Linux Audio

Check our new training course

Loading...
v4.6
 
   1/*
   2 * INET		An implementation of the TCP/IP protocol suite for the LINUX
   3 *		operating system.  INET is implemented using the  BSD Socket
   4 *		interface as the means of communication with the user level.
   5 *
   6 *		PACKET - implements raw packet sockets.
   7 *
   8 * Authors:	Ross Biro
   9 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10 *		Alan Cox, <gw4pts@gw4pts.ampr.org>
  11 *
  12 * Fixes:
  13 *		Alan Cox	:	verify_area() now used correctly
  14 *		Alan Cox	:	new skbuff lists, look ma no backlogs!
  15 *		Alan Cox	:	tidied skbuff lists.
  16 *		Alan Cox	:	Now uses generic datagram routines I
  17 *					added. Also fixed the peek/read crash
  18 *					from all old Linux datagram code.
  19 *		Alan Cox	:	Uses the improved datagram code.
  20 *		Alan Cox	:	Added NULL's for socket options.
  21 *		Alan Cox	:	Re-commented the code.
  22 *		Alan Cox	:	Use new kernel side addressing
  23 *		Rob Janssen	:	Correct MTU usage.
  24 *		Dave Platt	:	Counter leaks caused by incorrect
  25 *					interrupt locking and some slightly
  26 *					dubious gcc output. Can you read
  27 *					compiler: it said _VOLATILE_
  28 *	Richard Kooijman	:	Timestamp fixes.
  29 *		Alan Cox	:	New buffers. Use sk->mac.raw.
  30 *		Alan Cox	:	sendmsg/recvmsg support.
  31 *		Alan Cox	:	Protocol setting support
  32 *	Alexey Kuznetsov	:	Untied from IPv4 stack.
  33 *	Cyrus Durgin		:	Fixed kerneld for kmod.
  34 *	Michal Ostrowski        :       Module initialization cleanup.
  35 *         Ulises Alonso        :       Frame number limit removal and
  36 *                                      packet_set_ring memory leak.
  37 *		Eric Biederman	:	Allow for > 8 byte hardware addresses.
  38 *					The convention is that longer addresses
  39 *					will simply extend the hardware address
  40 *					byte arrays at the end of sockaddr_ll
  41 *					and packet_mreq.
  42 *		Johann Baudy	:	Added TX RING.
  43 *		Chetan Loke	:	Implemented TPACKET_V3 block abstraction
  44 *					layer.
  45 *					Copyright (C) 2011, <lokec@ccs.neu.edu>
  46 *
  47 *
  48 *		This program is free software; you can redistribute it and/or
  49 *		modify it under the terms of the GNU General Public License
  50 *		as published by the Free Software Foundation; either version
  51 *		2 of the License, or (at your option) any later version.
  52 *
  53 */
  54
 
 
 
 
  55#include <linux/types.h>
  56#include <linux/mm.h>
  57#include <linux/capability.h>
  58#include <linux/fcntl.h>
  59#include <linux/socket.h>
  60#include <linux/in.h>
  61#include <linux/inet.h>
  62#include <linux/netdevice.h>
  63#include <linux/if_packet.h>
  64#include <linux/wireless.h>
  65#include <linux/kernel.h>
  66#include <linux/kmod.h>
  67#include <linux/slab.h>
  68#include <linux/vmalloc.h>
  69#include <net/net_namespace.h>
  70#include <net/ip.h>
  71#include <net/protocol.h>
  72#include <linux/skbuff.h>
  73#include <net/sock.h>
  74#include <linux/errno.h>
  75#include <linux/timer.h>
  76#include <asm/uaccess.h>
  77#include <asm/ioctls.h>
  78#include <asm/page.h>
  79#include <asm/cacheflush.h>
  80#include <asm/io.h>
  81#include <linux/proc_fs.h>
  82#include <linux/seq_file.h>
  83#include <linux/poll.h>
  84#include <linux/module.h>
  85#include <linux/init.h>
  86#include <linux/mutex.h>
  87#include <linux/if_vlan.h>
  88#include <linux/virtio_net.h>
  89#include <linux/errqueue.h>
  90#include <linux/net_tstamp.h>
  91#include <linux/percpu.h>
  92#ifdef CONFIG_INET
  93#include <net/inet_common.h>
  94#endif
  95#include <linux/bpf.h>
 
 
  96
  97#include "internal.h"
  98
  99/*
 100   Assumptions:
 101   - if device has no dev->hard_header routine, it adds and removes ll header
 102     inside itself. In this case ll header is invisible outside of device,
 103     but higher levels still should reserve dev->hard_header_len.
 104     Some devices are enough clever to reallocate skb, when header
 105     will not fit to reserved space (tunnel), another ones are silly
 106     (PPP).
 
 
 
 107   - packet socket receives packets with pulled ll header,
 108     so that SOCK_RAW should push it back.
 109
 110On receive:
 111-----------
 112
 113Incoming, dev->hard_header!=NULL
 114   mac_header -> ll header
 115   data       -> data
 116
 117Outgoing, dev->hard_header!=NULL
 118   mac_header -> ll header
 119   data       -> ll header
 120
 121Incoming, dev->hard_header==NULL
 122   mac_header -> UNKNOWN position. It is very likely, that it points to ll
 123		 header.  PPP makes it, that is wrong, because introduce
 124		 assymetry between rx and tx paths.
 125   data       -> data
 126
 127Outgoing, dev->hard_header==NULL
 128   mac_header -> data. ll header is still not built!
 129   data       -> data
 130
 131Resume
 132  If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 
 133
 134
 135On transmit:
 136------------
 137
 138dev->hard_header != NULL
 139   mac_header -> ll header
 140   data       -> ll header
 141
 142dev->hard_header == NULL (ll header is added by device, we cannot control it)
 143   mac_header -> data
 144   data       -> data
 145
 146   We should set nh.raw on output to correct posistion,
 147   packet classifier depends on it.
 148 */
 149
 150/* Private packet socket structures. */
 151
 152/* identical to struct packet_mreq except it has
 153 * a longer address field.
 154 */
 155struct packet_mreq_max {
 156	int		mr_ifindex;
 157	unsigned short	mr_type;
 158	unsigned short	mr_alen;
 159	unsigned char	mr_address[MAX_ADDR_LEN];
 160};
 161
 162union tpacket_uhdr {
 163	struct tpacket_hdr  *h1;
 164	struct tpacket2_hdr *h2;
 165	struct tpacket3_hdr *h3;
 166	void *raw;
 167};
 168
 169static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 170		int closing, int tx_ring);
 171
 172#define V3_ALIGNMENT	(8)
 173
 174#define BLK_HDR_LEN	(ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 175
 176#define BLK_PLUS_PRIV(sz_of_priv) \
 177	(BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 178
 179#define PGV_FROM_VMALLOC 1
 180
 181#define BLOCK_STATUS(x)	((x)->hdr.bh1.block_status)
 182#define BLOCK_NUM_PKTS(x)	((x)->hdr.bh1.num_pkts)
 183#define BLOCK_O2FP(x)		((x)->hdr.bh1.offset_to_first_pkt)
 184#define BLOCK_LEN(x)		((x)->hdr.bh1.blk_len)
 185#define BLOCK_SNUM(x)		((x)->hdr.bh1.seq_num)
 186#define BLOCK_O2PRIV(x)	((x)->offset_to_priv)
 187#define BLOCK_PRIV(x)		((void *)((char *)(x) + BLOCK_O2PRIV(x)))
 188
 189struct packet_sock;
 190static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
 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_kbdq_core *,
 199			struct tpacket_block_desc *);
 200static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 201			struct packet_sock *);
 202static void prb_retire_current_block(struct tpacket_kbdq_core *,
 203		struct packet_sock *, unsigned int status);
 204static int prb_queue_frozen(struct tpacket_kbdq_core *);
 205static void prb_open_block(struct tpacket_kbdq_core *,
 206		struct tpacket_block_desc *);
 207static void prb_retire_rx_blk_timer_expired(unsigned long);
 208static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 209static void prb_init_blk_timer(struct packet_sock *,
 210		struct tpacket_kbdq_core *,
 211		void (*func) (unsigned long));
 212static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 213static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 214		struct tpacket3_hdr *);
 215static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 216		struct tpacket3_hdr *);
 217static void packet_flush_mclist(struct sock *sk);
 
 218
 219struct packet_skb_cb {
 220	union {
 221		struct sockaddr_pkt pkt;
 222		union {
 223			/* Trick: alias skb original length with
 224			 * ll.sll_family and ll.protocol in order
 225			 * to save room.
 226			 */
 227			unsigned int origlen;
 228			struct sockaddr_ll ll;
 229		};
 230	} sa;
 231};
 232
 233#define vio_le() virtio_legacy_is_little_endian()
 234
 235#define PACKET_SKB_CB(__skb)	((struct packet_skb_cb *)((__skb)->cb))
 236
 237#define GET_PBDQC_FROM_RB(x)	((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 238#define GET_PBLOCK_DESC(x, bid)	\
 239	((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 240#define GET_CURR_PBLOCK_DESC_FROM_CORE(x)	\
 241	((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 242#define GET_NEXT_PRB_BLK_NUM(x) \
 243	(((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 244	((x)->kactive_blk_num+1) : 0)
 245
 246static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 247static void __fanout_link(struct sock *sk, struct packet_sock *po);
 248
 249static int packet_direct_xmit(struct sk_buff *skb)
 
 250{
 251	struct net_device *dev = skb->dev;
 252	netdev_features_t features;
 253	struct netdev_queue *txq;
 254	int ret = NETDEV_TX_BUSY;
 255
 256	if (unlikely(!netif_running(dev) ||
 257		     !netif_carrier_ok(dev)))
 258		goto drop;
 259
 260	features = netif_skb_features(skb);
 261	if (skb_needs_linearize(skb, features) &&
 262	    __skb_linearize(skb))
 263		goto drop;
 264
 265	txq = skb_get_tx_queue(dev, skb);
 
 266
 267	local_bh_disable();
 
 
 
 268
 269	HARD_TX_LOCK(dev, txq, smp_processor_id());
 270	if (!netif_xmit_frozen_or_drv_stopped(txq))
 271		ret = netdev_start_xmit(skb, dev, txq, false);
 272	HARD_TX_UNLOCK(dev, txq);
 273
 274	local_bh_enable();
 
 
 275
 276	if (!dev_xmit_complete(ret))
 277		kfree_skb(skb);
 
 
 278
 279	return ret;
 280drop:
 281	atomic_long_inc(&dev->tx_dropped);
 282	kfree_skb(skb);
 283	return NET_XMIT_DROP;
 
 
 
 284}
 285
 286static struct net_device *packet_cached_dev_get(struct packet_sock *po)
 287{
 288	struct net_device *dev;
 289
 290	rcu_read_lock();
 291	dev = rcu_dereference(po->cached_dev);
 292	if (likely(dev))
 293		dev_hold(dev);
 294	rcu_read_unlock();
 295
 296	return dev;
 297}
 298
 299static void packet_cached_dev_assign(struct packet_sock *po,
 300				     struct net_device *dev)
 301{
 302	rcu_assign_pointer(po->cached_dev, dev);
 303}
 304
 305static void packet_cached_dev_reset(struct packet_sock *po)
 306{
 307	RCU_INIT_POINTER(po->cached_dev, NULL);
 308}
 309
 310static bool packet_use_direct_xmit(const struct packet_sock *po)
 311{
 312	return po->xmit == packet_direct_xmit;
 313}
 314
 315static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
 316{
 317	return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
 318}
 319
 320static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
 321{
 
 322	const struct net_device_ops *ops = dev->netdev_ops;
 
 323	u16 queue_index;
 324
 
 
 
 
 325	if (ops->ndo_select_queue) {
 326		queue_index = ops->ndo_select_queue(dev, skb, NULL,
 327						    __packet_pick_tx_queue);
 328		queue_index = netdev_cap_txqueue(dev, queue_index);
 329	} else {
 330		queue_index = __packet_pick_tx_queue(dev, skb);
 331	}
 332
 333	skb_set_queue_mapping(skb, queue_index);
 334}
 335
 336/* register_prot_hook must be invoked with the po->bind_lock held,
 337 * or from a context in which asynchronous accesses to the packet
 338 * socket is not possible (packet_create()).
 339 */
 340static void register_prot_hook(struct sock *sk)
 341{
 342	struct packet_sock *po = pkt_sk(sk);
 343
 344	if (!po->running) {
 345		if (po->fanout)
 346			__fanout_link(sk, po);
 347		else
 348			dev_add_pack(&po->prot_hook);
 349
 350		sock_hold(sk);
 351		po->running = 1;
 352	}
 353}
 354
 355/* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
 356 * held.   If the sync parameter is true, we will temporarily drop
 
 
 
 
 
 357 * the po->bind_lock and do a synchronize_net to make sure no
 358 * asynchronous packet processing paths still refer to the elements
 359 * of po->prot_hook.  If the sync parameter is false, it is the
 360 * callers responsibility to take care of this.
 361 */
 362static void __unregister_prot_hook(struct sock *sk, bool sync)
 363{
 364	struct packet_sock *po = pkt_sk(sk);
 365
 366	po->running = 0;
 
 
 367
 368	if (po->fanout)
 369		__fanout_unlink(sk, po);
 370	else
 371		__dev_remove_pack(&po->prot_hook);
 372
 373	__sock_put(sk);
 374
 375	if (sync) {
 376		spin_unlock(&po->bind_lock);
 377		synchronize_net();
 378		spin_lock(&po->bind_lock);
 379	}
 380}
 381
 382static void unregister_prot_hook(struct sock *sk, bool sync)
 383{
 384	struct packet_sock *po = pkt_sk(sk);
 385
 386	if (po->running)
 387		__unregister_prot_hook(sk, sync);
 388}
 389
 390static inline struct page * __pure pgv_to_page(void *addr)
 391{
 392	if (is_vmalloc_addr(addr))
 393		return vmalloc_to_page(addr);
 394	return virt_to_page(addr);
 395}
 396
 397static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 398{
 399	union tpacket_uhdr h;
 400
 
 
 401	h.raw = frame;
 402	switch (po->tp_version) {
 403	case TPACKET_V1:
 404		h.h1->tp_status = status;
 405		flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 406		break;
 407	case TPACKET_V2:
 408		h.h2->tp_status = status;
 409		flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 410		break;
 411	case TPACKET_V3:
 
 
 
 412	default:
 413		WARN(1, "TPACKET version not supported.\n");
 414		BUG();
 415	}
 416
 417	smp_wmb();
 418}
 419
 420static int __packet_get_status(struct packet_sock *po, void *frame)
 421{
 422	union tpacket_uhdr h;
 423
 424	smp_rmb();
 425
 
 
 426	h.raw = frame;
 427	switch (po->tp_version) {
 428	case TPACKET_V1:
 429		flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 430		return h.h1->tp_status;
 431	case TPACKET_V2:
 432		flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 433		return h.h2->tp_status;
 434	case TPACKET_V3:
 
 
 435	default:
 436		WARN(1, "TPACKET version not supported.\n");
 437		BUG();
 438		return 0;
 439	}
 440}
 441
 442static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
 443				   unsigned int flags)
 444{
 445	struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 446
 447	if (shhwtstamps &&
 448	    (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
 449	    ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
 450		return TP_STATUS_TS_RAW_HARDWARE;
 451
 452	if (ktime_to_timespec_cond(skb->tstamp, ts))
 
 453		return TP_STATUS_TS_SOFTWARE;
 454
 455	return 0;
 456}
 457
 458static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
 459				    struct sk_buff *skb)
 460{
 461	union tpacket_uhdr h;
 462	struct timespec ts;
 463	__u32 ts_status;
 464
 465	if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
 466		return 0;
 467
 468	h.raw = frame;
 
 
 
 
 
 
 
 469	switch (po->tp_version) {
 470	case TPACKET_V1:
 471		h.h1->tp_sec = ts.tv_sec;
 472		h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
 473		break;
 474	case TPACKET_V2:
 475		h.h2->tp_sec = ts.tv_sec;
 476		h.h2->tp_nsec = ts.tv_nsec;
 477		break;
 478	case TPACKET_V3:
 
 
 
 479	default:
 480		WARN(1, "TPACKET version not supported.\n");
 481		BUG();
 482	}
 483
 484	/* one flush is safe, as both fields always lie on the same cacheline */
 485	flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
 486	smp_wmb();
 487
 488	return ts_status;
 489}
 490
 491static void *packet_lookup_frame(struct packet_sock *po,
 492		struct packet_ring_buffer *rb,
 493		unsigned int position,
 494		int status)
 495{
 496	unsigned int pg_vec_pos, frame_offset;
 497	union tpacket_uhdr h;
 498
 499	pg_vec_pos = position / rb->frames_per_block;
 500	frame_offset = position % rb->frames_per_block;
 501
 502	h.raw = rb->pg_vec[pg_vec_pos].buffer +
 503		(frame_offset * rb->frame_size);
 504
 505	if (status != __packet_get_status(po, h.raw))
 506		return NULL;
 507
 508	return h.raw;
 509}
 510
 511static void *packet_current_frame(struct packet_sock *po,
 512		struct packet_ring_buffer *rb,
 513		int status)
 514{
 515	return packet_lookup_frame(po, rb, rb->head, status);
 516}
 517
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 518static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 519{
 520	del_timer_sync(&pkc->retire_blk_timer);
 521}
 522
 523static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 524		struct sk_buff_head *rb_queue)
 525{
 526	struct tpacket_kbdq_core *pkc;
 527
 528	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 529
 530	spin_lock_bh(&rb_queue->lock);
 531	pkc->delete_blk_timer = 1;
 532	spin_unlock_bh(&rb_queue->lock);
 533
 534	prb_del_retire_blk_timer(pkc);
 535}
 536
 537static void prb_init_blk_timer(struct packet_sock *po,
 538		struct tpacket_kbdq_core *pkc,
 539		void (*func) (unsigned long))
 540{
 541	init_timer(&pkc->retire_blk_timer);
 542	pkc->retire_blk_timer.data = (long)po;
 543	pkc->retire_blk_timer.function = func;
 544	pkc->retire_blk_timer.expires = jiffies;
 545}
 546
 547static void prb_setup_retire_blk_timer(struct packet_sock *po)
 548{
 549	struct tpacket_kbdq_core *pkc;
 550
 551	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 552	prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
 
 
 553}
 554
 555static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 556				int blk_size_in_bytes)
 557{
 558	struct net_device *dev;
 559	unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 560	struct ethtool_link_ksettings ecmd;
 561	int err;
 562
 563	rtnl_lock();
 564	dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 565	if (unlikely(!dev)) {
 566		rtnl_unlock();
 567		return DEFAULT_PRB_RETIRE_TOV;
 568	}
 569	err = __ethtool_get_link_ksettings(dev, &ecmd);
 570	rtnl_unlock();
 571	if (!err) {
 572		/*
 573		 * If the link speed is so slow you don't really
 574		 * need to worry about perf anyways
 575		 */
 576		if (ecmd.base.speed < SPEED_1000 ||
 577		    ecmd.base.speed == SPEED_UNKNOWN) {
 578			return DEFAULT_PRB_RETIRE_TOV;
 579		} else {
 580			msec = 1;
 581			div = ecmd.base.speed / 1000;
 582		}
 583	}
 584
 
