{
2023	struct virtio_net_hdr vnet_hdr;
2024
2025	if (*len < sizeof(vnet_hdr))
2026		return -EINVAL;
2027	*len -= sizeof(vnet_hdr);
2028
2029	if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2030		return -EINVAL;
2031
2032	return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2033}
2034
2035/*
2036 * This function makes lazy skb cloning in hope that most of packets
2037 * are discarded by BPF.
2038 *
2039 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2040 * and skb->cb are mangled. It works because (and until) packets
2041 * falling here are owned by current CPU. Output packets are cloned
2042 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2043 * sequencially, so that if we return skb to original state on exit,
2044 * we will not harm anyone.
2045 */
2046
2047static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2048		      struct packet_type *pt, struct net_device *orig_dev)
2049{
2050	struct sock *sk;
2051	struct sockaddr_ll *sll;
2052	struct packet_sock *po;
2053	u8 *skb_head = skb->data;
2054	int skb_len = skb->len;
2055	unsigned int snaplen, res;
2056	bool is_drop_n_account = false;
2057
2058	if (skb->pkt_type == PACKET_LOOPBACK)
2059		goto drop;
2060
2061	sk = pt->af_packet_priv;
2062	po = pkt_sk(sk);
2063
2064	if (!net_eq(dev_net(dev), sock_net(sk)))
2065		goto drop;
2066
2067	skb->dev = dev;
2068
2069	if (dev->header_ops) {
2070		/* The device has an explicit notion of ll header,
2071		 * exported to higher levels.
2072		 *
2073		 * Otherwise, the device hides details of its frame
2074		 * structure, so that corresponding packet head is
2075		 * never delivered to user.
2076		 */
2077		if (sk->sk_type != SOCK_DGRAM)
2078			skb_push(skb, skb->data - skb_mac_header(skb));
2079		else if (skb->pkt_type == PACKET_OUTGOING) {
2080			/* Special case: outgoing packets have ll header at head */
2081			skb_pull(skb, skb_network_offset(skb));
2082		}
2083	}
2084
2085	snaplen = skb->len;
2086
2087	res = run_filter(skb, sk, snaplen);
2088	if (!res)
2089		goto drop_n_restore;
2090	if (snaplen > res)
2091		snaplen = res;
2092
2093	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2094		goto drop_n_acct;
2095
2096	if (skb_shared(skb)) {
2097		struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2098		if (nskb == NULL)
2099			goto drop_n_acct;
2100
2101		if (skb_head != skb->data) {
2102			skb->data = skb_head;
2103			skb->len = skb_len;
2104		}
2105		consume_skb(skb);
2106		skb = nskb;
2107	}
2108
2109	sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2110
2111	sll = &PACKET_SKB_CB(skb)->sa.ll;
2112	sll->sll_hatype = dev->type;
2113	sll->sll_pkttype = skb->pkt_type;
2114	if (unlikely(po->origdev))
2115		sll->sll_ifindex = orig_dev->ifindex;
2116	else
2117		sll->sll_ifindex = dev->ifindex;
2118
2119	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2120
2121	/* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2122	 * Use their space for storing the original skb length.
2123	 */
2124	PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2125
2126	if (pskb_trim(skb, snaplen))
2127		goto drop_n_acct;
2128
2129	skb_set_owner_r(skb, sk);
2130	skb->dev = NULL;
2131	skb_dst_drop(skb);
2132
2133	/* drop conntrack reference */
2134	nf_reset_ct(skb);
2135
2136	spin_lock(&sk->sk_receive_queue.lock);
2137	po->stats.stats1.tp_packets++;
2138	sock_skb_set_dropcount(sk, skb);
 
2139	__skb_queue_tail(&sk->sk_receive_queue, skb);
2140	spin_unlock(&sk->sk_receive_queue.lock);
2141	sk->sk_data_ready(sk);
2142	return 0;
2143
2144drop_n_acct:
2145	is_drop_n_account = true;
2146	atomic_inc(&po->tp_drops);
2147	atomic_inc(&sk->sk_drops);
2148
2149drop_n_restore:
2150	if (skb_head != skb->data && skb_shared(skb)) {
2151		skb->data = skb_head;
2152		skb->len = skb_len;
2153	}
2154drop:
2155	if (!is_drop_n_account)
2156		consume_skb(skb);
2157	else
2158		kfree_skb(skb);
2159	return 0;
2160}
2161
2162static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2163		       struct packet_type *pt, struct net_device *orig_dev)
2164{
2165	struct sock *sk;
2166	struct packet_sock *po;
2167	struct sockaddr_ll *sll;
2168	union tpacket_uhdr h;
2169	u8 *skb_head = skb->data;
2170	int skb_len = skb->len;
2171	unsigned int snaplen, res;
2172	unsigned long status = TP_STATUS_USER;
2173	unsigned short macoff, hdrlen;
2174	unsigned int netoff;
2175	struct sk_buff *copy_skb = NULL;
2176	struct timespec64 ts;
2177	__u32 ts_status;
2178	bool is_drop_n_account = false;
2179	unsigned int slot_id = 0;
2180	bool do_vnet = false;
2181
2182	/* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2183	 * We may add members to them until current aligned size without forcing
2184	 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2185	 */
2186	BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2187	BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2188
2189	if (skb->pkt_type == PACKET_LOOPBACK)
2190		goto drop;
2191
2192	sk = pt->af_packet_priv;
2193	po = pkt_sk(sk);
2194
2195	if (!net_eq(dev_net(dev), sock_net(sk)))
2196		goto drop;
2197
2198	if (dev->header_ops) {
2199		if (sk->sk_type != SOCK_DGRAM)
2200			skb_push(skb, skb->data - skb_mac_header(skb));
2201		else if (skb->pkt_type == PACKET_OUTGOING) {
2202			/* Special case: outgoing packets have ll header at head */
2203			skb_pull(skb, skb_network_offset(skb));
2204		}
2205	}
2206
2207	snaplen = skb->len;
2208
2209	res = run_filter(skb, sk, snaplen);
2210	if (!res)
2211		goto drop_n_restore;
2212
2213	/* If we are flooded, just give up */
2214	if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2215		atomic_inc(&po->tp_drops);
2216		goto drop_n_restore;
2217	}
2218
2219	if (skb->ip_summed == CHECKSUM_PARTIAL)
2220		status |= TP_STATUS_CSUMNOTREADY;
2221	else if (skb->pkt_type != PACKET_OUTGOING &&
2222		 (skb->ip_summed == CHECKSUM_COMPLETE ||
2223		  skb_csum_unnecessary(skb)))
2224		status |= TP_STATUS_CSUM_VALID;
2225
2226	if (snaplen > res)
2227		snaplen = res;
2228
2229	if (sk->sk_type == SOCK_DGRAM) {
2230		macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2231				  po->tp_reserve;
2232	} else {
2233		unsigned int maclen = skb_network_offset(skb);
2234		netoff = TPACKET_ALIGN(po->tp_hdrlen +
2235				       (maclen < 16 ? 16 : maclen)) +
2236				       po->tp_reserve;
2237		if (po->has_vnet_hdr) {
2238			netoff += sizeof(struct virtio_net_hdr);
2239			do_vnet = true;
2240		}
2241		macoff = netoff - maclen;
2242	}
2243	if (netoff > USHRT_MAX) {
2244		atomic_inc(&po->tp_drops);
2245		goto drop_n_restore;
2246	}
2247	if (po->tp_version <= TPACKET_V2) {
2248		if (macoff + snaplen > po->rx_ring.frame_size) {
2249			if (po->copy_thresh &&
2250			    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2251				if (skb_shared(skb)) {
2252					copy_skb = skb_clone(skb, GFP_ATOMIC);
2253				} else {
2254					copy_skb = skb_get(skb);
2255					skb_head = skb->data;
2256				}
2257				if (copy_skb)
 
 
2258					skb_set_owner_r(copy_skb, sk);
 
2259			}
2260			snaplen = po->rx_ring.frame_size - macoff;
2261			if ((int)snaplen < 0) {
2262				snaplen = 0;
2263				do_vnet = false;
2264			}
2265		}
2266	} else if (unlikely(macoff + snaplen >
2267			    GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2268		u32 nval;
2269
2270		nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2271		pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2272			    snaplen, nval, macoff);
2273		snaplen = nval;
2274		if (unlikely((int)snaplen < 0)) {
2275			snaplen = 0;
2276			macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2277			do_vnet = false;
2278		}
2279	}
2280	spin_lock(&sk->sk_receive_queue.lock);
2281	h.raw = packet_current_rx_frame(po, skb,
2282					TP_STATUS_KERNEL, (macoff+snaplen));
2283	if (!h.raw)
2284		goto drop_n_account;
2285
2286	if (po->tp_version <= TPACKET_V2) {
2287		slot_id = po->rx_ring.head;
2288		if (test_bit(slot_id, po->rx_ring.rx_owner_map))
2289			goto drop_n_account;
2290		__set_bit(slot_id, po->rx_ring.rx_owner_map);
2291	}
2292
2293	if (do_vnet &&
2294	    virtio_net_hdr_from_skb(skb, h.raw + macoff -
2295				    sizeof(struct virtio_net_hdr),
2296				    vio_le(), true, 0)) {
2297		if (po->tp_version == TPACKET_V3)
2298			prb_clear_blk_fill_status(&po->rx_ring);
2299		goto drop_n_account;
2300	}
2301
2302	if (po->tp_version <= TPACKET_V2) {
2303		packet_increment_rx_head(po, &po->rx_ring);
2304	/*
2305	 * LOSING will be reported till you read the stats,
2306	 * because it's COR - Clear On Read.
2307	 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2308	 * at packet level.
2309	 */
2310		if (atomic_read(&po->tp_drops))
2311			status |= TP_STATUS_LOSING;
2312	}
2313
2314	po->stats.stats1.tp_packets++;
2315	if (copy_skb) {
2316		status |= TP_STATUS_COPY;
 
2317		__skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2318	}
2319	spin_unlock(&sk->sk_receive_queue.lock);
2320
2321	skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2322
2323	if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
 
 
 
 
 
2324		ktime_get_real_ts64(&ts);
2325
2326	status |= ts_status;
2327
2328	switch (po->tp_version) {
2329	case TPACKET_V1:
2330		h.h1->tp_len = skb->len;
2331		h.h1->tp_snaplen = snaplen;
2332		h.h1->tp_mac = macoff;
2333		h.h1->tp_net = netoff;
2334		h.h1->tp_sec = ts.tv_sec;
2335		h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2336		hdrlen = sizeof(*h.h1);
2337		break;
2338	case TPACKET_V2:
2339		h.h2->tp_len = skb->len;
2340		h.h2->tp_snaplen = snaplen;
2341		h.h2->tp_mac = macoff;
2342		h.h2->tp_net = netoff;
2343		h.h2->tp_sec = ts.tv_sec;
2344		h.h2->tp_nsec = ts.tv_nsec;
2345		if (skb_vlan_tag_present(skb)) {
2346			h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2347			h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2348			status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2349		} else {
2350			h.h2->tp_vlan_tci = 0;
2351			h.h2->tp_vlan_tpid = 0;
2352		}
2353		memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2354		hdrlen = sizeof(*h.h2);
2355		break;
2356	case TPACKET_V3:
2357		/* tp_nxt_offset,vlan are already populated above.
2358		 * So DONT clear those fields here
2359		 */
2360		h.h3->tp_status |= status;
2361		h.h3->tp_len = skb->len;
2362		h.h3->tp_snaplen = snaplen;
2363		h.h3->tp_mac = macoff;
2364		h.h3->tp_net = netoff;
2365		h.h3->tp_sec  = ts.tv_sec;
2366		h.h3->tp_nsec = ts.tv_nsec;
2367		memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2368		hdrlen = sizeof(*h.h3);
2369		break;
2370	default:
2371		BUG();
2372	}
2373
2374	sll = h.raw + TPACKET_ALIGN(hdrlen);
2375	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2376	sll->sll_family = AF_PACKET;
2377	sll->sll_hatype = dev->type;
2378	sll->sll_protocol = skb->protocol;
2379	sll->sll_pkttype = skb->pkt_type;
2380	if (unlikely(po->origdev))
2381		sll->sll_ifindex = orig_dev->ifindex;
2382	else
2383		sll->sll_ifindex = dev->ifindex;
2384
2385	smp_mb();
2386
2387#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2388	if (po->tp_version <= TPACKET_V2) {
2389		u8 *start, *end;
2390
2391		end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2392					macoff + snaplen);
2393
2394		for (start = h.raw; start < end; start += PAGE_SIZE)
2395			flush_dcache_page(pgv_to_page(start));
2396	}
2397	smp_wmb();
2398#endif
2399
2400	if (po->tp_version <= TPACKET_V2) {
2401		spin_lock(&sk->sk_receive_queue.lock);
2402		__packet_set_status(po, h.raw, status);
2403		__clear_bit(slot_id, po->rx_ring.rx_owner_map);
2404		spin_unlock(&sk->sk_receive_queue.lock);
2405		sk->sk_data_ready(sk);
2406	} else if (po->tp_version == TPACKET_V3) {
2407		prb_clear_blk_fill_status(&po->rx_ring);
2408	}
2409
2410drop_n_restore:
2411	if (skb_head != skb->data && skb_shared(skb)) {
2412		skb->data = skb_head;
2413		skb->len = skb_len;
2414	}
2415drop:
2416	if (!is_drop_n_account)
2417		consume_skb(skb);
2418	else
2419		kfree_skb(skb);
2420	return 0;
2421
2422drop_n_account:
2423	spin_unlock(&sk->sk_receive_queue.lock);
2424	atomic_inc(&po->tp_drops);
2425	is_drop_n_account = true;
2426
2427	sk->sk_data_ready(sk);
2428	kfree_skb(copy_skb);
2429	goto drop_n_restore;
2430}
2431
2432static void tpacket_destruct_skb(struct sk_buff *skb)
2433{
2434	struct packet_sock *po = pkt_sk(skb->sk);
2435
2436	if (likely(po->tx_ring.pg_vec)) {
2437		void *ph;
2438		__u32 ts;
2439
2440		ph = skb_zcopy_get_nouarg(skb);
2441		packet_dec_pending(&po->tx_ring);
2442
2443		ts = __packet_set_timestamp(po, ph, skb);
2444		__packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2445
2446		if (!packet_read_pending(&po->tx_ring))
2447			complete(&po->skb_completion);
2448	}
2449
2450	sock_wfree(skb);
2451}
2452
2453static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2454{
2455	if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2456	    (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2457	     __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2458	      __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2459		vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2460			 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2461			__virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2462
2463	if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2464		return -EINVAL;
2465
2466	return 0;
2467}
2468
2469static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2470				 struct virtio_net_hdr *vnet_hdr)
2471{
2472	if (*len < sizeof(*vnet_hdr))
2473		return -EINVAL;
2474	*len -= sizeof(*vnet_hdr);
2475
2476	if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2477		return -EFAULT;
2478
2479	return __packet_snd_vnet_parse(vnet_hdr, *len);
2480}
2481
2482static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2483		void *frame, struct net_device *dev, void *data, int tp_len,
2484		__be16 proto, unsigned char *addr, int hlen, int copylen,
2485		const struct sockcm_cookie *sockc)
2486{
2487	union tpacket_uhdr ph;
2488	int to_write, offset, len, nr_frags, len_max;
2489	struct socket *sock = po->sk.sk_socket;
2490	struct page *page;
2491	int err;
2492
2493	ph.raw = frame;
2494
2495	skb->protocol = proto;
2496	skb->dev = dev;
2497	skb->priority = po->sk.sk_priority;
2498	skb->mark = po->sk.sk_mark;
2499	skb->tstamp = sockc->transmit_time;
2500	skb_setup_tx_timestamp(skb, sockc->tsflags);
2501	skb_zcopy_set_nouarg(skb, ph.raw);
2502
2503	skb_reserve(skb, hlen);
2504	skb_reset_network_header(skb);
2505
2506	to_write = tp_len;
2507
2508	if (sock->type == SOCK_DGRAM) {
2509		err = dev_hard_header(skb, dev, ntohs(proto), addr,
2510				NULL, tp_len);
2511		if (unlikely(err < 0))
2512			return -EINVAL;
2513	} else if (copylen) {
2514		int hdrlen = min_t(int, copylen, tp_len);
2515
2516		skb_push(skb, dev->hard_header_len);
2517		skb_put(skb, copylen - dev->hard_header_len);
2518		err = skb_store_bits(skb, 0, data, hdrlen);
2519		if (unlikely(err))
2520			return err;
2521		if (!dev_validate_header(dev, skb->data, hdrlen))
2522			return -EINVAL;
2523
2524		data += hdrlen;
2525		to_write -= hdrlen;
2526	}
2527
2528	offset = offset_in_page(data);
2529	len_max = PAGE_SIZE - offset;
2530	len = ((to_write > len_max) ? len_max : to_write);
2531
2532	skb->data_len = to_write;
2533	skb->len += to_write;
2534	skb->truesize += to_write;
2535	refcount_add(to_write, &po->sk.sk_wmem_alloc);
2536
2537	while (likely(to_write)) {
2538		nr_frags = skb_shinfo(skb)->nr_frags;
2539
2540		if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2541			pr_err("Packet exceed the number of skb frags(%lu)\n",
2542			       MAX_SKB_FRAGS);
2543			return -EFAULT;
2544		}
2545
2546		page = pgv_to_page(data);
2547		data += len;
2548		flush_dcache_page(page);
2549		get_page(page);
2550		skb_fill_page_desc(skb, nr_frags, page, offset, len);
2551		to_write -= len;
2552		offset = 0;
2553		len_max = PAGE_SIZE;
2554		len = ((to_write > len_max) ? len_max : to_write);
2555	}
2556
2557	packet_parse_headers(skb, sock);
2558
2559	return tp_len;
2560}
2561
2562static int tpacket_parse_header(struct packet_sock *po, void *frame,
2563				int size_max, void **data)
2564{
2565	union tpacket_uhdr ph;
2566	int tp_len, off;
2567
2568	ph.raw = frame;
2569
2570	switch (po->tp_version) {
2571	case TPACKET_V3:
2572		if (ph.h3->tp_next_offset != 0) {
2573			pr_warn_once("variable sized slot not supported");
2574			return -EINVAL;
2575		}
2576		tp_len = ph.h3->tp_len;
2577		break;
2578	case TPACKET_V2:
2579		tp_len = ph.h2->tp_len;
2580		break;
2581	default:
2582		tp_len = ph.h1->tp_len;
2583		break;
2584	}
2585	if (unlikely(tp_len > size_max)) {
2586		pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2587		return -EMSGSIZE;
2588	}
2589
2590	if (unlikely(po->tp_tx_has_off)) {
2591		int off_min, off_max;
2592
2593		off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2594		off_max = po->tx_ring.frame_size - tp_len;
2595		if (po->sk.sk_type == SOCK_DGRAM) {
2596			switch (po->tp_version) {
2597			case TPACKET_V3:
2598				off = ph.h3->tp_net;
2599				break;
2600			case TPACKET_V2:
2601				off = ph.h2->tp_net;
2602				break;
2603			default:
2604				off = ph.h1->tp_net;
2605				break;
2606			}
2607		} else {
2608			switch (po->tp_version) {
2609			case TPACKET_V3:
2610				off = ph.h3->tp_mac;
2611				break;
2612			case TPACKET_V2:
2613				off = ph.h2->tp_mac;
2614				break;
2615			default:
2616				off = ph.h1->tp_mac;
2617				break;
2618			}
2619		}
2620		if (unlikely((off < off_min) || (off_max < off)))
2621			return -EINVAL;
2622	} else {
2623		off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2624	}
2625
2626	*data = frame + off;
2627	return tp_len;
2628}
2629
2630static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2631{
2632	struct sk_buff *skb = NULL;
2633	struct net_device *dev;
2634	struct virtio_net_hdr *vnet_hdr = NULL;
2635	struct sockcm_cookie sockc;
2636	__be16 proto;
2637	int err, reserve = 0;
2638	void *ph;
2639	DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2640	bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2641	unsigned char *addr = NULL;
2642	int tp_len, size_max;
2643	void *data;
2644	int len_sum = 0;
2645	int status = TP_STATUS_AVAILABLE;
2646	int hlen, tlen, copylen = 0;
2647	long timeo = 0;
2648
2649	mutex_lock(&po->pg_vec_lock);
2650
2651	/* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2652	 * we need to confirm it under protection of pg_vec_lock.
2653	 */
2654	if (unlikely(!po->tx_ring.pg_vec)) {
2655		err = -EBUSY;
2656		goto out;
2657	}
2658	if (likely(saddr == NULL)) {
2659		dev	= packet_cached_dev_get(po);
2660		proto	= po->num;
2661	} else {
2662		err = -EINVAL;
2663		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2664			goto out;
2665		if (msg->msg_namelen < (saddr->sll_halen
2666					+ offsetof(struct sockaddr_ll,
2667						sll_addr)))
2668			goto out;
2669		proto	= saddr->sll_protocol;
2670		dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2671		if (po->sk.sk_socket->type == SOCK_DGRAM) {
2672			if (dev && msg->msg_namelen < dev->addr_len +
2673				   offsetof(struct sockaddr_ll, sll_addr))
2674				goto out_put;
2675			addr = saddr->sll_addr;
2676		}
2677	}
2678
2679	err = -ENXIO;
2680	if (unlikely(dev == NULL))
2681		goto out;
2682	err = -ENETDOWN;
2683	if (unlikely(!(dev->flags & IFF_UP)))
2684		goto out_put;
2685
2686	sockcm_init(&sockc, &po->sk);
2687	if (msg->msg_controllen) {
2688		err = sock_cmsg_send(&po->sk, msg, &sockc);
2689		if (unlikely(err))
2690			goto out_put;
2691	}
2692
2693	if (po->sk.sk_socket->type == SOCK_RAW)
2694		reserve = dev->hard_header_len;
2695	size_max = po->tx_ring.frame_size
2696		- (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2697
2698	if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2699		size_max = dev->mtu + reserve + VLAN_HLEN;
2700
2701	reinit_completion(&po->skb_completion);
2702
2703	do {
2704		ph = packet_current_frame(po, &po->tx_ring,
2705					  TP_STATUS_SEND_REQUEST);
2706		if (unlikely(ph == NULL)) {
2707			if (need_wait && skb) {
2708				timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2709				timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2710				if (timeo <= 0) {
2711					err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2712					goto out_put;
2713				}
2714			}
2715			/* check for additional frames */
2716			continue;
2717		}
2718
2719		skb = NULL;
2720		tp_len = tpacket_parse_header(po, ph, size_max, &data);
2721		if (tp_len < 0)
2722			goto tpacket_error;
2723
2724		status = TP_STATUS_SEND_REQUEST;
2725		hlen = LL_RESERVED_SPACE(dev);
2726		tlen = dev->needed_tailroom;
2727		if (po->has_vnet_hdr) {
2728			vnet_hdr = data;
2729			data += sizeof(*vnet_hdr);
2730			tp_len -= sizeof(*vnet_hdr);
2731			if (tp_len < 0 ||
2732			    __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2733				tp_len = -EINVAL;
2734				goto tpacket_error;
2735			}
2736			copylen = __virtio16_to_cpu(vio_le(),
2737						    vnet_hdr->hdr_len);
2738		}
2739		copylen = max_t(int, copylen, dev->hard_header_len);
2740		skb = sock_alloc_send_skb(&po->sk,
2741				hlen + tlen + sizeof(struct sockaddr_ll) +
2742				(copylen - dev->hard_header_len),
2743				!need_wait, &err);
2744
2745		if (unlikely(skb == NULL)) {
2746			/* we assume the socket was initially writeable ... */
2747			if (likely(len_sum > 0))
2748				err = len_sum;
2749			goto out_status;
2750		}
2751		tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2752					  addr, hlen, copylen, &sockc);
2753		if (likely(tp_len >= 0) &&
2754		    tp_len > dev->mtu + reserve &&
2755		    !po->has_vnet_hdr &&
2756		    !packet_extra_vlan_len_allowed(dev, skb))
2757			tp_len = -EMSGSIZE;
2758
2759		if (unlikely(tp_len < 0)) {
2760tpacket_error:
2761			if (po->tp_loss) {
2762				__packet_set_status(po, ph,
2763						TP_STATUS_AVAILABLE);
2764				packet_increment_head(&po->tx_ring);
2765				kfree_skb(skb);
2766				continue;
2767			} else {
2768				status = TP_STATUS_WRONG_FORMAT;
2769				err = tp_len;
2770				goto out_status;
2771			}
2772		}
2773
2774		if (po->has_vnet_hdr) {
2775			if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2776				tp_len = -EINVAL;
2777				goto tpacket_error;
2778			}
2779			virtio_net_hdr_set_proto(skb, vnet_hdr);
2780		}
2781
2782		skb->destructor = tpacket_destruct_skb;
2783		__packet_set_status(po, ph, TP_STATUS_SENDING);
2784		packet_inc_pending(&po->tx_ring);
2785
2786		status = TP_STATUS_SEND_REQUEST;
2787		err = po->xmit(skb);
2788		if (unlikely(err > 0)) {
2789			err = net_xmit_errno(err);
 
2790			if (err && __packet_get_status(po, ph) ==
2791				   TP_STATUS_AVAILABLE) {
2792				/* skb was destructed already */
2793				skb = NULL;
2794				goto out_status;
2795			}
2796			/*
2797			 * skb was dropped but not destructed yet;
2798			 * let's treat it like congestion or err < 0
2799			 */
2800			err = 0;
2801		}
2802		packet_increment_head(&po->tx_ring);
2803		len_sum += tp_len;
2804	} while (likely((ph != NULL) ||
2805		/* Note: packet_read_pending() might be slow if we have
2806		 * to call it as it's per_cpu variable, but in fast-path
2807		 * we already short-circuit the loop with the first
2808		 * condition, and luckily don't have to go that path
2809		 * anyway.
2810		 */
2811		 (need_wait && packet_read_pending(&po->tx_ring))));
2812
2813	err = len_sum;
2814	goto out_put;
2815
2816out_status:
2817	__packet_set_status(po, ph, status);
2818	kfree_skb(skb);
2819out_put:
2820	dev_put(dev);
2821out:
2822	mutex_unlock(&po->pg_vec_lock);
2823	return err;
2824}
2825
2826static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2827				        size_t reserve, size_t len,
2828				        size_t linear, int noblock,
2829				        int *err)
2830{
2831	struct sk_buff *skb;
2832
2833	/* Under a page?  Don't bother with paged skb. */
2834	if (prepad + len < PAGE_SIZE || !linear)
2835		linear = len;
2836
2837	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2838				   err, 0);
2839	if (!skb)
2840		return NULL;
2841
2842	skb_reserve(skb, reserve);
2843	skb_put(skb, linear);
2844	skb->data_len = len - linear;
2845	skb->len += len - linear;
2846
2847	return skb;
2848}
2849
2850static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2851{
2852	struct sock *sk = sock->sk;
2853	DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2854	struct sk_buff *skb;
2855	struct net_device *dev;
2856	__be16 proto;
2857	unsigned char *addr = NULL;
2858	int err, reserve = 0;
2859	struct sockcm_cookie sockc;
2860	struct virtio_net_hdr vnet_hdr = { 0 };
2861	int offset = 0;
2862	struct packet_sock *po = pkt_sk(sk);
2863	bool has_vnet_hdr = false;
2864	int hlen, tlen, linear;
2865	int extra_len = 0;
2866
2867	/*
2868	 *	Get and verify the address.
2869	 */
2870
2871	if (likely(saddr == NULL)) {
2872		dev	= packet_cached_dev_get(po);
2873		proto	= po->num;
2874	} else {
2875		err = -EINVAL;
2876		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2877			goto out;
2878		if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2879			goto out;
2880		proto	= saddr->sll_protocol;
2881		dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2882		if (sock->type == SOCK_DGRAM) {
2883			if (dev && msg->msg_namelen < dev->addr_len +
2884				   offsetof(struct sockaddr_ll, sll_addr))
2885				goto out_unlock;
2886			addr = saddr->sll_addr;
2887		}
2888	}
2889
2890	err = -ENXIO;
2891	if (unlikely(dev == NULL))
2892		goto out_unlock;
2893	err = -ENETDOWN;
2894	if (unlikely(!(dev->flags & IFF_UP)))
2895		goto out_unlock;
2896
2897	sockcm_init(&sockc, sk);
2898	sockc.mark = sk->sk_mark;
2899	if (msg->msg_controllen) {
2900		err = sock_cmsg_send(sk, msg, &sockc);
2901		if (unlikely(err))
2902			goto out_unlock;
2903	}
2904
2905	if (sock->type == SOCK_RAW)
2906		reserve = dev->hard_header_len;
2907	if (po->has_vnet_hdr) {
2908		err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2909		if (err)
2910			goto out_unlock;
2911		has_vnet_hdr = true;
2912	}
2913
2914	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2915		if (!netif_supports_nofcs(dev)) {
2916			err = -EPROTONOSUPPORT;
2917			goto out_unlock;
2918		}
2919		extra_len = 4; /* We're doing our own CRC */
2920	}
2921
2922	err = -EMSGSIZE;
2923	if (!vnet_hdr.gso_type &&
2924	    (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2925		goto out_unlock;
2926
2927	err = -ENOBUFS;
2928	hlen = LL_RESERVED_SPACE(dev);
2929	tlen = dev->needed_tailroom;
2930	linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2931	linear = max(linear, min_t(int, len, dev->hard_header_len));
2932	skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2933			       msg->msg_flags & MSG_DONTWAIT, &err);
2934	if (skb == NULL)
2935		goto out_unlock;
2936
2937	skb_reset_network_header(skb);
2938
2939	err = -EINVAL;
2940	if (sock->type == SOCK_DGRAM) {
2941		offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2942		if (unlikely(offset < 0))
2943			goto out_free;
2944	} else if (reserve) {
2945		skb_reserve(skb, -reserve);
2946		if (len < reserve + sizeof(struct ipv6hdr) &&
2947		    dev->min_header_len != dev->hard_header_len)
2948			skb_reset_network_header(skb);
2949	}
2950
2951	/* Returns -EFAULT on error */
2952	err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2953	if (err)
2954		goto out_free;
2955
2956	if (sock->type == SOCK_RAW &&
2957	    !dev_validate_header(dev, skb->data, len)) {
2958		err = -EINVAL;
2959		goto out_free;
2960	}
2961
2962	skb_setup_tx_timestamp(skb, sockc.tsflags);
2963
2964	if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2965	    !packet_extra_vlan_len_allowed(dev, skb)) {
2966		err = -EMSGSIZE;
2967		goto out_free;
2968	}
2969
2970	skb->protocol = proto;
2971	skb->dev = dev;
2972	skb->priority = sk->sk_priority;
2973	skb->mark = sockc.mark;
2974	skb->tstamp = sockc.transmit_time;
2975
 