 585	mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 586
 587	if (div)
 588		mbits /= div;
 589
 590	tmo = mbits * msec;
 591
 592	if (div)
 593		return tmo+1;
 594	return tmo;
 595}
 596
 597static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 598			union tpacket_req_u *req_u)
 599{
 600	p1->feature_req_word = req_u->req3.tp_feature_req_word;
 601}
 602
 603static void init_prb_bdqc(struct packet_sock *po,
 604			struct packet_ring_buffer *rb,
 605			struct pgv *pg_vec,
 606			union tpacket_req_u *req_u)
 607{
 608	struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
 609	struct tpacket_block_desc *pbd;
 610
 611	memset(p1, 0x0, sizeof(*p1));
 612
 613	p1->knxt_seq_num = 1;
 614	p1->pkbdq = pg_vec;
 615	pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 616	p1->pkblk_start	= pg_vec[0].buffer;
 617	p1->kblk_size = req_u->req3.tp_block_size;
 618	p1->knum_blocks	= req_u->req3.tp_block_nr;
 619	p1->hdrlen = po->tp_hdrlen;
 620	p1->version = po->tp_version;
 621	p1->last_kactive_blk_num = 0;
 622	po->stats.stats3.tp_freeze_q_cnt = 0;
 623	if (req_u->req3.tp_retire_blk_tov)
 624		p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 625	else
 626		p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 627						req_u->req3.tp_block_size);
 628	p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 629	p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 
 630
 631	p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
 632	prb_init_ft_ops(p1, req_u);
 633	prb_setup_retire_blk_timer(po);
 634	prb_open_block(p1, pbd);
 635}
 636
 637/*  Do NOT update the last_blk_num first.
 638 *  Assumes sk_buff_head lock is held.
 639 */
 640static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 641{
 642	mod_timer(&pkc->retire_blk_timer,
 643			jiffies + pkc->tov_in_jiffies);
 644	pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 645}
 646
 647/*
 648 * Timer logic:
 649 * 1) We refresh the timer only when we open a block.
 650 *    By doing this we don't waste cycles refreshing the timer
 651 *	  on packet-by-packet basis.
 652 *
 653 * With a 1MB block-size, on a 1Gbps line, it will take
 654 * i) ~8 ms to fill a block + ii) memcpy etc.
 655 * In this cut we are not accounting for the memcpy time.
 656 *
 657 * So, if the user sets the 'tmo' to 10ms then the timer
 658 * will never fire while the block is still getting filled
 659 * (which is what we want). However, the user could choose
 660 * to close a block early and that's fine.
 661 *
 662 * But when the timer does fire, we check whether or not to refresh it.
 663 * Since the tmo granularity is in msecs, it is not too expensive
 664 * to refresh the timer, lets say every '8' msecs.
 665 * Either the user can set the 'tmo' or we can derive it based on
 666 * a) line-speed and b) block-size.
 667 * prb_calc_retire_blk_tmo() calculates the tmo.
 668 *
 669 */
 670static void prb_retire_rx_blk_timer_expired(unsigned long data)
 671{
 672	struct packet_sock *po = (struct packet_sock *)data;
 
 673	struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 674	unsigned int frozen;
 675	struct tpacket_block_desc *pbd;
 676
 677	spin_lock(&po->sk.sk_receive_queue.lock);
 678
 679	frozen = prb_queue_frozen(pkc);
 680	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 681
 682	if (unlikely(pkc->delete_blk_timer))
 683		goto out;
 684
 685	/* We only need to plug the race when the block is partially filled.
 686	 * tpacket_rcv:
 687	 *		lock(); increment BLOCK_NUM_PKTS; unlock()
 688	 *		copy_bits() is in progress ...
 689	 *		timer fires on other cpu:
 690	 *		we can't retire the current block because copy_bits
 691	 *		is in progress.
 692	 *
 693	 */
 694	if (BLOCK_NUM_PKTS(pbd)) {
 695		while (atomic_read(&pkc->blk_fill_in_prog)) {
 696			/* Waiting for skb_copy_bits to finish... */
 697			cpu_relax();
 698		}
 699	}
 700
 701	if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 702		if (!frozen) {
 703			if (!BLOCK_NUM_PKTS(pbd)) {
 704				/* An empty block. Just refresh the timer. */
 705				goto refresh_timer;
 706			}
 707			prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 708			if (!prb_dispatch_next_block(pkc, po))
 709				goto refresh_timer;
 710			else
 711				goto out;
 712		} else {
 713			/* Case 1. Queue was frozen because user-space was
 714			 *	   lagging behind.
 715			 */
 716			if (prb_curr_blk_in_use(pkc, pbd)) {
 717				/*
 718				 * Ok, user-space is still behind.
 719				 * So just refresh the timer.
 720				 */
 721				goto refresh_timer;
 722			} else {
 723			       /* Case 2. queue was frozen,user-space caught up,
 724				* now the link went idle && the timer fired.
 725				* We don't have a block to close.So we open this
 726				* block and restart the timer.
 727				* opening a block thaws the queue,restarts timer
 728				* Thawing/timer-refresh is a side effect.
 729				*/
 730				prb_open_block(pkc, pbd);
 731				goto out;
 732			}
 733		}
 734	}
 735
 736refresh_timer:
 737	_prb_refresh_rx_retire_blk_timer(pkc);
 738
 739out:
 740	spin_unlock(&po->sk.sk_receive_queue.lock);
 741}
 742
 743static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 744		struct tpacket_block_desc *pbd1, __u32 status)
 745{
 746	/* Flush everything minus the block header */
 747
 748#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 749	u8 *start, *end;
 750
 751	start = (u8 *)pbd1;
 752
 753	/* Skip the block header(we know header WILL fit in 4K) */
 754	start += PAGE_SIZE;
 755
 756	end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 757	for (; start < end; start += PAGE_SIZE)
 758		flush_dcache_page(pgv_to_page(start));
 759
 760	smp_wmb();
 761#endif
 762
 763	/* Now update the block status. */
 764
 765	BLOCK_STATUS(pbd1) = status;
 766
 767	/* Flush the block header */
 768
 769#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 770	start = (u8 *)pbd1;
 771	flush_dcache_page(pgv_to_page(start));
 772
 773	smp_wmb();
 774#endif
 775}
 776
 777/*
 778 * Side effect:
 779 *
 780 * 1) flush the block
 781 * 2) Increment active_blk_num
 782 *
 783 * Note:We DONT refresh the timer on purpose.
 784 *	Because almost always the next block will be opened.
 785 */
 786static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 787		struct tpacket_block_desc *pbd1,
 788		struct packet_sock *po, unsigned int stat)
 789{
 790	__u32 status = TP_STATUS_USER | stat;
 791
 792	struct tpacket3_hdr *last_pkt;
 793	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 794	struct sock *sk = &po->sk;
 795
 796	if (po->stats.stats3.tp_drops)
 797		status |= TP_STATUS_LOSING;
 798
 799	last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 800	last_pkt->tp_next_offset = 0;
 801
 802	/* Get the ts of the last pkt */
 803	if (BLOCK_NUM_PKTS(pbd1)) {
 804		h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 805		h1->ts_last_pkt.ts_nsec	= last_pkt->tp_nsec;
 806	} else {
 807		/* Ok, we tmo'd - so get the current time.
 808		 *
 809		 * It shouldn't really happen as we don't close empty
 810		 * blocks. See prb_retire_rx_blk_timer_expired().
 811		 */
 812		struct timespec ts;
 813		getnstimeofday(&ts);
 814		h1->ts_last_pkt.ts_sec = ts.tv_sec;
 815		h1->ts_last_pkt.ts_nsec	= ts.tv_nsec;
 816	}
 817
 818	smp_wmb();
 819
 820	/* Flush the block */
 821	prb_flush_block(pkc1, pbd1, status);
 822
 823	sk->sk_data_ready(sk);
 824
 825	pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 826}
 827
 828static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 829{
 830	pkc->reset_pending_on_curr_blk = 0;
 831}
 832
 833/*
 834 * Side effect of opening a block:
 835 *
 836 * 1) prb_queue is thawed.
 837 * 2) retire_blk_timer is refreshed.
 838 *
 839 */
 840static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 841	struct tpacket_block_desc *pbd1)
 842{
 843	struct timespec ts;
 844	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 845
 846	smp_rmb();
 847
 848	/* We could have just memset this but we will lose the
 849	 * flexibility of making the priv area sticky
 850	 */
 851
 852	BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 853	BLOCK_NUM_PKTS(pbd1) = 0;
 854	BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 855
 856	getnstimeofday(&ts);
 857
 858	h1->ts_first_pkt.ts_sec = ts.tv_sec;
 859	h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 860
 861	pkc1->pkblk_start = (char *)pbd1;
 862	pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 863
 864	BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 865	BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 866
 867	pbd1->version = pkc1->version;
 868	pkc1->prev = pkc1->nxt_offset;
 869	pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 870
 871	prb_thaw_queue(pkc1);
 872	_prb_refresh_rx_retire_blk_timer(pkc1);
 873
 874	smp_wmb();
 875}
 876
 877/*
 878 * Queue freeze logic:
 879 * 1) Assume tp_block_nr = 8 blocks.
 880 * 2) At time 't0', user opens Rx ring.
 881 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 882 * 4) user-space is either sleeping or processing block '0'.
 883 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 884 *    it will close block-7,loop around and try to fill block '0'.
 885 *    call-flow:
 886 *    __packet_lookup_frame_in_block
 887 *      prb_retire_current_block()
 888 *      prb_dispatch_next_block()
 889 *        |->(BLOCK_STATUS == USER) evaluates to true
 890 *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 891 * 6) Now there are two cases:
 892 *    6.1) Link goes idle right after the queue is frozen.
 893 *         But remember, the last open_block() refreshed the timer.
 894 *         When this timer expires,it will refresh itself so that we can
 895 *         re-open block-0 in near future.
 896 *    6.2) Link is busy and keeps on receiving packets. This is a simple
 897 *         case and __packet_lookup_frame_in_block will check if block-0
 898 *         is free and can now be re-used.
 899 */
 900static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 901				  struct packet_sock *po)
 902{
 903	pkc->reset_pending_on_curr_blk = 1;
 904	po->stats.stats3.tp_freeze_q_cnt++;
 905}
 906
 907#define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 908
 909/*
 910 * If the next block is free then we will dispatch it
 911 * and return a good offset.
 912 * Else, we will freeze the queue.
 913 * So, caller must check the return value.
 914 */
 915static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 916		struct packet_sock *po)
 917{
 918	struct tpacket_block_desc *pbd;
 919
 920	smp_rmb();
 921
 922	/* 1. Get current block num */
 923	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 924
 925	/* 2. If this block is currently in_use then freeze the queue */
 926	if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 927		prb_freeze_queue(pkc, po);
 928		return NULL;
 929	}
 930
 931	/*
 932	 * 3.
 933	 * open this block and return the offset where the first packet
 934	 * needs to get stored.
 935	 */
 936	prb_open_block(pkc, pbd);
 937	return (void *)pkc->nxt_offset;
 938}
 939
 940static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 941		struct packet_sock *po, unsigned int status)
 942{
 943	struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 944
 945	/* retire/close the current block */
 946	if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 947		/*
 948		 * Plug the case where copy_bits() is in progress on
 949		 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 950		 * have space to copy the pkt in the current block and
 951		 * called prb_retire_current_block()
 952		 *
 953		 * We don't need to worry about the TMO case because
 954		 * the timer-handler already handled this case.
 955		 */
 956		if (!(status & TP_STATUS_BLK_TMO)) {
 957			while (atomic_read(&pkc->blk_fill_in_prog)) {
 958				/* Waiting for skb_copy_bits to finish... */
 959				cpu_relax();
 960			}
 961		}
 962		prb_close_block(pkc, pbd, po, status);
 963		return;
 964	}
 965}
 966
 967static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
 968				      struct tpacket_block_desc *pbd)
 969{
 970	return TP_STATUS_USER & BLOCK_STATUS(pbd);
 971}
 972
 973static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 974{
 975	return pkc->reset_pending_on_curr_blk;
 976}
 977
 978static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 
 979{
 980	struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 981	atomic_dec(&pkc->blk_fill_in_prog);
 
 982}
 983
 984static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 985			struct tpacket3_hdr *ppd)
 986{
 987	ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
 988}
 989
 990static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 991			struct tpacket3_hdr *ppd)
 992{
 993	ppd->hv1.tp_rxhash = 0;
 994}
 995
 996static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
 997			struct tpacket3_hdr *ppd)
 998{
 
 
 999	if (skb_vlan_tag_present(pkc->skb)) {
1000		ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1001		ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1002		ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
 
 
 
 
1003	} else {
1004		ppd->hv1.tp_vlan_tci = 0;
1005		ppd->hv1.tp_vlan_tpid = 0;
1006		ppd->tp_status = TP_STATUS_AVAILABLE;
1007	}
1008}
1009
1010static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1011			struct tpacket3_hdr *ppd)
1012{
1013	ppd->hv1.tp_padding = 0;
1014	prb_fill_vlan_info(pkc, ppd);
1015
1016	if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1017		prb_fill_rxhash(pkc, ppd);
1018	else
1019		prb_clear_rxhash(pkc, ppd);
1020}
1021
1022static void prb_fill_curr_block(char *curr,
1023				struct tpacket_kbdq_core *pkc,
1024				struct tpacket_block_desc *pbd,
1025				unsigned int len)
 
1026{
1027	struct tpacket3_hdr *ppd;
1028
1029	ppd  = (struct tpacket3_hdr *)curr;
1030	ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1031	pkc->prev = curr;
1032	pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1033	BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1034	BLOCK_NUM_PKTS(pbd) += 1;
1035	atomic_inc(&pkc->blk_fill_in_prog);
1036	prb_run_all_ft_ops(pkc, ppd);
1037}
1038
1039/* Assumes caller has the sk->rx_queue.lock */
1040static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1041					    struct sk_buff *skb,
1042						int status,
1043					    unsigned int len
1044					    )
1045{
1046	struct tpacket_kbdq_core *pkc;
1047	struct tpacket_block_desc *pbd;
1048	char *curr, *end;
1049
1050	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1051	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1052
1053	/* Queue is frozen when user space is lagging behind */
1054	if (prb_queue_frozen(pkc)) {
1055		/*
1056		 * Check if that last block which caused the queue to freeze,
1057		 * is still in_use by user-space.
1058		 */
1059		if (prb_curr_blk_in_use(pkc, pbd)) {
1060			/* Can't record this packet */
1061			return NULL;
1062		} else {
1063			/*
1064			 * Ok, the block was released by user-space.
1065			 * Now let's open that block.
1066			 * opening a block also thaws the queue.
1067			 * Thawing is a side effect.
1068			 */
1069			prb_open_block(pkc, pbd);
1070		}
1071	}
1072
1073	smp_mb();
1074	curr = pkc->nxt_offset;
1075	pkc->skb = skb;
1076	end = (char *)pbd + pkc->kblk_size;
1077
1078	/* first try the current block */
1079	if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1080		prb_fill_curr_block(curr, pkc, pbd, len);
1081		return (void *)curr;
1082	}
1083
1084	/* Ok, close the current block */
1085	prb_retire_current_block(pkc, po, 0);
1086
1087	/* Now, try to dispatch the next block */
1088	curr = (char *)prb_dispatch_next_block(pkc, po);
1089	if (curr) {
1090		pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1091		prb_fill_curr_block(curr, pkc, pbd, len);
1092		return (void *)curr;
1093	}
1094
1095	/*
1096	 * No free blocks are available.user_space hasn't caught up yet.
1097	 * Queue was just frozen and now this packet will get dropped.
1098	 */
1099	return NULL;
1100}
1101
1102static void *packet_current_rx_frame(struct packet_sock *po,
1103					    struct sk_buff *skb,
1104					    int status, unsigned int len)
1105{
1106	char *curr = NULL;
1107	switch (po->tp_version) {
1108	case TPACKET_V1:
1109	case TPACKET_V2:
1110		curr = packet_lookup_frame(po, &po->rx_ring,
1111					po->rx_ring.head, status);
1112		return curr;
1113	case TPACKET_V3:
1114		return __packet_lookup_frame_in_block(po, skb, status, len);
1115	default:
1116		WARN(1, "TPACKET version not supported\n");
1117		BUG();
1118		return NULL;
1119	}
1120}
1121
1122static void *prb_lookup_block(struct packet_sock *po,
1123				     struct packet_ring_buffer *rb,
1124				     unsigned int idx,
1125				     int status)
1126{
1127	struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1128	struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1129
1130	if (status != BLOCK_STATUS(pbd))
1131		return NULL;
1132	return pbd;
1133}
1134
1135static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1136{
1137	unsigned int prev;
1138	if (rb->prb_bdqc.kactive_blk_num)
1139		prev = rb->prb_bdqc.kactive_blk_num-1;
1140	else
1141		prev = rb->prb_bdqc.knum_blocks-1;
1142	return prev;
1143}
1144
1145/* Assumes caller has held the rx_queue.lock */
1146static void *__prb_previous_block(struct packet_sock *po,
1147					 struct packet_ring_buffer *rb,
1148					 int status)
1149{
1150	unsigned int previous = prb_previous_blk_num(rb);
1151	return prb_lookup_block(po, rb, previous, status);
1152}
1153
1154static void *packet_previous_rx_frame(struct packet_sock *po,
1155					     struct packet_ring_buffer *rb,
1156					     int status)
1157{
1158	if (po->tp_version <= TPACKET_V2)
1159		return packet_previous_frame(po, rb, status);
1160
1161	return __prb_previous_block(po, rb, status);
1162}
1163
1164static void packet_increment_rx_head(struct packet_sock *po,
1165					    struct packet_ring_buffer *rb)
1166{
1167	switch (po->tp_version) {
1168	case TPACKET_V1:
1169	case TPACKET_V2:
1170		return packet_increment_head(rb);
1171	case TPACKET_V3:
1172	default:
1173		WARN(1, "TPACKET version not supported.\n");
1174		BUG();
1175		return;
1176	}
1177}
1178
1179static void *packet_previous_frame(struct packet_sock *po,
1180		struct packet_ring_buffer *rb,
1181		int status)
1182{
1183	unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1184	return packet_lookup_frame(po, rb, previous, status);
1185}
1186
1187static void packet_increment_head(struct packet_ring_buffer *buff)
1188{
1189	buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1190}
1191
1192static void packet_inc_pending(struct packet_ring_buffer *rb)
1193{
1194	this_cpu_inc(*rb->pending_refcnt);
1195}
1196
1197static void packet_dec_pending(struct packet_ring_buffer *rb)
1198{
1199	this_cpu_dec(*rb->pending_refcnt);
1200}
1201
1202static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1203{
1204	unsigned int refcnt = 0;
1205	int cpu;
1206
1207	/* We don't use pending refcount in rx_ring. */
1208	if (rb->pending_refcnt == NULL)
1209		return 0;
1210
1211	for_each_possible_cpu(cpu)
1212		refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1213
1214	return refcnt;
1215}
1216
1217static int packet_alloc_pending(struct packet_sock *po)
1218{
1219	po->rx_ring.pending_refcnt = NULL;
1220
1221	po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1222	if (unlikely(po->tx_ring.pending_refcnt == NULL))
1223		return -ENOBUFS;
1224
1225	return 0;
1226}
1227
1228static void packet_free_pending(struct packet_sock *po)
1229{
1230	free_percpu(po->tx_ring.pending_refcnt);
1231}
1232
1233#define ROOM_POW_OFF	2
1234#define ROOM_NONE	0x0
1235#define ROOM_LOW	0x1
1236#define ROOM_NORMAL	0x2
1237
1238static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1239{
1240	int idx, len;
1241
1242	len = po->rx_ring.frame_max + 1;
1243	idx = po->rx_ring.head;
1244	if (pow_off)
1245		idx += len >> pow_off;
1246	if (idx >= len)
1247		idx -= len;
1248	return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1249}
1250
1251static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1252{
1253	int idx, len;
1254
1255	len = po->rx_ring.prb_bdqc.knum_blocks;
1256	idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1257	if (pow_off)
1258		idx += len >> pow_off;
1259	if (idx >= len)
1260		idx -= len;
1261	return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1262}
1263
1264static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
 