 
 
 
 
2976	if (has_vnet_hdr) {
2977		err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2978		if (err)
2979			goto out_free;
2980		len += sizeof(vnet_hdr);
2981		virtio_net_hdr_set_proto(skb, &vnet_hdr);
2982	}
2983
2984	packet_parse_headers(skb, sock);
2985
2986	if (unlikely(extra_len == 4))
2987		skb->no_fcs = 1;
2988
2989	err = po->xmit(skb);
2990	if (err > 0 && (err = net_xmit_errno(err)) != 0)
2991		goto out_unlock;
 
 
 
 
2992
2993	dev_put(dev);
2994
2995	return len;
2996
2997out_free:
2998	kfree_skb(skb);
2999out_unlock:
3000	if (dev)
3001		dev_put(dev);
3002out:
3003	return err;
3004}
3005
3006static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
3007{
3008	struct sock *sk = sock->sk;
3009	struct packet_sock *po = pkt_sk(sk);
3010
3011	if (po->tx_ring.pg_vec)
 
 
 
3012		return tpacket_snd(po, msg);
3013	else
3014		return packet_snd(sock, msg, len);
3015}
3016
3017/*
3018 *	Close a PACKET socket. This is fairly simple. We immediately go
3019 *	to 'closed' state and remove our protocol entry in the device list.
3020 */
3021
3022static int packet_release(struct socket *sock)
3023{
3024	struct sock *sk = sock->sk;
3025	struct packet_sock *po;
3026	struct packet_fanout *f;
3027	struct net *net;
3028	union tpacket_req_u req_u;
3029
3030	if (!sk)
3031		return 0;
3032
3033	net = sock_net(sk);
3034	po = pkt_sk(sk);
3035
3036	mutex_lock(&net->packet.sklist_lock);
3037	sk_del_node_init_rcu(sk);
3038	mutex_unlock(&net->packet.sklist_lock);
3039
3040	preempt_disable();
3041	sock_prot_inuse_add(net, sk->sk_prot, -1);
3042	preempt_enable();
3043
3044	spin_lock(&po->bind_lock);
3045	unregister_prot_hook(sk, false);
3046	packet_cached_dev_reset(po);
3047
3048	if (po->prot_hook.dev) {
3049		dev_put(po->prot_hook.dev);
3050		po->prot_hook.dev = NULL;
3051	}
3052	spin_unlock(&po->bind_lock);
3053
3054	packet_flush_mclist(sk);
3055
3056	lock_sock(sk);
3057	if (po->rx_ring.pg_vec) {
3058		memset(&req_u, 0, sizeof(req_u));
3059		packet_set_ring(sk, &req_u, 1, 0);
3060	}
3061
3062	if (po->tx_ring.pg_vec) {
3063		memset(&req_u, 0, sizeof(req_u));
3064		packet_set_ring(sk, &req_u, 1, 1);
3065	}
3066	release_sock(sk);
3067
3068	f = fanout_release(sk);
3069
3070	synchronize_net();
3071
3072	kfree(po->rollover);
3073	if (f) {
3074		fanout_release_data(f);
3075		kfree(f);
3076	}
3077	/*
3078	 *	Now the socket is dead. No more input will appear.
3079	 */
3080	sock_orphan(sk);
3081	sock->sk = NULL;
3082
3083	/* Purge queues */
3084
3085	skb_queue_purge(&sk->sk_receive_queue);
3086	packet_free_pending(po);
3087	sk_refcnt_debug_release(sk);
3088
3089	sock_put(sk);
3090	return 0;
3091}
3092
3093/*
3094 *	Attach a packet hook.
3095 */
3096
3097static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3098			  __be16 proto)
3099{
3100	struct packet_sock *po = pkt_sk(sk);
3101	struct net_device *dev_curr;
3102	__be16 proto_curr;
3103	bool need_rehook;
3104	struct net_device *dev = NULL;
3105	int ret = 0;
3106	bool unlisted = false;
 
 
3107
3108	lock_sock(sk);
3109	spin_lock(&po->bind_lock);
3110	rcu_read_lock();
3111
3112	if (po->fanout) {
3113		ret = -EINVAL;
3114		goto out_unlock;
3115	}
3116
3117	if (name) {
3118		dev = dev_get_by_name_rcu(sock_net(sk), name);
3119		if (!dev) {
3120			ret = -ENODEV;
3121			goto out_unlock;
3122		}
3123	} else if (ifindex) {
3124		dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3125		if (!dev) {
3126			ret = -ENODEV;
3127			goto out_unlock;
3128		}
3129	}
3130
3131	if (dev)
3132		dev_hold(dev);
3133
3134	proto_curr = po->prot_hook.type;
3135	dev_curr = po->prot_hook.dev;
3136
3137	need_rehook = proto_curr != proto || dev_curr != dev;
3138
3139	if (need_rehook) {
 
3140		if (po->running) {
3141			rcu_read_unlock();
3142			/* prevents packet_notifier() from calling
3143			 * register_prot_hook()
3144			 */
3145			po->num = 0;
3146			__unregister_prot_hook(sk, true);
3147			rcu_read_lock();
3148			dev_curr = po->prot_hook.dev;
3149			if (dev)
3150				unlisted = !dev_get_by_index_rcu(sock_net(sk),
3151								 dev->ifindex);
3152		}
3153
3154		BUG_ON(po->running);
3155		po->num = proto;
3156		po->prot_hook.type = proto;
3157
 
 
3158		if (unlikely(unlisted)) {
3159			dev_put(dev);
3160			po->prot_hook.dev = NULL;
3161			po->ifindex = -1;
3162			packet_cached_dev_reset(po);
3163		} else {
 
 
3164			po->prot_hook.dev = dev;
3165			po->ifindex = dev ? dev->ifindex : 0;
3166			packet_cached_dev_assign(po, dev);
3167		}
 
3168	}
3169	if (dev_curr)
3170		dev_put(dev_curr);
3171
3172	if (proto == 0 || !need_rehook)
3173		goto out_unlock;
3174
3175	if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3176		register_prot_hook(sk);
3177	} else {
3178		sk->sk_err = ENETDOWN;
3179		if (!sock_flag(sk, SOCK_DEAD))
3180			sk->sk_error_report(sk);
3181	}
3182
3183out_unlock:
3184	rcu_read_unlock();
3185	spin_unlock(&po->bind_lock);
3186	release_sock(sk);
3187	return ret;
3188}
3189
3190/*
3191 *	Bind a packet socket to a device
3192 */
3193
3194static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3195			    int addr_len)
3196{
3197	struct sock *sk = sock->sk;
3198	char name[sizeof(uaddr->sa_data) + 1];
3199
3200	/*
3201	 *	Check legality
3202	 */
3203
3204	if (addr_len != sizeof(struct sockaddr))
3205		return -EINVAL;
3206	/* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3207	 * zero-terminated.
3208	 */
3209	memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3210	name[sizeof(uaddr->sa_data)] = 0;
3211
3212	return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3213}
3214
3215static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3216{
3217	struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3218	struct sock *sk = sock->sk;
3219
3220	/*
3221	 *	Check legality
3222	 */
3223
3224	if (addr_len < sizeof(struct sockaddr_ll))
3225		return -EINVAL;
3226	if (sll->sll_family != AF_PACKET)
3227		return -EINVAL;
3228
3229	return packet_do_bind(sk, NULL, sll->sll_ifindex,
3230			      sll->sll_protocol ? : pkt_sk(sk)->num);
3231}
3232
3233static struct proto packet_proto = {
3234	.name	  = "PACKET",
3235	.owner	  = THIS_MODULE,
3236	.obj_size = sizeof(struct packet_sock),
3237};
3238
3239/*
3240 *	Create a packet of type SOCK_PACKET.
3241 */
3242
3243static int packet_create(struct net *net, struct socket *sock, int protocol,
3244			 int kern)
3245{
3246	struct sock *sk;
3247	struct packet_sock *po;
3248	__be16 proto = (__force __be16)protocol; /* weird, but documented */
3249	int err;
3250
3251	if (!ns_capable(net->user_ns, CAP_NET_RAW))
3252		return -EPERM;
3253	if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3254	    sock->type != SOCK_PACKET)
3255		return -ESOCKTNOSUPPORT;
3256
3257	sock->state = SS_UNCONNECTED;
3258
3259	err = -ENOBUFS;
3260	sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3261	if (sk == NULL)
3262		goto out;
3263
3264	sock->ops = &packet_ops;
3265	if (sock->type == SOCK_PACKET)
3266		sock->ops = &packet_ops_spkt;
3267
3268	sock_init_data(sock, sk);
3269
3270	po = pkt_sk(sk);
3271	init_completion(&po->skb_completion);
3272	sk->sk_family = PF_PACKET;
3273	po->num = proto;
3274	po->xmit = dev_queue_xmit;
3275
3276	err = packet_alloc_pending(po);
3277	if (err)
3278		goto out2;
3279
3280	packet_cached_dev_reset(po);
3281
3282	sk->sk_destruct = packet_sock_destruct;
3283	sk_refcnt_debug_inc(sk);
3284
3285	/*
3286	 *	Attach a protocol block
3287	 */
3288
3289	spin_lock_init(&po->bind_lock);
3290	mutex_init(&po->pg_vec_lock);
3291	po->rollover = NULL;
3292	po->prot_hook.func = packet_rcv;
3293
3294	if (sock->type == SOCK_PACKET)
3295		po->prot_hook.func = packet_rcv_spkt;
3296
3297	po->prot_hook.af_packet_priv = sk;
 
3298
3299	if (proto) {
3300		po->prot_hook.type = proto;
3301		__register_prot_hook(sk);
3302	}
3303
3304	mutex_lock(&net->packet.sklist_lock);
3305	sk_add_node_tail_rcu(sk, &net->packet.sklist);
3306	mutex_unlock(&net->packet.sklist_lock);
3307
3308	preempt_disable();
3309	sock_prot_inuse_add(net, &packet_proto, 1);
3310	preempt_enable();
3311
3312	return 0;
3313out2:
3314	sk_free(sk);
3315out:
3316	return err;
3317}
3318
3319/*
3320 *	Pull a packet from our receive queue and hand it to the user.
3321 *	If necessary we block.
3322 */
3323
3324static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3325			  int flags)
3326{
3327	struct sock *sk = sock->sk;
3328	struct sk_buff *skb;
3329	int copied, err;
3330	int vnet_hdr_len = 0;
3331	unsigned int origlen = 0;
3332
3333	err = -EINVAL;
3334	if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3335		goto out;
3336
3337#if 0
3338	/* What error should we return now? EUNATTACH? */
3339	if (pkt_sk(sk)->ifindex < 0)
3340		return -ENODEV;
3341#endif
3342
3343	if (flags & MSG_ERRQUEUE) {
3344		err = sock_recv_errqueue(sk, msg, len,
3345					 SOL_PACKET, PACKET_TX_TIMESTAMP);
3346		goto out;
3347	}
3348
3349	/*
3350	 *	Call the generic datagram receiver. This handles all sorts
3351	 *	of horrible races and re-entrancy so we can forget about it
3352	 *	in the protocol layers.
3353	 *
3354	 *	Now it will return ENETDOWN, if device have just gone down,
3355	 *	but then it will block.
3356	 */
3357
3358	skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3359
3360	/*
3361	 *	An error occurred so return it. Because skb_recv_datagram()
3362	 *	handles the blocking we don't see and worry about blocking
3363	 *	retries.
3364	 */
3365
3366	if (skb == NULL)
3367		goto out;
3368
3369	packet_rcv_try_clear_pressure(pkt_sk(sk));
3370
3371	if (pkt_sk(sk)->has_vnet_hdr) {
3372		err = packet_rcv_vnet(msg, skb, &len);
3373		if (err)
3374			goto out_free;
3375		vnet_hdr_len = sizeof(struct virtio_net_hdr);
3376	}
3377
3378	/* You lose any data beyond the buffer you gave. If it worries
3379	 * a user program they can ask the device for its MTU
3380	 * anyway.
3381	 */
3382	copied = skb->len;
3383	if (copied > len) {
3384		copied = len;
3385		msg->msg_flags |= MSG_TRUNC;
3386	}
3387
3388	err = skb_copy_datagram_msg(skb, 0, msg, copied);
3389	if (err)
3390		goto out_free;
3391
3392	if (sock->type != SOCK_PACKET) {
3393		struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3394
3395		/* Original length was stored in sockaddr_ll fields */
3396		origlen = PACKET_SKB_CB(skb)->sa.origlen;
3397		sll->sll_family = AF_PACKET;
3398		sll->sll_protocol = skb->protocol;
3399	}
3400
3401	sock_recv_ts_and_drops(msg, sk, skb);
3402
3403	if (msg->msg_name) {
 
 
3404		int copy_len;
3405
3406		/* If the address length field is there to be filled
3407		 * in, we fill it in now.
3408		 */
3409		if (sock->type == SOCK_PACKET) {
3410			__sockaddr_check_size(sizeof(struct sockaddr_pkt));
3411			msg->msg_namelen = sizeof(struct sockaddr_pkt);
3412			copy_len = msg->msg_namelen;
3413		} else {
3414			struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3415
3416			msg->msg_namelen = sll->sll_halen +
3417				offsetof(struct sockaddr_ll, sll_addr);
3418			copy_len = msg->msg_namelen;
3419			if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3420				memset(msg->msg_name +
3421				       offsetof(struct sockaddr_ll, sll_addr),
3422				       0, sizeof(sll->sll_addr));
3423				msg->msg_namelen = sizeof(struct sockaddr_ll);
3424			}
3425		}
 
 
 
 
3426		memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3427	}
3428
3429	if (pkt_sk(sk)->auxdata) {
3430		struct tpacket_auxdata aux;
3431
3432		aux.tp_status = TP_STATUS_USER;
3433		if (skb->ip_summed == CHECKSUM_PARTIAL)
3434			aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3435		else if (skb->pkt_type != PACKET_OUTGOING &&
3436			 (skb->ip_summed == CHECKSUM_COMPLETE ||
3437			  skb_csum_unnecessary(skb)))
3438			aux.tp_status |= TP_STATUS_CSUM_VALID;
3439
3440		aux.tp_len = origlen;
3441		aux.tp_snaplen = skb->len;
3442		aux.tp_mac = 0;
3443		aux.tp_net = skb_network_offset(skb);
3444		if (skb_vlan_tag_present(skb)) {
3445			aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3446			aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3447			aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3448		} else {
3449			aux.tp_vlan_tci = 0;
3450			aux.tp_vlan_tpid = 0;
3451		}
3452		put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3453	}
3454
3455	/*
3456	 *	Free or return the buffer as appropriate. Again this
3457	 *	hides all the races and re-entrancy issues from us.
3458	 */
3459	err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3460
3461out_free:
3462	skb_free_datagram(sk, skb);
3463out:
3464	return err;
3465}
3466
3467static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3468			       int peer)
3469{
3470	struct net_device *dev;
3471	struct sock *sk	= sock->sk;
3472
3473	if (peer)
3474		return -EOPNOTSUPP;
3475
3476	uaddr->sa_family = AF_PACKET;
3477	memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3478	rcu_read_lock();
3479	dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3480	if (dev)
3481		strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3482	rcu_read_unlock();
3483
3484	return sizeof(*uaddr);
3485}
3486
3487static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3488			  int peer)
3489{
3490	struct net_device *dev;
3491	struct sock *sk = sock->sk;
3492	struct packet_sock *po = pkt_sk(sk);
3493	DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
 
3494
3495	if (peer)
3496		return -EOPNOTSUPP;
3497
 
3498	sll->sll_family = AF_PACKET;
3499	sll->sll_ifindex = po->ifindex;
3500	sll->sll_protocol = po->num;
3501	sll->sll_pkttype = 0;
3502	rcu_read_lock();
3503	dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3504	if (dev) {
3505		sll->sll_hatype = dev->type;
3506		sll->sll_halen = dev->addr_len;
3507		memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3508	} else {
3509		sll->sll_hatype = 0;	/* Bad: we have no ARPHRD_UNSPEC */
3510		sll->sll_halen = 0;
3511	}
3512	rcu_read_unlock();
3513
3514	return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3515}
3516
3517static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3518			 int what)
3519{
3520	switch (i->type) {
3521	case PACKET_MR_MULTICAST:
3522		if (i->alen != dev->addr_len)
3523			return -EINVAL;
3524		if (what > 0)
3525			return dev_mc_add(dev, i->addr);
3526		else
3527			return dev_mc_del(dev, i->addr);
3528		break;
3529	case PACKET_MR_PROMISC:
3530		return dev_set_promiscuity(dev, what);
3531	case PACKET_MR_ALLMULTI:
3532		return dev_set_allmulti(dev, what);
3533	case PACKET_MR_UNICAST:
3534		if (i->alen != dev->addr_len)
3535			return -EINVAL;
3536		if (what > 0)
3537			return dev_uc_add(dev, i->addr);
3538		else
3539			return dev_uc_del(dev, i->addr);
3540		break;
3541	default:
3542		break;
3543	}
3544	return 0;
3545}
3546
3547static void packet_dev_mclist_delete(struct net_device *dev,
3548				     struct packet_mclist **mlp)
3549{
3550	struct packet_mclist *ml;
3551
3552	while ((ml = *mlp) != NULL) {
3553		if (ml->ifindex == dev->ifindex) {
3554			packet_dev_mc(dev, ml, -1);
3555			*mlp = ml->next;
3556			kfree(ml);
3557		} else
3558			mlp = &ml->next;
3559	}
3560}
3561
3562static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3563{
3564	struct packet_sock *po = pkt_sk(sk);
3565	struct packet_mclist *ml, *i;
3566	struct net_device *dev;
3567	int err;
3568
3569	rtnl_lock();
3570
3571	err = -ENODEV;
3572	dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3573	if (!dev)
3574		goto done;
3575
3576	err = -EINVAL;
3577	if (mreq->mr_alen > dev->addr_len)
3578		goto done;
3579
3580	err = -ENOBUFS;
3581	i = kmalloc(sizeof(*i), GFP_KERNEL);
3582	if (i == NULL)
3583		goto done;
3584
3585	err = 0;
3586	for (ml = po->mclist; ml; ml = ml->next) {
3587		if (ml->ifindex == mreq->mr_ifindex &&
3588		    ml->type == mreq->mr_type &&
3589		    ml->alen == mreq->mr_alen &&
3590		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3591			ml->count++;
3592			/* Free the new element ... */
3593			kfree(i);
3594			goto done;
3595		}
3596	}
3597
3598	i->type = mreq->mr_type;
3599	i->ifindex = mreq->mr_ifindex;
3600	i->alen = mreq->mr_alen;
3601	memcpy(i->addr, mreq->mr_address, i->alen);
3602	memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3603	i->count = 1;
3604	i->next = po->mclist;
3605	po->mclist = i;
3606	err = packet_dev_mc(dev, i, 1);
3607	if (err) {
3608		po->mclist = i->next;
3609		kfree(i);
3610	}
3611
3612done:
3613	rtnl_unlock();
3614	return err;
3615}
3616
3617static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3618{
3619	struct packet_mclist *ml, **mlp;
3620
3621	rtnl_lock();
3622
3623	for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3624		if (ml->ifindex == mreq->mr_ifindex &&
3625		    ml->type == mreq->mr_type &&
3626		    ml->alen == mreq->mr_alen &&
3627		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3628			if (--ml->count == 0) {
3629				struct net_device *dev;
3630				*mlp = ml->next;
3631				dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3632				if (dev)
3633					packet_dev_mc(dev, ml, -1);
3634				kfree(ml);
3635			}
3636			break;
3637		}
3638	}
3639	rtnl_unlock();
3640	return 0;
3641}
3642
3643static void packet_flush_mclist(struct sock *sk)
3644{
3645	struct packet_sock *po = pkt_sk(sk);
3646	struct packet_mclist *ml;
3647
3648	if (!po->mclist)
3649		return;
3650
3651	rtnl_lock();
3652	while ((ml = po->mclist) != NULL) {
3653		struct net_device *dev;
3654
3655		po->mclist = ml->next;
3656		dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3657		if (dev != NULL)
3658			packet_dev_mc(dev, ml, -1);
3659		kfree(ml);
3660	}
3661	rtnl_unlock();
3662}
3663
3664static int
3665packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval,
3666		  unsigned int optlen)
3667{
3668	struct sock *sk = sock->sk;
3669	struct packet_sock *po = pkt_sk(sk);
3670	int ret;
3671
3672	if (level != SOL_PACKET)
3673		return -ENOPROTOOPT;
3674
3675	switch (optname) {
3676	case PACKET_ADD_MEMBERSHIP:
3677	case PACKET_DROP_MEMBERSHIP:
3678	{
3679		struct packet_mreq_max mreq;
3680		int len = optlen;
3681		memset(&mreq, 0, sizeof(mreq));
3682		if (len < sizeof(struct packet_mreq))
3683			return -EINVAL;
3684		if (len > sizeof(mreq))
3685			len = sizeof(mreq);
3686		if (copy_from_sockptr(&mreq, optval, len))
3687			return -EFAULT;
3688		if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3689			return -EINVAL;
3690		if (optname == PACKET_ADD_MEMBERSHIP)
3691			ret = packet_mc_add(sk, &mreq);
3692		else
3693			ret = packet_mc_drop(sk, &mreq);
3694		return ret;
3695	}
3696
3697	case PACKET_RX_RING:
3698	case PACKET_TX_RING:
3699	{
3700		union tpacket_req_u req_u;
3701		int len;
3702
3703		lock_sock(sk);
3704		switch (po->tp_version) {
3705		case TPACKET_V1:
3706		case TPACKET_V2:
3707			len = sizeof(req_u.req);
3708			break;
3709		case TPACKET_V3:
3710		default:
3711			len = sizeof(req_u.req3);
3712			break;
3713		}
3714		if (optlen < len) {
3715			ret = -EINVAL;
3716		} else {
3717			if (copy_from_sockptr(&req_u.req, optval, len))
3718				ret = -EFAULT;
3719			else
3720				ret = packet_set_ring(sk, &req_u, 0,
3721						    optname == PACKET_TX_RING);
3722		}
3723		release_sock(sk);
3724		return ret;
3725	}
3726	case PACKET_COPY_THRESH:
3727	{
3728		int val;
3729
3730		if (optlen != sizeof(val))
3731			return -EINVAL;
3732		if (copy_from_sockptr(&val, optval, sizeof(val)))
3733			return -EFAULT;
3734
3735		pkt_sk(sk)->copy_thresh = val;
3736		return 0;
3737	}
3738	case PACKET_VERSION:
3739	{
3740		int val;
3741
3742		if (optlen != sizeof(val))
3743			return -EINVAL;
3744		if (copy_from_sockptr(&val, optval, sizeof(val)))
3745			return -EFAULT;
3746		switch (val) {
3747		case TPACKET_V1:
3748		case TPACKET_V2:
3749		case TPACKET_V3:
3750			break;
3751		default:
3752			return -EINVAL;
3753		}
3754		lock_sock(sk);
3755		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3756			ret = -EBUSY;
3757		} else {
3758			po->tp_version = val;
3759			ret = 0;
3760		}
3761		release_sock(sk);
3762		return ret;
3763	}
3764	case PACKET_RESERVE:
3765	{
3766		unsigned int val;
3767
3768		if (optlen != sizeof(val))
3769			return -EINVAL;
3770		if (copy_from_sockptr(&val, optval, sizeof(val)))
3771			return -EFAULT;
3772		if (val > INT_MAX)
3773			return -EINVAL;
3774		lock_sock(sk);
3775		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3776			ret = -EBUSY;
3777		} else {
3778			po->tp_reserve = val;
3779			ret = 0;
3780		}
3781		release_sock(sk);
3782		return ret;
3783	}
3784	case PACKET_LOSS:
3785	{
3786		unsigned int val;
3787
3788		if (optlen != sizeof(val))
3789			return -EINVAL;
3790		if (copy_from_sockptr(&val, optval, sizeof(val)))
3791			return -EFAULT;
3792
3793		lock_sock(sk);
3794		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3795			ret = -EBUSY;
3796		} else {
3797			po->tp_loss = !!val;
3798			ret = 0;
3799		}
3800		release_sock(sk);
3801		return ret;
3802	}
3803	case PACKET_AUXDATA:
3804	{
3805		int val;
3806
3807		if (optlen < sizeof(val))
3808			return -EINVAL;
3809		if (copy_from_sockptr(&val, optval, sizeof(val)))
3810			return -EFAULT;
3811
3812		lock_sock(sk);
3813		po->auxdata = !!val;
3814		release_sock(sk);
3815		return 0;
3816	}
3817	case PACKET_ORIGDEV:
3818	{
3819		int val;
3820
3821		if (optlen < sizeof(val))
3822			return -EINVAL;
3823		if (copy_from_sockptr(&val, optval, sizeof(val)))
3824			return -EFAULT;
3825
3826		lock_sock(sk);
3827		po->origdev = !!val;
3828		release_sock(sk);
3829		return 0;
3830	}
3831	case PACKET_VNET_HDR:
3832	{
3833		int val;
3834
3835		if (sock->type != SOCK_RAW)
3836			return -EINVAL;
3837		if (optlen < sizeof(val))
3838			return -EINVAL;
3839		if (copy_from_sockptr(&val, optval, sizeof(val)))
3840			return -EFAULT;
3841
3842		lock_sock(sk);
3843		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3844			ret = -EBUSY;
3845		} else {
3846			po->has_vnet_hdr = !!val;
3847			ret = 0;
3848		}
3849		release_sock(sk);
3850		return ret;
3851	}
3852	case PACKET_TIMESTAMP:
3853	{
3854		int val;
3855
3856		if (optlen != sizeof(val))
3857			return -EINVAL;
3858		if (copy_from_sockptr(&val, optval, sizeof(val)))
3859			return -EFAULT;
3860
3861		po->tp_tstamp = val;
3862		return 0;
3863	}
3864	case PACKET_FANOUT:
3865	{
3866		int val;
3867
3868		if (optlen != sizeof(val))
3869			return -EINVAL;
3870		if (copy_from_sockptr(&val, optval, sizeof(val)))
3871			return -EFAULT;
3872
3873		return fanout_add(sk, val & 0xffff, val >> 16);
3874	}
3875	case PACKET_FANOUT_DATA:
3876	{
3877		if (!po->fanout)
 