1265{
1266	struct sock *sk = &po->sk;
1267	int ret = ROOM_NONE;
1268
1269	if (po->prot_hook.func != tpacket_rcv) {
1270		int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1271					  - (skb ? skb->truesize : 0);
1272		if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
 
 
1273			return ROOM_NORMAL;
1274		else if (avail > 0)
1275			return ROOM_LOW;
1276		else
1277			return ROOM_NONE;
1278	}
1279
1280	if (po->tp_version == TPACKET_V3) {
1281		if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1282			ret = ROOM_NORMAL;
1283		else if (__tpacket_v3_has_room(po, 0))
1284			ret = ROOM_LOW;
1285	} else {
1286		if (__tpacket_has_room(po, ROOM_POW_OFF))
1287			ret = ROOM_NORMAL;
1288		else if (__tpacket_has_room(po, 0))
1289			ret = ROOM_LOW;
1290	}
1291
1292	return ret;
1293}
1294
1295static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1296{
 
1297	int ret;
1298	bool has_room;
1299
1300	spin_lock_bh(&po->sk.sk_receive_queue.lock);
1301	ret = __packet_rcv_has_room(po, skb);
1302	has_room = ret == ROOM_NORMAL;
1303	if (po->pressure == has_room)
1304		po->pressure = !has_room;
1305	spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1306
1307	return ret;
1308}
1309
 
 
 
 
 
 
 
1310static void packet_sock_destruct(struct sock *sk)
1311{
1312	skb_queue_purge(&sk->sk_error_queue);
1313
1314	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1315	WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1316
1317	if (!sock_flag(sk, SOCK_DEAD)) {
1318		pr_err("Attempt to release alive packet socket: %p\n", sk);
1319		return;
1320	}
1321
1322	sk_refcnt_debug_dec(sk);
1323}
1324
1325static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1326{
1327	u32 rxhash;
 
1328	int i, count = 0;
1329
1330	rxhash = skb_get_hash(skb);
1331	for (i = 0; i < ROLLOVER_HLEN; i++)
1332		if (po->rollover->history[i] == rxhash)
1333			count++;
1334
1335	po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
 
 
 
 
 
1336	return count > (ROLLOVER_HLEN >> 1);
1337}
1338
1339static unsigned int fanout_demux_hash(struct packet_fanout *f,
1340				      struct sk_buff *skb,
1341				      unsigned int num)
1342{
1343	return reciprocal_scale(skb_get_hash(skb), num);
1344}
1345
1346static unsigned int fanout_demux_lb(struct packet_fanout *f,
1347				    struct sk_buff *skb,
1348				    unsigned int num)
1349{
1350	unsigned int val = atomic_inc_return(&f->rr_cur);
1351
1352	return val % num;
1353}
1354
1355static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1356				     struct sk_buff *skb,
1357				     unsigned int num)
1358{
1359	return smp_processor_id() % num;
1360}
1361
1362static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1363				     struct sk_buff *skb,
1364				     unsigned int num)
1365{
1366	return prandom_u32_max(num);
1367}
1368
1369static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1370					  struct sk_buff *skb,
1371					  unsigned int idx, bool try_self,
1372					  unsigned int num)
1373{
1374	struct packet_sock *po, *po_next, *po_skip = NULL;
1375	unsigned int i, j, room = ROOM_NONE;
1376
1377	po = pkt_sk(f->arr[idx]);
1378
1379	if (try_self) {
1380		room = packet_rcv_has_room(po, skb);
1381		if (room == ROOM_NORMAL ||
1382		    (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1383			return idx;
1384		po_skip = po;
1385	}
1386
1387	i = j = min_t(int, po->rollover->sock, num - 1);
1388	do {
1389		po_next = pkt_sk(f->arr[i]);
1390		if (po_next != po_skip && !po_next->pressure &&
 
1391		    packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1392			if (i != j)
1393				po->rollover->sock = i;
1394			atomic_long_inc(&po->rollover->num);
1395			if (room == ROOM_LOW)
1396				atomic_long_inc(&po->rollover->num_huge);
1397			return i;
1398		}
1399
1400		if (++i == num)
1401			i = 0;
1402	} while (i != j);
1403
1404	atomic_long_inc(&po->rollover->num_failed);
1405	return idx;
1406}
1407
1408static unsigned int fanout_demux_qm(struct packet_fanout *f,
1409				    struct sk_buff *skb,
1410				    unsigned int num)
1411{
1412	return skb_get_queue_mapping(skb) % num;
1413}
1414
1415static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1416				     struct sk_buff *skb,
1417				     unsigned int num)
1418{
1419	struct bpf_prog *prog;
1420	unsigned int ret = 0;
1421
1422	rcu_read_lock();
1423	prog = rcu_dereference(f->bpf_prog);
1424	if (prog)
1425		ret = bpf_prog_run_clear_cb(prog, skb) % num;
1426	rcu_read_unlock();
1427
1428	return ret;
1429}
1430
1431static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1432{
1433	return f->flags & (flag >> 8);
1434}
1435
1436static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1437			     struct packet_type *pt, struct net_device *orig_dev)
1438{
1439	struct packet_fanout *f = pt->af_packet_priv;
1440	unsigned int num = READ_ONCE(f->num_members);
1441	struct net *net = read_pnet(&f->net);
1442	struct packet_sock *po;
1443	unsigned int idx;
1444
1445	if (!net_eq(dev_net(dev), net) || !num) {
1446		kfree_skb(skb);
1447		return 0;
1448	}
1449
1450	if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1451		skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1452		if (!skb)
1453			return 0;
1454	}
1455	switch (f->type) {
1456	case PACKET_FANOUT_HASH:
1457	default:
1458		idx = fanout_demux_hash(f, skb, num);
1459		break;
1460	case PACKET_FANOUT_LB:
1461		idx = fanout_demux_lb(f, skb, num);
1462		break;
1463	case PACKET_FANOUT_CPU:
1464		idx = fanout_demux_cpu(f, skb, num);
1465		break;
1466	case PACKET_FANOUT_RND:
1467		idx = fanout_demux_rnd(f, skb, num);
1468		break;
1469	case PACKET_FANOUT_QM:
1470		idx = fanout_demux_qm(f, skb, num);
1471		break;
1472	case PACKET_FANOUT_ROLLOVER:
1473		idx = fanout_demux_rollover(f, skb, 0, false, num);
1474		break;
1475	case PACKET_FANOUT_CBPF:
1476	case PACKET_FANOUT_EBPF:
1477		idx = fanout_demux_bpf(f, skb, num);
1478		break;
1479	}
1480
1481	if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1482		idx = fanout_demux_rollover(f, skb, idx, true, num);
1483
1484	po = pkt_sk(f->arr[idx]);
1485	return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1486}
1487
1488DEFINE_MUTEX(fanout_mutex);
1489EXPORT_SYMBOL_GPL(fanout_mutex);
1490static LIST_HEAD(fanout_list);
 
1491
1492static void __fanout_link(struct sock *sk, struct packet_sock *po)
1493{
1494	struct packet_fanout *f = po->fanout;
1495
1496	spin_lock(&f->lock);
1497	f->arr[f->num_members] = sk;
1498	smp_wmb();
1499	f->num_members++;
 
 
1500	spin_unlock(&f->lock);
1501}
1502
1503static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1504{
1505	struct packet_fanout *f = po->fanout;
1506	int i;
1507
1508	spin_lock(&f->lock);
1509	for (i = 0; i < f->num_members; i++) {
1510		if (f->arr[i] == sk)
 
1511			break;
1512	}
1513	BUG_ON(i >= f->num_members);
1514	f->arr[i] = f->arr[f->num_members - 1];
 
 
1515	f->num_members--;
 
 
1516	spin_unlock(&f->lock);
1517}
1518
1519static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1520{
1521	if (sk->sk_family != PF_PACKET)
1522		return false;
1523
1524	return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1525}
1526
1527static void fanout_init_data(struct packet_fanout *f)
1528{
1529	switch (f->type) {
1530	case PACKET_FANOUT_LB:
1531		atomic_set(&f->rr_cur, 0);
1532		break;
1533	case PACKET_FANOUT_CBPF:
1534	case PACKET_FANOUT_EBPF:
1535		RCU_INIT_POINTER(f->bpf_prog, NULL);
1536		break;
1537	}
1538}
1539
1540static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1541{
1542	struct bpf_prog *old;
1543
1544	spin_lock(&f->lock);
1545	old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1546	rcu_assign_pointer(f->bpf_prog, new);
1547	spin_unlock(&f->lock);
1548
1549	if (old) {
1550		synchronize_net();
1551		bpf_prog_destroy(old);
1552	}
1553}
1554
1555static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1556				unsigned int len)
1557{
1558	struct bpf_prog *new;
1559	struct sock_fprog fprog;
1560	int ret;
1561
1562	if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1563		return -EPERM;
1564	if (len != sizeof(fprog))
1565		return -EINVAL;
1566	if (copy_from_user(&fprog, data, len))
1567		return -EFAULT;
1568
1569	ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1570	if (ret)
1571		return ret;
1572
1573	__fanout_set_data_bpf(po->fanout, new);
1574	return 0;
1575}
1576
1577static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1578				unsigned int len)
1579{
1580	struct bpf_prog *new;
1581	u32 fd;
1582
1583	if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1584		return -EPERM;
1585	if (len != sizeof(fd))
1586		return -EINVAL;
1587	if (copy_from_user(&fd, data, len))
1588		return -EFAULT;
1589
1590	new = bpf_prog_get(fd);
1591	if (IS_ERR(new))
1592		return PTR_ERR(new);
1593	if (new->type != BPF_PROG_TYPE_SOCKET_FILTER) {
1594		bpf_prog_put(new);
1595		return -EINVAL;
1596	}
1597
1598	__fanout_set_data_bpf(po->fanout, new);
1599	return 0;
1600}
1601
1602static int fanout_set_data(struct packet_sock *po, char __user *data,
1603			   unsigned int len)
1604{
1605	switch (po->fanout->type) {
1606	case PACKET_FANOUT_CBPF:
1607		return fanout_set_data_cbpf(po, data, len);
1608	case PACKET_FANOUT_EBPF:
1609		return fanout_set_data_ebpf(po, data, len);
1610	default:
1611		return -EINVAL;
1612	};
1613}
1614
1615static void fanout_release_data(struct packet_fanout *f)
1616{
1617	switch (f->type) {
1618	case PACKET_FANOUT_CBPF:
1619	case PACKET_FANOUT_EBPF:
1620		__fanout_set_data_bpf(f, NULL);
1621	};
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1622}
1623
1624static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1625{
 
1626	struct packet_sock *po = pkt_sk(sk);
 
1627	struct packet_fanout *f, *match;
1628	u8 type = type_flags & 0xff;
1629	u8 flags = type_flags >> 8;
 
1630	int err;
1631
1632	switch (type) {
1633	case PACKET_FANOUT_ROLLOVER:
1634		if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1635			return -EINVAL;
 
1636	case PACKET_FANOUT_HASH:
1637	case PACKET_FANOUT_LB:
1638	case PACKET_FANOUT_CPU:
1639	case PACKET_FANOUT_RND:
1640	case PACKET_FANOUT_QM:
1641	case PACKET_FANOUT_CBPF:
1642	case PACKET_FANOUT_EBPF:
1643		break;
1644	default:
1645		return -EINVAL;
1646	}
1647
1648	if (!po->running)
1649		return -EINVAL;
1650
 
1651	if (po->fanout)
1652		return -EALREADY;
1653
1654	if (type == PACKET_FANOUT_ROLLOVER ||
1655	    (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1656		po->rollover = kzalloc(sizeof(*po->rollover), GFP_KERNEL);
1657		if (!po->rollover)
1658			return -ENOMEM;
1659		atomic_long_set(&po->rollover->num, 0);
1660		atomic_long_set(&po->rollover->num_huge, 0);
1661		atomic_long_set(&po->rollover->num_failed, 0);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1662	}
1663
1664	mutex_lock(&fanout_mutex);
1665	match = NULL;
1666	list_for_each_entry(f, &fanout_list, list) {
1667		if (f->id == id &&
1668		    read_pnet(&f->net) == sock_net(sk)) {
1669			match = f;
1670			break;
1671		}
1672	}
1673	err = -EINVAL;
1674	if (match && match->flags != flags)
1675		goto out;
1676	if (!match) {
 
 
 
 
 
 
 
 
 
1677		err = -ENOMEM;
1678		match = kzalloc(sizeof(*match), GFP_KERNEL);
 
1679		if (!match)
1680			goto out;
1681		write_pnet(&match->net, sock_net(sk));
1682		match->id = id;
1683		match->type = type;
1684		match->flags = flags;
1685		INIT_LIST_HEAD(&match->list);
1686		spin_lock_init(&match->lock);
1687		atomic_set(&match->sk_ref, 0);
1688		fanout_init_data(match);
1689		match->prot_hook.type = po->prot_hook.type;
1690		match->prot_hook.dev = po->prot_hook.dev;
1691		match->prot_hook.func = packet_rcv_fanout;
1692		match->prot_hook.af_packet_priv = match;
 
1693		match->prot_hook.id_match = match_fanout_group;
1694		dev_add_pack(&match->prot_hook);
 
1695		list_add(&match->list, &fanout_list);
1696	}
1697	err = -EINVAL;
1698	if (match->type == type &&
 
 
 
1699	    match->prot_hook.type == po->prot_hook.type &&
1700	    match->prot_hook.dev == po->prot_hook.dev) {
1701		err = -ENOSPC;
1702		if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1703			__dev_remove_pack(&po->prot_hook);
1704			po->fanout = match;
1705			atomic_inc(&match->sk_ref);
1706			__fanout_link(sk, po);
 
 
 
 
 
 
1707			err = 0;
1708		}
1709	}
 
 
 
 
 
 
 
1710out:
 
1711	mutex_unlock(&fanout_mutex);
1712	if (err) {
1713		kfree(po->rollover);
1714		po->rollover = NULL;
1715	}
1716	return err;
1717}
1718
1719static void fanout_release(struct sock *sk)
 
 
 
 
 
1720{
1721	struct packet_sock *po = pkt_sk(sk);
1722	struct packet_fanout *f;
1723
1724	f = po->fanout;
1725	if (!f)
1726		return;
1727
1728	mutex_lock(&fanout_mutex);
1729	po->fanout = NULL;
 
 
1730
1731	if (atomic_dec_and_test(&f->sk_ref)) {
1732		list_del(&f->list);
1733		dev_remove_pack(&f->prot_hook);
1734		fanout_release_data(f);
1735		kfree(f);
1736	}
1737	mutex_unlock(&fanout_mutex);
1738
1739	if (po->rollover)
1740		kfree_rcu(po->rollover, rcu);
1741}
1742
1743static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1744					  struct sk_buff *skb)
1745{
1746	/* Earlier code assumed this would be a VLAN pkt, double-check
1747	 * this now that we have the actual packet in hand. We can only
1748	 * do this check on Ethernet devices.
1749	 */
1750	if (unlikely(dev->type != ARPHRD_ETHER))
1751		return false;
1752
1753	skb_reset_mac_header(skb);
1754	return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1755}
1756
1757static const struct proto_ops packet_ops;
1758
1759static const struct proto_ops packet_ops_spkt;
1760
1761static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1762			   struct packet_type *pt, struct net_device *orig_dev)
1763{
1764	struct sock *sk;
1765	struct sockaddr_pkt *spkt;
1766
1767	/*
1768	 *	When we registered the protocol we saved the socket in the data
1769	 *	field for just this event.
1770	 */
1771
1772	sk = pt->af_packet_priv;
1773
1774	/*
1775	 *	Yank back the headers [hope the device set this
1776	 *	right or kerboom...]
1777	 *
1778	 *	Incoming packets have ll header pulled,
1779	 *	push it back.
1780	 *
1781	 *	For outgoing ones skb->data == skb_mac_header(skb)
1782	 *	so that this procedure is noop.
1783	 */
1784
1785	if (skb->pkt_type == PACKET_LOOPBACK)
1786		goto out;
1787
1788	if (!net_eq(dev_net(dev), sock_net(sk)))
1789		goto out;
1790
1791	skb = skb_share_check(skb, GFP_ATOMIC);
1792	if (skb == NULL)
1793		goto oom;
1794
1795	/* drop any routing info */
1796	skb_dst_drop(skb);
1797
1798	/* drop conntrack reference */
1799	nf_reset(skb);
1800
1801	spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1802
1803	skb_push(skb, skb->data - skb_mac_header(skb));
1804
1805	/*
1806	 *	The SOCK_PACKET socket receives _all_ frames.
1807	 */
1808
1809	spkt->spkt_family = dev->type;
1810	strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1811	spkt->spkt_protocol = skb->protocol;
1812
1813	/*
1814	 *	Charge the memory to the socket. This is done specifically
1815	 *	to prevent sockets using all the memory up.
1816	 */
1817
1818	if (sock_queue_rcv_skb(sk, skb) == 0)
1819		return 0;
1820
1821out:
1822	kfree_skb(skb);
1823oom:
1824	return 0;
1825}
1826
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1827
1828/*
1829 *	Output a raw packet to a device layer. This bypasses all the other
1830 *	protocol layers and you must therefore supply it with a complete frame
1831 */
1832
1833static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1834			       size_t len)
1835{
1836	struct sock *sk = sock->sk;
1837	DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1838	struct sk_buff *skb = NULL;
1839	struct net_device *dev;
 