3878			return -EINVAL;
3879
3880		return fanout_set_data(po, optval, optlen);
3881	}
3882	case PACKET_IGNORE_OUTGOING:
3883	{
3884		int val;
3885
3886		if (optlen != sizeof(val))
3887			return -EINVAL;
3888		if (copy_from_sockptr(&val, optval, sizeof(val)))
3889			return -EFAULT;
3890		if (val < 0 || val > 1)
3891			return -EINVAL;
3892
3893		po->prot_hook.ignore_outgoing = !!val;
3894		return 0;
3895	}
3896	case PACKET_TX_HAS_OFF:
3897	{
3898		unsigned int val;
3899
3900		if (optlen != sizeof(val))
3901			return -EINVAL;
3902		if (copy_from_sockptr(&val, optval, sizeof(val)))
3903			return -EFAULT;
3904
3905		lock_sock(sk);
3906		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3907			ret = -EBUSY;
3908		} else {
3909			po->tp_tx_has_off = !!val;
3910			ret = 0;
3911		}
3912		release_sock(sk);
3913		return 0;
3914	}
3915	case PACKET_QDISC_BYPASS:
3916	{
3917		int val;
3918
3919		if (optlen != sizeof(val))
3920			return -EINVAL;
3921		if (copy_from_sockptr(&val, optval, sizeof(val)))
3922			return -EFAULT;
3923
3924		po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3925		return 0;
3926	}
3927	default:
3928		return -ENOPROTOOPT;
3929	}
3930}
3931
3932static int packet_getsockopt(struct socket *sock, int level, int optname,
3933			     char __user *optval, int __user *optlen)
3934{
3935	int len;
3936	int val, lv = sizeof(val);
3937	struct sock *sk = sock->sk;
3938	struct packet_sock *po = pkt_sk(sk);
3939	void *data = &val;
3940	union tpacket_stats_u st;
3941	struct tpacket_rollover_stats rstats;
3942	int drops;
3943
3944	if (level != SOL_PACKET)
3945		return -ENOPROTOOPT;
3946
3947	if (get_user(len, optlen))
3948		return -EFAULT;
3949
3950	if (len < 0)
3951		return -EINVAL;
3952
3953	switch (optname) {
3954	case PACKET_STATISTICS:
3955		spin_lock_bh(&sk->sk_receive_queue.lock);
3956		memcpy(&st, &po->stats, sizeof(st));
3957		memset(&po->stats, 0, sizeof(po->stats));
3958		spin_unlock_bh(&sk->sk_receive_queue.lock);
3959		drops = atomic_xchg(&po->tp_drops, 0);
3960
3961		if (po->tp_version == TPACKET_V3) {
3962			lv = sizeof(struct tpacket_stats_v3);
3963			st.stats3.tp_drops = drops;
3964			st.stats3.tp_packets += drops;
3965			data = &st.stats3;
3966		} else {
3967			lv = sizeof(struct tpacket_stats);
3968			st.stats1.tp_drops = drops;
3969			st.stats1.tp_packets += drops;
3970			data = &st.stats1;
3971		}
3972
3973		break;
3974	case PACKET_AUXDATA:
3975		val = po->auxdata;
3976		break;
3977	case PACKET_ORIGDEV:
3978		val = po->origdev;
3979		break;
3980	case PACKET_VNET_HDR:
3981		val = po->has_vnet_hdr;
3982		break;
3983	case PACKET_VERSION:
3984		val = po->tp_version;
3985		break;
3986	case PACKET_HDRLEN:
3987		if (len > sizeof(int))
3988			len = sizeof(int);
3989		if (len < sizeof(int))
3990			return -EINVAL;
3991		if (copy_from_user(&val, optval, len))
3992			return -EFAULT;
3993		switch (val) {
3994		case TPACKET_V1:
3995			val = sizeof(struct tpacket_hdr);
3996			break;
3997		case TPACKET_V2:
3998			val = sizeof(struct tpacket2_hdr);
3999			break;
4000		case TPACKET_V3:
4001			val = sizeof(struct tpacket3_hdr);
4002			break;
4003		default:
4004			return -EINVAL;
4005		}
4006		break;
4007	case PACKET_RESERVE:
4008		val = po->tp_reserve;
4009		break;
4010	case PACKET_LOSS:
4011		val = po->tp_loss;
4012		break;
4013	case PACKET_TIMESTAMP:
4014		val = po->tp_tstamp;
4015		break;
4016	case PACKET_FANOUT:
4017		val = (po->fanout ?
4018		       ((u32)po->fanout->id |
4019			((u32)po->fanout->type << 16) |
4020			((u32)po->fanout->flags << 24)) :
4021		       0);
4022		break;
4023	case PACKET_IGNORE_OUTGOING:
4024		val = po->prot_hook.ignore_outgoing;
4025		break;
4026	case PACKET_ROLLOVER_STATS:
4027		if (!po->rollover)
4028			return -EINVAL;
4029		rstats.tp_all = atomic_long_read(&po->rollover->num);
4030		rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4031		rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4032		data = &rstats;
4033		lv = sizeof(rstats);
4034		break;
4035	case PACKET_TX_HAS_OFF:
4036		val = po->tp_tx_has_off;
4037		break;
4038	case PACKET_QDISC_BYPASS:
4039		val = packet_use_direct_xmit(po);
4040		break;
4041	default:
4042		return -ENOPROTOOPT;
4043	}
4044
4045	if (len > lv)
4046		len = lv;
4047	if (put_user(len, optlen))
4048		return -EFAULT;
4049	if (copy_to_user(optval, data, len))
4050		return -EFAULT;
4051	return 0;
4052}
4053
4054static int packet_notifier(struct notifier_block *this,
4055			   unsigned long msg, void *ptr)
4056{
4057	struct sock *sk;
4058	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4059	struct net *net = dev_net(dev);
4060
4061	rcu_read_lock();
4062	sk_for_each_rcu(sk, &net->packet.sklist) {
4063		struct packet_sock *po = pkt_sk(sk);
4064
4065		switch (msg) {
4066		case NETDEV_UNREGISTER:
4067			if (po->mclist)
4068				packet_dev_mclist_delete(dev, &po->mclist);
4069			fallthrough;
4070
4071		case NETDEV_DOWN:
4072			if (dev->ifindex == po->ifindex) {
4073				spin_lock(&po->bind_lock);
4074				if (po->running) {
4075					__unregister_prot_hook(sk, false);
4076					sk->sk_err = ENETDOWN;
4077					if (!sock_flag(sk, SOCK_DEAD))
4078						sk->sk_error_report(sk);
4079				}
4080				if (msg == NETDEV_UNREGISTER) {
4081					packet_cached_dev_reset(po);
4082					po->ifindex = -1;
4083					if (po->prot_hook.dev)
4084						dev_put(po->prot_hook.dev);
4085					po->prot_hook.dev = NULL;
4086				}
4087				spin_unlock(&po->bind_lock);
4088			}
4089			break;
4090		case NETDEV_UP:
4091			if (dev->ifindex == po->ifindex) {
4092				spin_lock(&po->bind_lock);
4093				if (po->num)
4094					register_prot_hook(sk);
4095				spin_unlock(&po->bind_lock);
4096			}
4097			break;
4098		}
4099	}
4100	rcu_read_unlock();
4101	return NOTIFY_DONE;
4102}
4103
4104
4105static int packet_ioctl(struct socket *sock, unsigned int cmd,
4106			unsigned long arg)
4107{
4108	struct sock *sk = sock->sk;
4109
4110	switch (cmd) {
4111	case SIOCOUTQ:
4112	{
4113		int amount = sk_wmem_alloc_get(sk);
4114
4115		return put_user(amount, (int __user *)arg);
4116	}
4117	case SIOCINQ:
4118	{
4119		struct sk_buff *skb;
4120		int amount = 0;
4121
4122		spin_lock_bh(&sk->sk_receive_queue.lock);
4123		skb = skb_peek(&sk->sk_receive_queue);
4124		if (skb)
4125			amount = skb->len;
4126		spin_unlock_bh(&sk->sk_receive_queue.lock);
4127		return put_user(amount, (int __user *)arg);
4128	}
4129#ifdef CONFIG_INET
4130	case SIOCADDRT:
4131	case SIOCDELRT:
4132	case SIOCDARP:
4133	case SIOCGARP:
4134	case SIOCSARP:
4135	case SIOCGIFADDR:
4136	case SIOCSIFADDR:
4137	case SIOCGIFBRDADDR:
4138	case SIOCSIFBRDADDR:
4139	case SIOCGIFNETMASK:
4140	case SIOCSIFNETMASK:
4141	case SIOCGIFDSTADDR:
4142	case SIOCSIFDSTADDR:
4143	case SIOCSIFFLAGS:
4144		return inet_dgram_ops.ioctl(sock, cmd, arg);
4145#endif
4146
4147	default:
4148		return -ENOIOCTLCMD;
4149	}
4150	return 0;
4151}
4152
4153static __poll_t packet_poll(struct file *file, struct socket *sock,
4154				poll_table *wait)
4155{
4156	struct sock *sk = sock->sk;
4157	struct packet_sock *po = pkt_sk(sk);
4158	__poll_t mask = datagram_poll(file, sock, wait);
4159
4160	spin_lock_bh(&sk->sk_receive_queue.lock);
4161	if (po->rx_ring.pg_vec) {
4162		if (!packet_previous_rx_frame(po, &po->rx_ring,
4163			TP_STATUS_KERNEL))
4164			mask |= EPOLLIN | EPOLLRDNORM;
4165	}
4166	packet_rcv_try_clear_pressure(po);
4167	spin_unlock_bh(&sk->sk_receive_queue.lock);
4168	spin_lock_bh(&sk->sk_write_queue.lock);
4169	if (po->tx_ring.pg_vec) {
4170		if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4171			mask |= EPOLLOUT | EPOLLWRNORM;
4172	}
4173	spin_unlock_bh(&sk->sk_write_queue.lock);
4174	return mask;
4175}
4176
4177
4178/* Dirty? Well, I still did not learn better way to account
4179 * for user mmaps.
4180 */
4181
4182static void packet_mm_open(struct vm_area_struct *vma)
4183{
4184	struct file *file = vma->vm_file;
4185	struct socket *sock = file->private_data;
4186	struct sock *sk = sock->sk;
4187
4188	if (sk)
4189		atomic_inc(&pkt_sk(sk)->mapped);
4190}
4191
4192static void packet_mm_close(struct vm_area_struct *vma)
4193{
4194	struct file *file = vma->vm_file;
4195	struct socket *sock = file->private_data;
4196	struct sock *sk = sock->sk;
4197
4198	if (sk)
4199		atomic_dec(&pkt_sk(sk)->mapped);
4200}
4201
4202static const struct vm_operations_struct packet_mmap_ops = {
4203	.open	=	packet_mm_open,
4204	.close	=	packet_mm_close,
4205};
4206
4207static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4208			unsigned int len)
4209{
4210	int i;
4211
4212	for (i = 0; i < len; i++) {
4213		if (likely(pg_vec[i].buffer)) {
4214			if (is_vmalloc_addr(pg_vec[i].buffer))
4215				vfree(pg_vec[i].buffer);
4216			else
4217				free_pages((unsigned long)pg_vec[i].buffer,
4218					   order);
4219			pg_vec[i].buffer = NULL;
4220		}
4221	}
4222	kfree(pg_vec);
4223}
4224
4225static char *alloc_one_pg_vec_page(unsigned long order)
4226{
4227	char *buffer;
4228	gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4229			  __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4230
4231	buffer = (char *) __get_free_pages(gfp_flags, order);
4232	if (buffer)
4233		return buffer;
4234
4235	/* __get_free_pages failed, fall back to vmalloc */
4236	buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4237	if (buffer)
4238		return buffer;
4239
4240	/* vmalloc failed, lets dig into swap here */
4241	gfp_flags &= ~__GFP_NORETRY;
4242	buffer = (char *) __get_free_pages(gfp_flags, order);
4243	if (buffer)
4244		return buffer;
4245
4246	/* complete and utter failure */
4247	return NULL;
4248}
4249
4250static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4251{
4252	unsigned int block_nr = req->tp_block_nr;
4253	struct pgv *pg_vec;
4254	int i;
4255
4256	pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4257	if (unlikely(!pg_vec))
4258		goto out;
4259
4260	for (i = 0; i < block_nr; i++) {
4261		pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4262		if (unlikely(!pg_vec[i].buffer))
4263			goto out_free_pgvec;
4264	}
4265
4266out:
4267	return pg_vec;
4268
4269out_free_pgvec:
4270	free_pg_vec(pg_vec, order, block_nr);
4271	pg_vec = NULL;
4272	goto out;
4273}
4274
4275static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4276		int closing, int tx_ring)
4277{
4278	struct pgv *pg_vec = NULL;
4279	struct packet_sock *po = pkt_sk(sk);
4280	unsigned long *rx_owner_map = NULL;
4281	int was_running, order = 0;
4282	struct packet_ring_buffer *rb;
4283	struct sk_buff_head *rb_queue;
4284	__be16 num;
4285	int err;
4286	/* Added to avoid minimal code churn */
4287	struct tpacket_req *req = &req_u->req;
4288
4289	rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4290	rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4291
4292	err = -EBUSY;
4293	if (!closing) {
4294		if (atomic_read(&po->mapped))
4295			goto out;
4296		if (packet_read_pending(rb))
4297			goto out;
4298	}
4299
4300	if (req->tp_block_nr) {
4301		unsigned int min_frame_size;
4302
4303		/* Sanity tests and some calculations */
4304		err = -EBUSY;
4305		if (unlikely(rb->pg_vec))
4306			goto out;
4307
4308		switch (po->tp_version) {
4309		case TPACKET_V1:
4310			po->tp_hdrlen = TPACKET_HDRLEN;
4311			break;
4312		case TPACKET_V2:
4313			po->tp_hdrlen = TPACKET2_HDRLEN;
4314			break;
4315		case TPACKET_V3:
4316			po->tp_hdrlen = TPACKET3_HDRLEN;
4317			break;
4318		}
4319
4320		err = -EINVAL;
4321		if (unlikely((int)req->tp_block_size <= 0))
4322			goto out;
4323		if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4324			goto out;
4325		min_frame_size = po->tp_hdrlen + po->tp_reserve;
4326		if (po->tp_version >= TPACKET_V3 &&
4327		    req->tp_block_size <
4328		    BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4329			goto out;
4330		if (unlikely(req->tp_frame_size < min_frame_size))
4331			goto out;
4332		if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4333			goto out;
4334
4335		rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4336		if (unlikely(rb->frames_per_block == 0))
4337			goto out;
4338		if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4339			goto out;
4340		if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4341					req->tp_frame_nr))
4342			goto out;
4343
4344		err = -ENOMEM;
4345		order = get_order(req->tp_block_size);
4346		pg_vec = alloc_pg_vec(req, order);
4347		if (unlikely(!pg_vec))
4348			goto out;
4349		switch (po->tp_version) {
4350		case TPACKET_V3:
4351			/* Block transmit is not supported yet */
4352			if (!tx_ring) {
4353				init_prb_bdqc(po, rb, pg_vec, req_u);
4354			} else {
4355				struct tpacket_req3 *req3 = &req_u->req3;
4356
4357				if (req3->tp_retire_blk_tov ||
4358				    req3->tp_sizeof_priv ||
4359				    req3->tp_feature_req_word) {
4360					err = -EINVAL;
4361					goto out_free_pg_vec;
4362				}
4363			}
4364			break;
4365		default:
4366			if (!tx_ring) {
4367				rx_owner_map = bitmap_alloc(req->tp_frame_nr,
4368					GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
4369				if (!rx_owner_map)
4370					goto out_free_pg_vec;
4371			}
4372			break;
4373		}
4374	}
4375	/* Done */
4376	else {
4377		err = -EINVAL;
4378		if (unlikely(req->tp_frame_nr))
4379			goto out;
4380	}
4381
4382
4383	/* Detach socket from network */
4384	spin_lock(&po->bind_lock);
4385	was_running = po->running;
4386	num = po->num;
4387	if (was_running) {
4388		po->num = 0;
4389		__unregister_prot_hook(sk, false);
4390	}
4391	spin_unlock(&po->bind_lock);
4392
4393	synchronize_net();
4394
4395	err = -EBUSY;
4396	mutex_lock(&po->pg_vec_lock);
4397	if (closing || atomic_read(&po->mapped) == 0) {
4398		err = 0;
4399		spin_lock_bh(&rb_queue->lock);
4400		swap(rb->pg_vec, pg_vec);
4401		if (po->tp_version <= TPACKET_V2)
4402			swap(rb->rx_owner_map, rx_owner_map);
4403		rb->frame_max = (req->tp_frame_nr - 1);
4404		rb->head = 0;
4405		rb->frame_size = req->tp_frame_size;
4406		spin_unlock_bh(&rb_queue->lock);
4407
4408		swap(rb->pg_vec_order, order);
4409		swap(rb->pg_vec_len, req->tp_block_nr);
4410
4411		rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4412		po->prot_hook.func = (po->rx_ring.pg_vec) ?
4413						tpacket_rcv : packet_rcv;
4414		skb_queue_purge(rb_queue);
4415		if (atomic_read(&po->mapped))
4416			pr_err("packet_mmap: vma is busy: %d\n",
4417			       atomic_read(&po->mapped));
4418	}
4419	mutex_unlock(&po->pg_vec_lock);
4420
4421	spin_lock(&po->bind_lock);
4422	if (was_running) {
4423		po->num = num;
4424		register_prot_hook(sk);
4425	}
4426	spin_unlock(&po->bind_lock);
4427	if (pg_vec && (po->tp_version > TPACKET_V2)) {
4428		/* Because we don't support block-based V3 on tx-ring */
4429		if (!tx_ring)
4430			prb_shutdown_retire_blk_timer(po, rb_queue);
4431	}
4432
4433out_free_pg_vec:
4434	bitmap_free(rx_owner_map);
4435	if (pg_vec)
4436		free_pg_vec(pg_vec, order, req->tp_block_nr);
 
4437out:
4438	return err;
4439}
4440
4441static int packet_mmap(struct file *file, struct socket *sock,
4442		struct vm_area_struct *vma)
4443{
4444	struct sock *sk = sock->sk;
4445	struct packet_sock *po = pkt_sk(sk);
4446	unsigned long size, expected_size;
4447	struct packet_ring_buffer *rb;
4448	unsigned long start;
4449	int err = -EINVAL;
4450	int i;
4451
4452	if (vma->vm_pgoff)
4453		return -EINVAL;
4454
4455	mutex_lock(&po->pg_vec_lock);
4456
4457	expected_size = 0;
4458	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4459		if (rb->pg_vec) {
4460			expected_size += rb->pg_vec_len
4461						* rb->pg_vec_pages
4462						* PAGE_SIZE;
4463		}
4464	}
4465
4466	if (expected_size == 0)
4467		goto out;
4468
4469	size = vma->vm_end - vma->vm_start;
4470	if (size != expected_size)
4471		goto out;
4472
4473	start = vma->vm_start;
4474	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4475		if (rb->pg_vec == NULL)
4476			continue;
4477
4478		for (i = 0; i < rb->pg_vec_len; i++) {
4479			struct page *page;
4480			void *kaddr = rb->pg_vec[i].buffer;
4481			int pg_num;
4482
4483			for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4484				page = pgv_to_page(kaddr);
4485				err = vm_insert_page(vma, start, page);
4486				if (unlikely(err))
4487					goto out;
4488				start += PAGE_SIZE;
4489				kaddr += PAGE_SIZE;
4490			}
4491		}
4492	}
4493
4494	atomic_inc(&po->mapped);
4495	vma->vm_ops = &packet_mmap_ops;
4496	err = 0;
4497
4498out:
4499	mutex_unlock(&po->pg_vec_lock);
4500	return err;
4501}
4502
4503static const struct proto_ops packet_ops_spkt = {
4504	.family =	PF_PACKET,
4505	.owner =	THIS_MODULE,
4506	.release =	packet_release,
4507	.bind =		packet_bind_spkt,
4508	.connect =	sock_no_connect,
4509	.socketpair =	sock_no_socketpair,
4510	.accept =	sock_no_accept,
4511	.getname =	packet_getname_spkt,
4512	.poll =		datagram_poll,
4513	.ioctl =	packet_ioctl,
4514	.gettstamp =	sock_gettstamp,
4515	.listen =	sock_no_listen,
4516	.shutdown =	sock_no_shutdown,
4517	.sendmsg =	packet_sendmsg_spkt,
4518	.recvmsg =	packet_recvmsg,
4519	.mmap =		sock_no_mmap,
4520	.sendpage =	sock_no_sendpage,
4521};
4522
4523static const struct proto_ops packet_ops = {
4524	.family =	PF_PACKET,
4525	.owner =	THIS_MODULE,
4526	.release =	packet_release,
4527	.bind =		packet_bind,
4528	.connect =	sock_no_connect,
4529	.socketpair =	sock_no_socketpair,
4530	.accept =	sock_no_accept,
4531	.getname =	packet_getname,
4532	.poll =		packet_poll,
4533	.ioctl =	packet_ioctl,
4534	.gettstamp =	sock_gettstamp,
4535	.listen =	sock_no_listen,
4536	.shutdown =	sock_no_shutdown,
4537	.setsockopt =	packet_setsockopt,
4538	.getsockopt =	packet_getsockopt,
4539	.sendmsg =	packet_sendmsg,
4540	.recvmsg =	packet_recvmsg,
4541	.mmap =		packet_mmap,
4542	.sendpage =	sock_no_sendpage,
4543};
4544
4545static const struct net_proto_family packet_family_ops = {
4546	.family =	PF_PACKET,
4547	.create =	packet_create,
4548	.owner	=	THIS_MODULE,
4549};
4550
4551static struct notifier_block packet_netdev_notifier = {
4552	.notifier_call =	packet_notifier,
4553};
4554
4555#ifdef CONFIG_PROC_FS
4556
4557static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4558	__acquires(RCU)
4559{
4560	struct net *net = seq_file_net(seq);
4561
4562	rcu_read_lock();
4563	return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4564}
4565
4566static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4567{
4568	struct net *net = seq_file_net(seq);
4569	return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4570}
4571
4572static void packet_seq_stop(struct seq_file *seq, void *v)
4573	__releases(RCU)
4574{
4575	rcu_read_unlock();
4576}
4577
4578static int packet_seq_show(struct seq_file *seq, void *v)
4579{
4580	if (v == SEQ_START_TOKEN)
4581		seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
 
 
4582	else {
4583		struct sock *s = sk_entry(v);
4584		const struct packet_sock *po = pkt_sk(s);
4585
4586		seq_printf(seq,
4587			   "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4588			   s,
4589			   refcount_read(&s->sk_refcnt),
4590			   s->sk_type,
4591			   ntohs(po->num),
4592			   po->ifindex,
4593			   po->running,
4594			   atomic_read(&s->sk_rmem_alloc),
4595			   from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4596			   sock_i_ino(s));
4597	}
4598
4599	return 0;
4600}
4601
4602static const struct seq_operations packet_seq_ops = {
4603	.start	= packet_seq_start,
4604	.next	= packet_seq_next,
4605	.stop	= packet_seq_stop,
4606	.show	= packet_seq_show,
4607};
4608#endif
4609
4610static int __net_init packet_net_init(struct net *net)
4611{
4612	mutex_init(&net->packet.sklist_lock);
4613	INIT_HLIST_HEAD(&net->packet.sklist);
4614
 
4615	if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4616			sizeof(struct seq_net_private)))
4617		return -ENOMEM;
 
4618
4619	return 0;
4620}
4621
4622static void __net_exit packet_net_exit(struct net *net)
4623{
4624	remove_proc_entry("packet", net->proc_net);
4625	WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4626}
4627
4628static struct pernet_operations packet_net_ops = {
4629	.init = packet_net_init,
4630	.exit = packet_net_exit,
4631};
4632
4633
4634static void __exit packet_exit(void)
4635{
4636	unregister_netdevice_notifier(&packet_netdev_notifier);
4637	unregister_pernet_subsys(&packet_net_ops);
4638	sock_unregister(PF_PACKET);
4639	proto_unregister(&packet_proto);
 
 
4640}
4641
4642static int __init packet_init(void)
4643{
4644	int rc;
4645
4646	rc = proto_register(&packet_proto, 0);
4647	if (rc)
4648		goto out;
4649	rc = sock_register(&packet_family_ops);
4650	if (rc)
4651		goto out_proto;
4652	rc = register_pernet_subsys(&packet_net_ops);
4653	if (rc)
4654		goto out_sock;
4655	rc = register_netdevice_notifier(&packet_netdev_notifier);
4656	if (rc)
4657		goto out_pernet;
 
 
 
 
 
 
4658
4659	return 0;
4660
4661out_pernet:
4662	unregister_pernet_subsys(&packet_net_ops);
4663out_sock:
4664	sock_unregister(PF_PACKET);
4665out_proto:
4666	proto_unregister(&packet_proto);
 
 
 
 
4667out:
4668	return rc;
4669}
4670
4671module_init(packet_init);
4672module_exit(packet_exit);
4673MODULE_LICENSE("GPL");
4674MODULE_ALIAS_NETPROTO(PF_PACKET);
v6.2
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * INET		An implementation of the TCP/IP protocol suite for the LINUX
   4 *		operating system.  INET is implemented using the  BSD Socket
   5 *		interface as the means of communication with the user level.
   6 *
   7 *		PACKET - implements raw packet sockets.
   8 *
   9 * Authors:	Ross Biro
  10 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  11 *		Alan Cox, <gw4pts@gw4pts.ampr.org>
  12 *
  13 * Fixes:
  14 *		Alan Cox	:	verify_area() now used correctly
  15 *		Alan Cox	:	new skbuff lists, look ma no backlogs!
  16 *		Alan Cox	:	tidied skbuff lists.
  17 *		Alan Cox	:	Now uses generic datagram routines I
  18 *					added. Also fixed the peek/read crash
  19 *					from all old Linux datagram code.
  20 *		Alan Cox	:	Uses the improved datagram code.
  21 *		Alan Cox	:	Added NULL's for socket options.
  22 *		Alan Cox	:	Re-commented the code.
  23 *		Alan Cox	:	Use new kernel side addressing
  24 *		Rob Janssen	:	Correct MTU usage.
  25 *		Dave Platt	:	Counter leaks caused by incorrect
  26 *					interrupt locking and some slightly
  27 *					dubious gcc output. Can you read
  28 *					compiler: it said _VOLATILE_
  29 *	Richard Kooijman	:	Timestamp fixes.
  30 *		Alan Cox	:	New buffers. Use sk->mac.raw.
  31 *		Alan Cox	:	sendmsg/recvmsg support.
  32 *		Alan Cox	:	Protocol setting support
  33 *	Alexey Kuznetsov	:	Untied from IPv4 stack.
  34 *	Cyrus Durgin		:	Fixed kerneld for kmod.
  35 *	Michal Ostrowski        :       Module initialization cleanup.
  36 *         Ulises Alonso        :       Frame number limit removal and
  37 *                                      packet_set_ring memory leak.
  38 *		Eric Biederman	:	Allow for > 8 byte hardware addresses.
  39 *					The convention is that longer addresses
  40 *					will simply extend the hardware address
  41 *					byte arrays at the end of sockaddr_ll
  42 *					and packet_mreq.
  43 *		Johann Baudy	:	Added TX RING.
  44 *		Chetan Loke	:	Implemented TPACKET_V3 block abstraction
  45 *					layer.
  46 *					Copyright (C) 2011, <lokec@ccs.neu.edu>
  47 */
  48
  49#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  50
  51#include <linux/ethtool.h>
  52#include <linux/filter.h>
  53#include <linux/types.h>
  54#include <linux/mm.h>
  55#include <linux/capability.h>
  56#include <linux/fcntl.h>
  57#include <linux/socket.h>
  58#include <linux/in.h>
  59#include <linux/inet.h>
  60#include <linux/netdevice.h>
  61#include <linux/if_packet.h>
  62#include <linux/wireless.h>
  63#include <linux/kernel.h>
  64#include <linux/kmod.h>
  65#include <linux/slab.h>
  66#include <linux/vmalloc.h>
  67#include <net/net_namespace.h>
  68#include <net/ip.h>
  69#include <net/protocol.h>
  70#include <linux/skbuff.h>
  71#include <net/sock.h>
  72#include <linux/errno.h>
  73#include <linux/timer.h>
  74#include <linux/uaccess.h>
  75#include <asm/ioctls.h>
  76#include <asm/page.h>
  77#include <asm/cacheflush.h>
  78#include <asm/io.h>
  79#include <linux/proc_fs.h>
  80#include <linux/seq_file.h>
  81#include <linux/poll.h>
  82#include <linux/module.h>
  83#include <linux/init.h>
  84#include <linux/mutex.h>
  85#include <linux/if_vlan.h>
  86#include <linux/virtio_net.h>
  87#include <linux/errqueue.h>
  88#include <linux/net_tstamp.h>
  89#include <linux/percpu.h>
  90#ifdef CONFIG_INET
  91#include <net/inet_common.h>
  92#endif
  93#include <linux/bpf.h>
  94#include <net/compat.h>
  95#include <linux/netfilter_netdev.h>
  96
  97#include "internal.h"
  98
  99/*
 100   Assumptions:
 101   - If the device has no dev->header_ops->create, there is no LL header
 102     visible above the device. In this case, its hard_header_len should be 0.
 103     The device may prepend its own header internally. In this case, its
 104     needed_headroom should be set to the space needed for it to add its
 105     internal header.
 106     For example, a WiFi driver pretending to be an Ethernet driver should
 107     set its hard_header_len to be the Ethernet header length, and set its
 108     needed_headroom to be (the real WiFi header length - the fake Ethernet
 109     header length).
 110   - packet socket receives packets with pulled ll header,
 111     so that SOCK_RAW should push it back.
 112
 113On receive:
 114-----------
 115
 116Incoming, dev_has_header(dev) == true
 117   mac_header -> ll header
 118   data       -> data
 119
 120Outgoing, dev_has_header(dev) == true
 121   mac_header -> ll header
 122   data       -> ll header
 123
 124Incoming, dev_has_header(dev) == false
 125   mac_header -> data
 126     However drivers often make it point to the ll header.
 127     This is incorrect because the ll header should be invisible to us.
 128   data       -> data
 129
 130Outgoing, dev_has_header(dev) == false
 131   mac_header -> data. ll header is invisible to us.
 132   data       -> data
 133
 134Resume
 135  If dev_has_header(dev) == false we are unable to restore the ll header,
 136    because it is invisible to us.
 137
 138
 139On transmit:
 140------------
 141
 142dev_has_header(dev) == true
 143   mac_header -> ll header
 144   data       -> ll header
 145
 146dev_has_header(dev) == false (ll header is invisible to us)
 147   mac_header -> data
 148   data       -> data
 149
 150   We should set network_header on output to the correct position,
 151   packet classifier depends on it.
 152 */
 153
 154/* Private packet socket structures. */
 155
 156/* identical to struct packet_mreq except it has
 157 * a longer address field.
 158 */
 159struct packet_mreq_max {
 160	int		mr_ifindex;
 161	unsigned short	mr_type;
 162	unsigned short	mr_alen;
 163	unsigned char	mr_address[MAX_ADDR_LEN];
 164};
 165
 166union tpacket_uhdr {
 167	struct tpacket_hdr  *h1;
 168	struct tpacket2_hdr *h2;
 169	struct tpacket3_hdr *h3;
 170	void *raw;
 171};
 172
 173static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 174		int closing, int tx_ring);
 175
 176#define V3_ALIGNMENT	(8)
 177
 178#define BLK_HDR_LEN	(ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 179
 180#define BLK_PLUS_PRIV(sz_of_priv) \
 181	(BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 182
 183#define BLOCK_STATUS(x)	((x)->hdr.bh1.block_status)
 184#define BLOCK_NUM_PKTS(x)	((x)->hdr.bh1.num_pkts)
 185#define BLOCK_O2FP(x)		((x)->hdr.bh1.offset_to_first_pkt)
 186#define BLOCK_LEN(x)		((x)->hdr.bh1.blk_len)
 187#define BLOCK_SNUM(x)		((x)->hdr.bh1.seq_num)
 188#define BLOCK_O2PRIV(x)	((x)->offset_to_priv)
 
 189
 190struct packet_sock;
 191static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 192		       struct packet_type *pt, struct net_device *orig_dev);
 193
 194static void *packet_previous_frame(struct packet_sock *po,
 195		struct packet_ring_buffer *rb,
 196		int status);
 197static void packet_increment_head(struct packet_ring_buffer *buff);
 198static int prb_curr_blk_in_use(struct tpacket_block_desc *);
 199static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 200			struct packet_sock *);
 201static void prb_retire_current_block(struct tpacket_kbdq_core *,
 202		struct packet_sock *, unsigned int status);
 203static int prb_queue_frozen(struct tpacket_kbdq_core *);
 204static void prb_open_block(struct tpacket_kbdq_core *,
 205		struct tpacket_block_desc *);
 206static void prb_retire_rx_blk_timer_expired(struct timer_list *);
 207static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 208static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 209static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 210		struct tpacket3_hdr *);
 211static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 212		struct tpacket3_hdr *);
 213static void packet_flush_mclist(struct sock *sk);
 214static u16 packet_pick_tx_queue(struct sk_buff *skb);
 215
 216struct packet_skb_cb {
 217	union {
 218		struct sockaddr_pkt pkt;
 219		union {
 220			/* Trick: alias skb original length with
 221			 * ll.sll_family and ll.protocol in order
 222			 * to save room.
 223			 */
 224			unsigned int origlen;
 225			struct sockaddr_ll ll;
 226		};
 227	} sa;
 228};
 229
 230#define vio_le() virtio_legacy_is_little_endian()
 231
 232#define PACKET_SKB_CB(__skb)	((struct packet_skb_cb *)((__skb)->cb))
 233
 234#define GET_PBDQC_FROM_RB(x)	((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 235#define GET_PBLOCK_DESC(x, bid)	\
 236	((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 237#define GET_CURR_PBLOCK_DESC_FROM_CORE(x)	\
 238	((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 239#define GET_NEXT_PRB_BLK_NUM(x) \
 240	(((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 241	((x)->kactive_blk_num+1) : 0)
 242
 243static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 244static void __fanout_link(struct sock *sk, struct packet_sock *po);
 245
 246#ifdef CONFIG_NETFILTER_EGRESS
 247static noinline struct sk_buff *nf_hook_direct_egress(struct sk_buff *skb)
 248{
 249	struct sk_buff *next, *head = NULL, *tail;
 250	int rc;
 251
 252	rcu_read_lock();
 253	for (; skb != NULL; skb = next) {
 254		next = skb->next;
 255		skb_mark_not_on_list(skb);
 256
 257		if (!nf_hook_egress(skb, &rc, skb->dev))
 258			continue;
 259
 260		if (!head)
 261			head = skb;
 262		else
 263			tail->next = skb;
 264
 265		tail = skb;
 266	}
 267	rcu_read_unlock();
 268
 269	return head;
 270}
 271#endif
 272
 273static int packet_direct_xmit(struct sk_buff *skb)
 274{
 275#ifdef CONFIG_NETFILTER_EGRESS
 276	if (nf_hook_egress_active()) {
 277		skb = nf_hook_direct_egress(skb);
 278		if (!skb)
 279			return NET_XMIT_DROP;
 280	}
 281#endif
 282	return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
 283}
 284
 285static struct net_device *packet_cached_dev_get(struct packet_sock *po)
 286{
 287	struct net_device *dev;
 288
 289	rcu_read_lock();
 290	dev = rcu_dereference(po->cached_dev);
 291	dev_hold(dev);
 