1840	__be16 proto = 0;
1841	int err;
1842	int extra_len = 0;
1843
1844	/*
1845	 *	Get and verify the address.
1846	 */
1847
1848	if (saddr) {
1849		if (msg->msg_namelen < sizeof(struct sockaddr))
1850			return -EINVAL;
1851		if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1852			proto = saddr->spkt_protocol;
1853	} else
1854		return -ENOTCONN;	/* SOCK_PACKET must be sent giving an address */
1855
1856	/*
1857	 *	Find the device first to size check it
1858	 */
1859
1860	saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1861retry:
1862	rcu_read_lock();
1863	dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1864	err = -ENODEV;
1865	if (dev == NULL)
1866		goto out_unlock;
1867
1868	err = -ENETDOWN;
1869	if (!(dev->flags & IFF_UP))
1870		goto out_unlock;
1871
1872	/*
1873	 * You may not queue a frame bigger than the mtu. This is the lowest level
1874	 * raw protocol and you must do your own fragmentation at this level.
1875	 */
1876
1877	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1878		if (!netif_supports_nofcs(dev)) {
1879			err = -EPROTONOSUPPORT;
1880			goto out_unlock;
1881		}
1882		extra_len = 4; /* We're doing our own CRC */
1883	}
1884
1885	err = -EMSGSIZE;
1886	if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1887		goto out_unlock;
1888
1889	if (!skb) {
1890		size_t reserved = LL_RESERVED_SPACE(dev);
1891		int tlen = dev->needed_tailroom;
1892		unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1893
1894		rcu_read_unlock();
1895		skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1896		if (skb == NULL)
1897			return -ENOBUFS;
1898		/* FIXME: Save some space for broken drivers that write a hard
1899		 * header at transmission time by themselves. PPP is the notable
1900		 * one here. This should really be fixed at the driver level.
1901		 */
1902		skb_reserve(skb, reserved);
1903		skb_reset_network_header(skb);
1904
1905		/* Try to align data part correctly */
1906		if (hhlen) {
1907			skb->data -= hhlen;
1908			skb->tail -= hhlen;
1909			if (len < hhlen)
1910				skb_reset_network_header(skb);
1911		}
1912		err = memcpy_from_msg(skb_put(skb, len), msg, len);
1913		if (err)
1914			goto out_free;
1915		goto retry;
1916	}
1917
1918	if (!dev_validate_header(dev, skb->data, len)) {
1919		err = -EINVAL;
1920		goto out_unlock;
1921	}
1922	if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1923	    !packet_extra_vlan_len_allowed(dev, skb)) {
1924		err = -EMSGSIZE;
1925		goto out_unlock;
1926	}
1927
 
 
 
 
 
 
 
1928	skb->protocol = proto;
1929	skb->dev = dev;
1930	skb->priority = sk->sk_priority;
1931	skb->mark = sk->sk_mark;
1932
1933	sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1934
1935	if (unlikely(extra_len == 4))
1936		skb->no_fcs = 1;
1937
1938	skb_probe_transport_header(skb, 0);
1939
1940	dev_queue_xmit(skb);
1941	rcu_read_unlock();
1942	return len;
1943
1944out_unlock:
1945	rcu_read_unlock();
1946out_free:
1947	kfree_skb(skb);
1948	return err;
1949}
1950
1951static unsigned int run_filter(struct sk_buff *skb,
1952			       const struct sock *sk,
1953			       unsigned int res)
1954{
1955	struct sk_filter *filter;
1956
1957	rcu_read_lock();
1958	filter = rcu_dereference(sk->sk_filter);
1959	if (filter != NULL)
1960		res = bpf_prog_run_clear_cb(filter->prog, skb);
1961	rcu_read_unlock();
1962
1963	return res;
1964}
1965
1966static int __packet_rcv_vnet(const struct sk_buff *skb,
1967			     struct virtio_net_hdr *vnet_hdr)
1968{
1969	*vnet_hdr = (const struct virtio_net_hdr) { 0 };
1970
1971	if (skb_is_gso(skb)) {
1972		struct skb_shared_info *sinfo = skb_shinfo(skb);
1973
1974		/* This is a hint as to how much should be linear. */
1975		vnet_hdr->hdr_len =
1976			__cpu_to_virtio16(vio_le(), skb_headlen(skb));
1977		vnet_hdr->gso_size =
1978			__cpu_to_virtio16(vio_le(), sinfo->gso_size);
1979
1980		if (sinfo->gso_type & SKB_GSO_TCPV4)
1981			vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1982		else if (sinfo->gso_type & SKB_GSO_TCPV6)
1983			vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1984		else if (sinfo->gso_type & SKB_GSO_UDP)
1985			vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
1986		else if (sinfo->gso_type & SKB_GSO_FCOE)
1987			return -EINVAL;
1988		else
1989			BUG();
1990
1991		if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1992			vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1993	} else
1994		vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
1995
1996	if (skb->ip_summed == CHECKSUM_PARTIAL) {
1997		vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1998		vnet_hdr->csum_start = __cpu_to_virtio16(vio_le(),
1999				  skb_checksum_start_offset(skb));
2000		vnet_hdr->csum_offset = __cpu_to_virtio16(vio_le(),
2001						 skb->csum_offset);
2002	} else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2003		vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID;
2004	} /* else everything is zero */
2005
2006	return 0;
2007}
2008
2009static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2010			   size_t *len)
2011{
2012	struct virtio_net_hdr vnet_hdr;
2013
2014	if (*len < sizeof(vnet_hdr))
2015		return -EINVAL;
2016	*len -= sizeof(vnet_hdr);
2017
2018	if (__packet_rcv_vnet(skb, &vnet_hdr))
2019		return -EINVAL;
2020
2021	return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2022}
2023
2024/*
2025 * This function makes lazy skb cloning in hope that most of packets
2026 * are discarded by BPF.
2027 *
2028 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2029 * and skb->cb are mangled. It works because (and until) packets
2030 * falling here are owned by current CPU. Output packets are cloned
2031 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2032 * sequencially, so that if we return skb to original state on exit,
2033 * we will not harm anyone.
2034 */
2035
2036static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2037		      struct packet_type *pt, struct net_device *orig_dev)
2038{
2039	struct sock *sk;
 
2040	struct sockaddr_ll *sll;
2041	struct packet_sock *po;
2042	u8 *skb_head = skb->data;
2043	int skb_len = skb->len;
2044	unsigned int snaplen, res;
2045
2046	if (skb->pkt_type == PACKET_LOOPBACK)
2047		goto drop;
2048
2049	sk = pt->af_packet_priv;
2050	po = pkt_sk(sk);
2051
2052	if (!net_eq(dev_net(dev), sock_net(sk)))
2053		goto drop;
2054
2055	skb->dev = dev;
2056
2057	if (dev->header_ops) {
2058		/* The device has an explicit notion of ll header,
2059		 * exported to higher levels.
2060		 *
2061		 * Otherwise, the device hides details of its frame
2062		 * structure, so that corresponding packet head is
2063		 * never delivered to user.
2064		 */
2065		if (sk->sk_type != SOCK_DGRAM)
2066			skb_push(skb, skb->data - skb_mac_header(skb));
2067		else if (skb->pkt_type == PACKET_OUTGOING) {
2068			/* Special case: outgoing packets have ll header at head */
2069			skb_pull(skb, skb_network_offset(skb));
2070		}
2071	}
2072
2073	snaplen = skb->len;
2074
2075	res = run_filter(skb, sk, snaplen);
2076	if (!res)
2077		goto drop_n_restore;
2078	if (snaplen > res)
2079		snaplen = res;
2080
2081	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2082		goto drop_n_acct;
2083
2084	if (skb_shared(skb)) {
2085		struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2086		if (nskb == NULL)
2087			goto drop_n_acct;
2088
2089		if (skb_head != skb->data) {
2090			skb->data = skb_head;
2091			skb->len = skb_len;
2092		}
2093		consume_skb(skb);
2094		skb = nskb;
2095	}
2096
2097	sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2098
2099	sll = &PACKET_SKB_CB(skb)->sa.ll;
2100	sll->sll_hatype = dev->type;
2101	sll->sll_pkttype = skb->pkt_type;
2102	if (unlikely(po->origdev))
2103		sll->sll_ifindex = orig_dev->ifindex;
2104	else
2105		sll->sll_ifindex = dev->ifindex;
2106
2107	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2108
2109	/* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2110	 * Use their space for storing the original skb length.
2111	 */
2112	PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2113
2114	if (pskb_trim(skb, snaplen))
2115		goto drop_n_acct;
2116
2117	skb_set_owner_r(skb, sk);
2118	skb->dev = NULL;
2119	skb_dst_drop(skb);
2120
2121	/* drop conntrack reference */
2122	nf_reset(skb);
2123
2124	spin_lock(&sk->sk_receive_queue.lock);
2125	po->stats.stats1.tp_packets++;
2126	sock_skb_set_dropcount(sk, skb);
 
2127	__skb_queue_tail(&sk->sk_receive_queue, skb);
2128	spin_unlock(&sk->sk_receive_queue.lock);
2129	sk->sk_data_ready(sk);
2130	return 0;
2131
2132drop_n_acct:
2133	spin_lock(&sk->sk_receive_queue.lock);
2134	po->stats.stats1.tp_drops++;
2135	atomic_inc(&sk->sk_drops);
2136	spin_unlock(&sk->sk_receive_queue.lock);
2137
2138drop_n_restore:
2139	if (skb_head != skb->data && skb_shared(skb)) {
2140		skb->data = skb_head;
2141		skb->len = skb_len;
2142	}
2143drop:
2144	consume_skb(skb);
2145	return 0;
2146}
2147
2148static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2149		       struct packet_type *pt, struct net_device *orig_dev)
2150{
2151	struct sock *sk;
 
2152	struct packet_sock *po;
2153	struct sockaddr_ll *sll;
2154	union tpacket_uhdr h;
2155	u8 *skb_head = skb->data;
2156	int skb_len = skb->len;
2157	unsigned int snaplen, res;
2158	unsigned long status = TP_STATUS_USER;
2159	unsigned short macoff, netoff, hdrlen;
 
2160	struct sk_buff *copy_skb = NULL;
2161	struct timespec ts;
2162	__u32 ts_status;
 
 
2163
2164	/* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2165	 * We may add members to them until current aligned size without forcing
2166	 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2167	 */
2168	BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2169	BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2170
2171	if (skb->pkt_type == PACKET_LOOPBACK)
2172		goto drop;
2173
2174	sk = pt->af_packet_priv;
2175	po = pkt_sk(sk);
2176
2177	if (!net_eq(dev_net(dev), sock_net(sk)))
2178		goto drop;
2179
2180	if (dev->header_ops) {
2181		if (sk->sk_type != SOCK_DGRAM)
2182			skb_push(skb, skb->data - skb_mac_header(skb));
2183		else if (skb->pkt_type == PACKET_OUTGOING) {
2184			/* Special case: outgoing packets have ll header at head */
2185			skb_pull(skb, skb_network_offset(skb));
2186		}
2187	}
2188
2189	snaplen = skb->len;
2190
2191	res = run_filter(skb, sk, snaplen);
2192	if (!res)
2193		goto drop_n_restore;
2194
 
 
 
 
 
 
2195	if (skb->ip_summed == CHECKSUM_PARTIAL)
2196		status |= TP_STATUS_CSUMNOTREADY;
2197	else if (skb->pkt_type != PACKET_OUTGOING &&
2198		 (skb->ip_summed == CHECKSUM_COMPLETE ||
2199		  skb_csum_unnecessary(skb)))
2200		status |= TP_STATUS_CSUM_VALID;
 
 
2201
2202	if (snaplen > res)
2203		snaplen = res;
2204
2205	if (sk->sk_type == SOCK_DGRAM) {
2206		macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2207				  po->tp_reserve;
2208	} else {
2209		unsigned int maclen = skb_network_offset(skb);
2210		netoff = TPACKET_ALIGN(po->tp_hdrlen +
2211				       (maclen < 16 ? 16 : maclen)) +
2212				       po->tp_reserve;
2213		if (po->has_vnet_hdr)
2214			netoff += sizeof(struct virtio_net_hdr);
 
2215		macoff = netoff - maclen;
2216	}
 
 
 
 
2217	if (po->tp_version <= TPACKET_V2) {
2218		if (macoff + snaplen > po->rx_ring.frame_size) {
2219			if (po->copy_thresh &&
2220			    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2221				if (skb_shared(skb)) {
2222					copy_skb = skb_clone(skb, GFP_ATOMIC);
2223				} else {
2224					copy_skb = skb_get(skb);
2225					skb_head = skb->data;
2226				}
2227				if (copy_skb)
 
 
2228					skb_set_owner_r(copy_skb, sk);
 
2229			}
2230			snaplen = po->rx_ring.frame_size - macoff;
2231			if ((int)snaplen < 0)
2232				snaplen = 0;
 
 
2233		}
2234	} else if (unlikely(macoff + snaplen >
2235			    GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2236		u32 nval;
2237
2238		nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2239		pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2240			    snaplen, nval, macoff);
2241		snaplen = nval;
2242		if (unlikely((int)snaplen < 0)) {
2243			snaplen = 0;
2244			macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
 
2245		}
2246	}
2247	spin_lock(&sk->sk_receive_queue.lock);
2248	h.raw = packet_current_rx_frame(po, skb,
2249					TP_STATUS_KERNEL, (macoff+snaplen));
2250	if (!h.raw)
2251		goto drop_n_account;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2252	if (po->tp_version <= TPACKET_V2) {
2253		packet_increment_rx_head(po, &po->rx_ring);
2254	/*
2255	 * LOSING will be reported till you read the stats,
2256	 * because it's COR - Clear On Read.
2257	 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2258	 * at packet level.
2259	 */
2260		if (po->stats.stats1.tp_drops)
2261			status |= TP_STATUS_LOSING;
2262	}
 
2263	po->stats.stats1.tp_packets++;
2264	if (copy_skb) {
2265		status |= TP_STATUS_COPY;
 
2266		__skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2267	}
2268	spin_unlock(&sk->sk_receive_queue.lock);
2269
2270	if (po->has_vnet_hdr) {
2271		if (__packet_rcv_vnet(skb, h.raw + macoff -
2272					   sizeof(struct virtio_net_hdr))) {
2273			spin_lock(&sk->sk_receive_queue.lock);
2274			goto drop_n_account;
2275		}
2276	}
2277
2278	skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2279
2280	if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2281		getnstimeofday(&ts);
 
 
 
 
 
 
2282
2283	status |= ts_status;
2284
2285	switch (po->tp_version) {
2286	case TPACKET_V1:
2287		h.h1->tp_len = skb->len;
2288		h.h1->tp_snaplen = snaplen;
2289		h.h1->tp_mac = macoff;
2290		h.h1->tp_net = netoff;
2291		h.h1->tp_sec = ts.tv_sec;
2292		h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2293		hdrlen = sizeof(*h.h1);
2294		break;
2295	case TPACKET_V2:
2296		h.h2->tp_len = skb->len;
2297		h.h2->tp_snaplen = snaplen;
2298		h.h2->tp_mac = macoff;
2299		h.h2->tp_net = netoff;
2300		h.h2->tp_sec = ts.tv_sec;
2301		h.h2->tp_nsec = ts.tv_nsec;
2302		if (skb_vlan_tag_present(skb)) {
2303			h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2304			h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2305			status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
 
 
 
 
2306		} else {
2307			h.h2->tp_vlan_tci = 0;
2308			h.h2->tp_vlan_tpid = 0;
2309		}
2310		memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2311		hdrlen = sizeof(*h.h2);
2312		break;
2313	case TPACKET_V3:
2314		/* tp_nxt_offset,vlan are already populated above.
2315		 * So DONT clear those fields here
2316		 */
2317		h.h3->tp_status |= status;
2318		h.h3->tp_len = skb->len;
2319		h.h3->tp_snaplen = snaplen;
2320		h.h3->tp_mac = macoff;
2321		h.h3->tp_net = netoff;
2322		h.h3->tp_sec  = ts.tv_sec;
2323		h.h3->tp_nsec = ts.tv_nsec;
2324		memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2325		hdrlen = sizeof(*h.h3);
2326		break;
2327	default:
2328		BUG();
2329	}
2330
2331	sll = h.raw + TPACKET_ALIGN(hdrlen);
2332	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2333	sll->sll_family = AF_PACKET;
2334	sll->sll_hatype = dev->type;
2335	sll->sll_protocol = skb->protocol;
 
2336	sll->sll_pkttype = skb->pkt_type;
2337	if (unlikely(po->origdev))
2338		sll->sll_ifindex = orig_dev->ifindex;
2339	else
2340		sll->sll_ifindex = dev->ifindex;
2341
2342	smp_mb();
2343
2344#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2345	if (po->tp_version <= TPACKET_V2) {
2346		u8 *start, *end;
2347
2348		end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2349					macoff + snaplen);
2350
2351		for (start = h.raw; start < end; start += PAGE_SIZE)
2352			flush_dcache_page(pgv_to_page(start));
2353	}
2354	smp_wmb();
2355#endif
2356
2357	if (po->tp_version <= TPACKET_V2) {
 
2358		__packet_set_status(po, h.raw, status);
 
 
2359		sk->sk_data_ready(sk);
2360	} else {
2361		prb_clear_blk_fill_status(&po->rx_ring);
2362	}
2363
2364drop_n_restore:
2365	if (skb_head != skb->data && skb_shared(skb)) {
2366		skb->data = skb_head;
2367		skb->len = skb_len;
2368	}
2369drop:
2370	kfree_skb(skb);
2371	return 0;
2372
2373drop_n_account:
2374	po->stats.stats1.tp_drops++;
2375	spin_unlock(&sk->sk_receive_queue.lock);
 
 
2376
2377	sk->sk_data_ready(sk);
2378	kfree_skb(copy_skb);
2379	goto drop_n_restore;
2380}
2381
2382static void tpacket_destruct_skb(struct sk_buff *skb)
2383{
2384	struct packet_sock *po = pkt_sk(skb->sk);
2385
2386	if (likely(po->tx_ring.pg_vec)) {
2387		void *ph;
2388		__u32 ts;
2389
2390		ph = skb_shinfo(skb)->destructor_arg;
2391		packet_dec_pending(&po->tx_ring);
2392
2393		ts = __packet_set_timestamp(po, ph, skb);
2394		__packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
 
 
2395	}
2396
2397	sock_wfree(skb);
2398}
2399
2400static void tpacket_set_protocol(const struct net_device *dev,
2401				 struct sk_buff *skb)
2402{
2403	if (dev->type == ARPHRD_ETHER) {
2404		skb_reset_mac_header(skb);
2405		skb->protocol = eth_hdr(skb)->h_proto;
2406	}
2407}
2408
2409static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2410{
2411	unsigned short gso_type = 0;
2412
2413	if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2414	    (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2415	     __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2416	      __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2417		vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2418			 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2419			__virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2420
2421	if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2422		return -EINVAL;
2423
2424	if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2425		switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2426		case VIRTIO_NET_HDR_GSO_TCPV4:
2427			gso_type = SKB_GSO_TCPV4;
2428			break;
2429		case VIRTIO_NET_HDR_GSO_TCPV6:
2430			gso_type = SKB_GSO_TCPV6;
2431			break;
2432		case VIRTIO_NET_HDR_GSO_UDP:
2433			gso_type = SKB_GSO_UDP;
2434			break;
2435		default:
2436			return -EINVAL;
2437		}
2438
2439		if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
2440			gso_type |= SKB_GSO_TCP_ECN;
2441
2442		if (vnet_hdr->gso_size == 0)
2443			return -EINVAL;
2444	}
2445
2446	vnet_hdr->gso_type = gso_type;	/* changes type, temporary storage */
2447	return 0;
2448}
2449
2450static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2451				 struct virtio_net_hdr *vnet_hdr)
2452{
2453	int n;
2454
2455	if (*len < sizeof(*vnet_hdr))
2456		return -EINVAL;
2457	*len -= sizeof(*vnet_hdr);
2458
2459	n = copy_from_iter(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter);
2460	if (n != sizeof(*vnet_hdr))
2461		return -EFAULT;
2462
2463	return __packet_snd_vnet_parse(vnet_hdr, *len);
2464}
2465
2466static int packet_snd_vnet_gso(struct sk_buff *skb,
2467			       struct virtio_net_hdr *vnet_hdr)
2468{
2469	if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2470		u16 s = __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start);
2471		u16 o = __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset);
2472
2473		if (!skb_partial_csum_set(skb, s, o))
2474			return -EINVAL;
2475	}
2476
2477	skb_shinfo(skb)->gso_size =
2478		__virtio16_to_cpu(vio_le(), vnet_hdr->gso_size);
2479	skb_shinfo(skb)->gso_type = vnet_hdr->gso_type;
2480
2481	/* Header must be checked, and gso_segs computed. */
2482	skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2483	skb_shinfo(skb)->gso_segs = 0;
2484	return 0;
2485}
2486
2487static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2488		void *frame, struct net_device *dev, void *data, int tp_len,
2489		__be16 proto, unsigned char *addr, int hlen, int copylen)
 
2490{
2491	union tpacket_uhdr ph;
2492	int to_write, offset, len, nr_frags, len_max;
2493	struct socket *sock = po->sk.sk_socket;
2494	struct page *page;
2495	int err;
2496
2497	ph.raw = frame;
2498
2499	skb->protocol = proto;
2500	skb->dev = dev;
2501	skb->priority = po->sk.sk_priority;
2502	skb->mark = po->sk.sk_mark;
2503	sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2504	skb_shinfo(skb)->destructor_arg = ph.raw;
 
2505
2506	skb_reserve(skb, hlen);
2507	skb_reset_network_header(skb);
2508
2509	to_write = tp_len;
2510
2511	if (sock->type == SOCK_DGRAM) {
2512		err = dev_hard_header(skb, dev, ntohs(proto), addr,
2513				NULL, tp_len);
2514		if (unlikely(err < 0))
2515			return -EINVAL;
2516	} else if (copylen) {
2517		int hdrlen = min_t(int, copylen, tp_len);
2518
2519		skb_push(skb, dev->hard_header_len);
2520		skb_put(skb, copylen - dev->hard_header_len);
2521		err = skb_store_bits(skb, 0, data, hdrlen);
2522		if (unlikely(err))
2523			return err;
2524		if (!dev_validate_header(dev, skb->data, hdrlen))
2525			return -EINVAL;
2526		if (!skb->protocol)
2527			tpacket_set_protocol(dev, skb);
2528
2529		data += hdrlen;
2530		to_write -= hdrlen;
2531	}
2532
2533	offset = offset_in_page(data);
2534	len_max = PAGE_SIZE - offset;
2535	len = ((to_write > len_max) ? len_max : to_write);
2536
2537	skb->data_len = to_write;
2538	skb->len += to_write;
2539	skb->truesize += to_write;
2540	atomic_add(to_write, &po->sk.sk_wmem_alloc);
2541
2542	while (likely(to_write)) {
2543		nr_frags = skb_shinfo(skb)->nr_frags;
2544
2545		if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2546			pr_err("Packet exceed the number of skb frags(%lu)\n",
2547			       MAX_SKB_FRAGS);
2548			return -EFAULT;
2549		}
2550
2551		page = pgv_to_page(data);
2552		data += len;
2553		flush_dcache_page(page);
2554		get_page(page);
2555		skb_fill_page_desc(skb, nr_frags, page, offset, len);
2556		to_write -= len;
2557		offset = 0;
2558		len_max = PAGE_SIZE;
2559		len = ((to_write > len_max) ? len_max : to_write);
2560	}
2561
2562	skb_probe_transport_header(skb, 0);
2563
2564	return tp_len;
2565}
2566
2567static int tpacket_parse_header(struct packet_sock *po, void *frame,
2568				int size_max, void **data)
2569{
2570	union tpacket_uhdr ph;
2571	int tp_len, off;
2572
2573	ph.raw = frame;
2574
2575	switch (po->tp_version) {
 
 
 
 
 
 
 
2576	case TPACKET_V2:
2577		tp_len = ph.h2->tp_len;
2578		break;
2579	default:
2580		tp_len = ph.h1->tp_len;
2581		break;
2582	}
2583	if (unlikely(tp_len > size_max)) {
2584		pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2585		return -EMSGSIZE;
2586	}
2587
2588	if (unlikely(po->tp_tx_has_off)) {
2589		int off_min, off_max;
2590
2591		off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2592		off_max = po->tx_ring.frame_size - tp_len;
2593		if (po->sk.sk_type == SOCK_DGRAM) {
2594			switch (po->tp_version) {
 
 
 
2595			case TPACKET_V2:
2596				off = ph.h2->tp_net;
2597				break;
2598			default:
2599				off = ph.h1->tp_net;
2600				break;
2601			}
2602		} else {
2603			switch (po->tp_version) {
 
 
 
2604			case TPACKET_V2:
2605				off = ph.h2->tp_mac;
2606				break;
2607			default:
2608				off = ph.h1->tp_mac;
2609				break;
2610			}
2611		}
2612		if (unlikely((off < off_min) || (off_max < off)))
2613			return -EINVAL;
2614	} else {
2615		off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2616	}
2617
2618	*data = frame + off;
2619	return tp_len;
2620}
2621
2622static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2623{
2624	struct sk_buff *skb;
2625	struct net_device *dev;
2626	struct virtio_net_hdr *vnet_hdr = NULL;
 
2627	__be16 proto;
2628	int err, reserve = 0;
2629	void *ph;
2630	DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2631	bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
 
 
2632	int tp_len, size_max;
2633	unsigned char *addr;
2634	void *data;
2635	int len_sum = 0;
2636	int status = TP_STATUS_AVAILABLE;
2637	int hlen, tlen, copylen = 0;
 
2638
2639	mutex_lock(&po->pg_vec_lock);
2640
 
 
 
 
 
 
 
2641	if (likely(saddr == NULL)) {
2642		dev	= packet_cached_dev_get(po);
2643		proto	= po->num;
2644		addr	= NULL;
2645	} else {
2646		err = -EINVAL;
2647		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2648			goto out;
2649		if (msg->msg_namelen < (saddr->sll_halen
2650					+ offsetof(struct sockaddr_ll,
2651						sll_addr)))
2652			goto out;
2653		proto	= saddr->sll_protocol;
2654		addr	= saddr->sll_addr;
2655		dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
 
 
 
 
 
 
2656	}
2657
2658	err = -ENXIO;
2659	if (unlikely(dev == NULL))
2660		goto out;
2661	err = -ENETDOWN;
2662	if (unlikely(!(dev->flags & IFF_UP)))
2663		goto out_put;
2664
 
 
 
 
 
 
 
2665	if (po->sk.sk_socket->type == SOCK_RAW)
2666		reserve = dev->hard_header_len;
2667	size_max = po->tx_ring.frame_size
2668		- (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2669
2670	if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2671		size_max = dev->mtu + reserve + VLAN_HLEN;
2672
 
 
2673	do {
2674		ph = packet_current_frame(po, &po->tx_ring,
2675					  TP_STATUS_SEND_REQUEST);
2676		if (unlikely(ph == NULL)) {
2677			if (need_wait && need_resched())
2678				schedule();
 
 
 
 
 
 
 
2679			continue;
2680		}
2681
2682		skb = NULL;
2683		tp_len = tpacket_parse_header(po, ph, size_max, &data);
2684		if (tp_len < 0)
2685			goto tpacket_error;
2686
2687		status = TP_STATUS_SEND_REQUEST;
2688		hlen = LL_RESERVED_SPACE(dev);
2689		tlen = dev->needed_tailroom;
2690		if (po->has_vnet_hdr) {
2691			vnet_hdr = data;
2692			data += sizeof(*vnet_hdr);
2693			tp_len -= sizeof(*vnet_hdr);
2694			if (tp_len < 0 ||
2695			    __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2696				tp_len = -EINVAL;
2697				goto tpacket_error;
2698			}
2699			copylen = __virtio16_to_cpu(vio_le(),
2700						    vnet_hdr->hdr_len);
2701		}
2702		copylen = max_t(int, copylen, dev->hard_header_len);
2703		skb = sock_alloc_send_skb(&po->sk,
2704				hlen + tlen + sizeof(struct sockaddr_ll) +
2705				(copylen - dev->hard_header_len),
2706				!need_wait, &err);
2707
2708		if (unlikely(skb == NULL)) {
2709			/* we assume the socket was initially writeable ... */
2710			if (likely(len_sum > 0))
2711				err = len_sum;
2712			goto out_status;
2713		}
2714		tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2715					  addr, hlen, copylen);
2716		if (likely(tp_len >= 0) &&
2717		    tp_len > dev->mtu + reserve &&
2718		    !po->has_vnet_hdr &&
2719		    !packet_extra_vlan_len_allowed(dev, skb))
2720			tp_len = -EMSGSIZE;
2721
2722		if (unlikely(tp_len < 0)) {
2723tpacket_error:
2724			if (po->tp_loss) {
2725				__packet_set_status(po, ph,
2726						TP_STATUS_AVAILABLE);
2727				packet_increment_head(&po->tx_ring);
2728				kfree_skb(skb);
2729				continue;
2730			} else {
2731				status = TP_STATUS_WRONG_FORMAT;
2732				err = tp_len;
2733				goto out_status;
2734			}
2735		}
2736
2737		if (po->has_vnet_hdr && packet_snd_vnet_gso(skb, vnet_hdr)) {
2738			tp_len = -EINVAL;
2739			goto tpacket_error;
 
 
 
2740		}
2741
2742		packet_pick_tx_queue(dev, skb);
2743
2744		skb->destructor = tpacket_destruct_skb;
2745		__packet_set_status(po, ph, TP_STATUS_SENDING);
2746		packet_inc_pending(&po->tx_ring);
2747
2748		status = TP_STATUS_SEND_REQUEST;
2749		err = po->xmit(skb);
2750		if (unlikely(err > 0)) {
2751			err = net_xmit_errno(err);
 
2752			if (err && __packet_get_status(po, ph) ==
2753				   TP_STATUS_AVAILABLE) {
2754				/* skb was destructed already */
2755				skb = NULL;
2756				goto out_status;
2757			}
2758			/*
2759			 * skb was dropped but not destructed yet;
2760			 * let's treat it like congestion or err < 0
2761			 */
2762			err = 0;
2763		}
2764		packet_increment_head(&po->tx_ring);
2765		len_sum += tp_len;
2766	} while (likely((ph != NULL) ||
2767		/* Note: packet_read_pending() might be slow if we have
2768		 * to call it as it's per_cpu variable, but in fast-path
2769		 * we already short-circuit the loop with the first
2770		 * condition, and luckily don't have to go that path
2771		 * anyway.
2772		 */
2773		 (need_wait && packet_read_pending(&po->tx_ring))));
2774
2775	err = len_sum;
2776	goto out_put;
2777
2778out_status:
2779	__packet_set_status(po, ph, status);
2780	kfree_skb(skb);
2781out_put:
2782	dev_put(dev);
2783out:
2784	mutex_unlock(&po->pg_vec_lock);
2785	return err;
2786}
2787
2788static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2789				        size_t reserve, size_t len,
2790				        size_t linear, int noblock,
2791				        int *err)
2792{
2793	struct sk_buff *skb;
2794
2795	/* Under a page?  Don't bother with paged skb. */
2796	if (prepad + len < PAGE_SIZE || !linear)
2797		linear = len;
2798
 
 
2799	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2800				   err, 0);
2801	if (!skb)
2802		return NULL;
2803
2804	skb_reserve(skb, reserve);
2805	skb_put(skb, linear);
2806	skb->data_len = len - linear;
2807	skb->len += len - linear;
2808
2809	return skb;
2810}
2811
2812static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2813{
2814	struct sock *sk = sock->sk;
2815	DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2816	struct sk_buff *skb;
2817	struct net_device *dev;
2818	__be16 proto;
2819	unsigned char *addr;
2820	int err, reserve = 0;
2821	struct sockcm_cookie sockc;
2822	struct virtio_net_hdr vnet_hdr = { 0 };
2823	int offset = 0;
2824	struct packet_sock *po = pkt_sk(sk);
2825	int hlen, tlen;
 
2826	int extra_len = 0;
2827
2828	/*
2829	 *	Get and verify the address.
2830	 */
2831
2832	if (likely(saddr == NULL)) {
2833		dev	= packet_cached_dev_get(po);
2834		proto	= po->num;
2835		addr	= NULL;
2836	} else {
2837		err = -EINVAL;
2838		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2839			goto out;
2840		if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2841			goto out;
2842		proto	= saddr->sll_protocol;
2843		addr	= saddr->sll_addr;
2844		dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
 
 
 
 
 
 
2845	}
2846
2847	err = -ENXIO;
2848	if (unlikely(dev == NULL))
2849		goto out_unlock;
2850	err = -ENETDOWN;
2851	if (unlikely(!(dev->flags & IFF_UP)))
2852		goto out_unlock;
2853
2854	sockc.mark = sk->sk_mark;
 
2855	if (msg->msg_controllen) {
2856		err = sock_cmsg_send(sk, msg, &sockc);
2857		if (unlikely(err))
2858			goto out_unlock;
2859	}
2860
2861	if (sock->type == SOCK_RAW)
2862		reserve = dev->hard_header_len;
2863	if (po->has_vnet_hdr) {
2864		err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2865		if (err)
2866			goto out_unlock;
2867	}
2868
2869	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2870		if (!netif_supports_nofcs(dev)) {
2871			err = -EPROTONOSUPPORT;
2872			goto out_unlock;
2873		}
2874		extra_len = 4; /* We're doing our own CRC */
2875	}
2876
2877	err = -EMSGSIZE;
2878	if (!vnet_hdr.gso_type &&
2879	    (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2880		goto out_unlock;
2881
2882	err = -ENOBUFS;
2883	hlen = LL_RESERVED_SPACE(dev);
2884	tlen = dev->needed_tailroom;
2885	skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
2886			       __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len),
 
2887			       msg->msg_flags & MSG_DONTWAIT, &err);
2888	if (skb == NULL)
2889		goto out_unlock;
2890
2891	skb_set_network_header(skb, reserve);
2892
2893	err = -EINVAL;
2894	if (sock->type == SOCK_DGRAM) {
2895		offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2896		if (unlikely(offset < 0))
2897			goto out_free;
 
 
 
 
 
2898	}
2899
2900	/* Returns -EFAULT on error */
2901	err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2902	if (err)
2903		goto out_free;
2904
2905	if (sock->type == SOCK_RAW &&
2906	    !dev_validate_header(dev, skb->data, len)) {
2907		err = -EINVAL;
2908		goto out_free;
2909	}
2910
2911	sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2912
2913	if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2914	    !packet_extra_vlan_len_allowed(dev, skb)) {
2915		err = -EMSGSIZE;
2916		goto out_free;
2917	}
2918
2919	skb->protocol = proto;
2920	skb->dev = dev;
2921	skb->priority = sk->sk_priority;
2922	skb->mark = sockc.mark;
 
 
 
 
2923
2924	packet_pick_tx_queue(dev, skb);
2925
2926	if (po->has_vnet_hdr) {
2927		err = packet_snd_vnet_gso(skb, &vnet_hdr);
2928		if (err)
2929			goto out_free;
2930		len += sizeof(vnet_hdr);
 
2931	}
2932
2933	skb_probe_transport_header(skb, reserve);
2934
2935	if (unlikely(extra_len == 4))
2936		skb->no_fcs = 1;
2937
2938	err = po->xmit(skb);
2939	if (err > 0 && (err = net_xmit_errno(err)) != 0)
2940		goto out_unlock;
 
 
 
2941
2942	dev_put(dev);
2943
2944	return len;
2945
2946out_free:
2947	kfree_skb(skb);
2948out_unlock:
2949	if (dev)
2950		dev_put(dev);
2951out:
2952	return err;
2953}
2954
2955static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2956{
2957	struct sock *sk = sock->sk;
2958	struct packet_sock *po = pkt_sk(sk);
2959
2960	if (po->tx_ring.pg_vec)
 
 
 
2961		return tpacket_snd(po, msg);
2962	else
2963		return packet_snd(sock, msg, len);
2964}
2965
2966/*
2967 *	Close a PACKET socket. This is fairly simple. We immediately go
2968 *	to 'closed' state and remove our protocol entry in the device list.
2969 */
2970
2971static int packet_release(struct socket *sock)
2972{
2973	struct sock *sk = sock->sk;
2974	struct packet_sock *po;
 
2975	struct net *net;
2976	union tpacket_req_u req_u;
2977
2978	if (!sk)
2979		return 0;
2980
2981	net = sock_net(sk);
2982	po = pkt_sk(sk);
2983
2984	mutex_lock(&net->packet.sklist_lock);
2985	sk_del_node_init_rcu(sk);
2986	mutex_unlock(&net->packet.sklist_lock);
2987
2988	preempt_disable();
2989	sock_prot_inuse_add(net, sk->sk_prot, -1);
2990	preempt_enable();
2991
2992	spin_lock(&po->bind_lock);
2993	unregister_prot_hook(sk, false);
2994	packet_cached_dev_reset(po);
2995
2996	if (po->prot_hook.dev) {
2997		dev_put(po->prot_hook.dev);
2998		po->prot_hook.dev = NULL;
2999	}
3000	spin_unlock(&po->bind_lock);
3001
3002	packet_flush_mclist(sk);
3003
 
3004	if (po->rx_ring.pg_vec) {
3005		memset(&req_u, 0, sizeof(req_u));
3006		packet_set_ring(sk, &req_u, 1, 0);
3007	}
3008
3009	if (po->tx_ring.pg_vec) {
3010		memset(&req_u, 0, sizeof(req_u));
3011		packet_set_ring(sk, &req_u, 1, 1);
3012	}
 
3013
3014	fanout_release(sk);
3015
3016	synchronize_net();
 
 
 
 
 
 
3017	/*
3018	 *	Now the socket is dead. No more input will appear.
3019	 */
3020	sock_orphan(sk);
3021	sock->sk = NULL;
3022
3023	/* Purge queues */
3024
3025	skb_queue_purge(&sk->sk_receive_queue);
3026	packet_free_pending(po);
3027	sk_refcnt_debug_release(sk);
3028
3029	sock_put(sk);
3030	return 0;
3031}
3032
3033/*
3034 *	Attach a packet hook.
3035 */
3036
3037static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3038			  __be16 proto)
3039{
3040	struct packet_sock *po = pkt_sk(sk);
3041	struct net_device *dev_curr;
3042	__be16 proto_curr;
3043	bool need_rehook;
3044	struct net_device *dev = NULL;
3045	int ret = 0;
3046	bool unlisted = false;
3047
3048	if (po->fanout)
3049		return -EINVAL;
3050
3051	lock_sock(sk);
3052	spin_lock(&po->bind_lock);
 
 
 
3053	rcu_read_lock();
3054
 
 
 
 
 
3055	if (name) {
3056		dev = dev_get_by_name_rcu(sock_net(sk), name);
3057		if (!dev) {
3058			ret = -ENODEV;
3059			goto out_unlock;
3060		}
3061	} else if (ifindex) {
3062		dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3063		if (!dev) {
3064			ret = -ENODEV;
3065			goto out_unlock;
3066		}
3067	}
3068
3069	if (dev)
3070		dev_hold(dev);
3071
3072	proto_curr = po->prot_hook.type;
3073	dev_curr = po->prot_hook.dev;
3074
3075	need_rehook = proto_curr != proto || dev_curr != dev;
3076
3077	if (need_rehook) {
3078		if (po->running) {
 