 292	rcu_read_unlock();
 293
 294	return dev;
 295}
 296
 297static void packet_cached_dev_assign(struct packet_sock *po,
 298				     struct net_device *dev)
 299{
 300	rcu_assign_pointer(po->cached_dev, dev);
 301}
 302
 303static void packet_cached_dev_reset(struct packet_sock *po)
 304{
 305	RCU_INIT_POINTER(po->cached_dev, NULL);
 306}
 307
 308static bool packet_use_direct_xmit(const struct packet_sock *po)
 309{
 310	return po->xmit == packet_direct_xmit;
 311}
 312
 313static u16 packet_pick_tx_queue(struct sk_buff *skb)
 314{
 315	struct net_device *dev = skb->dev;
 316	const struct net_device_ops *ops = dev->netdev_ops;
 317	int cpu = raw_smp_processor_id();
 318	u16 queue_index;
 319
 320#ifdef CONFIG_XPS
 321	skb->sender_cpu = cpu + 1;
 322#endif
 323	skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues);
 324	if (ops->ndo_select_queue) {
 325		queue_index = ops->ndo_select_queue(dev, skb, NULL);
 326		queue_index = netdev_cap_txqueue(dev, queue_index);
 327	} else {
 328		queue_index = netdev_pick_tx(dev, skb, NULL);
 329	}
 330
 331	return queue_index;
 332}
 333
 334/* __register_prot_hook must be invoked through register_prot_hook
 335 * or from a context in which asynchronous accesses to the packet
 336 * socket is not possible (packet_create()).
 337 */
 338static void __register_prot_hook(struct sock *sk)
 339{
 340	struct packet_sock *po = pkt_sk(sk);
 341
 342	if (!po->running) {
 343		if (po->fanout)
 344			__fanout_link(sk, po);
 345		else
 346			dev_add_pack(&po->prot_hook);
 347
 348		sock_hold(sk);
 349		po->running = 1;
 350	}
 351}
 352
 353static void register_prot_hook(struct sock *sk)
 354{
 355	lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
 356	__register_prot_hook(sk);
 357}
 358
 359/* If the sync parameter is true, we will temporarily drop
 360 * the po->bind_lock and do a synchronize_net to make sure no
 361 * asynchronous packet processing paths still refer to the elements
 362 * of po->prot_hook.  If the sync parameter is false, it is the
 363 * callers responsibility to take care of this.
 364 */
 365static void __unregister_prot_hook(struct sock *sk, bool sync)
 366{
 367	struct packet_sock *po = pkt_sk(sk);
 368
 369	lockdep_assert_held_once(&po->bind_lock);
 370
 371	po->running = 0;
 372
 373	if (po->fanout)
 374		__fanout_unlink(sk, po);
 375	else
 376		__dev_remove_pack(&po->prot_hook);
 377
 378	__sock_put(sk);
 379
 380	if (sync) {
 381		spin_unlock(&po->bind_lock);
 382		synchronize_net();
 383		spin_lock(&po->bind_lock);
 384	}
 385}
 386
 387static void unregister_prot_hook(struct sock *sk, bool sync)
 388{
 389	struct packet_sock *po = pkt_sk(sk);
 390
 391	if (po->running)
 392		__unregister_prot_hook(sk, sync);
 393}
 394
 395static inline struct page * __pure pgv_to_page(void *addr)
 396{
 397	if (is_vmalloc_addr(addr))
 398		return vmalloc_to_page(addr);
 399	return virt_to_page(addr);
 400}
 401
 402static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 403{
 404	union tpacket_uhdr h;
 405
 406	h.raw = frame;
 407	switch (po->tp_version) {
 408	case TPACKET_V1:
 409		h.h1->tp_status = status;
 410		flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 411		break;
 412	case TPACKET_V2:
 413		h.h2->tp_status = status;
 414		flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 415		break;
 416	case TPACKET_V3:
 417		h.h3->tp_status = status;
 418		flush_dcache_page(pgv_to_page(&h.h3->tp_status));
 419		break;
 420	default:
 421		WARN(1, "TPACKET version not supported.\n");
 422		BUG();
 423	}
 424
 425	smp_wmb();
 426}
 427
 428static int __packet_get_status(const struct packet_sock *po, void *frame)
 429{
 430	union tpacket_uhdr h;
 431
 432	smp_rmb();
 433
 434	h.raw = frame;
 435	switch (po->tp_version) {
 436	case TPACKET_V1:
 437		flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 438		return h.h1->tp_status;
 439	case TPACKET_V2:
 440		flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 441		return h.h2->tp_status;
 442	case TPACKET_V3:
 443		flush_dcache_page(pgv_to_page(&h.h3->tp_status));
 444		return h.h3->tp_status;
 445	default:
 446		WARN(1, "TPACKET version not supported.\n");
 447		BUG();
 448		return 0;
 449	}
 450}
 451
 452static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec64 *ts,
 453				   unsigned int flags)
 454{
 455	struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 456
 457	if (shhwtstamps &&
 458	    (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
 459	    ktime_to_timespec64_cond(shhwtstamps->hwtstamp, ts))
 460		return TP_STATUS_TS_RAW_HARDWARE;
 461
 462	if ((flags & SOF_TIMESTAMPING_SOFTWARE) &&
 463	    ktime_to_timespec64_cond(skb_tstamp(skb), ts))
 464		return TP_STATUS_TS_SOFTWARE;
 465
 466	return 0;
 467}
 468
 469static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
 470				    struct sk_buff *skb)
 471{
 472	union tpacket_uhdr h;
 473	struct timespec64 ts;
 474	__u32 ts_status;
 475
 476	if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
 477		return 0;
 478
 479	h.raw = frame;
 480	/*
 481	 * versions 1 through 3 overflow the timestamps in y2106, since they
 482	 * all store the seconds in a 32-bit unsigned integer.
 483	 * If we create a version 4, that should have a 64-bit timestamp,
 484	 * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit
 485	 * nanoseconds.
 486	 */
 487	switch (po->tp_version) {
 488	case TPACKET_V1:
 489		h.h1->tp_sec = ts.tv_sec;
 490		h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
 491		break;
 492	case TPACKET_V2:
 493		h.h2->tp_sec = ts.tv_sec;
 494		h.h2->tp_nsec = ts.tv_nsec;
 495		break;
 496	case TPACKET_V3:
 497		h.h3->tp_sec = ts.tv_sec;
 498		h.h3->tp_nsec = ts.tv_nsec;
 499		break;
 500	default:
 501		WARN(1, "TPACKET version not supported.\n");
 502		BUG();
 503	}
 504
 505	/* one flush is safe, as both fields always lie on the same cacheline */
 506	flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
 507	smp_wmb();
 508
 509	return ts_status;
 510}
 511
 512static void *packet_lookup_frame(const struct packet_sock *po,
 513				 const struct packet_ring_buffer *rb,
 514				 unsigned int position,
 515				 int status)
 516{
 517	unsigned int pg_vec_pos, frame_offset;
 518	union tpacket_uhdr h;
 519
 520	pg_vec_pos = position / rb->frames_per_block;
 521	frame_offset = position % rb->frames_per_block;
 522
 523	h.raw = rb->pg_vec[pg_vec_pos].buffer +
 524		(frame_offset * rb->frame_size);
 525
 526	if (status != __packet_get_status(po, h.raw))
 527		return NULL;
 528
 529	return h.raw;
 530}
 531
 532static void *packet_current_frame(struct packet_sock *po,
 533		struct packet_ring_buffer *rb,
 534		int status)
 535{
 536	return packet_lookup_frame(po, rb, rb->head, status);
 537}
 538
 539static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 540{
 541	del_timer_sync(&pkc->retire_blk_timer);
 542}
 543
 544static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 545		struct sk_buff_head *rb_queue)
 546{
 547	struct tpacket_kbdq_core *pkc;
 548
 549	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 550
 551	spin_lock_bh(&rb_queue->lock);
 552	pkc->delete_blk_timer = 1;
 553	spin_unlock_bh(&rb_queue->lock);
 554
 555	prb_del_retire_blk_timer(pkc);
 556}
 557
 558static void prb_setup_retire_blk_timer(struct packet_sock *po)
 559{
 560	struct tpacket_kbdq_core *pkc;
 561
 562	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 563	timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
 564		    0);
 565	pkc->retire_blk_timer.expires = jiffies;
 566}
 567
 568static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 569				int blk_size_in_bytes)
 570{
 571	struct net_device *dev;
 572	unsigned int mbits, div;
 573	struct ethtool_link_ksettings ecmd;
 574	int err;
 575
 576	rtnl_lock();
 577	dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 578	if (unlikely(!dev)) {
 579		rtnl_unlock();
 580		return DEFAULT_PRB_RETIRE_TOV;
 581	}
 582	err = __ethtool_get_link_ksettings(dev, &ecmd);
 583	rtnl_unlock();
 584	if (err)
 585		return DEFAULT_PRB_RETIRE_TOV;
 586
 587	/* If the link speed is so slow you don't really
 588	 * need to worry about perf anyways
 589	 */
 590	if (ecmd.base.speed < SPEED_1000 ||
 591	    ecmd.base.speed == SPEED_UNKNOWN)
 592		return DEFAULT_PRB_RETIRE_TOV;
 593
 594	div = ecmd.base.speed / 1000;
 595	mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 596
 597	if (div)
 598		mbits /= div;
 599
 600	if (div)
 601		return mbits + 1;
 602	return mbits;
 603}
 604
 605static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 606			union tpacket_req_u *req_u)
 607{
 608	p1->feature_req_word = req_u->req3.tp_feature_req_word;
 609}
 610
 611static void init_prb_bdqc(struct packet_sock *po,
 612			struct packet_ring_buffer *rb,
 613			struct pgv *pg_vec,
 614			union tpacket_req_u *req_u)
 615{
 616	struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
 617	struct tpacket_block_desc *pbd;
 618
 619	memset(p1, 0x0, sizeof(*p1));
 620
 621	p1->knxt_seq_num = 1;
 622	p1->pkbdq = pg_vec;
 623	pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 624	p1->pkblk_start	= pg_vec[0].buffer;
 625	p1->kblk_size = req_u->req3.tp_block_size;
 626	p1->knum_blocks	= req_u->req3.tp_block_nr;
 627	p1->hdrlen = po->tp_hdrlen;
 628	p1->version = po->tp_version;
 629	p1->last_kactive_blk_num = 0;
 630	po->stats.stats3.tp_freeze_q_cnt = 0;
 631	if (req_u->req3.tp_retire_blk_tov)
 632		p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 633	else
 634		p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 635						req_u->req3.tp_block_size);
 636	p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 637	p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 638	rwlock_init(&p1->blk_fill_in_prog_lock);
 639
 640	p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
 641	prb_init_ft_ops(p1, req_u);
 642	prb_setup_retire_blk_timer(po);
 643	prb_open_block(p1, pbd);
 644}
 645
 646/*  Do NOT update the last_blk_num first.
 647 *  Assumes sk_buff_head lock is held.
 648 */
 649static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 650{
 651	mod_timer(&pkc->retire_blk_timer,
 652			jiffies + pkc->tov_in_jiffies);
 653	pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 654}
 655
 656/*
 657 * Timer logic:
 658 * 1) We refresh the timer only when we open a block.
 659 *    By doing this we don't waste cycles refreshing the timer
 660 *	  on packet-by-packet basis.
 661 *
 662 * With a 1MB block-size, on a 1Gbps line, it will take
 663 * i) ~8 ms to fill a block + ii) memcpy etc.
 664 * In this cut we are not accounting for the memcpy time.
 665 *
 666 * So, if the user sets the 'tmo' to 10ms then the timer
 667 * will never fire while the block is still getting filled
 668 * (which is what we want). However, the user could choose
 669 * to close a block early and that's fine.
 670 *
 671 * But when the timer does fire, we check whether or not to refresh it.
 672 * Since the tmo granularity is in msecs, it is not too expensive
 673 * to refresh the timer, lets say every '8' msecs.
 674 * Either the user can set the 'tmo' or we can derive it based on
 675 * a) line-speed and b) block-size.
 676 * prb_calc_retire_blk_tmo() calculates the tmo.
 677 *
 678 */
 679static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
 680{
 681	struct packet_sock *po =
 682		from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
 683	struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 684	unsigned int frozen;
 685	struct tpacket_block_desc *pbd;
 686
 687	spin_lock(&po->sk.sk_receive_queue.lock);
 688
 689	frozen = prb_queue_frozen(pkc);
 690	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 691
 692	if (unlikely(pkc->delete_blk_timer))
 693		goto out;
 694
 695	/* We only need to plug the race when the block is partially filled.
 696	 * tpacket_rcv:
 697	 *		lock(); increment BLOCK_NUM_PKTS; unlock()
 698	 *		copy_bits() is in progress ...
 699	 *		timer fires on other cpu:
 700	 *		we can't retire the current block because copy_bits
 701	 *		is in progress.
 702	 *
 703	 */
 704	if (BLOCK_NUM_PKTS(pbd)) {
 705		/* Waiting for skb_copy_bits to finish... */
 706		write_lock(&pkc->blk_fill_in_prog_lock);
 707		write_unlock(&pkc->blk_fill_in_prog_lock);
 708	}
 709
 710	if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 711		if (!frozen) {
 712			if (!BLOCK_NUM_PKTS(pbd)) {
 713				/* An empty block. Just refresh the timer. */
 714				goto refresh_timer;
 715			}
 716			prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 717			if (!prb_dispatch_next_block(pkc, po))
 718				goto refresh_timer;
 719			else
 720				goto out;
 721		} else {
 722			/* Case 1. Queue was frozen because user-space was
 723			 *	   lagging behind.
 724			 */
 725			if (prb_curr_blk_in_use(pbd)) {
 726				/*
 727				 * Ok, user-space is still behind.
 728				 * So just refresh the timer.
 729				 */
 730				goto refresh_timer;
 731			} else {
 732			       /* Case 2. queue was frozen,user-space caught up,
 733				* now the link went idle && the timer fired.
 734				* We don't have a block to close.So we open this
 735				* block and restart the timer.
 736				* opening a block thaws the queue,restarts timer
 737				* Thawing/timer-refresh is a side effect.
 738				*/
 739				prb_open_block(pkc, pbd);
 740				goto out;
 741			}
 742		}
 743	}
 744
 745refresh_timer:
 746	_prb_refresh_rx_retire_blk_timer(pkc);
 747
 748out:
 749	spin_unlock(&po->sk.sk_receive_queue.lock);
 750}
 751
 752static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 753		struct tpacket_block_desc *pbd1, __u32 status)
 754{
 755	/* Flush everything minus the block header */
 756
 757#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 758	u8 *start, *end;
 759
 760	start = (u8 *)pbd1;
 761
 762	/* Skip the block header(we know header WILL fit in 4K) */
 763	start += PAGE_SIZE;
 764
 765	end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 766	for (; start < end; start += PAGE_SIZE)
 767		flush_dcache_page(pgv_to_page(start));
 768
 769	smp_wmb();
 770#endif
 771
 772	/* Now update the block status. */
 773
 774	BLOCK_STATUS(pbd1) = status;
 775
 776	/* Flush the block header */
 777
 778#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 779	start = (u8 *)pbd1;
 780	flush_dcache_page(pgv_to_page(start));
 781
 782	smp_wmb();
 783#endif
 784}
 785
 786/*
 787 * Side effect:
 788 *
 789 * 1) flush the block
 790 * 2) Increment active_blk_num
 791 *
 792 * Note:We DONT refresh the timer on purpose.
 793 *	Because almost always the next block will be opened.
 794 */
 795static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 796		struct tpacket_block_desc *pbd1,
 797		struct packet_sock *po, unsigned int stat)
 798{
 799	__u32 status = TP_STATUS_USER | stat;
 800
 801	struct tpacket3_hdr *last_pkt;
 802	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 803	struct sock *sk = &po->sk;
 804
 805	if (atomic_read(&po->tp_drops))
 806		status |= TP_STATUS_LOSING;
 807
 808	last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 809	last_pkt->tp_next_offset = 0;
 810
 811	/* Get the ts of the last pkt */
 812	if (BLOCK_NUM_PKTS(pbd1)) {
 813		h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 814		h1->ts_last_pkt.ts_nsec	= last_pkt->tp_nsec;
 815	} else {
 816		/* Ok, we tmo'd - so get the current time.
 817		 *
 818		 * It shouldn't really happen as we don't close empty
 819		 * blocks. See prb_retire_rx_blk_timer_expired().
 820		 */
 821		struct timespec64 ts;
 822		ktime_get_real_ts64(&ts);
 823		h1->ts_last_pkt.ts_sec = ts.tv_sec;
 824		h1->ts_last_pkt.ts_nsec	= ts.tv_nsec;
 825	}
 826
 827	smp_wmb();
 828
 829	/* Flush the block */
 830	prb_flush_block(pkc1, pbd1, status);
 831
 832	sk->sk_data_ready(sk);
 833
 834	pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 835}
 836
 837static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 838{
 839	pkc->reset_pending_on_curr_blk = 0;
 840}
 841
 842/*
 843 * Side effect of opening a block:
 844 *
 845 * 1) prb_queue is thawed.
 846 * 2) retire_blk_timer is refreshed.
 847 *
 848 */
 849static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 850	struct tpacket_block_desc *pbd1)
 851{
 852	struct timespec64 ts;
 853	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 854
 855	smp_rmb();
 856
 857	/* We could have just memset this but we will lose the
 858	 * flexibility of making the priv area sticky
 859	 */
 860
 861	BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 862	BLOCK_NUM_PKTS(pbd1) = 0;
 863	BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 864
 865	ktime_get_real_ts64(&ts);
 866
 867	h1->ts_first_pkt.ts_sec = ts.tv_sec;
 868	h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 869
 870	pkc1->pkblk_start = (char *)pbd1;
 871	pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 872
 873	BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 874	BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 875
 876	pbd1->version = pkc1->version;
 877	pkc1->prev = pkc1->nxt_offset;
 878	pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 879
 880	prb_thaw_queue(pkc1);
 881	_prb_refresh_rx_retire_blk_timer(pkc1);
 882
 883	smp_wmb();
 884}
 885
 886/*
 887 * Queue freeze logic:
 888 * 1) Assume tp_block_nr = 8 blocks.
 889 * 2) At time 't0', user opens Rx ring.
 890 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 891 * 4) user-space is either sleeping or processing block '0'.
 892 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 893 *    it will close block-7,loop around and try to fill block '0'.
 894 *    call-flow:
 895 *    __packet_lookup_frame_in_block
 896 *      prb_retire_current_block()
 897 *      prb_dispatch_next_block()
 898 *        |->(BLOCK_STATUS == USER) evaluates to true
 899 *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 900 * 6) Now there are two cases:
 901 *    6.1) Link goes idle right after the queue is frozen.
 902 *         But remember, the last open_block() refreshed the timer.
 903 *         When this timer expires,it will refresh itself so that we can
 904 *         re-open block-0 in near future.
 905 *    6.2) Link is busy and keeps on receiving packets. This is a simple
 906 *         case and __packet_lookup_frame_in_block will check if block-0
 907 *         is free and can now be re-used.
 908 */
 909static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 910				  struct packet_sock *po)
 911{
 912	pkc->reset_pending_on_curr_blk = 1;
 913	po->stats.stats3.tp_freeze_q_cnt++;
 914}
 915
 916#define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 917
 918/*
 919 * If the next block is free then we will dispatch it
 920 * and return a good offset.
 921 * Else, we will freeze the queue.
 922 * So, caller must check the return value.
 923 */
 924static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 925		struct packet_sock *po)
 926{
 927	struct tpacket_block_desc *pbd;
 928
 929	smp_rmb();
 930
 931	/* 1. Get current block num */
 932	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 933
 934	/* 2. If this block is currently in_use then freeze the queue */
 935	if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 936		prb_freeze_queue(pkc, po);
 937		return NULL;
 938	}
 939
 940	/*
 941	 * 3.
 942	 * open this block and return the offset where the first packet
 943	 * needs to get stored.
 944	 */
 945	prb_open_block(pkc, pbd);
 946	return (void *)pkc->nxt_offset;
 947}
 948
 949static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 950		struct packet_sock *po, unsigned int status)
 951{
 952	struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 953
 954	/* retire/close the current block */
 955	if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 956		/*
 957		 * Plug the case where copy_bits() is in progress on
 958		 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 959		 * have space to copy the pkt in the current block and
 960		 * called prb_retire_current_block()
 961		 *
 962		 * We don't need to worry about the TMO case because
 963		 * the timer-handler already handled this case.
 964		 */
 965		if (!(status & TP_STATUS_BLK_TMO)) {
 966			/* Waiting for skb_copy_bits to finish... */
 967			write_lock(&pkc->blk_fill_in_prog_lock);
 968			write_unlock(&pkc->blk_fill_in_prog_lock);
 969		}
 970		prb_close_block(pkc, pbd, po, status);
 971		return;
 972	}
 973}
 974
 975static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
 976{
 977	return TP_STATUS_USER & BLOCK_STATUS(pbd);
 978}
 979
 980static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 981{
 982	return pkc->reset_pending_on_curr_blk;
 983}
 984
 985static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 986	__releases(&pkc->blk_fill_in_prog_lock)
 987{
 988	struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 989
 990	read_unlock(&pkc->blk_fill_in_prog_lock);
 991}
 992
 993static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 994			struct tpacket3_hdr *ppd)
 995{
 996	ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
 997}
 998
 999static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
1000			struct tpacket3_hdr *ppd)
1001{
1002	ppd->hv1.tp_rxhash = 0;
1003}
1004
1005static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1006			struct tpacket3_hdr *ppd)
1007{
1008	if (skb_vlan_tag_present(pkc->skb)) {
1009		ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1010		ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1011		ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1012	} else {
1013		ppd->hv1.tp_vlan_tci = 0;
1014		ppd->hv1.tp_vlan_tpid = 0;
1015		ppd->tp_status = TP_STATUS_AVAILABLE;
1016	}
1017}
1018
1019static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1020			struct tpacket3_hdr *ppd)
1021{
1022	ppd->hv1.tp_padding = 0;
1023	prb_fill_vlan_info(pkc, ppd);
1024
1025	if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1026		prb_fill_rxhash(pkc, ppd);
1027	else
1028		prb_clear_rxhash(pkc, ppd);
1029}
1030
1031static void prb_fill_curr_block(char *curr,
1032				struct tpacket_kbdq_core *pkc,
1033				struct tpacket_block_desc *pbd,
1034				unsigned int len)
1035	__acquires(&pkc->blk_fill_in_prog_lock)
1036{
1037	struct tpacket3_hdr *ppd;
1038
1039	ppd  = (struct tpacket3_hdr *)curr;
1040	ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1041	pkc->prev = curr;
1042	pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1043	BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1044	BLOCK_NUM_PKTS(pbd) += 1;
1045	read_lock(&pkc->blk_fill_in_prog_lock);
1046	prb_run_all_ft_ops(pkc, ppd);
1047}
1048
1049/* Assumes caller has the sk->rx_queue.lock */
1050static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1051					    struct sk_buff *skb,
1052					    unsigned int len
1053					    )
1054{
1055	struct tpacket_kbdq_core *pkc;
1056	struct tpacket_block_desc *pbd;
1057	char *curr, *end;
1058
1059	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1060	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1061
1062	/* Queue is frozen when user space is lagging behind */
1063	if (prb_queue_frozen(pkc)) {
1064		/*
1065		 * Check if that last block which caused the queue to freeze,
1066		 * is still in_use by user-space.
1067		 */
1068		if (prb_curr_blk_in_use(pbd)) {
1069			/* Can't record this packet */
1070			return NULL;
1071		} else {
1072			/*
1073			 * Ok, the block was released by user-space.
1074			 * Now let's open that block.
1075			 * opening a block also thaws the queue.
1076			 * Thawing is a side effect.
1077			 */
1078			prb_open_block(pkc, pbd);
1079		}
1080	}
1081
1082	smp_mb();
1083	curr = pkc->nxt_offset;
1084	pkc->skb = skb;
1085	end = (char *)pbd + pkc->kblk_size;
1086
1087	/* first try the current block */
1088	if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1089		prb_fill_curr_block(curr, pkc, pbd, len);
1090		return (void *)curr;
1091	}
1092
1093	/* Ok, close the current block */
1094	prb_retire_current_block(pkc, po, 0);
1095
1096	/* Now, try to dispatch the next block */
1097	curr = (char *)prb_dispatch_next_block(pkc, po);
1098	if (curr) {
1099		pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1100		prb_fill_curr_block(curr, pkc, pbd, len);
1101		return (void *)curr;
1102	}
1103
1104	/*
1105	 * No free blocks are available.user_space hasn't caught up yet.
1106	 * Queue was just frozen and now this packet will get dropped.
1107	 */
1108	return NULL;
1109}
1110
1111static void *packet_current_rx_frame(struct packet_sock *po,
1112					    struct sk_buff *skb,
1113					    int status, unsigned int len)
1114{
1115	char *curr = NULL;
1116	switch (po->tp_version) {
1117	case TPACKET_V1:
1118	case TPACKET_V2:
1119		curr = packet_lookup_frame(po, &po->rx_ring,
1120					po->rx_ring.head, status);
1121		return curr;
1122	case TPACKET_V3:
1123		return __packet_lookup_frame_in_block(po, skb, len);
1124	default:
1125		WARN(1, "TPACKET version not supported\n");
1126		BUG();
1127		return NULL;
1128	}
1129}
1130
1131static void *prb_lookup_block(const struct packet_sock *po,
1132			      const struct packet_ring_buffer *rb,
1133			      unsigned int idx,
1134			      int status)
1135{
1136	struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1137	struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1138
1139	if (status != BLOCK_STATUS(pbd))
1140		return NULL;
1141	return pbd;
1142}
1143
1144static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1145{
1146	unsigned int prev;
1147	if (rb->prb_bdqc.kactive_blk_num)
1148		prev = rb->prb_bdqc.kactive_blk_num-1;
1149	else
1150		prev = rb->prb_bdqc.knum_blocks-1;
1151	return prev;
1152}
1153
1154/* Assumes caller has held the rx_queue.lock */
1155static void *__prb_previous_block(struct packet_sock *po,
1156					 struct packet_ring_buffer *rb,
1157					 int status)
1158{
1159	unsigned int previous = prb_previous_blk_num(rb);
1160	return prb_lookup_block(po, rb, previous, status);
1161}
1162
1163static void *packet_previous_rx_frame(struct packet_sock *po,
1164					     struct packet_ring_buffer *rb,
1165					     int status)
1166{
1167	if (po->tp_version <= TPACKET_V2)
1168		return packet_previous_frame(po, rb, status);
1169
1170	return __prb_previous_block(po, rb, status);
1171}
1172
1173static void packet_increment_rx_head(struct packet_sock *po,
1174					    struct packet_ring_buffer *rb)
1175{
1176	switch (po->tp_version) {
1177	case TPACKET_V1:
1178	case TPACKET_V2:
1179		return packet_increment_head(rb);
1180	case TPACKET_V3:
1181	default:
1182		WARN(1, "TPACKET version not supported.\n");
1183		BUG();
1184		return;
1185	}
1186}
1187
1188static void *packet_previous_frame(struct packet_sock *po,
1189		struct packet_ring_buffer *rb,
1190		int status)
1191{
1192	unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1193	return packet_lookup_frame(po, rb, previous, status);
1194}
1195
1196static void packet_increment_head(struct packet_ring_buffer *buff)
1197{
1198	buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1199}
1200
1201static void packet_inc_pending(struct packet_ring_buffer *rb)
1202{
1203	this_cpu_inc(*rb->pending_refcnt);
1204}
1205
1206static void packet_dec_pending(struct packet_ring_buffer *rb)
1207{
1208	this_cpu_dec(*rb->pending_refcnt);
1209}
1210
1211static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1212{
1213	unsigned int refcnt = 0;
1214	int cpu;
1215
1216	/* We don't use pending refcount in rx_ring. */
1217	if (rb->pending_refcnt == NULL)
1218		return 0;
1219
1220	for_each_possible_cpu(cpu)
1221		refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1222
1223	return refcnt;
1224}
1225
1226static int packet_alloc_pending(struct packet_sock *po)
1227{
1228	po->rx_ring.pending_refcnt = NULL;
1229
1230	po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1231	if (unlikely(po->tx_ring.pending_refcnt == NULL))
1232		return -ENOBUFS;
1233
1234	return 0;
1235}
1236
1237static void packet_free_pending(struct packet_sock *po)
1238{
1239	free_percpu(po->tx_ring.pending_refcnt);
1240}
1241
1242#define ROOM_POW_OFF	2
1243#define ROOM_NONE	0x0
1244#define ROOM_LOW	0x1
1245#define ROOM_NORMAL	0x2
1246
1247static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1248{
1249	int idx, len;
1250
1251	len = READ_ONCE(po->rx_ring.frame_max) + 1;
1252	idx = READ_ONCE(po->rx_ring.head);
1253	if (pow_off)
1254		idx += len >> pow_off;
1255	if (idx >= len)
1256		idx -= len;
1257	return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1258}
1259
1260static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1261{
1262	int idx, len;
1263
1264	len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1265	idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1266	if (pow_off)
1267		idx += len >> pow_off;
1268	if (idx >= len)
1269		idx -= len;
1270	return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1271}
1272
1273static int __packet_rcv_has_room(const struct packet_sock *po,
1274				 const struct sk_buff *skb)
1275{
1276	const struct sock *sk = &po->sk;
1277	int ret = ROOM_NONE;
1278
1279	if (po->prot_hook.func != tpacket_rcv) {
1280		int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1281		int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1282				   - (skb ? skb->truesize : 0);
1283
1284		if (avail > (rcvbuf >> ROOM_POW_OFF))
1285			return ROOM_NORMAL;
1286		else if (avail > 0)
1287			return ROOM_LOW;
1288		else
1289			return ROOM_NONE;
1290	}
1291
1292	if (po->tp_version == TPACKET_V3) {
1293		if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1294			ret = ROOM_NORMAL;
1295		else if (__tpacket_v3_has_room(po, 0))
1296			ret = ROOM_LOW;
1297	} else {
1298		if (__tpacket_has_room(po, ROOM_POW_OFF))
1299			ret = ROOM_NORMAL;
1300		else if (__tpacket_has_room(po, 0))
1301			ret = ROOM_LOW;
1302	}
1303
1304	return ret;
1305}
1306
1307static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1308{
1309	int pressure, ret;
1310
1311	ret = __packet_rcv_has_room(po, skb);
1312	pressure = ret != ROOM_NORMAL;
1313
1314	if (READ_ONCE(po->pressure) != pressure)
1315		WRITE_ONCE(po->pressure, pressure);
1316
1317	return ret;
1318}
1319
1320static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1321{
1322	if (READ_ONCE(po->pressure) &&
1323	    __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1324		WRITE_ONCE(po->pressure,  0);
1325}
1326
1327static void packet_sock_destruct(struct sock *sk)
1328{
1329	skb_queue_purge(&sk->sk_error_queue);
1330
1331	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1332	WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1333
1334	if (!sock_flag(sk, SOCK_DEAD)) {
1335		pr_err("Attempt to release alive packet socket: %p\n", sk);
1336		return;
1337	}
1338
1339	sk_refcnt_debug_dec(sk);
1340}
1341
1342static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1343{
1344	u32 *history = po->rollover->history;
1345	u32 victim, rxhash;
1346	int i, count = 0;
1347
1348	rxhash = skb_get_hash(skb);
1349	for (i = 0; i < ROLLOVER_HLEN; i++)
1350		if (READ_ONCE(history[i]) == rxhash)
1351			count++;
1352
1353	victim = get_random_u32_below(ROLLOVER_HLEN);
1354
1355	/* Avoid dirtying the cache line if possible */
1356	if (READ_ONCE(history[victim]) != rxhash)
1357		WRITE_ONCE(history[victim], rxhash);
1358
1359	return count > (ROLLOVER_HLEN >> 1);
1360}
1361
1362static unsigned int fanout_demux_hash(struct packet_fanout *f,
1363				      struct sk_buff *skb,
1364				      unsigned int num)
1365{
1366	return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1367}
1368
1369static unsigned int fanout_demux_lb(struct packet_fanout *f,
1370				    struct sk_buff *skb,
1371				    unsigned int num)
1372{
1373	unsigned int val = atomic_inc_return(&f->rr_cur);
1374
1375	return val % num;
1376}
1377
1378static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1379				     struct sk_buff *skb,
1380				     unsigned int num)
1381{
1382	return smp_processor_id() % num;
1383}
1384
1385static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1386				     struct sk_buff *skb,
1387				     unsigned int num)
1388{
1389	return get_random_u32_below(num);
1390}
1391
1392static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1393					  struct sk_buff *skb,
1394					  unsigned int idx, bool try_self,
1395					  unsigned int num)
1396{
1397	struct packet_sock *po, *po_next, *po_skip = NULL;
1398	unsigned int i, j, room = ROOM_NONE;
1399
1400	po = pkt_sk(rcu_dereference(f->arr[idx]));
1401
1402	if (try_self) {
1403		room = packet_rcv_has_room(po, skb);
1404		if (room == ROOM_NORMAL ||
1405		    (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1406			return idx;
1407		po_skip = po;
1408	}
1409
1410	i = j = min_t(int, po->rollover->sock, num - 1);
1411	do {
1412		po_next = pkt_sk(rcu_dereference(f->arr[i]));
1413		if (po_next != po_skip && !READ_ONCE(po_next->pressure) &&
1414		    packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1415			if (i != j)
1416				po->rollover->sock = i;
1417			atomic_long_inc(&po->rollover->num);
1418			if (room == ROOM_LOW)
1419				atomic_long_inc(&po->rollover->num_huge);
1420			return i;
1421		}
1422
1423		if (++i == num)
1424			i = 0;
1425	} while (i != j);
1426
1427	atomic_long_inc(&po->rollover->num_failed);
1428	return idx;
1429}
1430
1431static unsigned int fanout_demux_qm(struct packet_fanout *f,
1432				    struct sk_buff *skb,
1433				    unsigned int num)
1434{
1435	return skb_get_queue_mapping(skb) % num;
1436}
1437
1438static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1439				     struct sk_buff *skb,
1440				     unsigned int num)
1441{
1442	struct bpf_prog *prog;
1443	unsigned int ret = 0;
1444
1445	rcu_read_lock();
1446	prog = rcu_dereference(f->bpf_prog);
1447	if (prog)
1448		ret = bpf_prog_run_clear_cb(prog, skb) % num;
1449	rcu_read_unlock();
1450
1451	return ret;
1452}
1453
1454static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1455{
1456	return f->flags & (flag >> 8);
1457}
1458
1459static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1460			     struct packet_type *pt, struct net_device *orig_dev)
1461{
1462	struct packet_fanout *f = pt->af_packet_priv;
1463	unsigned int num = READ_ONCE(f->num_members);
1464	struct net *net = read_pnet(&f->net);
1465	struct packet_sock *po;
1466	unsigned int idx;
1467
1468	if (!net_eq(dev_net(dev), net) || !num) {
1469		kfree_skb(skb);
1470		return 0;
1471	}
1472
1473	if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1474		skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1475		if (!skb)
1476			return 0;
1477	}
1478	switch (f->type) {
1479	case PACKET_FANOUT_HASH:
1480	default:
1481		idx = fanout_demux_hash(f, skb, num);
1482		break;
1483	case PACKET_FANOUT_LB:
1484		idx = fanout_demux_lb(f, skb, num);
1485		break;
1486	case PACKET_FANOUT_CPU:
1487		idx = fanout_demux_cpu(f, skb, num);
1488		break;
1489	case PACKET_FANOUT_RND:
1490		idx = fanout_demux_rnd(f, skb, num);
1491		break;
1492	case PACKET_FANOUT_QM:
1493		idx = fanout_demux_qm(f, skb, num);
1494		break;
1495	case PACKET_FANOUT_ROLLOVER:
1496		idx = fanout_demux_rollover(f, skb, 0, false, num);
1497		break;
1498	case PACKET_FANOUT_CBPF:
1499	case PACKET_FANOUT_EBPF:
1500		idx = fanout_demux_bpf(f, skb, num);
1501		break;
1502	}
1503
1504	if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1505		idx = fanout_demux_rollover(f, skb, idx, true, num);
1506
1507	po = pkt_sk(rcu_dereference(f->arr[idx]));
1508	return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1509}
1510
1511DEFINE_MUTEX(fanout_mutex);
1512EXPORT_SYMBOL_GPL(fanout_mutex);
1513static LIST_HEAD(fanout_list);
1514static u16 fanout_next_id;
1515
1516static void __fanout_link(struct sock *sk, struct packet_sock *po)
1517{
1518	struct packet_fanout *f = po->fanout;
1519
1520	spin_lock(&f->lock);
1521	rcu_assign_pointer(f->arr[f->num_members], sk);
1522	smp_wmb();
1523	f->num_members++;
1524	if (f->num_members == 1)
1525		dev_add_pack(&f->prot_hook);
1526	spin_unlock(&f->lock);
1527}
1528
1529static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1530{
1531	struct packet_fanout *f = po->fanout;
1532	int i;
1533
1534	spin_lock(&f->lock);
1535	for (i = 0; i < f->num_members; i++) {
1536		if (rcu_dereference_protected(f->arr[i],
1537					      lockdep_is_held(&f->lock)) == sk)
1538			break;
1539	}
1540	BUG_ON(i >= f->num_members);
1541	rcu_assign_pointer(f->arr[i],
1542			   rcu_dereference_protected(f->arr[f->num_members - 1],
1543						     lockdep_is_held(&f->lock)));
1544	f->num_members--;
1545	if (f->num_members == 0)
1546		__dev_remove_pack(&f->prot_hook);
1547	spin_unlock(&f->lock);
1548}
1549
1550static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1551{
1552	if (sk->sk_family != PF_PACKET)
1553		return false;
1554
1555	return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1556}
1557
1558static void fanout_init_data(struct packet_fanout *f)
1559{
1560	switch (f->type) {
1561	case PACKET_FANOUT_LB:
1562		atomic_set(&f->rr_cur, 0);
1563		break;
1564	case PACKET_FANOUT_CBPF:
1565	case PACKET_FANOUT_EBPF:
1566		RCU_INIT_POINTER(f->bpf_prog, NULL);
1567		break;
1568	}
1569}
1570
1571static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1572{
1573	struct bpf_prog *old;
1574
1575	spin_lock(&f->lock);
1576	old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1577	rcu_assign_pointer(f->bpf_prog, new);
1578	spin_unlock(&f->lock);
1579
1580	if (old) {
1581		synchronize_net();
1582		bpf_prog_destroy(old);
1583	}
1584}
1585
1586static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data,
1587				unsigned int len)
1588{
1589	struct bpf_prog *new;
1590	struct sock_fprog fprog;
1591	int ret;
1592
1593	if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1594		return -EPERM;
1595
1596	ret = copy_bpf_fprog_from_user(&fprog, data, len);
1597	if (ret)
1598		return ret;
1599
1600	ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1601	if (ret)
1602		return ret;
1603
1604	__fanout_set_data_bpf(po->fanout, new);
1605	return 0;
1606}
1607
1608static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data,
1609				unsigned int len)
1610{
1611	struct bpf_prog *new;
1612	u32 fd;
1613
1614	if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1615		return -EPERM;
1616	if (len != sizeof(fd))
1617		return -EINVAL;
1618	if (copy_from_sockptr(&fd, data, len))
1619		return -EFAULT;
1620
1621	new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1622	if (IS_ERR(new))
1623		return PTR_ERR(new);
1624
1625	__fanout_set_data_bpf(po->fanout, new);
1626	return 0;
1627}
1628
1629static int fanout_set_data(struct packet_sock *po, sockptr_t data,
1630			   unsigned int len)
1631{
1632	switch (po->fanout->type) {
1633	case PACKET_FANOUT_CBPF:
1634		return fanout_set_data_cbpf(po, data, len);
1635	case PACKET_FANOUT_EBPF:
1636		return fanout_set_data_ebpf(po, data, len);
1637	default:
1638		return -EINVAL;
1639	}
1640}
1641
1642static void fanout_release_data(struct packet_fanout *f)
1643{
1644	switch (f->type) {
1645	case PACKET_FANOUT_CBPF:
1646	case PACKET_FANOUT_EBPF:
1647		__fanout_set_data_bpf(f, NULL);
1648	}
1649}
1650
1651static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1652{
1653	struct packet_fanout *f;
1654
1655	list_for_each_entry(f, &fanout_list, list) {
1656		if (f->id == candidate_id &&
1657		    read_pnet(&f->net) == sock_net(sk)) {
1658			return false;
1659		}
1660	}
1661	return true;
1662}
1663
1664static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1665{
1666	u16 id = fanout_next_id;
1667
1668	do {
1669		if (__fanout_id_is_free(sk, id)) {
1670			*new_id = id;
1671			fanout_next_id = id + 1;
1672			return true;
1673		}
1674
1675		id++;
1676	} while (id != fanout_next_id);
1677
1678	return false;
1679}
1680
1681static int fanout_add(struct sock *sk, struct fanout_args *args)
1682{
1683	struct packet_rollover *rollover = NULL;
1684	struct packet_sock *po = pkt_sk(sk);
1685	u16 type_flags = args->type_flags;
1686	struct packet_fanout *f, *match;
1687	u8 type = type_flags & 0xff;
1688	u8 flags = type_flags >> 8;
1689	u16 id = args->id;
1690	int err;
1691
1692	switch (type) {
1693	case PACKET_FANOUT_ROLLOVER:
1694		if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1695			return -EINVAL;
1696		break;
1697	case PACKET_FANOUT_HASH:
1698	case PACKET_FANOUT_LB:
1699	case PACKET_FANOUT_CPU:
1700	case PACKET_FANOUT_RND:
1701	case PACKET_FANOUT_QM:
1702	case PACKET_FANOUT_CBPF:
1703	case PACKET_FANOUT_EBPF:
1704		break;
1705	default:
1706		return -EINVAL;
1707	}
1708
1709	mutex_lock(&fanout_mutex);
1710
1711	err = -EALREADY;
1712	if (po->fanout)
1713		goto out;
1714
1715	if (type == PACKET_FANOUT_ROLLOVER ||
1716	    (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1717		err = -ENOMEM;
1718		rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1719		if (!rollover)
1720			goto out;
1721		atomic_long_set(&rollover->num, 0);
1722		atomic_long_set(&rollover->num_huge, 0);
1723		atomic_long_set(&rollover->num_failed, 0);
1724	}
1725
1726	if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1727		if (id != 0) {
1728			err = -EINVAL;
1729			goto out;
1730		}
1731		if (!fanout_find_new_id(sk, &id)) {
1732			err = -ENOMEM;
1733			goto out;
1734		}
1735		/* ephemeral flag for the first socket in the group: drop it */
1736		flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1737	}
1738
1739	match = NULL;
1740	list_for_each_entry(f, &fanout_list, list) {
1741		if (f->id == id &&
1742		    read_pnet(&f->net) == sock_net(sk)) {
1743			match = f;
1744			break;
1745		}
1746	}
1747	err = -EINVAL;
1748	if (match) {
1749		if (match->flags != flags)
1750			goto out;
1751		if (args->max_num_members &&
1752		    args->max_num_members != match->max_num_members)
1753			goto out;
1754	} else {
1755		if (args->max_num_members > PACKET_FANOUT_MAX)
1756			goto out;
1757		if (!args->max_num_members)
1758			/* legacy PACKET_FANOUT_MAX */
1759			args->max_num_members = 256;
1760		err = -ENOMEM;
1761		match = kvzalloc(struct_size(match, arr, args->max_num_members),
1762				 GFP_KERNEL);
1763		if (!match)
1764			goto out;
1765		write_pnet(&match->net, sock_net(sk));
1766		match->id = id;
1767		match->type = type;
1768		match->flags = flags;
1769		INIT_LIST_HEAD(&match->list);
1770		spin_lock_init(&match->lock);
1771		refcount_set(&match->sk_ref, 0);
1772		fanout_init_data(match);
1773		match->prot_hook.type = po->prot_hook.type;
1774		match->prot_hook.dev = po->prot_hook.dev;
1775		match->prot_hook.func = packet_rcv_fanout;
1776		match->prot_hook.af_packet_priv = match;
1777		match->prot_hook.af_packet_net = read_pnet(&match->net);
1778		match->prot_hook.id_match = match_fanout_group;
1779		match->max_num_members = args->max_num_members;
1780		match->prot_hook.ignore_outgoing = type_flags & PACKET_FANOUT_FLAG_IGNORE_OUTGOING;
1781		list_add(&match->list, &fanout_list);
1782	}
1783	err = -EINVAL;
1784
1785	spin_lock(&po->bind_lock);
1786	if (po->running &&
1787	    match->type == type &&
1788	    match->prot_hook.type == po->prot_hook.type &&
1789	    match->prot_hook.dev == po->prot_hook.dev) {
1790		err = -ENOSPC;
1791		if (refcount_read(&match->sk_ref) < match->max_num_members) {
1792			__dev_remove_pack(&po->prot_hook);
1793
1794			/* Paired with packet_setsockopt(PACKET_FANOUT_DATA) */
1795			WRITE_ONCE(po->fanout, match);
1796
1797			po->rollover = rollover;
1798			rollover = NULL;
1799			refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1800			__fanout_link(sk, po);
1801			err = 0;
1802		}
1803	}
1804	spin_unlock(&po->bind_lock);
1805
1806	if (err && !refcount_read(&match->sk_ref)) {
1807		list_del(&match->list);
1808		kvfree(match);
1809	}
1810
1811out:
1812	kfree(rollover);
1813	mutex_unlock(&fanout_mutex);
1814	return err;
1815}
1816
1817/* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1818 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1819 * It is the responsibility of the caller to call fanout_release_data() and
1820 * free the returned packet_fanout (after synchronize_net())
1821 */
1822static struct packet_fanout *fanout_release(struct sock *sk)
1823{
1824	struct packet_sock *po = pkt_sk(sk);
1825	struct packet_fanout *f;
1826
1827	mutex_lock(&fanout_mutex);
1828	f = po->fanout;
1829	if (f) {
1830		po->fanout = NULL;
1831
1832		if (refcount_dec_and_test(&f->sk_ref))
1833			list_del(&f->list);
1834		else
1835			f = NULL;
1836	}
1837	mutex_unlock(&fanout_mutex);
1838
1839	return f;
1840}
1841
1842static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1843					  struct sk_buff *skb)
1844{
1845	/* Earlier code assumed this would be a VLAN pkt, double-check
1846	 * this now that we have the actual packet in hand. We can only
1847	 * do this check on Ethernet devices.
1848	 */
1849	if (unlikely(dev->type != ARPHRD_ETHER))
1850		return false;
1851
1852	skb_reset_mac_header(skb);
1853	return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1854}
1855
1856static const struct proto_ops packet_ops;
1857
1858static const struct proto_ops packet_ops_spkt;
1859
1860static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1861			   struct packet_type *pt, struct net_device *orig_dev)
1862{
1863	struct sock *sk;
1864	struct sockaddr_pkt *spkt;
1865
1866	/*
1867	 *	When we registered the protocol we saved the socket in the data
1868	 *	field for just this event.
1869	 */
1870
1871	sk = pt->af_packet_priv;
1872
1873	/*
1874	 *	Yank back the headers [hope the device set this
1875	 *	right or kerboom...]
1876	 *
1877	 *	Incoming packets have ll header pulled,
1878	 *	push it back.
1879	 *
1880	 *	For outgoing ones skb->data == skb_mac_header(skb)
1881	 *	so that this procedure is noop.
1882	 */
1883
1884	if (skb->pkt_type == PACKET_LOOPBACK)
1885		goto out;
1886
1887	if (!net_eq(dev_net(dev), sock_net(sk)))
1888		goto out;
1889
1890	skb = skb_share_check(skb, GFP_ATOMIC);
1891	if (skb == NULL)
1892		goto oom;
1893
1894	/* drop any routing info */
1895	skb_dst_drop(skb);
1896
1897	/* drop conntrack reference */
1898	nf_reset_ct(skb);
1899
1900	spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1901
1902	skb_push(skb, skb->data - skb_mac_header(skb));
1903
1904	/*
1905	 *	The SOCK_PACKET socket receives _all_ frames.
1906	 */
1907
1908	spkt->spkt_family = dev->type;
1909	strscpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1910	spkt->spkt_protocol = skb->protocol;
1911
1912	/*
1913	 *	Charge the memory to the socket. This is done specifically
1914	 *	to prevent sockets using all the memory up.
1915	 */
1916
1917	if (sock_queue_rcv_skb(sk, skb) == 0)
1918		return 0;
1919
1920out:
1921	kfree_skb(skb);
1922oom:
1923	return 0;
1924}
1925
1926static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1927{
1928	int depth;
1929
1930	if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1931	    sock->type == SOCK_RAW) {
1932		skb_reset_mac_header(skb);
1933		skb->protocol = dev_parse_header_protocol(skb);
1934	}
1935
1936	/* Move network header to the right position for VLAN tagged packets */
1937	if (likely(skb->dev->type == ARPHRD_ETHER) &&
1938	    eth_type_vlan(skb->protocol) &&
1939	    __vlan_get_protocol(skb, skb->protocol, &depth) != 0) {
1940		if (pskb_may_pull(skb, depth))
1941			skb_set_network_header(skb, depth);
1942	}
1943
1944	skb_probe_transport_header(skb);
1945}
1946
1947/*
1948 *	Output a raw packet to a device layer. This bypasses all the other
1949 *	protocol layers and you must therefore supply it with a complete frame
1950 */
1951
1952static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1953			       size_t len)
1954{
1955	struct sock *sk = sock->sk;
1956	DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1957	struct sk_buff *skb = NULL;
1958	struct net_device *dev;
1959	struct sockcm_cookie sockc;
1960	__be16 proto = 0;
1961	int err;
1962	int extra_len = 0;
1963
1964	/*
1965	 *	Get and verify the address.
1966	 */
1967
1968	if (saddr) {
1969		if (msg->msg_namelen < sizeof(struct sockaddr))
1970			return -EINVAL;
1971		if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1972			proto = saddr->spkt_protocol;
1973	} else
1974		return -ENOTCONN;	/* SOCK_PACKET must be sent giving an address */
1975
1976	/*
1977	 *	Find the device first to size check it
1978	 */
1979
1980	saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1981retry:
1982	rcu_read_lock();
1983	dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1984	err = -ENODEV;
1985	if (dev == NULL)
1986		goto out_unlock;
1987
1988	err = -ENETDOWN;
1989	if (!(dev->flags & IFF_UP))
1990		goto out_unlock;
1991
1992	/*
1993	 * You may not queue a frame bigger than the mtu. This is the lowest level
1994	 * raw protocol and you must do your own fragmentation at this level.
1995	 */
1996
1997	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1998		if (!netif_supports_nofcs(dev)) {
1999			err = -EPROTONOSUPPORT;
2000			goto out_unlock;
2001		}
2002		extra_len = 4; /* We're doing our own CRC */
2003	}
2004
2005	err = -EMSGSIZE;
2006	if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
2007		goto out_unlock;
2008
2009	if (!skb) {
2010		size_t reserved = LL_RESERVED_SPACE(dev);
2011		int tlen = dev->needed_tailroom;
2012		unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
2013
2014		rcu_read_unlock();
2015		skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
2016		if (skb == NULL)
2017			return -ENOBUFS;
2018		/* FIXME: Save some space for broken drivers that write a hard
2019		 * header at transmission time by themselves. PPP is the notable
2020		 * one here. This should really be fixed at the driver level.
2021		 */
2022		skb_reserve(skb, reserved);
2023		skb_reset_network_header(skb);
2024
2025		/* Try to align data part correctly */
2026		if (hhlen) {
2027			skb->data -= hhlen;
2028			skb->tail -= hhlen;
2029			if (len < hhlen)
2030				skb_reset_network_header(skb);
2031		}
2032		err = memcpy_from_msg(skb_put(skb, len), msg, len);
2033		if (err)
2034			goto out_free;
2035		goto retry;
2036	}
2037
2038	if (!dev_validate_header(dev, skb->data, len)) {
2039		err = -EINVAL;
2040		goto out_unlock;
2041	}
2042	if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
2043	    !packet_extra_vlan_len_allowed(dev, skb)) {
2044		err = -EMSGSIZE;
2045		goto out_unlock;
2046	}
2047
2048	sockcm_init(&sockc, sk);
2049	if (msg->msg_controllen) {
2050		err = sock_cmsg_send(sk, msg, &sockc);
2051		if (unlikely(err))
2052			goto out_unlock;
2053	}
2054
2055	skb->protocol = proto;
2056	skb->dev = dev;
2057	skb->priority = sk->sk_priority;
2058	skb->mark = sk->sk_mark;
2059	skb->tstamp = sockc.transmit_time;
2060
2061	skb_setup_tx_timestamp(skb, sockc.tsflags);
2062
2063	if (unlikely(extra_len == 4))
2064		skb->no_fcs = 1;
2065
2066	packet_parse_headers(skb, sock);
2067
2068	dev_queue_xmit(skb);
2069	rcu_read_unlock();
2070	return len;
2071
2072out_unlock:
2073	rcu_read_unlock();
2074out_free:
2075	kfree_skb(skb);
2076	return err;
2077}
2078
2079static unsigned int run_filter(struct sk_buff *skb,
2080			       const struct sock *sk,
2081			       unsigned int res)
2082{
2083	struct sk_filter *filter;
2084
2085	rcu_read_lock();
2086	filter = rcu_dereference(sk->sk_filter);
2087	if (filter != NULL)
2088		res = bpf_prog_run_clear_cb(filter->prog, skb);
2089	rcu_read_unlock();
2090
2091	return res;
2092}
2093
2094static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2095			   size_t *len)
2096{
2097	struct virtio_net_hdr vnet_hdr;
2098
2099	if (*len < sizeof(vnet_hdr))
2100		return -EINVAL;
2101	*len -= sizeof(vnet_hdr);
2102
2103	if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2104		return -EINVAL;
2105
2106	return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2107}
2108
2109/*
2110 * This function makes lazy skb cloning in hope that most of packets
2111 * are discarded by BPF.
2112 *
2113 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2114 * and skb->cb are mangled. It works because (and until) packets
2115 * falling here are owned by current CPU. Output packets are cloned
2116 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2117 * sequentially, so that if we return skb to original state on exit,
2118 * we will not harm anyone.
2119 */
2120
2121static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2122		      struct packet_type *pt, struct net_device *orig_dev)
2123{
2124	struct sock *sk;
2125	struct sockaddr_ll *sll;
2126	struct packet_sock *po;
2127	u8 *skb_head = skb->data;
2128	int skb_len = skb->len;
2129	unsigned int snaplen, res;
2130	bool is_drop_n_account = false;
2131
2132	if (skb->pkt_type == PACKET_LOOPBACK)
2133		goto drop;
2134
2135	sk = pt->af_packet_priv;
2136	po = pkt_sk(sk);
2137
2138	if (!net_eq(dev_net(dev), sock_net(sk)))
2139		goto drop;
2140
2141	skb->dev = dev;
2142
2143	if (dev_has_header(dev)) {
2144		/* The device has an explicit notion of ll header,
2145		 * exported to higher levels.
2146		 *
2147		 * Otherwise, the device hides details of its frame
2148		 * structure, so that corresponding packet head is
2149		 * never delivered to user.
2150		 */
2151		if (sk->sk_type != SOCK_DGRAM)
2152			skb_push(skb, skb->data - skb_mac_header(skb));
2153		else if (skb->pkt_type == PACKET_OUTGOING) {
2154			/* Special case: outgoing packets have ll header at head */
2155			skb_pull(skb, skb_network_offset(skb));
2156		}
2157	}
2158
2159	snaplen = skb->len;
2160
2161	res = run_filter(skb, sk, snaplen);
2162	if (!res)
2163		goto drop_n_restore;
2164	if (snaplen > res)
2165		snaplen = res;
2166
2167	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2168		goto drop_n_acct;
2169
2170	if (skb_shared(skb)) {
2171		struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2172		if (nskb == NULL)
2173			goto drop_n_acct;
2174
2175		if (skb_head != skb->data) {
2176			skb->data = skb_head;
2177			skb->len = skb_len;
2178		}
2179		consume_skb(skb);
2180		skb = nskb;
2181	}
2182
2183	sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2184
2185	sll = &PACKET_SKB_CB(skb)->sa.ll;
2186	sll->sll_hatype = dev->type;
2187	sll->sll_pkttype = skb->pkt_type;
2188	if (unlikely(po->origdev))
2189		sll->sll_ifindex = orig_dev->ifindex;
2190	else
2191		sll->sll_ifindex = dev->ifindex;
2192
2193	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2194
2195	/* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2196	 * Use their space for storing the original skb length.
2197	 */
2198	PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2199
2200	if (pskb_trim(skb, snaplen))
2201		goto drop_n_acct;
2202
2203	skb_set_owner_r(skb, sk);
2204	skb->dev = NULL;
2205	skb_dst_drop(skb);
2206
2207	/* drop conntrack reference */
2208	nf_reset_ct(skb);
2209
2210	spin_lock(&sk->sk_receive_queue.lock);
2211	po->stats.stats1.tp_packets++;
2212	sock_skb_set_dropcount(sk, skb);
2213	skb_clear_delivery_time(skb);
2214	__skb_queue_tail(&sk->sk_receive_queue, skb);
2215	spin_unlock(&sk->sk_receive_queue.lock);
2216	sk->sk_data_ready(sk);
2217	return 0;
2218
2219drop_n_acct:
2220	is_drop_n_account = true;
2221	atomic_inc(&po->tp_drops);
2222	atomic_inc(&sk->sk_drops);
2223
2224drop_n_restore:
2225	if (skb_head != skb->data && skb_shared(skb)) {
2226		skb->data = skb_head;
2227		skb->len = skb_len;
2228	}
2229drop:
2230	if (!is_drop_n_account)
2231		consume_skb(skb);
2232	else
2233		kfree_skb(skb);
2234	return 0;
2235}
2236
2237static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2238		       struct packet_type *pt, struct net_device *orig_dev)
2239{
2240	struct sock *sk;
2241	struct packet_sock *po;
2242	struct sockaddr_ll *sll;
2243	union tpacket_uhdr h;
2244	u8 *skb_head = skb->data;
2245	int skb_len = skb->len;
2246	unsigned int snaplen, res;
2247	unsigned long status = TP_STATUS_USER;
2248	unsigned short macoff, hdrlen;
2249	unsigned int netoff;
2250	struct sk_buff *copy_skb = NULL;
2251	struct timespec64 ts;
2252	__u32 ts_status;
2253	bool is_drop_n_account = false;
2254	unsigned int slot_id = 0;
2255	bool do_vnet = false;
2256
2257	/* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2258	 * We may add members to them until current aligned size without forcing
2259	 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2260	 */
2261	BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2262	BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2263
2264	if (skb->pkt_type == PACKET_LOOPBACK)
2265		goto drop;
2266
2267	sk = pt->af_packet_priv;
2268	po = pkt_sk(sk);
2269
2270	if (!net_eq(dev_net(dev), sock_net(sk)))
2271		goto drop;
2272
2273	if (dev_has_header(dev)) {
2274		if (sk->sk_type != SOCK_DGRAM)
2275			skb_push(skb, skb->data - skb_mac_header(skb));
2276		else if (skb->pkt_type == PACKET_OUTGOING) {
2277			/* Special case: outgoing packets have ll header at head */
2278			skb_pull(skb, skb_network_offset(skb));
2279		}
2280	}
2281
2282	snaplen = skb->len;
2283
2284	res = run_filter(skb, sk, snaplen);
2285	if (!res)
2286		goto drop_n_restore;
2287
2288	/* If we are flooded, just give up */
2289	if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2290		atomic_inc(&po->tp_drops);
2291		goto drop_n_restore;
2292	}
2293
2294	if (skb->ip_summed == CHECKSUM_PARTIAL)
2295		status |= TP_STATUS_CSUMNOTREADY;
2296	else if (skb->pkt_type != PACKET_OUTGOING &&
2297		 skb_csum_unnecessary(skb))
 
2298		status |= TP_STATUS_CSUM_VALID;
2299
2300	if (snaplen > res)
2301		snaplen = res;
2302
2303	if (sk->sk_type == SOCK_DGRAM) {
2304		macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2305				  po->tp_reserve;
2306	} else {
2307		unsigned int maclen = skb_network_offset(skb);
2308		netoff = TPACKET_ALIGN(po->tp_hdrlen +
2309				       (maclen < 16 ? 16 : maclen)) +
2310				       po->tp_reserve;
2311		if (po->has_vnet_hdr) {
2312			netoff += sizeof(struct virtio_net_hdr);
2313			do_vnet = true;
2314		}
2315		macoff = netoff - maclen;
2316	}
2317	if (netoff > USHRT_MAX) {
2318		atomic_inc(&po->tp_drops);
2319		goto drop_n_restore;
2320	}
2321	if (po->tp_version <= TPACKET_V2) {
2322		if (macoff + snaplen > po->rx_ring.frame_size) {
2323			if (po->copy_thresh &&
2324			    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2325				if (skb_shared(skb)) {
2326					copy_skb = skb_clone(skb, GFP_ATOMIC);
2327				} else {
2328					copy_skb = skb_get(skb);
2329					skb_head = skb->data;
2330				}
2331				if (copy_skb) {
2332					memset(&PACKET_SKB_CB(copy_skb)->sa.ll, 0,
2333					       sizeof(PACKET_SKB_CB(copy_skb)->sa.ll));
2334					skb_set_owner_r(copy_skb, sk);
2335				}
2336			}
2337			snaplen = po->rx_ring.frame_size - macoff;
2338			if ((int)snaplen < 0) {
2339				snaplen = 0;
2340				do_vnet = false;
2341			}
2342		}
2343	} else if (unlikely(macoff + snaplen >
2344			    GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2345		u32 nval;
2346
2347		nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2348		pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2349			    snaplen, nval, macoff);
2350		snaplen = nval;
2351		if (unlikely((int)snaplen < 0)) {
2352			snaplen = 0;
2353			macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2354			do_vnet = false;
2355		}
2356	}
2357	spin_lock(&sk->sk_receive_queue.lock);
2358	h.raw = packet_current_rx_frame(po, skb,
2359					TP_STATUS_KERNEL, (macoff+snaplen));
2360	if (!h.raw)
2361		goto drop_n_account;
2362
2363	if (po->tp_version <= TPACKET_V2) {
2364		slot_id = po->rx_ring.head;
2365		if (test_bit(slot_id, po->rx_ring.rx_owner_map))
2366			goto drop_n_account;
2367		__set_bit(slot_id, po->rx_ring.rx_owner_map);
2368	}
2369
2370	if (do_vnet &&
2371	    virtio_net_hdr_from_skb(skb, h.raw + macoff -
2372				    sizeof(struct virtio_net_hdr),
2373				    vio_le(), true, 0)) {
2374		if (po->tp_version == TPACKET_V3)
2375			prb_clear_blk_fill_status(&po->rx_ring);
2376		goto drop_n_account;
2377	}
2378
2379	if (po->tp_version <= TPACKET_V2) {
2380		packet_increment_rx_head(po, &po->rx_ring);
2381	/*
2382	 * LOSING will be reported till you read the stats,
2383	 * because it's COR - Clear On Read.
2384	 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2385	 * at packet level.
2386	 */
2387		if (atomic_read(&po->tp_drops))
2388			status |= TP_STATUS_LOSING;
2389	}
2390
2391	po->stats.stats1.tp_packets++;
2392	if (copy_skb) {
2393		status |= TP_STATUS_COPY;
2394		skb_clear_delivery_time(copy_skb);
2395		__skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2396	}
2397	spin_unlock(&sk->sk_receive_queue.lock);
2398
2399	skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2400
2401	/* Always timestamp; prefer an existing software timestamp taken
2402	 * closer to the time of capture.
2403	 */
2404	ts_status = tpacket_get_timestamp(skb, &ts,
2405					  po->tp_tstamp | SOF_TIMESTAMPING_SOFTWARE);
2406	if (!ts_status)
2407		ktime_get_real_ts64(&ts);
2408
2409	status |= ts_status;
2410
2411	switch (po->tp_version) {
2412	case TPACKET_V1:
2413		h.h1->tp_len = skb->len;
2414		h.h1->tp_snaplen = snaplen;
2415		h.h1->tp_mac = macoff;
2416		h.h1->tp_net = netoff;
2417		h.h1->tp_sec = ts.tv_sec;
2418		h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2419		hdrlen = sizeof(*h.h1);
2420		break;
2421	case TPACKET_V2:
2422		h.h2->tp_len = skb->len;
2423		h.h2->tp_snaplen = snaplen;
2424		h.h2->tp_mac = macoff;
2425		h.h2->tp_net = netoff;
2426		h.h2->tp_sec = ts.tv_sec;
2427		h.h2->tp_nsec = ts.tv_nsec;
2428		if (skb_vlan_tag_present(skb)) {
2429			h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2430			h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2431			status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2432		} else {
2433			h.h2->tp_vlan_tci = 0;
2434			h.h2->tp_vlan_tpid = 0;
2435		}
2436		memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2437		hdrlen = sizeof(*h.h2);
2438		break;
2439	case TPACKET_V3:
2440		/* tp_nxt_offset,vlan are already populated above.
2441		 * So DONT clear those fields here
2442		 */
2443		h.h3->tp_status |= status;
2444		h.h3->tp_len = skb->len;
2445		h.h3->tp_snaplen = snaplen;
2446		h.h3->tp_mac = macoff;
2447		h.h3->tp_net = netoff;
2448		h.h3->tp_sec  = ts.tv_sec;
2449		h.h3->tp_nsec = ts.tv_nsec;
2450		memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2451		hdrlen = sizeof(*h.h3);
2452		break;
2453	default:
2454		BUG();
2455	}
2456
2457	sll = h.raw + TPACKET_ALIGN(hdrlen);
2458	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2459	sll->sll_family = AF_PACKET;
2460	sll->sll_hatype = dev->type;
2461	sll->sll_protocol = skb->protocol;
2462	sll->sll_pkttype = skb->pkt_type;
2463	if (unlikely(po->origdev))
2464		sll->sll_ifindex = orig_dev->ifindex;
2465	else
2466		sll->sll_ifindex = dev->ifindex;
2467
2468	smp_mb();
2469
2470#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2471	if (po->tp_version <= TPACKET_V2) {
2472		u8 *start, *end;
2473
2474		end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2475					macoff + snaplen);
2476
2477		for (start = h.raw; start < end; start += PAGE_SIZE)
2478			flush_dcache_page(pgv_to_page(start));
2479	}
2480	smp_wmb();
2481#endif
2482
2483	if (po->tp_version <= TPACKET_V2) {
2484		spin_lock(&sk->sk_receive_queue.lock);
2485		__packet_set_status(po, h.raw, status);
2486		__clear_bit(slot_id, po->rx_ring.rx_owner_map);
2487		spin_unlock(&sk->sk_receive_queue.lock);
2488		sk->sk_data_ready(sk);
2489	} else if (po->tp_version == TPACKET_V3) {
2490		prb_clear_blk_fill_status(&po->rx_ring);
2491	}
2492
2493drop_n_restore:
2494	if (skb_head != skb->data && skb_shared(skb)) {
2495		skb->data = skb_head;
2496		skb->len = skb_len;
2497	}
2498drop:
2499	if (!is_drop_n_account)
2500		consume_skb(skb);
2501	else
2502		kfree_skb(skb);
2503	return 0;
2504
2505drop_n_account:
2506	spin_unlock(&sk->sk_receive_queue.lock);
2507	atomic_inc(&po->tp_drops);
2508	is_drop_n_account = true;
2509
2510	sk->sk_data_ready(sk);
2511	kfree_skb(copy_skb);
2512	goto drop_n_restore;
2513}
2514
2515static void tpacket_destruct_skb(struct sk_buff *skb)
2516{
2517	struct packet_sock *po = pkt_sk(skb->sk);
2518
2519	if (likely(po->tx_ring.pg_vec)) {
2520		void *ph;
2521		__u32 ts;
2522
2523		ph = skb_zcopy_get_nouarg(skb);
2524		packet_dec_pending(&po->tx_ring);
2525
2526		ts = __packet_set_timestamp(po, ph, skb);
2527		__packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2528
2529		if (!packet_read_pending(&po->tx_ring))
2530			complete(&po->skb_completion);
2531	}
2532
2533	sock_wfree(skb);
2534}
2535
2536static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2537{
2538	if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2539	    (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2540	     __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2541	      __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2542		vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2543			 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2544			__virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2545
2546	if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2547		return -EINVAL;
2548
2549	return 0;
2550}
2551
2552static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2553				 struct virtio_net_hdr *vnet_hdr)
2554{
2555	if (*len < sizeof(*vnet_hdr))
2556		return -EINVAL;
2557	*len -= sizeof(*vnet_hdr);
2558
2559	if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2560		return -EFAULT;
2561
2562	return __packet_snd_vnet_parse(vnet_hdr, *len);
2563}
2564
2565static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2566		void *frame, struct net_device *dev, void *data, int tp_len,
2567		__be16 proto, unsigned char *addr, int hlen, int copylen,
2568		const struct sockcm_cookie *sockc)
2569{
2570	union tpacket_uhdr ph;
2571	int to_write, offset, len, nr_frags, len_max;
2572	struct socket *sock = po->sk.sk_socket;
2573	struct page *page;
2574	int err;
2575
2576	ph.raw = frame;
2577
2578	skb->protocol = proto;
2579	skb->dev = dev;
2580	skb->priority = po->sk.sk_priority;
2581	skb->mark = po->sk.sk_mark;
2582	skb->tstamp = sockc->transmit_time;
2583	skb_setup_tx_timestamp(skb, sockc->tsflags);
2584	skb_zcopy_set_nouarg(skb, ph.raw);
2585
2586	skb_reserve(skb, hlen);
2587	skb_reset_network_header(skb);
2588
2589	to_write = tp_len;
2590
2591	if (sock->type == SOCK_DGRAM) {
2592		err = dev_hard_header(skb, dev, ntohs(proto), addr,
2593				NULL, tp_len);
2594		if (unlikely(err < 0))
2595			return -EINVAL;
2596	} else if (copylen) {
2597		int hdrlen = min_t(int, copylen, tp_len);
2598
2599		skb_push(skb, dev->hard_header_len);
2600		skb_put(skb, copylen - dev->hard_header_len);
2601		err = skb_store_bits(skb, 0, data, hdrlen);
2602		if (unlikely(err))
2603			return err;
2604		if (!dev_validate_header(dev, skb->data, hdrlen))
2605			return -EINVAL;
2606
2607		data += hdrlen;
2608		to_write -= hdrlen;
2609	}
2610
2611	offset = offset_in_page(data);
2612	len_max = PAGE_SIZE - offset;
2613	len = ((to_write > len_max) ? len_max : to_write);
2614
2615	skb->data_len = to_write;
2616	skb->len += to_write;
2617	skb->truesize += to_write;
2618	refcount_add(to_write, &po->sk.sk_wmem_alloc);
2619
2620	while (likely(to_write)) {
2621		nr_frags = skb_shinfo(skb)->nr_frags;
2622
2623		if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2624			pr_err("Packet exceed the number of skb frags(%lu)\n",
2625			       MAX_SKB_FRAGS);
2626			return -EFAULT;
2627		}
2628
2629		page = pgv_to_page(data);
2630		data += len;
2631		flush_dcache_page(page);
2632		get_page(page);
2633		skb_fill_page_desc(skb, nr_frags, page, offset, len);
2634		to_write -= len;
2635		offset = 0;
2636		len_max = PAGE_SIZE;
2637		len = ((to_write > len_max) ? len_max : to_write);
2638	}
2639
2640	packet_parse_headers(skb, sock);
2641
2642	return tp_len;
2643}
2644
2645static int tpacket_parse_header(struct packet_sock *po, void *frame,
2646				int size_max, void **data)
2647{
2648	union tpacket_uhdr ph;
2649	int tp_len, off;
2650
2651	ph.raw = frame;
2652
2653	switch (po->tp_version) {
2654	case TPACKET_V3:
2655		if (ph.h3->tp_next_offset != 0) {
2656			pr_warn_once("variable sized slot not supported");
2657			return -EINVAL;
2658		}
2659		tp_len = ph.h3->tp_len;
2660		break;
2661	case TPACKET_V2:
2662		tp_len = ph.h2->tp_len;
2663		break;
2664	default:
2665		tp_len = ph.h1->tp_len;
2666		break;
2667	}
2668	if (unlikely(tp_len > size_max)) {
2669		pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2670		return -EMSGSIZE;
2671	}
2672
2673	if (unlikely(po->tp_tx_has_off)) {
2674		int off_min, off_max;
2675
2676		off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2677		off_max = po->tx_ring.frame_size - tp_len;
2678		if (po->sk.sk_type == SOCK_DGRAM) {
2679			switch (po->tp_version) {
2680			case TPACKET_V3:
2681				off = ph.h3->tp_net;
2682				break;
2683			case TPACKET_V2:
2684				off = ph.h2->tp_net;
2685				break;
2686			default:
2687				off = ph.h1->tp_net;
2688				break;
2689			}
2690		} else {
2691			switch (po->tp_version) {
2692			case TPACKET_V3:
2693				off = ph.h3->tp_mac;
2694				break;
2695			case TPACKET_V2:
2696				off = ph.h2->tp_mac;
2697				break;
2698			default:
2699				off = ph.h1->tp_mac;
2700				break;
2701			}
2702		}
2703		if (unlikely((off < off_min) || (off_max < off)))
2704			return -EINVAL;
2705	} else {
2706		off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2707	}
2708
2709	*data = frame + off;
2710	return tp_len;
2711}
2712
2713static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2714{
2715	struct sk_buff *skb = NULL;
2716	struct net_device *dev;
2717	struct virtio_net_hdr *vnet_hdr = NULL;
2718	struct sockcm_cookie sockc;
2719	__be16 proto;
2720	int err, reserve = 0;
2721	void *ph;
2722	DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2723	bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2724	unsigned char *addr = NULL;
2725	int tp_len, size_max;
2726	void *data;
2727	int len_sum = 0;
2728	int status = TP_STATUS_AVAILABLE;
2729	int hlen, tlen, copylen = 0;
2730	long timeo = 0;
2731
2732	mutex_lock(&po->pg_vec_lock);
2733
2734	/* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2735	 * we need to confirm it under protection of pg_vec_lock.
2736	 */
2737	if (unlikely(!po->tx_ring.pg_vec)) {
2738		err = -EBUSY;
2739		goto out;
2740	}
2741	if (likely(saddr == NULL)) {
2742		dev	= packet_cached_dev_get(po);
2743		proto	= READ_ONCE(po->num);
2744	} else {
2745		err = -EINVAL;
2746		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2747			goto out;
2748		if (msg->msg_namelen < (saddr->sll_halen
2749					+ offsetof(struct sockaddr_ll,
2750						sll_addr)))
2751			goto out;
2752		proto	= saddr->sll_protocol;
2753		dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2754		if (po->sk.sk_socket->type == SOCK_DGRAM) {
2755			if (dev && msg->msg_namelen < dev->addr_len +
2756				   offsetof(struct sockaddr_ll, sll_addr))
2757				goto out_put;
2758			addr = saddr->sll_addr;
2759		}
2760	}
2761
2762	err = -ENXIO;
2763	if (unlikely(dev == NULL))
2764		goto out;
2765	err = -ENETDOWN;
2766	if (unlikely(!(dev->flags & IFF_UP)))
2767		goto out_put;
2768
2769	sockcm_init(&sockc, &po->sk);
2770	if (msg->msg_controllen) {
2771		err = sock_cmsg_send(&po->sk, msg, &sockc);
2772		if (unlikely(err))
2773			goto out_put;
2774	}
2775
2776	if (po->sk.sk_socket->type == SOCK_RAW)
2777		reserve = dev->hard_header_len;
2778	size_max = po->tx_ring.frame_size
2779		- (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2780
2781	if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2782		size_max = dev->mtu + reserve + VLAN_HLEN;
2783
2784	reinit_completion(&po->skb_completion);
2785
2786	do {
2787		ph = packet_current_frame(po, &po->tx_ring,
2788					  TP_STATUS_SEND_REQUEST);
2789		if (unlikely(ph == NULL)) {
2790			if (need_wait && skb) {
2791				timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2792				timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2793				if (timeo <= 0) {
2794					err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2795					goto out_put;
2796				}
2797			}
2798			/* check for additional frames */
2799			continue;
2800		}
2801
2802		skb = NULL;
2803		tp_len = tpacket_parse_header(po, ph, size_max, &data);
2804		if (tp_len < 0)
2805			goto tpacket_error;
2806
2807		status = TP_STATUS_SEND_REQUEST;
2808		hlen = LL_RESERVED_SPACE(dev);
2809		tlen = dev->needed_tailroom;
2810		if (po->has_vnet_hdr) {
2811			vnet_hdr = data;
2812			data += sizeof(*vnet_hdr);
2813			tp_len -= sizeof(*vnet_hdr);
2814			if (tp_len < 0 ||
2815			    __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2816				tp_len = -EINVAL;
2817				goto tpacket_error;
2818			}
2819			copylen = __virtio16_to_cpu(vio_le(),
2820						    vnet_hdr->hdr_len);
2821		}
2822		copylen = max_t(int, copylen, dev->hard_header_len);
2823		skb = sock_alloc_send_skb(&po->sk,
2824				hlen + tlen + sizeof(struct sockaddr_ll) +
2825				(copylen - dev->hard_header_len),
2826				!need_wait, &err);
2827
2828		if (unlikely(skb == NULL)) {
2829			/* we assume the socket was initially writeable ... */
2830			if (likely(len_sum > 0))
2831				err = len_sum;
2832			goto out_status;
2833		}
2834		tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2835					  addr, hlen, copylen, &sockc);
2836		if (likely(tp_len >= 0) &&
2837		    tp_len > dev->mtu + reserve &&
2838		    !po->has_vnet_hdr &&
2839		    !packet_extra_vlan_len_allowed(dev, skb))
2840			tp_len = -EMSGSIZE;
2841
2842		if (unlikely(tp_len < 0)) {
2843tpacket_error:
2844			if (po->tp_loss) {
2845				__packet_set_status(po, ph,
2846						TP_STATUS_AVAILABLE);
2847				packet_increment_head(&po->tx_ring);
2848				kfree_skb(skb);
2849				continue;
2850			} else {
2851				status = TP_STATUS_WRONG_FORMAT;
2852				err = tp_len;
2853				goto out_status;
2854			}
2855		}
2856
2857		if (po->has_vnet_hdr) {
2858			if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2859				tp_len = -EINVAL;
2860				goto tpacket_error;
2861			}
2862			virtio_net_hdr_set_proto(skb, vnet_hdr);
2863		}
2864
2865		skb->destructor = tpacket_destruct_skb;
2866		__packet_set_status(po, ph, TP_STATUS_SENDING);
2867		packet_inc_pending(&po->tx_ring);
2868
2869		status = TP_STATUS_SEND_REQUEST;
2870		err = po->xmit(skb);
2871		if (unlikely(err != 0)) {
2872			if (err > 0)
2873				err = net_xmit_errno(err);
2874			if (err && __packet_get_status(po, ph) ==
2875				   TP_STATUS_AVAILABLE) {
2876				/* skb was destructed already */
2877				skb = NULL;
2878				goto out_status;
2879			}
2880			/*
2881			 * skb was dropped but not destructed yet;
2882			 * let's treat it like congestion or err < 0
2883			 */
2884			err = 0;
2885		}
2886		packet_increment_head(&po->tx_ring);
2887		len_sum += tp_len;
2888	} while (likely((ph != NULL) ||
2889		/* Note: packet_read_pending() might be slow if we have
2890		 * to call it as it's per_cpu variable, but in fast-path
2891		 * we already short-circuit the loop with the first
2892		 * condition, and luckily don't have to go that path
2893		 * anyway.
2894		 */
2895		 (need_wait && packet_read_pending(&po->tx_ring))));
2896
2897	err = len_sum;
2898	goto out_put;
2899
2900out_status:
2901	__packet_set_status(po, ph, status);
2902	kfree_skb(skb);
2903out_put:
2904	dev_put(dev);
2905out:
2906	mutex_unlock(&po->pg_vec_lock);
2907	return err;
2908}
2909
2910static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2911				        size_t reserve, size_t len,
2912				        size_t linear, int noblock,
2913				        int *err)
2914{
2915	struct sk_buff *skb;
2916
2917	/* Under a page?  Don't bother with paged skb. */
2918	if (prepad + len < PAGE_SIZE || !linear)
2919		linear = len;
2920
2921	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2922				   err, 0);
2923	if (!skb)
2924		return NULL;
2925
2926	skb_reserve(skb, reserve);
2927	skb_put(skb, linear);
2928	skb->data_len = len - linear;
2929	skb->len += len - linear;
2930
2931	return skb;
2932}
2933
2934static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2935{
2936	struct sock *sk = sock->sk;
2937	DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2938	struct sk_buff *skb;
2939	struct net_device *dev;
2940	__be16 proto;
2941	unsigned char *addr = NULL;
2942	int err, reserve = 0;
2943	struct sockcm_cookie sockc;
2944	struct virtio_net_hdr vnet_hdr = { 0 };
2945	int offset = 0;
2946	struct packet_sock *po = pkt_sk(sk);
2947	bool has_vnet_hdr = false;
2948	int hlen, tlen, linear;
2949	int extra_len = 0;
2950
2951	/*
2952	 *	Get and verify the address.
2953	 */
2954
2955	if (likely(saddr == NULL)) {
2956		dev	= packet_cached_dev_get(po);
2957		proto	= READ_ONCE(po->num);
2958	} else {
2959		err = -EINVAL;
2960		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2961			goto out;
2962		if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2963			goto out;
2964		proto	= saddr->sll_protocol;
2965		dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2966		if (sock->type == SOCK_DGRAM) {
2967			if (dev && msg->msg_namelen < dev->addr_len +
2968				   offsetof(struct sockaddr_ll, sll_addr))
2969				goto out_unlock;
2970			addr = saddr->sll_addr;
2971		}
2972	}
2973
2974	err = -ENXIO;
2975	if (unlikely(dev == NULL))
2976		goto out_unlock;
2977	err = -ENETDOWN;
2978	if (unlikely(!(dev->flags & IFF_UP)))
2979		goto out_unlock;
2980
2981	sockcm_init(&sockc, sk);
2982	sockc.mark = sk->sk_mark;
2983	if (msg->msg_controllen) {
2984		err = sock_cmsg_send(sk, msg, &sockc);
2985		if (unlikely(err))
2986			goto out_unlock;
2987	}
2988
2989	if (sock->type == SOCK_RAW)
2990		reserve = dev->hard_header_len;
2991	if (po->has_vnet_hdr) {
2992		err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2993		if (err)
2994			goto out_unlock;
2995		has_vnet_hdr = true;
2996	}
2997
2998	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2999		if (!netif_supports_nofcs(dev)) {
3000			err = -EPROTONOSUPPORT;
3001			goto out_unlock;
3002		}
3003		extra_len = 4; /* We're doing our own CRC */
3004	}
3005
3006	err = -EMSGSIZE;
3007	if (!vnet_hdr.gso_type &&
3008	    (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
3009		goto out_unlock;
3010
3011	err = -ENOBUFS;
3012	hlen = LL_RESERVED_SPACE(dev);
3013	tlen = dev->needed_tailroom;
3014	linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
3015	linear = max(linear, min_t(int, len, dev->hard_header_len));
3016	skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
3017			       msg->msg_flags & MSG_DONTWAIT, &err);
3018	if (skb == NULL)
3019		goto out_unlock;
3020
3021	skb_reset_network_header(skb);
3022
3023	err = -EINVAL;
3024	if (sock->type == SOCK_DGRAM) {
3025		offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
3026		if (unlikely(offset < 0))
3027			goto out_free;
3028	} else if (reserve) {
3029		skb_reserve(skb, -reserve);
3030		if (len < reserve + sizeof(struct ipv6hdr) &&
3031		    dev->min_header_len != dev->hard_header_len)
3032			skb_reset_network_header(skb);
3033	}
3034
3035	/* Returns -EFAULT on error */
3036	err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
3037	if (err)
3038		goto out_free;
3039
3040	if ((sock->type == SOCK_RAW &&
3041	     !dev_validate_header(dev, skb->data, len)) || !skb->len) {
3042		err = -EINVAL;
3043		goto out_free;
3044	}
3045
3046	skb_setup_tx_timestamp(skb, sockc.tsflags);
3047
3048	if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
3049	    !packet_extra_vlan_len_allowed(dev, skb)) {
3050		err = -EMSGSIZE;
3051		goto out_free;
3052	}
3053
3054	skb->protocol = proto;
3055	skb->dev = dev;
3056	skb->priority = sk->sk_priority;
3057	skb->mark = sockc.mark;
3058	skb->tstamp = sockc.transmit_time;
3059
3060	if (unlikely(extra_len == 4))
3061		skb->no_fcs = 1;
3062
3063	packet_parse_headers(skb, sock);
3064
3065	if (has_vnet_hdr) {
3066		err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
3067		if (err)
3068			goto out_free;
3069		len += sizeof(vnet_hdr);
3070		virtio_net_hdr_set_proto(skb, &vnet_hdr);
3071	}
3072
 