3079			rcu_read_unlock();
 
 
 
 
3080			__unregister_prot_hook(sk, true);
3081			rcu_read_lock();
3082			dev_curr = po->prot_hook.dev;
3083			if (dev)
3084				unlisted = !dev_get_by_index_rcu(sock_net(sk),
3085								 dev->ifindex);
3086		}
3087
3088		po->num = proto;
 
3089		po->prot_hook.type = proto;
3090
 
 
3091		if (unlikely(unlisted)) {
3092			dev_put(dev);
3093			po->prot_hook.dev = NULL;
3094			po->ifindex = -1;
3095			packet_cached_dev_reset(po);
3096		} else {
 
 
3097			po->prot_hook.dev = dev;
3098			po->ifindex = dev ? dev->ifindex : 0;
3099			packet_cached_dev_assign(po, dev);
3100		}
 
3101	}
3102	if (dev_curr)
3103		dev_put(dev_curr);
3104
3105	if (proto == 0 || !need_rehook)
3106		goto out_unlock;
3107
3108	if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3109		register_prot_hook(sk);
3110	} else {
3111		sk->sk_err = ENETDOWN;
3112		if (!sock_flag(sk, SOCK_DEAD))
3113			sk->sk_error_report(sk);
3114	}
3115
3116out_unlock:
3117	rcu_read_unlock();
3118	spin_unlock(&po->bind_lock);
3119	release_sock(sk);
3120	return ret;
3121}
3122
3123/*
3124 *	Bind a packet socket to a device
3125 */
3126
3127static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3128			    int addr_len)
3129{
3130	struct sock *sk = sock->sk;
3131	char name[15];
3132
3133	/*
3134	 *	Check legality
3135	 */
3136
3137	if (addr_len != sizeof(struct sockaddr))
3138		return -EINVAL;
3139	strlcpy(name, uaddr->sa_data, sizeof(name));
 
 
 
 
3140
3141	return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3142}
3143
3144static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3145{
3146	struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3147	struct sock *sk = sock->sk;
3148
3149	/*
3150	 *	Check legality
3151	 */
3152
3153	if (addr_len < sizeof(struct sockaddr_ll))
3154		return -EINVAL;
3155	if (sll->sll_family != AF_PACKET)
3156		return -EINVAL;
3157
3158	return packet_do_bind(sk, NULL, sll->sll_ifindex,
3159			      sll->sll_protocol ? : pkt_sk(sk)->num);
3160}
3161
3162static struct proto packet_proto = {
3163	.name	  = "PACKET",
3164	.owner	  = THIS_MODULE,
3165	.obj_size = sizeof(struct packet_sock),
3166};
3167
3168/*
3169 *	Create a packet of type SOCK_PACKET.
3170 */
3171
3172static int packet_create(struct net *net, struct socket *sock, int protocol,
3173			 int kern)
3174{
3175	struct sock *sk;
3176	struct packet_sock *po;
3177	__be16 proto = (__force __be16)protocol; /* weird, but documented */
3178	int err;
3179
3180	if (!ns_capable(net->user_ns, CAP_NET_RAW))
3181		return -EPERM;
3182	if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3183	    sock->type != SOCK_PACKET)
3184		return -ESOCKTNOSUPPORT;
3185
3186	sock->state = SS_UNCONNECTED;
3187
3188	err = -ENOBUFS;
3189	sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3190	if (sk == NULL)
3191		goto out;
3192
3193	sock->ops = &packet_ops;
3194	if (sock->type == SOCK_PACKET)
3195		sock->ops = &packet_ops_spkt;
3196
 
 
 
 
 
3197	sock_init_data(sock, sk);
3198
3199	po = pkt_sk(sk);
3200	sk->sk_family = PF_PACKET;
3201	po->num = proto;
3202	po->xmit = dev_queue_xmit;
3203
3204	err = packet_alloc_pending(po);
3205	if (err)
3206		goto out2;
3207
3208	packet_cached_dev_reset(po);
3209
3210	sk->sk_destruct = packet_sock_destruct;
3211	sk_refcnt_debug_inc(sk);
3212
3213	/*
3214	 *	Attach a protocol block
3215	 */
3216
3217	spin_lock_init(&po->bind_lock);
3218	mutex_init(&po->pg_vec_lock);
3219	po->rollover = NULL;
3220	po->prot_hook.func = packet_rcv;
3221
3222	if (sock->type == SOCK_PACKET)
3223		po->prot_hook.func = packet_rcv_spkt;
3224
3225	po->prot_hook.af_packet_priv = sk;
 
3226
3227	if (proto) {
3228		po->prot_hook.type = proto;
3229		register_prot_hook(sk);
3230	}
3231
3232	mutex_lock(&net->packet.sklist_lock);
3233	sk_add_node_rcu(sk, &net->packet.sklist);
3234	mutex_unlock(&net->packet.sklist_lock);
3235
3236	preempt_disable();
3237	sock_prot_inuse_add(net, &packet_proto, 1);
3238	preempt_enable();
3239
3240	return 0;
3241out2:
3242	sk_free(sk);
3243out:
3244	return err;
3245}
3246
3247/*
3248 *	Pull a packet from our receive queue and hand it to the user.
3249 *	If necessary we block.
3250 */
3251
3252static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3253			  int flags)
3254{
3255	struct sock *sk = sock->sk;
3256	struct sk_buff *skb;
3257	int copied, err;
3258	int vnet_hdr_len = 0;
3259	unsigned int origlen = 0;
3260
3261	err = -EINVAL;
3262	if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3263		goto out;
3264
3265#if 0
3266	/* What error should we return now? EUNATTACH? */
3267	if (pkt_sk(sk)->ifindex < 0)
3268		return -ENODEV;
3269#endif
3270
3271	if (flags & MSG_ERRQUEUE) {
3272		err = sock_recv_errqueue(sk, msg, len,
3273					 SOL_PACKET, PACKET_TX_TIMESTAMP);
3274		goto out;
3275	}
3276
3277	/*
3278	 *	Call the generic datagram receiver. This handles all sorts
3279	 *	of horrible races and re-entrancy so we can forget about it
3280	 *	in the protocol layers.
3281	 *
3282	 *	Now it will return ENETDOWN, if device have just gone down,
3283	 *	but then it will block.
3284	 */
3285
3286	skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3287
3288	/*
3289	 *	An error occurred so return it. Because skb_recv_datagram()
3290	 *	handles the blocking we don't see and worry about blocking
3291	 *	retries.
3292	 */
3293
3294	if (skb == NULL)
3295		goto out;
3296
3297	if (pkt_sk(sk)->pressure)
3298		packet_rcv_has_room(pkt_sk(sk), NULL);
3299
3300	if (pkt_sk(sk)->has_vnet_hdr) {
3301		err = packet_rcv_vnet(msg, skb, &len);
3302		if (err)
3303			goto out_free;
3304		vnet_hdr_len = sizeof(struct virtio_net_hdr);
3305	}
3306
3307	/* You lose any data beyond the buffer you gave. If it worries
3308	 * a user program they can ask the device for its MTU
3309	 * anyway.
3310	 */
3311	copied = skb->len;
3312	if (copied > len) {
3313		copied = len;
3314		msg->msg_flags |= MSG_TRUNC;
3315	}
3316
3317	err = skb_copy_datagram_msg(skb, 0, msg, copied);
3318	if (err)
3319		goto out_free;
3320
3321	if (sock->type != SOCK_PACKET) {
3322		struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3323
3324		/* Original length was stored in sockaddr_ll fields */
3325		origlen = PACKET_SKB_CB(skb)->sa.origlen;
3326		sll->sll_family = AF_PACKET;
3327		sll->sll_protocol = skb->protocol;
 
3328	}
3329
3330	sock_recv_ts_and_drops(msg, sk, skb);
3331
3332	if (msg->msg_name) {
 
 
 
 
3333		/* If the address length field is there to be filled
3334		 * in, we fill it in now.
3335		 */
3336		if (sock->type == SOCK_PACKET) {
3337			__sockaddr_check_size(sizeof(struct sockaddr_pkt));
3338			msg->msg_namelen = sizeof(struct sockaddr_pkt);
 
3339		} else {
3340			struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3341
3342			msg->msg_namelen = sll->sll_halen +
3343				offsetof(struct sockaddr_ll, sll_addr);
 
 
 
 
 
 
 
 
 
 
 
3344		}
3345		memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3346		       msg->msg_namelen);
3347	}
3348
3349	if (pkt_sk(sk)->auxdata) {
3350		struct tpacket_auxdata aux;
3351
3352		aux.tp_status = TP_STATUS_USER;
3353		if (skb->ip_summed == CHECKSUM_PARTIAL)
3354			aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3355		else if (skb->pkt_type != PACKET_OUTGOING &&
3356			 (skb->ip_summed == CHECKSUM_COMPLETE ||
3357			  skb_csum_unnecessary(skb)))
3358			aux.tp_status |= TP_STATUS_CSUM_VALID;
 
 
3359
3360		aux.tp_len = origlen;
3361		aux.tp_snaplen = skb->len;
3362		aux.tp_mac = 0;
3363		aux.tp_net = skb_network_offset(skb);
3364		if (skb_vlan_tag_present(skb)) {
3365			aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3366			aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3367			aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3368		} else {
3369			aux.tp_vlan_tci = 0;
3370			aux.tp_vlan_tpid = 0;
3371		}
3372		put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3373	}
3374
3375	/*
3376	 *	Free or return the buffer as appropriate. Again this
3377	 *	hides all the races and re-entrancy issues from us.
3378	 */
3379	err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3380
3381out_free:
3382	skb_free_datagram(sk, skb);
3383out:
3384	return err;
3385}
3386
3387static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3388			       int *uaddr_len, int peer)
3389{
3390	struct net_device *dev;
3391	struct sock *sk	= sock->sk;
3392
3393	if (peer)
3394		return -EOPNOTSUPP;
3395
3396	uaddr->sa_family = AF_PACKET;
3397	memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3398	rcu_read_lock();
3399	dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3400	if (dev)
3401		strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3402	rcu_read_unlock();
3403	*uaddr_len = sizeof(*uaddr);
3404
3405	return 0;
3406}
3407
3408static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3409			  int *uaddr_len, int peer)
3410{
3411	struct net_device *dev;
3412	struct sock *sk = sock->sk;
3413	struct packet_sock *po = pkt_sk(sk);
3414	DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
 
3415
3416	if (peer)
3417		return -EOPNOTSUPP;
3418
 
3419	sll->sll_family = AF_PACKET;
3420	sll->sll_ifindex = po->ifindex;
3421	sll->sll_protocol = po->num;
3422	sll->sll_pkttype = 0;
3423	rcu_read_lock();
3424	dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3425	if (dev) {
3426		sll->sll_hatype = dev->type;
3427		sll->sll_halen = dev->addr_len;
3428		memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
 
 
 
 
 
3429	} else {
3430		sll->sll_hatype = 0;	/* Bad: we have no ARPHRD_UNSPEC */
3431		sll->sll_halen = 0;
3432	}
3433	rcu_read_unlock();
3434	*uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3435
3436	return 0;
3437}
3438
3439static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3440			 int what)
3441{
3442	switch (i->type) {
3443	case PACKET_MR_MULTICAST:
3444		if (i->alen != dev->addr_len)
3445			return -EINVAL;
3446		if (what > 0)
3447			return dev_mc_add(dev, i->addr);
3448		else
3449			return dev_mc_del(dev, i->addr);
3450		break;
3451	case PACKET_MR_PROMISC:
3452		return dev_set_promiscuity(dev, what);
3453	case PACKET_MR_ALLMULTI:
3454		return dev_set_allmulti(dev, what);
3455	case PACKET_MR_UNICAST:
3456		if (i->alen != dev->addr_len)
3457			return -EINVAL;
3458		if (what > 0)
3459			return dev_uc_add(dev, i->addr);
3460		else
3461			return dev_uc_del(dev, i->addr);
3462		break;
3463	default:
3464		break;
3465	}
3466	return 0;
3467}
3468
3469static void packet_dev_mclist_delete(struct net_device *dev,
3470				     struct packet_mclist **mlp)
3471{
3472	struct packet_mclist *ml;
3473
3474	while ((ml = *mlp) != NULL) {
3475		if (ml->ifindex == dev->ifindex) {
3476			packet_dev_mc(dev, ml, -1);
3477			*mlp = ml->next;
3478			kfree(ml);
3479		} else
3480			mlp = &ml->next;
3481	}
3482}
3483
3484static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3485{
3486	struct packet_sock *po = pkt_sk(sk);
3487	struct packet_mclist *ml, *i;
3488	struct net_device *dev;
3489	int err;
3490
3491	rtnl_lock();
3492
3493	err = -ENODEV;
3494	dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3495	if (!dev)
3496		goto done;
3497
3498	err = -EINVAL;
3499	if (mreq->mr_alen > dev->addr_len)
3500		goto done;
3501
3502	err = -ENOBUFS;
3503	i = kmalloc(sizeof(*i), GFP_KERNEL);
3504	if (i == NULL)
3505		goto done;
3506
3507	err = 0;
3508	for (ml = po->mclist; ml; ml = ml->next) {
3509		if (ml->ifindex == mreq->mr_ifindex &&
3510		    ml->type == mreq->mr_type &&
3511		    ml->alen == mreq->mr_alen &&
3512		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3513			ml->count++;
3514			/* Free the new element ... */
3515			kfree(i);
3516			goto done;
3517		}
3518	}
3519
3520	i->type = mreq->mr_type;
3521	i->ifindex = mreq->mr_ifindex;
3522	i->alen = mreq->mr_alen;
3523	memcpy(i->addr, mreq->mr_address, i->alen);
3524	memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3525	i->count = 1;
3526	i->next = po->mclist;
3527	po->mclist = i;
3528	err = packet_dev_mc(dev, i, 1);
3529	if (err) {
3530		po->mclist = i->next;
3531		kfree(i);
3532	}
3533
3534done:
3535	rtnl_unlock();
3536	return err;
3537}
3538
3539static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3540{
3541	struct packet_mclist *ml, **mlp;
3542
3543	rtnl_lock();
3544
3545	for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3546		if (ml->ifindex == mreq->mr_ifindex &&
3547		    ml->type == mreq->mr_type &&
3548		    ml->alen == mreq->mr_alen &&
3549		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3550			if (--ml->count == 0) {
3551				struct net_device *dev;
3552				*mlp = ml->next;
3553				dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3554				if (dev)
3555					packet_dev_mc(dev, ml, -1);
3556				kfree(ml);
3557			}
3558			break;
3559		}
3560	}
3561	rtnl_unlock();
3562	return 0;
3563}
3564
3565static void packet_flush_mclist(struct sock *sk)
3566{
3567	struct packet_sock *po = pkt_sk(sk);
3568	struct packet_mclist *ml;
3569
3570	if (!po->mclist)
3571		return;
3572
3573	rtnl_lock();
3574	while ((ml = po->mclist) != NULL) {
3575		struct net_device *dev;
3576
3577		po->mclist = ml->next;
3578		dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3579		if (dev != NULL)
3580			packet_dev_mc(dev, ml, -1);
3581		kfree(ml);
3582	}
3583	rtnl_unlock();
3584}
3585
3586static int
3587packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
 
3588{
3589	struct sock *sk = sock->sk;
3590	struct packet_sock *po = pkt_sk(sk);
3591	int ret;
3592
3593	if (level != SOL_PACKET)
3594		return -ENOPROTOOPT;
3595
3596	switch (optname) {
3597	case PACKET_ADD_MEMBERSHIP:
3598	case PACKET_DROP_MEMBERSHIP:
3599	{
3600		struct packet_mreq_max mreq;
3601		int len = optlen;
3602		memset(&mreq, 0, sizeof(mreq));
3603		if (len < sizeof(struct packet_mreq))
3604			return -EINVAL;
3605		if (len > sizeof(mreq))
3606			len = sizeof(mreq);
3607		if (copy_from_user(&mreq, optval, len))
3608			return -EFAULT;
3609		if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3610			return -EINVAL;
3611		if (optname == PACKET_ADD_MEMBERSHIP)
3612			ret = packet_mc_add(sk, &mreq);
3613		else
3614			ret = packet_mc_drop(sk, &mreq);
3615		return ret;
3616	}
3617
3618	case PACKET_RX_RING:
3619	case PACKET_TX_RING:
3620	{
3621		union tpacket_req_u req_u;
3622		int len;
3623
 
 
3624		switch (po->tp_version) {
3625		case TPACKET_V1:
3626		case TPACKET_V2:
3627			len = sizeof(req_u.req);
 
 
 
 
3628			break;
3629		case TPACKET_V3:
3630		default:
3631			len = sizeof(req_u.req3);
 
 
 
 
3632			break;
3633		}
3634		if (optlen < len)
3635			return -EINVAL;
3636		if (copy_from_user(&req_u.req, optval, len))
3637			return -EFAULT;
3638		return packet_set_ring(sk, &req_u, 0,
3639			optname == PACKET_TX_RING);
3640	}
3641	case PACKET_COPY_THRESH:
3642	{
3643		int val;
3644
3645		if (optlen != sizeof(val))
3646			return -EINVAL;
3647		if (copy_from_user(&val, optval, sizeof(val)))
3648			return -EFAULT;
3649
3650		pkt_sk(sk)->copy_thresh = val;
3651		return 0;
3652	}
3653	case PACKET_VERSION:
3654	{
3655		int val;
3656
3657		if (optlen != sizeof(val))
3658			return -EINVAL;
3659		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3660			return -EBUSY;
3661		if (copy_from_user(&val, optval, sizeof(val)))
3662			return -EFAULT;
3663		switch (val) {
3664		case TPACKET_V1:
3665		case TPACKET_V2:
3666		case TPACKET_V3:
3667			po->tp_version = val;
3668			return 0;
3669		default:
3670			return -EINVAL;
3671		}
 
 
 
 
 
 
 
 
 
3672	}
3673	case PACKET_RESERVE:
3674	{
3675		unsigned int val;
3676
3677		if (optlen != sizeof(val))
3678			return -EINVAL;
3679		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3680			return -EBUSY;
3681		if (copy_from_user(&val, optval, sizeof(val)))
3682			return -EFAULT;
3683		po->tp_reserve = val;
3684		return 0;
 
 
 
 
 
 
 
 
 
3685	}
3686	case PACKET_LOSS:
3687	{
3688		unsigned int val;
3689
3690		if (optlen != sizeof(val))
3691			return -EINVAL;
3692		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3693			return -EBUSY;
3694		if (copy_from_user(&val, optval, sizeof(val)))
3695			return -EFAULT;
3696		po->tp_loss = !!val;
3697		return 0;
 
 
 
 
 
 
 
 
3698	}
3699	case PACKET_AUXDATA:
3700	{
3701		int val;
3702
3703		if (optlen < sizeof(val))
3704			return -EINVAL;
3705		if (copy_from_user(&val, optval, sizeof(val)))
3706			return -EFAULT;
3707
3708		po->auxdata = !!val;
3709		return 0;
3710	}
3711	case PACKET_ORIGDEV:
3712	{
3713		int val;
3714
3715		if (optlen < sizeof(val))
3716			return -EINVAL;
3717		if (copy_from_user(&val, optval, sizeof(val)))
3718			return -EFAULT;
3719
3720		po->origdev = !!val;
3721		return 0;
3722	}
3723	case PACKET_VNET_HDR:
 
3724	{
3725		int val;
3726
3727		if (sock->type != SOCK_RAW)
3728			return -EINVAL;
3729		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3730			return -EBUSY;
3731		if (optlen < sizeof(val))
3732			return -EINVAL;
3733		if (copy_from_user(&val, optval, sizeof(val)))
3734			return -EFAULT;
3735
3736		po->has_vnet_hdr = !!val;
3737		return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3738	}
3739	case PACKET_TIMESTAMP:
3740	{
3741		int val;
3742
3743		if (optlen != sizeof(val))
3744			return -EINVAL;
3745		if (copy_from_user(&val, optval, sizeof(val)))
3746			return -EFAULT;
3747
3748		po->tp_tstamp = val;
3749		return 0;
3750	}
3751	case PACKET_FANOUT:
3752	{
3753		int val;
3754
3755		if (optlen != sizeof(val))
3756			return -EINVAL;
3757		if (copy_from_user(&val, optval, sizeof(val)))
3758			return -EFAULT;
3759
3760		return fanout_add(sk, val & 0xffff, val >> 16);
3761	}
3762	case PACKET_FANOUT_DATA:
3763	{
3764		if (!po->fanout)
 
3765			return -EINVAL;
3766
3767		return fanout_set_data(po, optval, optlen);
3768	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3769	case PACKET_TX_HAS_OFF:
3770	{
3771		unsigned int val;
3772
3773		if (optlen != sizeof(val))
3774			return -EINVAL;
3775		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3776			return -EBUSY;
3777		if (copy_from_user(&val, optval, sizeof(val)))
3778			return -EFAULT;
3779		po->tp_tx_has_off = !!val;
 
 
 
 
 
3780		return 0;
3781	}
3782	case PACKET_QDISC_BYPASS:
3783	{
3784		int val;
3785
3786		if (optlen != sizeof(val))
3787			return -EINVAL;
3788		if (copy_from_user(&val, optval, sizeof(val)))
3789			return -EFAULT;
3790
3791		po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3792		return 0;
3793	}
3794	default:
3795		return -ENOPROTOOPT;
3796	}
3797}
3798
3799static int packet_getsockopt(struct socket *sock, int level, int optname,
3800			     char __user *optval, int __user *optlen)
3801{
3802	int len;
3803	int val, lv = sizeof(val);
3804	struct sock *sk = sock->sk;
3805	struct packet_sock *po = pkt_sk(sk);
3806	void *data = &val;
3807	union tpacket_stats_u st;
3808	struct tpacket_rollover_stats rstats;
 
3809
3810	if (level != SOL_PACKET)
3811		return -ENOPROTOOPT;
3812
3813	if (get_user(len, optlen))
3814		return -EFAULT;
3815
3816	if (len < 0)
3817		return -EINVAL;
3818
3819	switch (optname) {
3820	case PACKET_STATISTICS:
3821		spin_lock_bh(&sk->sk_receive_queue.lock);
3822		memcpy(&st, &po->stats, sizeof(st));
3823		memset(&po->stats, 0, sizeof(po->stats));
3824		spin_unlock_bh(&sk->sk_receive_queue.lock);
 
3825
3826		if (po->tp_version == TPACKET_V3) {
3827			lv = sizeof(struct tpacket_stats_v3);
3828			st.stats3.tp_packets += st.stats3.tp_drops;
 
3829			data = &st.stats3;
3830		} else {
3831			lv = sizeof(struct tpacket_stats);
3832			st.stats1.tp_packets += st.stats1.tp_drops;
 
3833			data = &st.stats1;
3834		}
3835
3836		break;
3837	case PACKET_AUXDATA:
3838		val = po->auxdata;
3839		break;
3840	case PACKET_ORIGDEV:
3841		val = po->origdev;
3842		break;
3843	case PACKET_VNET_HDR:
3844		val = po->has_vnet_hdr;
 
 
 
 
 
 
3845		break;
3846	case PACKET_VERSION:
3847		val = po->tp_version;
3848		break;
3849	case PACKET_HDRLEN:
3850		if (len > sizeof(int))
3851			len = sizeof(int);
 
 
3852		if (copy_from_user(&val, optval, len))
3853			return -EFAULT;
3854		switch (val) {
3855		case TPACKET_V1:
3856			val = sizeof(struct tpacket_hdr);
3857			break;
3858		case TPACKET_V2:
3859			val = sizeof(struct tpacket2_hdr);
3860			break;
3861		case TPACKET_V3:
3862			val = sizeof(struct tpacket3_hdr);
3863			break;
3864		default:
3865			return -EINVAL;
3866		}
3867		break;
3868	case PACKET_RESERVE:
3869		val = po->tp_reserve;
3870		break;
3871	case PACKET_LOSS:
3872		val = po->tp_loss;
3873		break;
3874	case PACKET_TIMESTAMP:
3875		val = po->tp_tstamp;
3876		break;
3877	case PACKET_FANOUT:
3878		val = (po->fanout ?
3879		       ((u32)po->fanout->id |
3880			((u32)po->fanout->type << 16) |
3881			((u32)po->fanout->flags << 24)) :
3882		       0);
3883		break;
 
 
 
3884	case PACKET_ROLLOVER_STATS:
3885		if (!po->rollover)
3886			return -EINVAL;
3887		rstats.tp_all = atomic_long_read(&po->rollover->num);
3888		rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3889		rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3890		data = &rstats;
3891		lv = sizeof(rstats);
3892		break;
3893	case PACKET_TX_HAS_OFF:
3894		val = po->tp_tx_has_off;
3895		break;
3896	case PACKET_QDISC_BYPASS:
3897		val = packet_use_direct_xmit(po);
3898		break;
3899	default:
3900		return -ENOPROTOOPT;
3901	}
3902
3903	if (len > lv)
3904		len = lv;
3905	if (put_user(len, optlen))
3906		return -EFAULT;
3907	if (copy_to_user(optval, data, len))
3908		return -EFAULT;
3909	return 0;
3910}
3911
3912
3913static int packet_notifier(struct notifier_block *this,
3914			   unsigned long msg, void *ptr)
3915{
3916	struct sock *sk;
3917	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3918	struct net *net = dev_net(dev);
3919
3920	rcu_read_lock();
3921	sk_for_each_rcu(sk, &net->packet.sklist) {
3922		struct packet_sock *po = pkt_sk(sk);
3923
3924		switch (msg) {
3925		case NETDEV_UNREGISTER:
3926			if (po->mclist)
3927				packet_dev_mclist_delete(dev, &po->mclist);
3928			/* fallthrough */
3929
3930		case NETDEV_DOWN:
3931			if (dev->ifindex == po->ifindex) {
3932				spin_lock(&po->bind_lock);
3933				if (po->running) {
3934					__unregister_prot_hook(sk, false);
3935					sk->sk_err = ENETDOWN;
3936					if (!sock_flag(sk, SOCK_DEAD))
3937						sk->sk_error_report(sk);
3938				}
3939				if (msg == NETDEV_UNREGISTER) {
3940					packet_cached_dev_reset(po);
3941					po->ifindex = -1;
3942					if (po->prot_hook.dev)
3943						dev_put(po->prot_hook.dev);
3944					po->prot_hook.dev = NULL;
3945				}
3946				spin_unlock(&po->bind_lock);
3947			}
3948			break;
3949		case NETDEV_UP:
3950			if (dev->ifindex == po->ifindex) {
3951				spin_lock(&po->bind_lock);
3952				if (po->num)
3953					register_prot_hook(sk);
3954				spin_unlock(&po->bind_lock);
3955			}
3956			break;
3957		}
3958	}
3959	rcu_read_unlock();
3960	return NOTIFY_DONE;
3961}
3962
3963
3964static int packet_ioctl(struct socket *sock, unsigned int cmd,
3965			unsigned long arg)
3966{
3967	struct sock *sk = sock->sk;
3968
3969	switch (cmd) {
3970	case SIOCOUTQ:
3971	{
3972		int amount = sk_wmem_alloc_get(sk);
3973
3974		return put_user(amount, (int __user *)arg);
3975	}
3976	case SIOCINQ:
3977	{
3978		struct sk_buff *skb;
3979		int amount = 0;
3980
3981		spin_lock_bh(&sk->sk_receive_queue.lock);
3982		skb = skb_peek(&sk->sk_receive_queue);
3983		if (skb)
3984			amount = skb->len;
3985		spin_unlock_bh(&sk->sk_receive_queue.lock);
3986		return put_user(amount, (int __user *)arg);
3987	}
3988	case SIOCGSTAMP:
3989		return sock_get_timestamp(sk, (struct timeval __user *)arg);
3990	case SIOCGSTAMPNS:
3991		return sock_get_timestampns(sk, (struct timespec __user *)arg);
3992
3993#ifdef CONFIG_INET
3994	case SIOCADDRT:
3995	case SIOCDELRT:
3996	case SIOCDARP:
3997	case SIOCGARP:
3998	case SIOCSARP:
3999	case SIOCGIFADDR:
4000	case SIOCSIFADDR:
4001	case SIOCGIFBRDADDR:
4002	case SIOCSIFBRDADDR:
4003	case SIOCGIFNETMASK:
4004	case SIOCSIFNETMASK:
4005	case SIOCGIFDSTADDR:
4006	case SIOCSIFDSTADDR:
4007	case SIOCSIFFLAGS:
4008		return inet_dgram_ops.ioctl(sock, cmd, arg);
4009#endif
4010
4011	default:
4012		return -ENOIOCTLCMD;
4013	}
4014	return 0;
4015}
4016
4017static unsigned int packet_poll(struct file *file, struct socket *sock,
4018				poll_table *wait)
4019{
4020	struct sock *sk = sock->sk;
4021	struct packet_sock *po = pkt_sk(sk);
4022	unsigned int mask = datagram_poll(file, sock, wait);
4023
4024	spin_lock_bh(&sk->sk_receive_queue.lock);
4025	if (po->rx_ring.pg_vec) {
4026		if (!packet_previous_rx_frame(po, &po->rx_ring,
4027			TP_STATUS_KERNEL))
4028			mask |= POLLIN | POLLRDNORM;
4029	}
4030	if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
4031		po->pressure = 0;
4032	spin_unlock_bh(&sk->sk_receive_queue.lock);
4033	spin_lock_bh(&sk->sk_write_queue.lock);
4034	if (po->tx_ring.pg_vec) {
4035		if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4036			mask |= POLLOUT | POLLWRNORM;
4037	}
4038	spin_unlock_bh(&sk->sk_write_queue.lock);
4039	return mask;
4040}
4041
4042
4043/* Dirty? Well, I still did not learn better way to account
4044 * for user mmaps.
4045 */
4046
4047static void packet_mm_open(struct vm_area_struct *vma)
4048{
4049	struct file *file = vma->vm_file;
4050	struct socket *sock = file->private_data;
4051	struct sock *sk = sock->sk;
4052
4053	if (sk)
4054		atomic_inc(&pkt_sk(sk)->mapped);
4055}
4056
4057static void packet_mm_close(struct vm_area_struct *vma)
4058{
4059	struct file *file = vma->vm_file;
4060	struct socket *sock = file->private_data;
4061	struct sock *sk = sock->sk;
4062
4063	if (sk)
4064		atomic_dec(&pkt_sk(sk)->mapped);
4065}
4066
4067static const struct vm_operations_struct packet_mmap_ops = {
4068	.open	=	packet_mm_open,
4069	.close	=	packet_mm_close,
4070};
4071
4072static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4073			unsigned int len)
4074{
4075	int i;
4076
4077	for (i = 0; i < len; i++) {
4078		if (likely(pg_vec[i].buffer)) {
4079			if (is_vmalloc_addr(pg_vec[i].buffer))
4080				vfree(pg_vec[i].buffer);
4081			else
4082				free_pages((unsigned long)pg_vec[i].buffer,
4083					   order);
4084			pg_vec[i].buffer = NULL;
4085		}
4086	}
4087	kfree(pg_vec);
4088}
4089
4090static char *alloc_one_pg_vec_page(unsigned long order)
4091{
4092	char *buffer;
4093	gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4094			  __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4095
4096	buffer = (char *) __get_free_pages(gfp_flags, order);
4097	if (buffer)
4098		return buffer;
4099
4100	/* __get_free_pages failed, fall back to vmalloc */
4101	buffer = vzalloc((1 << order) * PAGE_SIZE);
4102	if (buffer)
4103		return buffer;
4104
4105	/* vmalloc failed, lets dig into swap here */
4106	gfp_flags &= ~__GFP_NORETRY;
4107	buffer = (char *) __get_free_pages(gfp_flags, order);
4108	if (buffer)
4109		return buffer;
4110
4111	/* complete and utter failure */
4112	return NULL;
4113}
4114
4115static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4116{
4117	unsigned int block_nr = req->tp_block_nr;
4118	struct pgv *pg_vec;
4119	int i;
4120
4121	pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4122	if (unlikely(!pg_vec))
4123		goto out;
4124
4125	for (i = 0; i < block_nr; i++) {
4126		pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4127		if (unlikely(!pg_vec[i].buffer))
4128			goto out_free_pgvec;
4129	}
4130
4131out:
4132	return pg_vec;
4133
4134out_free_pgvec:
4135	free_pg_vec(pg_vec, order, block_nr);
4136	pg_vec = NULL;
4137	goto out;
4138}
4139
4140static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4141		int closing, int tx_ring)
4142{
4143	struct pgv *pg_vec = NULL;
4144	struct packet_sock *po = pkt_sk(sk);
 
4145	int was_running, order = 0;
4146	struct packet_ring_buffer *rb;
4147	struct sk_buff_head *rb_queue;
4148	__be16 num;
4149	int err = -EINVAL;
4150	/* Added to avoid minimal code churn */
4151	struct tpacket_req *req = &req_u->req;
4152
4153	/* Opening a Tx-ring is NOT supported in TPACKET_V3 */
4154	if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
4155		net_warn_ratelimited("Tx-ring is not supported.\n");
4156		goto out;
4157	}
4158
4159	rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4160	rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4161
4162	err = -EBUSY;
4163	if (!closing) {
4164		if (atomic_read(&po->mapped))
4165			goto out;
4166		if (packet_read_pending(rb))
4167			goto out;
4168	}
4169
4170	if (req->tp_block_nr) {
 
 
4171		/* Sanity tests and some calculations */
4172		err = -EBUSY;
4173		if (unlikely(rb->pg_vec))
4174			goto out;
4175
4176		switch (po->tp_version) {
4177		case TPACKET_V1:
4178			po->tp_hdrlen = TPACKET_HDRLEN;
4179			break;
4180		case TPACKET_V2:
4181			po->tp_hdrlen = TPACKET2_HDRLEN;
4182			break;
4183		case TPACKET_V3:
4184			po->tp_hdrlen = TPACKET3_HDRLEN;
4185			break;
4186		}
4187
4188		err = -EINVAL;
4189		if (unlikely((int)req->tp_block_size <= 0))
4190			goto out;
4191		if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4192			goto out;
 
4193		if (po->tp_version >= TPACKET_V3 &&
4194		    (int)(req->tp_block_size -
4195			  BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
4196			goto out;
4197		if (unlikely(req->tp_frame_size < po->tp_hdrlen +
4198					po->tp_reserve))
4199			goto out;
4200		if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4201			goto out;
4202
4203		rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4204		if (unlikely(rb->frames_per_block == 0))
4205			goto out;
 
 
4206		if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4207					req->tp_frame_nr))
4208			goto out;
4209
4210		err = -ENOMEM;
4211		order = get_order(req->tp_block_size);
4212		pg_vec = alloc_pg_vec(req, order);
4213		if (unlikely(!pg_vec))
4214			goto out;
4215		switch (po->tp_version) {
4216		case TPACKET_V3:
4217		/* Transmit path is not supported. We checked
4218		 * it above but just being paranoid
4219		 */
4220			if (!tx_ring)
4221				init_prb_bdqc(po, rb, pg_vec, req_u);
 
 
 
 
 
 
 
 
 
 
4222			break;
4223		default:
 
 
 
 
 
 
4224			break;
4225		}
4226	}
4227	/* Done */
4228	else {
4229		err = -EINVAL;
4230		if (unlikely(req->tp_frame_nr))
4231			goto out;
4232	}
4233
4234	lock_sock(sk);
4235
4236	/* Detach socket from network */
4237	spin_lock(&po->bind_lock);
4238	was_running = po->running;
4239	num = po->num;
4240	if (was_running) {
4241		po->num = 0;
4242		__unregister_prot_hook(sk, false);
4243	}
4244	spin_unlock(&po->bind_lock);
4245
4246	synchronize_net();
4247
4248	err = -EBUSY;
4249	mutex_lock(&po->pg_vec_lock);
4250	if (closing || atomic_read(&po->mapped) == 0) {
4251		err = 0;
4252		spin_lock_bh(&rb_queue->lock);
4253		swap(rb->pg_vec, pg_vec);
 
 
4254		rb->frame_max = (req->tp_frame_nr - 1);
4255		rb->head = 0;
4256		rb->frame_size = req->tp_frame_size;
4257		spin_unlock_bh(&rb_queue->lock);
4258
4259		swap(rb->pg_vec_order, order);
4260		swap(rb->pg_vec_len, req->tp_block_nr);
4261
4262		rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4263		po->prot_hook.func = (po->rx_ring.pg_vec) ?
4264						tpacket_rcv : packet_rcv;
4265		skb_queue_purge(rb_queue);
4266		if (atomic_read(&po->mapped))
4267			pr_err("packet_mmap: vma is busy: %d\n",
4268			       atomic_read(&po->mapped));
4269	}
4270	mutex_unlock(&po->pg_vec_lock);
4271
4272	spin_lock(&po->bind_lock);
4273	if (was_running) {
4274		po->num = num;
4275		register_prot_hook(sk);
4276	}
4277	spin_unlock(&po->bind_lock);
4278	if (closing && (po->tp_version > TPACKET_V2)) {
4279		/* Because we don't support block-based V3 on tx-ring */
4280		if (!tx_ring)
4281			prb_shutdown_retire_blk_timer(po, rb_queue);
4282	}
4283	release_sock(sk);
4284
4285	if (pg_vec)
 
 
4286		free_pg_vec(pg_vec, order, req->tp_block_nr);
 