 
 
 
 
3073	err = po->xmit(skb);
3074	if (unlikely(err != 0)) {
3075		if (err > 0)
3076			err = net_xmit_errno(err);
3077		if (err)
3078			goto out_unlock;
3079	}
3080
3081	dev_put(dev);
3082
3083	return len;
3084
3085out_free:
3086	kfree_skb(skb);
3087out_unlock:
3088	dev_put(dev);
 
3089out:
3090	return err;
3091}
3092
3093static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
3094{
3095	struct sock *sk = sock->sk;
3096	struct packet_sock *po = pkt_sk(sk);
3097
3098	/* Reading tx_ring.pg_vec without holding pg_vec_lock is racy.
3099	 * tpacket_snd() will redo the check safely.
3100	 */
3101	if (data_race(po->tx_ring.pg_vec))
3102		return tpacket_snd(po, msg);
3103
3104	return packet_snd(sock, msg, len);
3105}
3106
3107/*
3108 *	Close a PACKET socket. This is fairly simple. We immediately go
3109 *	to 'closed' state and remove our protocol entry in the device list.
3110 */
3111
3112static int packet_release(struct socket *sock)
3113{
3114	struct sock *sk = sock->sk;
3115	struct packet_sock *po;
3116	struct packet_fanout *f;
3117	struct net *net;
3118	union tpacket_req_u req_u;
3119
3120	if (!sk)
3121		return 0;
3122
3123	net = sock_net(sk);
3124	po = pkt_sk(sk);
3125
3126	mutex_lock(&net->packet.sklist_lock);
3127	sk_del_node_init_rcu(sk);
3128	mutex_unlock(&net->packet.sklist_lock);
3129
 
3130	sock_prot_inuse_add(net, sk->sk_prot, -1);
 
3131
3132	spin_lock(&po->bind_lock);
3133	unregister_prot_hook(sk, false);
3134	packet_cached_dev_reset(po);
3135
3136	if (po->prot_hook.dev) {
3137		netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3138		po->prot_hook.dev = NULL;
3139	}
3140	spin_unlock(&po->bind_lock);
3141
3142	packet_flush_mclist(sk);
3143
3144	lock_sock(sk);
3145	if (po->rx_ring.pg_vec) {
3146		memset(&req_u, 0, sizeof(req_u));
3147		packet_set_ring(sk, &req_u, 1, 0);
3148	}
3149
3150	if (po->tx_ring.pg_vec) {
3151		memset(&req_u, 0, sizeof(req_u));
3152		packet_set_ring(sk, &req_u, 1, 1);
3153	}
3154	release_sock(sk);
3155
3156	f = fanout_release(sk);
3157
3158	synchronize_net();
3159
3160	kfree(po->rollover);
3161	if (f) {
3162		fanout_release_data(f);
3163		kvfree(f);
3164	}
3165	/*
3166	 *	Now the socket is dead. No more input will appear.
3167	 */
3168	sock_orphan(sk);
3169	sock->sk = NULL;
3170
3171	/* Purge queues */
3172
3173	skb_queue_purge(&sk->sk_receive_queue);
3174	packet_free_pending(po);
3175	sk_refcnt_debug_release(sk);
3176
3177	sock_put(sk);
3178	return 0;
3179}
3180
3181/*
3182 *	Attach a packet hook.
3183 */
3184
3185static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3186			  __be16 proto)
3187{
3188	struct packet_sock *po = pkt_sk(sk);
 
 
 
3189	struct net_device *dev = NULL;
 
3190	bool unlisted = false;
3191	bool need_rehook;
3192	int ret = 0;
3193
3194	lock_sock(sk);
3195	spin_lock(&po->bind_lock);
3196	rcu_read_lock();
3197
3198	if (po->fanout) {
3199		ret = -EINVAL;
3200		goto out_unlock;
3201	}
3202
3203	if (name) {
3204		dev = dev_get_by_name_rcu(sock_net(sk), name);
3205		if (!dev) {
3206			ret = -ENODEV;
3207			goto out_unlock;
3208		}
3209	} else if (ifindex) {
3210		dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3211		if (!dev) {
3212			ret = -ENODEV;
3213			goto out_unlock;
3214		}
3215	}
3216
3217	need_rehook = po->prot_hook.type != proto || po->prot_hook.dev != dev;
 
 
 
 
 
 
3218
3219	if (need_rehook) {
3220		dev_hold(dev);
3221		if (po->running) {
3222			rcu_read_unlock();
3223			/* prevents packet_notifier() from calling
3224			 * register_prot_hook()
3225			 */
3226			WRITE_ONCE(po->num, 0);
3227			__unregister_prot_hook(sk, true);
3228			rcu_read_lock();
 
3229			if (dev)
3230				unlisted = !dev_get_by_index_rcu(sock_net(sk),
3231								 dev->ifindex);
3232		}
3233
3234		BUG_ON(po->running);
3235		WRITE_ONCE(po->num, proto);
3236		po->prot_hook.type = proto;
3237
3238		netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3239
3240		if (unlikely(unlisted)) {
 
3241			po->prot_hook.dev = NULL;
3242			WRITE_ONCE(po->ifindex, -1);
3243			packet_cached_dev_reset(po);
3244		} else {
3245			netdev_hold(dev, &po->prot_hook.dev_tracker,
3246				    GFP_ATOMIC);
3247			po->prot_hook.dev = dev;
3248			WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0);
3249			packet_cached_dev_assign(po, dev);
3250		}
3251		dev_put(dev);
3252	}
 
 
3253
3254	if (proto == 0 || !need_rehook)
3255		goto out_unlock;
3256
3257	if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3258		register_prot_hook(sk);
3259	} else {
3260		sk->sk_err = ENETDOWN;
3261		if (!sock_flag(sk, SOCK_DEAD))
3262			sk_error_report(sk);
3263	}
3264
3265out_unlock:
3266	rcu_read_unlock();
3267	spin_unlock(&po->bind_lock);
3268	release_sock(sk);
3269	return ret;
3270}
3271
3272/*
3273 *	Bind a packet socket to a device
3274 */
3275
3276static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3277			    int addr_len)
3278{
3279	struct sock *sk = sock->sk;
3280	char name[sizeof(uaddr->sa_data_min) + 1];
3281
3282	/*
3283	 *	Check legality
3284	 */
3285
3286	if (addr_len != sizeof(struct sockaddr))
3287		return -EINVAL;
3288	/* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3289	 * zero-terminated.
3290	 */
3291	memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data_min));
3292	name[sizeof(uaddr->sa_data_min)] = 0;
3293
3294	return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3295}
3296
3297static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3298{
3299	struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3300	struct sock *sk = sock->sk;
3301
3302	/*
3303	 *	Check legality
3304	 */
3305
3306	if (addr_len < sizeof(struct sockaddr_ll))
3307		return -EINVAL;
3308	if (sll->sll_family != AF_PACKET)
3309		return -EINVAL;
3310
3311	return packet_do_bind(sk, NULL, sll->sll_ifindex,
3312			      sll->sll_protocol ? : pkt_sk(sk)->num);
3313}
3314
3315static struct proto packet_proto = {
3316	.name	  = "PACKET",
3317	.owner	  = THIS_MODULE,
3318	.obj_size = sizeof(struct packet_sock),
3319};
3320
3321/*
3322 *	Create a packet of type SOCK_PACKET.
3323 */
3324
3325static int packet_create(struct net *net, struct socket *sock, int protocol,
3326			 int kern)
3327{
3328	struct sock *sk;
3329	struct packet_sock *po;
3330	__be16 proto = (__force __be16)protocol; /* weird, but documented */
3331	int err;
3332
3333	if (!ns_capable(net->user_ns, CAP_NET_RAW))
3334		return -EPERM;
3335	if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3336	    sock->type != SOCK_PACKET)
3337		return -ESOCKTNOSUPPORT;
3338
3339	sock->state = SS_UNCONNECTED;
3340
3341	err = -ENOBUFS;
3342	sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3343	if (sk == NULL)
3344		goto out;
3345
3346	sock->ops = &packet_ops;
3347	if (sock->type == SOCK_PACKET)
3348		sock->ops = &packet_ops_spkt;
3349
3350	sock_init_data(sock, sk);
3351
3352	po = pkt_sk(sk);
3353	init_completion(&po->skb_completion);
3354	sk->sk_family = PF_PACKET;
3355	po->num = proto;
3356	po->xmit = dev_queue_xmit;
3357
3358	err = packet_alloc_pending(po);
3359	if (err)
3360		goto out2;
3361
3362	packet_cached_dev_reset(po);
3363
3364	sk->sk_destruct = packet_sock_destruct;
3365	sk_refcnt_debug_inc(sk);
3366
3367	/*
3368	 *	Attach a protocol block
3369	 */
3370
3371	spin_lock_init(&po->bind_lock);
3372	mutex_init(&po->pg_vec_lock);
3373	po->rollover = NULL;
3374	po->prot_hook.func = packet_rcv;
3375
3376	if (sock->type == SOCK_PACKET)
3377		po->prot_hook.func = packet_rcv_spkt;
3378
3379	po->prot_hook.af_packet_priv = sk;
3380	po->prot_hook.af_packet_net = sock_net(sk);
3381
3382	if (proto) {
3383		po->prot_hook.type = proto;
3384		__register_prot_hook(sk);
3385	}
3386
3387	mutex_lock(&net->packet.sklist_lock);
3388	sk_add_node_tail_rcu(sk, &net->packet.sklist);
3389	mutex_unlock(&net->packet.sklist_lock);
3390
 
3391	sock_prot_inuse_add(net, &packet_proto, 1);
 
3392
3393	return 0;
3394out2:
3395	sk_free(sk);
3396out:
3397	return err;
3398}
3399
3400/*
3401 *	Pull a packet from our receive queue and hand it to the user.
3402 *	If necessary we block.
3403 */
3404
3405static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3406			  int flags)
3407{
3408	struct sock *sk = sock->sk;
3409	struct sk_buff *skb;
3410	int copied, err;
3411	int vnet_hdr_len = 0;
3412	unsigned int origlen = 0;
3413
3414	err = -EINVAL;
3415	if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3416		goto out;
3417
3418#if 0
3419	/* What error should we return now? EUNATTACH? */
3420	if (pkt_sk(sk)->ifindex < 0)
3421		return -ENODEV;
3422#endif
3423
3424	if (flags & MSG_ERRQUEUE) {
3425		err = sock_recv_errqueue(sk, msg, len,
3426					 SOL_PACKET, PACKET_TX_TIMESTAMP);
3427		goto out;
3428	}
3429
3430	/*
3431	 *	Call the generic datagram receiver. This handles all sorts
3432	 *	of horrible races and re-entrancy so we can forget about it
3433	 *	in the protocol layers.
3434	 *
3435	 *	Now it will return ENETDOWN, if device have just gone down,
3436	 *	but then it will block.
3437	 */
3438
3439	skb = skb_recv_datagram(sk, flags, &err);
3440
3441	/*
3442	 *	An error occurred so return it. Because skb_recv_datagram()
3443	 *	handles the blocking we don't see and worry about blocking
3444	 *	retries.
3445	 */
3446
3447	if (skb == NULL)
3448		goto out;
3449
3450	packet_rcv_try_clear_pressure(pkt_sk(sk));
3451
3452	if (pkt_sk(sk)->has_vnet_hdr) {
3453		err = packet_rcv_vnet(msg, skb, &len);
3454		if (err)
3455			goto out_free;
3456		vnet_hdr_len = sizeof(struct virtio_net_hdr);
3457	}
3458
3459	/* You lose any data beyond the buffer you gave. If it worries
3460	 * a user program they can ask the device for its MTU
3461	 * anyway.
3462	 */
3463	copied = skb->len;
3464	if (copied > len) {
3465		copied = len;
3466		msg->msg_flags |= MSG_TRUNC;
3467	}
3468
3469	err = skb_copy_datagram_msg(skb, 0, msg, copied);
3470	if (err)
3471		goto out_free;
3472
3473	if (sock->type != SOCK_PACKET) {
3474		struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3475
3476		/* Original length was stored in sockaddr_ll fields */
3477		origlen = PACKET_SKB_CB(skb)->sa.origlen;
3478		sll->sll_family = AF_PACKET;
3479		sll->sll_protocol = skb->protocol;
3480	}
3481
3482	sock_recv_cmsgs(msg, sk, skb);
3483
3484	if (msg->msg_name) {
3485		const size_t max_len = min(sizeof(skb->cb),
3486					   sizeof(struct sockaddr_storage));
3487		int copy_len;
3488
3489		/* If the address length field is there to be filled
3490		 * in, we fill it in now.
3491		 */
3492		if (sock->type == SOCK_PACKET) {
3493			__sockaddr_check_size(sizeof(struct sockaddr_pkt));
3494			msg->msg_namelen = sizeof(struct sockaddr_pkt);
3495			copy_len = msg->msg_namelen;
3496		} else {
3497			struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3498
3499			msg->msg_namelen = sll->sll_halen +
3500				offsetof(struct sockaddr_ll, sll_addr);
3501			copy_len = msg->msg_namelen;
3502			if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3503				memset(msg->msg_name +
3504				       offsetof(struct sockaddr_ll, sll_addr),
3505				       0, sizeof(sll->sll_addr));
3506				msg->msg_namelen = sizeof(struct sockaddr_ll);
3507			}
3508		}
3509		if (WARN_ON_ONCE(copy_len > max_len)) {
3510			copy_len = max_len;
3511			msg->msg_namelen = copy_len;
3512		}
3513		memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3514	}
3515
3516	if (pkt_sk(sk)->auxdata) {
3517		struct tpacket_auxdata aux;
3518
3519		aux.tp_status = TP_STATUS_USER;
3520		if (skb->ip_summed == CHECKSUM_PARTIAL)
3521			aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3522		else if (skb->pkt_type != PACKET_OUTGOING &&
3523			 skb_csum_unnecessary(skb))
 
3524			aux.tp_status |= TP_STATUS_CSUM_VALID;
3525
3526		aux.tp_len = origlen;
3527		aux.tp_snaplen = skb->len;
3528		aux.tp_mac = 0;
3529		aux.tp_net = skb_network_offset(skb);
3530		if (skb_vlan_tag_present(skb)) {
3531			aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3532			aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3533			aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3534		} else {
3535			aux.tp_vlan_tci = 0;
3536			aux.tp_vlan_tpid = 0;
3537		}
3538		put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3539	}
3540
3541	/*
3542	 *	Free or return the buffer as appropriate. Again this
3543	 *	hides all the races and re-entrancy issues from us.
3544	 */
3545	err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3546
3547out_free:
3548	skb_free_datagram(sk, skb);
3549out:
3550	return err;
3551}
3552
3553static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3554			       int peer)
3555{
3556	struct net_device *dev;
3557	struct sock *sk	= sock->sk;
3558
3559	if (peer)
3560		return -EOPNOTSUPP;
3561
3562	uaddr->sa_family = AF_PACKET;
3563	memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data_min));
3564	rcu_read_lock();
3565	dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex));
3566	if (dev)
3567		strscpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data_min));
3568	rcu_read_unlock();
3569
3570	return sizeof(*uaddr);
3571}
3572
3573static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3574			  int peer)
3575{
3576	struct net_device *dev;
3577	struct sock *sk = sock->sk;
3578	struct packet_sock *po = pkt_sk(sk);
3579	DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3580	int ifindex;
3581
3582	if (peer)
3583		return -EOPNOTSUPP;
3584
3585	ifindex = READ_ONCE(po->ifindex);
3586	sll->sll_family = AF_PACKET;
3587	sll->sll_ifindex = ifindex;
3588	sll->sll_protocol = READ_ONCE(po->num);
3589	sll->sll_pkttype = 0;
3590	rcu_read_lock();
3591	dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3592	if (dev) {
3593		sll->sll_hatype = dev->type;
3594		sll->sll_halen = dev->addr_len;
3595		memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3596	} else {
3597		sll->sll_hatype = 0;	/* Bad: we have no ARPHRD_UNSPEC */
3598		sll->sll_halen = 0;
3599	}
3600	rcu_read_unlock();
3601
3602	return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3603}
3604
3605static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3606			 int what)
3607{
3608	switch (i->type) {
3609	case PACKET_MR_MULTICAST:
3610		if (i->alen != dev->addr_len)
3611			return -EINVAL;
3612		if (what > 0)
3613			return dev_mc_add(dev, i->addr);
3614		else
3615			return dev_mc_del(dev, i->addr);
3616		break;
3617	case PACKET_MR_PROMISC:
3618		return dev_set_promiscuity(dev, what);
3619	case PACKET_MR_ALLMULTI:
3620		return dev_set_allmulti(dev, what);
3621	case PACKET_MR_UNICAST:
3622		if (i->alen != dev->addr_len)
3623			return -EINVAL;
3624		if (what > 0)
3625			return dev_uc_add(dev, i->addr);
3626		else
3627			return dev_uc_del(dev, i->addr);
3628		break;
3629	default:
3630		break;
3631	}
3632	return 0;
3633}
3634
3635static void packet_dev_mclist_delete(struct net_device *dev,
3636				     struct packet_mclist **mlp)
3637{
3638	struct packet_mclist *ml;
3639
3640	while ((ml = *mlp) != NULL) {
3641		if (ml->ifindex == dev->ifindex) {
3642			packet_dev_mc(dev, ml, -1);
3643			*mlp = ml->next;
3644			kfree(ml);
3645		} else
3646			mlp = &ml->next;
3647	}
3648}
3649
3650static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3651{
3652	struct packet_sock *po = pkt_sk(sk);
3653	struct packet_mclist *ml, *i;
3654	struct net_device *dev;
3655	int err;
3656
3657	rtnl_lock();
3658
3659	err = -ENODEV;
3660	dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3661	if (!dev)
3662		goto done;
3663
3664	err = -EINVAL;
3665	if (mreq->mr_alen > dev->addr_len)
3666		goto done;
3667
3668	err = -ENOBUFS;
3669	i = kmalloc(sizeof(*i), GFP_KERNEL);
3670	if (i == NULL)
3671		goto done;
3672
3673	err = 0;
3674	for (ml = po->mclist; ml; ml = ml->next) {
3675		if (ml->ifindex == mreq->mr_ifindex &&
3676		    ml->type == mreq->mr_type &&
3677		    ml->alen == mreq->mr_alen &&
3678		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3679			ml->count++;
3680			/* Free the new element ... */
3681			kfree(i);
3682			goto done;
3683		}
3684	}
3685
3686	i->type = mreq->mr_type;
3687	i->ifindex = mreq->mr_ifindex;
3688	i->alen = mreq->mr_alen;
3689	memcpy(i->addr, mreq->mr_address, i->alen);
3690	memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3691	i->count = 1;
3692	i->next = po->mclist;
3693	po->mclist = i;
3694	err = packet_dev_mc(dev, i, 1);
3695	if (err) {
3696		po->mclist = i->next;
3697		kfree(i);
3698	}
3699
3700done:
3701	rtnl_unlock();
3702	return err;
3703}
3704
3705static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3706{
3707	struct packet_mclist *ml, **mlp;
3708
3709	rtnl_lock();
3710
3711	for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3712		if (ml->ifindex == mreq->mr_ifindex &&
3713		    ml->type == mreq->mr_type &&
3714		    ml->alen == mreq->mr_alen &&
3715		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3716			if (--ml->count == 0) {
3717				struct net_device *dev;
3718				*mlp = ml->next;
3719				dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3720				if (dev)
3721					packet_dev_mc(dev, ml, -1);
3722				kfree(ml);
3723			}
3724			break;
3725		}
3726	}
3727	rtnl_unlock();
3728	return 0;
3729}
3730
3731static void packet_flush_mclist(struct sock *sk)
3732{
3733	struct packet_sock *po = pkt_sk(sk);
3734	struct packet_mclist *ml;
3735
3736	if (!po->mclist)
3737		return;
3738
3739	rtnl_lock();
3740	while ((ml = po->mclist) != NULL) {
3741		struct net_device *dev;
3742
3743		po->mclist = ml->next;
3744		dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3745		if (dev != NULL)
3746			packet_dev_mc(dev, ml, -1);
3747		kfree(ml);
3748	}
3749	rtnl_unlock();
3750}
3751
3752static int
3753packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval,
3754		  unsigned int optlen)
3755{
3756	struct sock *sk = sock->sk;
3757	struct packet_sock *po = pkt_sk(sk);
3758	int ret;
3759
3760	if (level != SOL_PACKET)
3761		return -ENOPROTOOPT;
3762
3763	switch (optname) {
3764	case PACKET_ADD_MEMBERSHIP:
3765	case PACKET_DROP_MEMBERSHIP:
3766	{
3767		struct packet_mreq_max mreq;
3768		int len = optlen;
3769		memset(&mreq, 0, sizeof(mreq));
3770		if (len < sizeof(struct packet_mreq))
3771			return -EINVAL;
3772		if (len > sizeof(mreq))
3773			len = sizeof(mreq);
3774		if (copy_from_sockptr(&mreq, optval, len))
3775			return -EFAULT;
3776		if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3777			return -EINVAL;
3778		if (optname == PACKET_ADD_MEMBERSHIP)
3779			ret = packet_mc_add(sk, &mreq);
3780		else
3781			ret = packet_mc_drop(sk, &mreq);
3782		return ret;
3783	}
3784
3785	case PACKET_RX_RING:
3786	case PACKET_TX_RING:
3787	{
3788		union tpacket_req_u req_u;
3789		int len;
3790
3791		lock_sock(sk);
3792		switch (po->tp_version) {
3793		case TPACKET_V1:
3794		case TPACKET_V2:
3795			len = sizeof(req_u.req);
3796			break;
3797		case TPACKET_V3:
3798		default:
3799			len = sizeof(req_u.req3);
3800			break;
3801		}
3802		if (optlen < len) {
3803			ret = -EINVAL;
3804		} else {
3805			if (copy_from_sockptr(&req_u.req, optval, len))
3806				ret = -EFAULT;
3807			else
3808				ret = packet_set_ring(sk, &req_u, 0,
3809						    optname == PACKET_TX_RING);
3810		}
3811		release_sock(sk);
3812		return ret;
3813	}
3814	case PACKET_COPY_THRESH:
3815	{
3816		int val;
3817
3818		if (optlen != sizeof(val))
3819			return -EINVAL;
3820		if (copy_from_sockptr(&val, optval, sizeof(val)))
3821			return -EFAULT;
3822
3823		pkt_sk(sk)->copy_thresh = val;
3824		return 0;
3825	}
3826	case PACKET_VERSION:
3827	{
3828		int val;
3829
3830		if (optlen != sizeof(val))
3831			return -EINVAL;
3832		if (copy_from_sockptr(&val, optval, sizeof(val)))
3833			return -EFAULT;
3834		switch (val) {
3835		case TPACKET_V1:
3836		case TPACKET_V2:
3837		case TPACKET_V3:
3838			break;
3839		default:
3840			return -EINVAL;
3841		}
3842		lock_sock(sk);
3843		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3844			ret = -EBUSY;
3845		} else {
3846			po->tp_version = val;
3847			ret = 0;
3848		}
3849		release_sock(sk);
3850		return ret;
3851	}
3852	case PACKET_RESERVE:
3853	{
3854		unsigned int val;
3855
3856		if (optlen != sizeof(val))
3857			return -EINVAL;
3858		if (copy_from_sockptr(&val, optval, sizeof(val)))
3859			return -EFAULT;
3860		if (val > INT_MAX)
3861			return -EINVAL;
3862		lock_sock(sk);
3863		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3864			ret = -EBUSY;
3865		} else {
3866			po->tp_reserve = val;
3867			ret = 0;
3868		}
3869		release_sock(sk);
3870		return ret;
3871	}
3872	case PACKET_LOSS:
3873	{
3874		unsigned int val;
3875
3876		if (optlen != sizeof(val))
3877			return -EINVAL;
3878		if (copy_from_sockptr(&val, optval, sizeof(val)))
3879			return -EFAULT;
3880
3881		lock_sock(sk);
3882		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3883			ret = -EBUSY;
3884		} else {
3885			po->tp_loss = !!val;
3886			ret = 0;
3887		}
3888		release_sock(sk);
3889		return ret;
3890	}
3891	case PACKET_AUXDATA:
3892	{
3893		int val;
3894
3895		if (optlen < sizeof(val))
3896			return -EINVAL;
3897		if (copy_from_sockptr(&val, optval, sizeof(val)))
3898			return -EFAULT;
3899
3900		lock_sock(sk);
3901		po->auxdata = !!val;
3902		release_sock(sk);
3903		return 0;
3904	}
3905	case PACKET_ORIGDEV:
3906	{
3907		int val;
3908
3909		if (optlen < sizeof(val))
3910			return -EINVAL;
3911		if (copy_from_sockptr(&val, optval, sizeof(val)))
3912			return -EFAULT;
3913
3914		lock_sock(sk);
3915		po->origdev = !!val;
3916		release_sock(sk);
3917		return 0;
3918	}
3919	case PACKET_VNET_HDR:
3920	{
3921		int val;
3922
3923		if (sock->type != SOCK_RAW)
3924			return -EINVAL;
3925		if (optlen < sizeof(val))
3926			return -EINVAL;
3927		if (copy_from_sockptr(&val, optval, sizeof(val)))
3928			return -EFAULT;
3929
3930		lock_sock(sk);
3931		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3932			ret = -EBUSY;
3933		} else {
3934			po->has_vnet_hdr = !!val;
3935			ret = 0;
3936		}
3937		release_sock(sk);
3938		return ret;
3939	}
3940	case PACKET_TIMESTAMP:
3941	{
3942		int val;
3943
3944		if (optlen != sizeof(val))
3945			return -EINVAL;
3946		if (copy_from_sockptr(&val, optval, sizeof(val)))
3947			return -EFAULT;
3948
3949		po->tp_tstamp = val;
3950		return 0;
3951	}
3952	case PACKET_FANOUT:
3953	{
3954		struct fanout_args args = { 0 };
3955
3956		if (optlen != sizeof(int) && optlen != sizeof(args))
3957			return -EINVAL;
3958		if (copy_from_sockptr(&args, optval, optlen))
3959			return -EFAULT;
3960
3961		return fanout_add(sk, &args);
3962	}
3963	case PACKET_FANOUT_DATA:
3964	{
3965		/* Paired with the WRITE_ONCE() in fanout_add() */
3966		if (!READ_ONCE(po->fanout))
3967			return -EINVAL;
3968
3969		return fanout_set_data(po, optval, optlen);
3970	}
3971	case PACKET_IGNORE_OUTGOING:
3972	{
3973		int val;
3974
3975		if (optlen != sizeof(val))
3976			return -EINVAL;
3977		if (copy_from_sockptr(&val, optval, sizeof(val)))
3978			return -EFAULT;
3979		if (val < 0 || val > 1)
3980			return -EINVAL;
3981
3982		po->prot_hook.ignore_outgoing = !!val;
3983		return 0;
3984	}
3985	case PACKET_TX_HAS_OFF:
3986	{
3987		unsigned int val;
3988
3989		if (optlen != sizeof(val))
3990			return -EINVAL;
3991		if (copy_from_sockptr(&val, optval, sizeof(val)))
3992			return -EFAULT;
3993
3994		lock_sock(sk);
3995		if (!po->rx_ring.pg_vec && !po->tx_ring.pg_vec)
 
 
3996			po->tp_tx_has_off = !!val;
3997
 
3998		release_sock(sk);
3999		return 0;
4000	}
4001	case PACKET_QDISC_BYPASS:
4002	{
4003		int val;
4004
4005		if (optlen != sizeof(val))
4006			return -EINVAL;
4007		if (copy_from_sockptr(&val, optval, sizeof(val)))
4008			return -EFAULT;
4009
4010		po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
4011		return 0;
4012	}
4013	default:
4014		return -ENOPROTOOPT;
4015	}
4016}
4017
4018static int packet_getsockopt(struct socket *sock, int level, int optname,
4019			     char __user *optval, int __user *optlen)
4020{
4021	int len;
4022	int val, lv = sizeof(val);
4023	struct sock *sk = sock->sk;
4024	struct packet_sock *po = pkt_sk(sk);
4025	void *data = &val;
4026	union tpacket_stats_u st;
4027	struct tpacket_rollover_stats rstats;
4028	int drops;
4029
4030	if (level != SOL_PACKET)
4031		return -ENOPROTOOPT;
4032
4033	if (get_user(len, optlen))
4034		return -EFAULT;
4035
4036	if (len < 0)
4037		return -EINVAL;
4038
4039	switch (optname) {
4040	case PACKET_STATISTICS:
4041		spin_lock_bh(&sk->sk_receive_queue.lock);
4042		memcpy(&st, &po->stats, sizeof(st));
4043		memset(&po->stats, 0, sizeof(po->stats));
4044		spin_unlock_bh(&sk->sk_receive_queue.lock);
4045		drops = atomic_xchg(&po->tp_drops, 0);
4046
4047		if (po->tp_version == TPACKET_V3) {
4048			lv = sizeof(struct tpacket_stats_v3);
4049			st.stats3.tp_drops = drops;
4050			st.stats3.tp_packets += drops;
4051			data = &st.stats3;
4052		} else {
4053			lv = sizeof(struct tpacket_stats);
4054			st.stats1.tp_drops = drops;
4055			st.stats1.tp_packets += drops;
4056			data = &st.stats1;
4057		}
4058
4059		break;
4060	case PACKET_AUXDATA:
4061		val = po->auxdata;
4062		break;
4063	case PACKET_ORIGDEV:
4064		val = po->origdev;
4065		break;
4066	case PACKET_VNET_HDR:
4067		val = po->has_vnet_hdr;
4068		break;
4069	case PACKET_VERSION:
4070		val = po->tp_version;
4071		break;
4072	case PACKET_HDRLEN:
4073		if (len > sizeof(int))
4074			len = sizeof(int);
4075		if (len < sizeof(int))
4076			return -EINVAL;
4077		if (copy_from_user(&val, optval, len))
4078			return -EFAULT;
4079		switch (val) {
4080		case TPACKET_V1:
4081			val = sizeof(struct tpacket_hdr);
4082			break;
4083		case TPACKET_V2:
4084			val = sizeof(struct tpacket2_hdr);
4085			break;
4086		case TPACKET_V3:
4087			val = sizeof(struct tpacket3_hdr);
4088			break;
4089		default:
4090			return -EINVAL;
4091		}
4092		break;
4093	case PACKET_RESERVE:
4094		val = po->tp_reserve;
4095		break;
4096	case PACKET_LOSS:
4097		val = po->tp_loss;
4098		break;
4099	case PACKET_TIMESTAMP:
4100		val = po->tp_tstamp;
4101		break;
4102	case PACKET_FANOUT:
4103		val = (po->fanout ?
4104		       ((u32)po->fanout->id |
4105			((u32)po->fanout->type << 16) |
4106			((u32)po->fanout->flags << 24)) :
4107		       0);
4108		break;
4109	case PACKET_IGNORE_OUTGOING:
4110		val = po->prot_hook.ignore_outgoing;
4111		break;
4112	case PACKET_ROLLOVER_STATS:
4113		if (!po->rollover)
4114			return -EINVAL;
4115		rstats.tp_all = atomic_long_read(&po->rollover->num);
4116		rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4117		rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4118		data = &rstats;
4119		lv = sizeof(rstats);
4120		break;
4121	case PACKET_TX_HAS_OFF:
4122		val = po->tp_tx_has_off;
4123		break;
4124	case PACKET_QDISC_BYPASS:
4125		val = packet_use_direct_xmit(po);
4126		break;
4127	default:
4128		return -ENOPROTOOPT;
4129	}
4130
4131	if (len > lv)
4132		len = lv;
4133	if (put_user(len, optlen))
4134		return -EFAULT;
4135	if (copy_to_user(optval, data, len))
4136		return -EFAULT;
4137	return 0;
4138}
4139
4140static int packet_notifier(struct notifier_block *this,
4141			   unsigned long msg, void *ptr)
4142{
4143	struct sock *sk;
4144	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4145	struct net *net = dev_net(dev);
4146
4147	rcu_read_lock();
4148	sk_for_each_rcu(sk, &net->packet.sklist) {
4149		struct packet_sock *po = pkt_sk(sk);
4150
4151		switch (msg) {
4152		case NETDEV_UNREGISTER:
4153			if (po->mclist)
4154				packet_dev_mclist_delete(dev, &po->mclist);
4155			fallthrough;
4156
4157		case NETDEV_DOWN:
4158			if (dev->ifindex == po->ifindex) {
4159				spin_lock(&po->bind_lock);
4160				if (po->running) {
4161					__unregister_prot_hook(sk, false);
4162					sk->sk_err = ENETDOWN;
4163					if (!sock_flag(sk, SOCK_DEAD))
4164						sk_error_report(sk);
4165				}
4166				if (msg == NETDEV_UNREGISTER) {
4167					packet_cached_dev_reset(po);
4168					WRITE_ONCE(po->ifindex, -1);
4169					netdev_put(po->prot_hook.dev,
4170						   &po->prot_hook.dev_tracker);
4171					po->prot_hook.dev = NULL;
4172				}
4173				spin_unlock(&po->bind_lock);
4174			}
4175			break;
4176		case NETDEV_UP:
4177			if (dev->ifindex == po->ifindex) {
4178				spin_lock(&po->bind_lock);
4179				if (po->num)
4180					register_prot_hook(sk);
4181				spin_unlock(&po->bind_lock);
4182			}
4183			break;
4184		}
4185	}
4186	rcu_read_unlock();
4187	return NOTIFY_DONE;
4188}
4189
4190
4191static int packet_ioctl(struct socket *sock, unsigned int cmd,
4192			unsigned long arg)
4193{
4194	struct sock *sk = sock->sk;
4195
4196	switch (cmd) {
4197	case SIOCOUTQ:
4198	{
4199		int amount = sk_wmem_alloc_get(sk);
4200
4201		return put_user(amount, (int __user *)arg);
4202	}
4203	case SIOCINQ:
4204	{
4205		struct sk_buff *skb;
4206		int amount = 0;
4207
4208		spin_lock_bh(&sk->sk_receive_queue.lock);
4209		skb = skb_peek(&sk->sk_receive_queue);
4210		if (skb)
4211			amount = skb->len;
4212		spin_unlock_bh(&sk->sk_receive_queue.lock);
4213		return put_user(amount, (int __user *)arg);
4214	}
4215#ifdef CONFIG_INET
4216	case SIOCADDRT:
4217	case SIOCDELRT:
4218	case SIOCDARP:
4219	case SIOCGARP:
4220	case SIOCSARP:
4221	case SIOCGIFADDR:
4222	case SIOCSIFADDR:
4223	case SIOCGIFBRDADDR:
4224	case SIOCSIFBRDADDR:
4225	case SIOCGIFNETMASK:
4226	case SIOCSIFNETMASK:
4227	case SIOCGIFDSTADDR:
4228	case SIOCSIFDSTADDR:
4229	case SIOCSIFFLAGS:
4230		return inet_dgram_ops.ioctl(sock, cmd, arg);
4231#endif
4232
4233	default:
4234		return -ENOIOCTLCMD;
4235	}
4236	return 0;
4237}
4238
4239static __poll_t packet_poll(struct file *file, struct socket *sock,
4240				poll_table *wait)
4241{
4242	struct sock *sk = sock->sk;
4243	struct packet_sock *po = pkt_sk(sk);
4244	__poll_t mask = datagram_poll(file, sock, wait);
4245
4246	spin_lock_bh(&sk->sk_receive_queue.lock);
4247	if (po->rx_ring.pg_vec) {
4248		if (!packet_previous_rx_frame(po, &po->rx_ring,
4249			TP_STATUS_KERNEL))
4250			mask |= EPOLLIN | EPOLLRDNORM;
4251	}
4252	packet_rcv_try_clear_pressure(po);
4253	spin_unlock_bh(&sk->sk_receive_queue.lock);
4254	spin_lock_bh(&sk->sk_write_queue.lock);
4255	if (po->tx_ring.pg_vec) {
4256		if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4257			mask |= EPOLLOUT | EPOLLWRNORM;
4258	}
4259	spin_unlock_bh(&sk->sk_write_queue.lock);
4260	return mask;
4261}
4262
4263
4264/* Dirty? Well, I still did not learn better way to account
4265 * for user mmaps.
4266 */
4267
4268static void packet_mm_open(struct vm_area_struct *vma)
4269{
4270	struct file *file = vma->vm_file;
4271	struct socket *sock = file->private_data;
4272	struct sock *sk = sock->sk;
4273
4274	if (sk)
4275		atomic_inc(&pkt_sk(sk)->mapped);
4276}
4277
4278static void packet_mm_close(struct vm_area_struct *vma)
4279{
4280	struct file *file = vma->vm_file;
4281	struct socket *sock = file->private_data;
4282	struct sock *sk = sock->sk;
4283
4284	if (sk)
4285		atomic_dec(&pkt_sk(sk)->mapped);
4286}
4287
4288static const struct vm_operations_struct packet_mmap_ops = {
4289	.open	=	packet_mm_open,
4290	.close	=	packet_mm_close,
4291};
4292
4293static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4294			unsigned int len)
4295{
4296	int i;
4297
4298	for (i = 0; i < len; i++) {
4299		if (likely(pg_vec[i].buffer)) {
4300			if (is_vmalloc_addr(pg_vec[i].buffer))
4301				vfree(pg_vec[i].buffer);
4302			else
4303				free_pages((unsigned long)pg_vec[i].buffer,
4304					   order);
4305			pg_vec[i].buffer = NULL;
4306		}
4307	}
4308	kfree(pg_vec);
4309}
4310
4311static char *alloc_one_pg_vec_page(unsigned long order)
4312{
4313	char *buffer;
4314	gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4315			  __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4316
4317	buffer = (char *) __get_free_pages(gfp_flags, order);
4318	if (buffer)
4319		return buffer;
4320
4321	/* __get_free_pages failed, fall back to vmalloc */
4322	buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4323	if (buffer)
4324		return buffer;
4325
4326	/* vmalloc failed, lets dig into swap here */
4327	gfp_flags &= ~__GFP_NORETRY;
4328	buffer = (char *) __get_free_pages(gfp_flags, order);
4329	if (buffer)
4330		return buffer;
4331
4332	/* complete and utter failure */
4333	return NULL;
4334}
4335
4336static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4337{
4338	unsigned int block_nr = req->tp_block_nr;
4339	struct pgv *pg_vec;
4340	int i;
4341
4342	pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4343	if (unlikely(!pg_vec))
4344		goto out;
4345
4346	for (i = 0; i < block_nr; i++) {
4347		pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4348		if (unlikely(!pg_vec[i].buffer))
4349			goto out_free_pgvec;
4350	}
4351
4352out:
4353	return pg_vec;
4354
4355out_free_pgvec:
4356	free_pg_vec(pg_vec, order, block_nr);
4357	pg_vec = NULL;
4358	goto out;
4359}
4360
4361static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4362		int closing, int tx_ring)
4363{
4364	struct pgv *pg_vec = NULL;
4365	struct packet_sock *po = pkt_sk(sk);
4366	unsigned long *rx_owner_map = NULL;
4367	int was_running, order = 0;
4368	struct packet_ring_buffer *rb;
4369	struct sk_buff_head *rb_queue;
4370	__be16 num;
4371	int err;
4372	/* Added to avoid minimal code churn */
4373	struct tpacket_req *req = &req_u->req;
4374
4375	rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4376	rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4377
4378	err = -EBUSY;
4379	if (!closing) {
4380		if (atomic_read(&po->mapped))
4381			goto out;
4382		if (packet_read_pending(rb))
4383			goto out;
4384	}
4385
4386	if (req->tp_block_nr) {
4387		unsigned int min_frame_size;
4388
4389		/* Sanity tests and some calculations */
4390		err = -EBUSY;
4391		if (unlikely(rb->pg_vec))
4392			goto out;
4393
4394		switch (po->tp_version) {
4395		case TPACKET_V1:
4396			po->tp_hdrlen = TPACKET_HDRLEN;
4397			break;
4398		case TPACKET_V2:
4399			po->tp_hdrlen = TPACKET2_HDRLEN;
4400			break;
4401		case TPACKET_V3:
4402			po->tp_hdrlen = TPACKET3_HDRLEN;
4403			break;
4404		}
4405
4406		err = -EINVAL;
4407		if (unlikely((int)req->tp_block_size <= 0))
4408			goto out;
4409		if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4410			goto out;
4411		min_frame_size = po->tp_hdrlen + po->tp_reserve;
4412		if (po->tp_version >= TPACKET_V3 &&
4413		    req->tp_block_size <
4414		    BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4415			goto out;
4416		if (unlikely(req->tp_frame_size < min_frame_size))
4417			goto out;
4418		if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4419			goto out;
4420
4421		rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4422		if (unlikely(rb->frames_per_block == 0))
4423			goto out;
4424		if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4425			goto out;
4426		if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4427					req->tp_frame_nr))
4428			goto out;
4429
4430		err = -ENOMEM;
4431		order = get_order(req->tp_block_size);
4432		pg_vec = alloc_pg_vec(req, order);
4433		if (unlikely(!pg_vec))
4434			goto out;
4435		switch (po->tp_version) {
4436		case TPACKET_V3:
4437			/* Block transmit is not supported yet */
4438			if (!tx_ring) {
4439				init_prb_bdqc(po, rb, pg_vec, req_u);
4440			} else {
4441				struct tpacket_req3 *req3 = &req_u->req3;
4442
4443				if (req3->tp_retire_blk_tov ||
4444				    req3->tp_sizeof_priv ||
4445				    req3->tp_feature_req_word) {
4446					err = -EINVAL;
4447					goto out_free_pg_vec;
4448				}
4449			}
4450			break;
4451		default:
4452			if (!tx_ring) {
4453				rx_owner_map = bitmap_alloc(req->tp_frame_nr,
4454					GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
4455				if (!rx_owner_map)
4456					goto out_free_pg_vec;
4457			}
4458			break;
4459		}
4460	}
4461	/* Done */
4462	else {
4463		err = -EINVAL;
4464		if (unlikely(req->tp_frame_nr))
4465			goto out;
4466	}
4467
4468
4469	/* Detach socket from network */
4470	spin_lock(&po->bind_lock);
4471	was_running = po->running;
4472	num = po->num;
4473	if (was_running) {
4474		WRITE_ONCE(po->num, 0);
4475		__unregister_prot_hook(sk, false);
4476	}
4477	spin_unlock(&po->bind_lock);
4478
4479	synchronize_net();
4480
4481	err = -EBUSY;
4482	mutex_lock(&po->pg_vec_lock);
4483	if (closing || atomic_read(&po->mapped) == 0) {
4484		err = 0;
4485		spin_lock_bh(&rb_queue->lock);
4486		swap(rb->pg_vec, pg_vec);
4487		if (po->tp_version <= TPACKET_V2)
4488			swap(rb->rx_owner_map, rx_owner_map);
4489		rb->frame_max = (req->tp_frame_nr - 1);
4490		rb->head = 0;
4491		rb->frame_size = req->tp_frame_size;
4492		spin_unlock_bh(&rb_queue->lock);
4493
4494		swap(rb->pg_vec_order, order);
4495		swap(rb->pg_vec_len, req->tp_block_nr);
4496
4497		rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4498		po->prot_hook.func = (po->rx_ring.pg_vec) ?
4499						tpacket_rcv : packet_rcv;
4500		skb_queue_purge(rb_queue);
4501		if (atomic_read(&po->mapped))
4502			pr_err("packet_mmap: vma is busy: %d\n",
4503			       atomic_read(&po->mapped));
4504	}
4505	mutex_unlock(&po->pg_vec_lock);
4506
4507	spin_lock(&po->bind_lock);
4508	if (was_running) {
4509		WRITE_ONCE(po->num, num);
4510		register_prot_hook(sk);
4511	}
4512	spin_unlock(&po->bind_lock);
4513	if (pg_vec && (po->tp_version > TPACKET_V2)) {
4514		/* Because we don't support block-based V3 on tx-ring */
4515		if (!tx_ring)
4516			prb_shutdown_retire_blk_timer(po, rb_queue);
4517	}
4518
4519out_free_pg_vec:
4520	if (pg_vec) {
4521		bitmap_free(rx_owner_map);
4522		free_pg_vec(pg_vec, order, req->tp_block_nr);
4523	}
4524out:
4525	return err;
4526}
4527
4528static int packet_mmap(struct file *file, struct socket *sock,
4529		struct vm_area_struct *vma)
4530{
4531	struct sock *sk = sock->sk;
4532	struct packet_sock *po = pkt_sk(sk);
4533	unsigned long size, expected_size;
4534	struct packet_ring_buffer *rb;
4535	unsigned long start;
4536	int err = -EINVAL;
4537	int i;
4538
4539	if (vma->vm_pgoff)
4540		return -EINVAL;
4541
4542	mutex_lock(&po->pg_vec_lock);
4543
4544	expected_size = 0;
4545	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4546		if (rb->pg_vec) {
4547			expected_size += rb->pg_vec_len
4548						* rb->pg_vec_pages
4549						* PAGE_SIZE;
4550		}
4551	}
4552
4553	if (expected_size == 0)
4554		goto out;
4555
4556	size = vma->vm_end - vma->vm_start;
4557	if (size != expected_size)
4558		goto out;
4559
4560	start = vma->vm_start;
4561	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4562		if (rb->pg_vec == NULL)
4563			continue;
4564
4565		for (i = 0; i < rb->pg_vec_len; i++) {
4566			struct page *page;
4567			void *kaddr = rb->pg_vec[i].buffer;
4568			int pg_num;
4569
4570			for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4571				page = pgv_to_page(kaddr);
4572				err = vm_insert_page(vma, start, page);
4573				if (unlikely(err))
4574					goto out;
4575				start += PAGE_SIZE;
4576				kaddr += PAGE_SIZE;
4577			}
4578		}
4579	}
4580
4581	atomic_inc(&po->mapped);
4582	vma->vm_ops = &packet_mmap_ops;
4583	err = 0;
4584
4585out:
4586	mutex_unlock(&po->pg_vec_lock);
4587	return err;
4588}
4589
4590static const struct proto_ops packet_ops_spkt = {
4591	.family =	PF_PACKET,
4592	.owner =	THIS_MODULE,
4593	.release =	packet_release,
4594	.bind =		packet_bind_spkt,
4595	.connect =	sock_no_connect,
4596	.socketpair =	sock_no_socketpair,
4597	.accept =	sock_no_accept,
4598	.getname =	packet_getname_spkt,
4599	.poll =		datagram_poll,
4600	.ioctl =	packet_ioctl,
4601	.gettstamp =	sock_gettstamp,
4602	.listen =	sock_no_listen,
4603	.shutdown =	sock_no_shutdown,
4604	.sendmsg =	packet_sendmsg_spkt,
4605	.recvmsg =	packet_recvmsg,
4606	.mmap =		sock_no_mmap,
4607	.sendpage =	sock_no_sendpage,
4608};
4609
4610static const struct proto_ops packet_ops = {
4611	.family =	PF_PACKET,
4612	.owner =	THIS_MODULE,
4613	.release =	packet_release,
4614	.bind =		packet_bind,
4615	.connect =	sock_no_connect,
4616	.socketpair =	sock_no_socketpair,
4617	.accept =	sock_no_accept,
4618	.getname =	packet_getname,
4619	.poll =		packet_poll,
4620	.ioctl =	packet_ioctl,
4621	.gettstamp =	sock_gettstamp,
4622	.listen =	sock_no_listen,
4623	.shutdown =	sock_no_shutdown,
4624	.setsockopt =	packet_setsockopt,
4625	.getsockopt =	packet_getsockopt,
4626	.sendmsg =	packet_sendmsg,
4627	.recvmsg =	packet_recvmsg,
4628	.mmap =		packet_mmap,
4629	.sendpage =	sock_no_sendpage,
4630};
4631
4632static const struct net_proto_family packet_family_ops = {
4633	.family =	PF_PACKET,
4634	.create =	packet_create,
4635	.owner	=	THIS_MODULE,
4636};
4637
4638static struct notifier_block packet_netdev_notifier = {
4639	.notifier_call =	packet_notifier,
4640};
4641
4642#ifdef CONFIG_PROC_FS
4643
4644static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4645	__acquires(RCU)
4646{
4647	struct net *net = seq_file_net(seq);
4648
4649	rcu_read_lock();
4650	return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4651}
4652
4653static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4654{
4655	struct net *net = seq_file_net(seq);
4656	return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4657}
4658
4659static void packet_seq_stop(struct seq_file *seq, void *v)
4660	__releases(RCU)
4661{
4662	rcu_read_unlock();
4663}
4664
4665static int packet_seq_show(struct seq_file *seq, void *v)
4666{
4667	if (v == SEQ_START_TOKEN)
4668		seq_printf(seq,
4669			   "%*sRefCnt Type Proto  Iface R Rmem   User   Inode\n",
4670			   IS_ENABLED(CONFIG_64BIT) ? -17 : -9, "sk");
4671	else {
4672		struct sock *s = sk_entry(v);
4673		const struct packet_sock *po = pkt_sk(s);
4674
4675		seq_printf(seq,
4676			   "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4677			   s,
4678			   refcount_read(&s->sk_refcnt),
4679			   s->sk_type,
4680			   ntohs(READ_ONCE(po->num)),
4681			   READ_ONCE(po->ifindex),
4682			   po->running,
4683			   atomic_read(&s->sk_rmem_alloc),
4684			   from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4685			   sock_i_ino(s));
4686	}
4687
4688	return 0;
4689}
4690
4691static const struct seq_operations packet_seq_ops = {
4692	.start	= packet_seq_start,
4693	.next	= packet_seq_next,
4694	.stop	= packet_seq_stop,
4695	.show	= packet_seq_show,
4696};
4697#endif
4698
4699static int __net_init packet_net_init(struct net *net)
4700{
4701	mutex_init(&net->packet.sklist_lock);
4702	INIT_HLIST_HEAD(&net->packet.sklist);
4703
4704#ifdef CONFIG_PROC_FS
4705	if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4706			sizeof(struct seq_net_private)))
4707		return -ENOMEM;
4708#endif /* CONFIG_PROC_FS */
4709
4710	return 0;
4711}
4712
4713static void __net_exit packet_net_exit(struct net *net)
4714{
4715	remove_proc_entry("packet", net->proc_net);
4716	WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4717}
4718
4719static struct pernet_operations packet_net_ops = {
4720	.init = packet_net_init,
4721	.exit = packet_net_exit,
4722};
4723
4724
4725static void __exit packet_exit(void)
4726{
 
 
4727	sock_unregister(PF_PACKET);
4728	proto_unregister(&packet_proto);
4729	unregister_netdevice_notifier(&packet_netdev_notifier);
4730	unregister_pernet_subsys(&packet_net_ops);
4731}
4732
4733static int __init packet_init(void)
4734{
4735	int rc;
4736
 
 
 
 
 
 
4737	rc = register_pernet_subsys(&packet_net_ops);
4738	if (rc)
4739		goto out;
4740	rc = register_netdevice_notifier(&packet_netdev_notifier);
4741	if (rc)
4742		goto out_pernet;
4743	rc = proto_register(&packet_proto, 0);
4744	if (rc)
4745		goto out_notifier;
4746	rc = sock_register(&packet_family_ops);
4747	if (rc)
4748		goto out_proto;
4749
4750	return 0;
4751
 
 
 
 
4752out_proto:
4753	proto_unregister(&packet_proto);
4754out_notifier:
4755	unregister_netdevice_notifier(&packet_netdev_notifier);
4756out_pernet:
4757	unregister_pernet_subsys(&packet_net_ops);
4758out:
4759	return rc;
4760}
4761
4762module_init(packet_init);
4763module_exit(packet_exit);
4764MODULE_LICENSE("GPL");
4765MODULE_ALIAS_NETPROTO(PF_PACKET);