4287out:
4288	return err;
4289}
4290
4291static int packet_mmap(struct file *file, struct socket *sock,
4292		struct vm_area_struct *vma)
4293{
4294	struct sock *sk = sock->sk;
4295	struct packet_sock *po = pkt_sk(sk);
4296	unsigned long size, expected_size;
4297	struct packet_ring_buffer *rb;
4298	unsigned long start;
4299	int err = -EINVAL;
4300	int i;
4301
4302	if (vma->vm_pgoff)
4303		return -EINVAL;
4304
4305	mutex_lock(&po->pg_vec_lock);
4306
4307	expected_size = 0;
4308	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4309		if (rb->pg_vec) {
4310			expected_size += rb->pg_vec_len
4311						* rb->pg_vec_pages
4312						* PAGE_SIZE;
4313		}
4314	}
4315
4316	if (expected_size == 0)
4317		goto out;
4318
4319	size = vma->vm_end - vma->vm_start;
4320	if (size != expected_size)
4321		goto out;
4322
4323	start = vma->vm_start;
4324	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4325		if (rb->pg_vec == NULL)
4326			continue;
4327
4328		for (i = 0; i < rb->pg_vec_len; i++) {
4329			struct page *page;
4330			void *kaddr = rb->pg_vec[i].buffer;
4331			int pg_num;
4332
4333			for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4334				page = pgv_to_page(kaddr);
4335				err = vm_insert_page(vma, start, page);
4336				if (unlikely(err))
4337					goto out;
4338				start += PAGE_SIZE;
4339				kaddr += PAGE_SIZE;
4340			}
4341		}
4342	}
4343
4344	atomic_inc(&po->mapped);
4345	vma->vm_ops = &packet_mmap_ops;
4346	err = 0;
4347
4348out:
4349	mutex_unlock(&po->pg_vec_lock);
4350	return err;
4351}
4352
4353static const struct proto_ops packet_ops_spkt = {
4354	.family =	PF_PACKET,
4355	.owner =	THIS_MODULE,
4356	.release =	packet_release,
4357	.bind =		packet_bind_spkt,
4358	.connect =	sock_no_connect,
4359	.socketpair =	sock_no_socketpair,
4360	.accept =	sock_no_accept,
4361	.getname =	packet_getname_spkt,
4362	.poll =		datagram_poll,
4363	.ioctl =	packet_ioctl,
 
4364	.listen =	sock_no_listen,
4365	.shutdown =	sock_no_shutdown,
4366	.setsockopt =	sock_no_setsockopt,
4367	.getsockopt =	sock_no_getsockopt,
4368	.sendmsg =	packet_sendmsg_spkt,
4369	.recvmsg =	packet_recvmsg,
4370	.mmap =		sock_no_mmap,
4371	.sendpage =	sock_no_sendpage,
4372};
4373
4374static const struct proto_ops packet_ops = {
4375	.family =	PF_PACKET,
4376	.owner =	THIS_MODULE,
4377	.release =	packet_release,
4378	.bind =		packet_bind,
4379	.connect =	sock_no_connect,
4380	.socketpair =	sock_no_socketpair,
4381	.accept =	sock_no_accept,
4382	.getname =	packet_getname,
4383	.poll =		packet_poll,
4384	.ioctl =	packet_ioctl,
 
4385	.listen =	sock_no_listen,
4386	.shutdown =	sock_no_shutdown,
4387	.setsockopt =	packet_setsockopt,
4388	.getsockopt =	packet_getsockopt,
4389	.sendmsg =	packet_sendmsg,
4390	.recvmsg =	packet_recvmsg,
4391	.mmap =		packet_mmap,
4392	.sendpage =	sock_no_sendpage,
4393};
4394
4395static const struct net_proto_family packet_family_ops = {
4396	.family =	PF_PACKET,
4397	.create =	packet_create,
4398	.owner	=	THIS_MODULE,
4399};
4400
4401static struct notifier_block packet_netdev_notifier = {
4402	.notifier_call =	packet_notifier,
4403};
4404
4405#ifdef CONFIG_PROC_FS
4406
4407static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4408	__acquires(RCU)
4409{
4410	struct net *net = seq_file_net(seq);
4411
4412	rcu_read_lock();
4413	return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4414}
4415
4416static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4417{
4418	struct net *net = seq_file_net(seq);
4419	return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4420}
4421
4422static void packet_seq_stop(struct seq_file *seq, void *v)
4423	__releases(RCU)
4424{
4425	rcu_read_unlock();
4426}
4427
4428static int packet_seq_show(struct seq_file *seq, void *v)
4429{
4430	if (v == SEQ_START_TOKEN)
4431		seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
 
 
4432	else {
4433		struct sock *s = sk_entry(v);
4434		const struct packet_sock *po = pkt_sk(s);
4435
4436		seq_printf(seq,
4437			   "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4438			   s,
4439			   atomic_read(&s->sk_refcnt),
4440			   s->sk_type,
4441			   ntohs(po->num),
4442			   po->ifindex,
4443			   po->running,
4444			   atomic_read(&s->sk_rmem_alloc),
4445			   from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4446			   sock_i_ino(s));
4447	}
4448
4449	return 0;
4450}
4451
4452static const struct seq_operations packet_seq_ops = {
4453	.start	= packet_seq_start,
4454	.next	= packet_seq_next,
4455	.stop	= packet_seq_stop,
4456	.show	= packet_seq_show,
4457};
4458
4459static int packet_seq_open(struct inode *inode, struct file *file)
4460{
4461	return seq_open_net(inode, file, &packet_seq_ops,
4462			    sizeof(struct seq_net_private));
4463}
4464
4465static const struct file_operations packet_seq_fops = {
4466	.owner		= THIS_MODULE,
4467	.open		= packet_seq_open,
4468	.read		= seq_read,
4469	.llseek		= seq_lseek,
4470	.release	= seq_release_net,
4471};
4472
4473#endif
4474
4475static int __net_init packet_net_init(struct net *net)
4476{
4477	mutex_init(&net->packet.sklist_lock);
4478	INIT_HLIST_HEAD(&net->packet.sklist);
4479
4480	if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
 
 
4481		return -ENOMEM;
 
4482
4483	return 0;
4484}
4485
4486static void __net_exit packet_net_exit(struct net *net)
4487{
4488	remove_proc_entry("packet", net->proc_net);
 
4489}
4490
4491static struct pernet_operations packet_net_ops = {
4492	.init = packet_net_init,
4493	.exit = packet_net_exit,
4494};
4495
4496
4497static void __exit packet_exit(void)
4498{
4499	unregister_netdevice_notifier(&packet_netdev_notifier);
4500	unregister_pernet_subsys(&packet_net_ops);
4501	sock_unregister(PF_PACKET);
4502	proto_unregister(&packet_proto);
 
 
4503}
4504
4505static int __init packet_init(void)
4506{
4507	int rc = proto_register(&packet_proto, 0);
4508
4509	if (rc != 0)
 
4510		goto out;
 
 
 
 
 
 
 
 
 
 
 
4511
4512	sock_register(&packet_family_ops);
4513	register_pernet_subsys(&packet_net_ops);
4514	register_netdevice_notifier(&packet_netdev_notifier);
 
 
 
4515out:
4516	return rc;
4517}
4518
4519module_init(packet_init);
4520module_exit(packet_exit);
 
4521MODULE_LICENSE("GPL");
4522MODULE_ALIAS_NETPROTO(PF_PACKET);
v6.13.7
   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_xmit(const struct packet_sock *po, struct sk_buff *skb)
 274{
 275	if (!packet_sock_flag(po, PACKET_SOCK_QDISC_BYPASS))
 276		return dev_queue_xmit(skb);
 277
 278#ifdef CONFIG_NETFILTER_EGRESS
 279	if (nf_hook_egress_active()) {
 280		skb = nf_hook_direct_egress(skb);
 281		if (!skb)
 282			return NET_XMIT_DROP;
 283	}
 284#endif
 285	return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
 286}
 287
 288static struct net_device *packet_cached_dev_get(struct packet_sock *po)
 289{
 290	struct net_device *dev;
 291
 292	rcu_read_lock();
 293	dev = rcu_dereference(po->cached_dev);
 294	dev_hold(dev);
 
 295	rcu_read_unlock();
 296
 297	return dev;
 298}
 299
 300static void packet_cached_dev_assign(struct packet_sock *po,
 301				     struct net_device *dev)
 302{
 303	rcu_assign_pointer(po->cached_dev, dev);
 304}
 305
 306static void packet_cached_dev_reset(struct packet_sock *po)
 307{
 308	RCU_INIT_POINTER(po->cached_dev, NULL);
 309}
 310
 311static u16 packet_pick_tx_queue(struct sk_buff *skb)
 
 
 
 
 
 
 
 
 
 
 312{
 313	struct net_device *dev = skb->dev;
 314	const struct net_device_ops *ops = dev->netdev_ops;
 315	int cpu = raw_smp_processor_id();
 316	u16 queue_index;
 317
 318#ifdef CONFIG_XPS
 319	skb->sender_cpu = cpu + 1;
 320#endif
 321	skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues);
 322	if (ops->ndo_select_queue) {
 323		queue_index = ops->ndo_select_queue(dev, skb, NULL);
 
 324		queue_index = netdev_cap_txqueue(dev, queue_index);
 325	} else {
 326		queue_index = netdev_pick_tx(dev, skb, NULL);
 327	}
 328
 329	return queue_index;
 330}
 331
 332/* __register_prot_hook must be invoked through register_prot_hook
 333 * or from a context in which asynchronous accesses to the packet
 334 * socket is not possible (packet_create()).
 335 */
 336static void __register_prot_hook(struct sock *sk)
 337{
 338	struct packet_sock *po = pkt_sk(sk);
 339
 340	if (!packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
 341		if (po->fanout)
 342			__fanout_link(sk, po);
 343		else
 344			dev_add_pack(&po->prot_hook);
 345
 346		sock_hold(sk);
 347		packet_sock_flag_set(po, PACKET_SOCK_RUNNING, 1);
 348	}
 349}
 350
 351static void register_prot_hook(struct sock *sk)
 352{
 353	lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
 354	__register_prot_hook(sk);
 355}
 356
 357/* If the sync parameter is true, we will temporarily drop
 358 * the po->bind_lock and do a synchronize_net to make sure no
 359 * asynchronous packet processing paths still refer to the elements
 360 * of po->prot_hook.  If the sync parameter is false, it is the
 361 * callers responsibility to take care of this.
 362 */
 363static void __unregister_prot_hook(struct sock *sk, bool sync)
 364{
 365	struct packet_sock *po = pkt_sk(sk);
 366
 367	lockdep_assert_held_once(&po->bind_lock);
 368
 369	packet_sock_flag_set(po, PACKET_SOCK_RUNNING, 0);
 370
 371	if (po->fanout)
 372		__fanout_unlink(sk, po);
 373	else
 374		__dev_remove_pack(&po->prot_hook);
 375
 376	__sock_put(sk);
 377
 378	if (sync) {
 379		spin_unlock(&po->bind_lock);
 380		synchronize_net();
 381		spin_lock(&po->bind_lock);
 382	}
 383}
 384
 385static void unregister_prot_hook(struct sock *sk, bool sync)
 386{
 387	struct packet_sock *po = pkt_sk(sk);
 388
 389	if (packet_sock_flag(po, PACKET_SOCK_RUNNING))
 390		__unregister_prot_hook(sk, sync);
 391}
 392
 393static inline struct page * __pure pgv_to_page(void *addr)
 394{
 395	if (is_vmalloc_addr(addr))
 396		return vmalloc_to_page(addr);
 397	return virt_to_page(addr);
 398}
 399
 400static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 401{
 402	union tpacket_uhdr h;
 403
 404	/* WRITE_ONCE() are paired with READ_ONCE() in __packet_get_status */
 405
 406	h.raw = frame;
 407	switch (po->tp_version) {
 408	case TPACKET_V1:
 409		WRITE_ONCE(h.h1->tp_status, status);
 410		flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 411		break;
 412	case TPACKET_V2:
 413		WRITE_ONCE(h.h2->tp_status, status);
 414		flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 415		break;
 416	case TPACKET_V3:
 417		WRITE_ONCE(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	/* READ_ONCE() are paired with WRITE_ONCE() in __packet_set_status */
 435
 436	h.raw = frame;
 437	switch (po->tp_version) {
 438	case TPACKET_V1:
 439		flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 440		return READ_ONCE(h.h1->tp_status);
 441	case TPACKET_V2:
 442		flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 443		return READ_ONCE(h.h2->tp_status);
 444	case TPACKET_V3:
 445		flush_dcache_page(pgv_to_page(&h.h3->tp_status));
 446		return READ_ONCE(h.h3->tp_status);
 447	default:
 448		WARN(1, "TPACKET version not supported.\n");
 449		BUG();
 450		return 0;
 451	}
 452}
 453
 454static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec64 *ts,
 455				   unsigned int flags)
 456{
 457	struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 458
 459	if (shhwtstamps &&
 460	    (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
 461	    ktime_to_timespec64_cond(shhwtstamps->hwtstamp, ts))
 462		return TP_STATUS_TS_RAW_HARDWARE;
 463
 464	if ((flags & SOF_TIMESTAMPING_SOFTWARE) &&
 465	    ktime_to_timespec64_cond(skb_tstamp(skb), ts))
 466		return TP_STATUS_TS_SOFTWARE;
 467
 468	return 0;
 469}
 470
 471static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
 472				    struct sk_buff *skb)
 473{
 474	union tpacket_uhdr h;
 475	struct timespec64 ts;
 476	__u32 ts_status;
 477
 478	if (!(ts_status = tpacket_get_timestamp(skb, &ts, READ_ONCE(po->tp_tstamp))))
 479		return 0;
 480
 481	h.raw = frame;
 482	/*
 483	 * versions 1 through 3 overflow the timestamps in y2106, since they
 484	 * all store the seconds in a 32-bit unsigned integer.
 485	 * If we create a version 4, that should have a 64-bit timestamp,
 486	 * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit
 487	 * nanoseconds.
 488	 */
 489	switch (po->tp_version) {
 490	case TPACKET_V1:
 491		h.h1->tp_sec = ts.tv_sec;
 492		h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
 493		break;
 494	case TPACKET_V2:
 495		h.h2->tp_sec = ts.tv_sec;
 496		h.h2->tp_nsec = ts.tv_nsec;
 497		break;
 498	case TPACKET_V3:
 499		h.h3->tp_sec = ts.tv_sec;
 500		h.h3->tp_nsec = ts.tv_nsec;
 501		break;
 502	default:
 503		WARN(1, "TPACKET version not supported.\n");
 504		BUG();
 505	}
 506
 507	/* one flush is safe, as both fields always lie on the same cacheline */
 508	flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
 509	smp_wmb();
 510
 511	return ts_status;
 512}
 513
 514static void *packet_lookup_frame(const struct packet_sock *po,
 515				 const struct packet_ring_buffer *rb,
 516				 unsigned int position,
 517				 int status)
 518{
 519	unsigned int pg_vec_pos, frame_offset;
 520	union tpacket_uhdr h;
 521
 522	pg_vec_pos = position / rb->frames_per_block;
 523	frame_offset = position % rb->frames_per_block;
 524
 525	h.raw = rb->pg_vec[pg_vec_pos].buffer +
 526		(frame_offset * rb->frame_size);
 527
 528	if (status != __packet_get_status(po, h.raw))
 529		return NULL;
 530
 531	return h.raw;
 532}
 533
 534static void *packet_current_frame(struct packet_sock *po,
 535		struct packet_ring_buffer *rb,
 536		int status)
 537{
 538	return packet_lookup_frame(po, rb, rb->head, status);
 539}
 540
 541static u16 vlan_get_tci(const struct sk_buff *skb, struct net_device *dev)
 542{
 543	struct vlan_hdr vhdr, *vh;
 544	unsigned int header_len;
 545
 546	if (!dev)
 547		return 0;
 548
 549	/* In the SOCK_DGRAM scenario, skb data starts at the network
 550	 * protocol, which is after the VLAN headers. The outer VLAN
 551	 * header is at the hard_header_len offset in non-variable
 552	 * length link layer headers. If it's a VLAN device, the
 553	 * min_header_len should be used to exclude the VLAN header
 554	 * size.
 555	 */
 556	if (dev->min_header_len == dev->hard_header_len)
 557		header_len = dev->hard_header_len;
 558	else if (is_vlan_dev(dev))
 559		header_len = dev->min_header_len;
 560	else
 561		return 0;
 562
 563	vh = skb_header_pointer(skb, skb_mac_offset(skb) + header_len,
 564				sizeof(vhdr), &vhdr);
 565	if (unlikely(!vh))
 566		return 0;
 567
 568	return ntohs(vh->h_vlan_TCI);
 569}
 570
 571static __be16 vlan_get_protocol_dgram(const struct sk_buff *skb)
 572{
 573	__be16 proto = skb->protocol;
 574
 575	if (unlikely(eth_type_vlan(proto)))
 576		proto = __vlan_get_protocol_offset(skb, proto,
 577						   skb_mac_offset(skb), NULL);
 578
 579	return proto;
 580}
 581
 582static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 583{
 584	del_timer_sync(&pkc->retire_blk_timer);
 585}
 586
 587static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 588		struct sk_buff_head *rb_queue)
 589{
 590	struct tpacket_kbdq_core *pkc;
 591
 592	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 593
 594	spin_lock_bh(&rb_queue->lock);
 595	pkc->delete_blk_timer = 1;
 596	spin_unlock_bh(&rb_queue->lock);
 597
 598	prb_del_retire_blk_timer(pkc);
 599}
 600
 
 
 
 
 
 
 
 
 
 
 601static void prb_setup_retire_blk_timer(struct packet_sock *po)
 602{
 603	struct tpacket_kbdq_core *pkc;
 604
 605	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 606	timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
 607		    0);
 608	pkc->retire_blk_timer.expires = jiffies;
 609}
 610
 611static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 612				int blk_size_in_bytes)
 613{
 614	struct net_device *dev;
 615	unsigned int mbits, div;
 616	struct ethtool_link_ksettings ecmd;
 617	int err;
 618
 619	rtnl_lock();
 620	dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 621	if (unlikely(!dev)) {
 622		rtnl_unlock();
 623		return DEFAULT_PRB_RETIRE_TOV;
 624	}
 625	err = __ethtool_get_link_ksettings(dev, &ecmd);
 626	rtnl_unlock();
 627	if (err)
 628		return DEFAULT_PRB_RETIRE_TOV;
 629
 630	/* If the link speed is so slow you don't really
 631	 * need to worry about perf anyways
 632	 */
 633	if (ecmd.base.speed < SPEED_1000 ||
 634	    ecmd.base.speed == SPEED_UNKNOWN)
 635		return DEFAULT_PRB_RETIRE_TOV;
 
 
 
 
 636
 637	div = ecmd.base.speed / 1000;
 638	mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 639
 640	if (div)
 641		mbits /= div;
 642
 
 
 643	if (div)
 644		return mbits + 1;
 645	return mbits;
 646}
 647
 648static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 649			union tpacket_req_u *req_u)
 650{
 651	p1->feature_req_word = req_u->req3.tp_feature_req_word;
 652}
 653
 654static void init_prb_bdqc(struct packet_sock *po,
 655			struct packet_ring_buffer *rb,
 656			struct pgv *pg_vec,
 657			union tpacket_req_u *req_u)
 658{
 659	struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
 660	struct tpacket_block_desc *pbd;
 661
 662	memset(p1, 0x0, sizeof(*p1));
 663
 664	p1->knxt_seq_num = 1;
 665	p1->pkbdq = pg_vec;
 666	pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 667	p1->pkblk_start	= pg_vec[0].buffer;
 668	p1->kblk_size = req_u->req3.tp_block_size;
 669	p1->knum_blocks	= req_u->req3.tp_block_nr;
 670	p1->hdrlen = po->tp_hdrlen;
 671	p1->version = po->tp_version;
 672	p1->last_kactive_blk_num = 0;
 673	po->stats.stats3.tp_freeze_q_cnt = 0;
 674	if (req_u->req3.tp_retire_blk_tov)
 675		p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 676	else
 677		p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 678						req_u->req3.tp_block_size);
 679	p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 680	p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 681	rwlock_init(&p1->blk_fill_in_prog_lock);
 682
 683	p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
 684	prb_init_ft_ops(p1, req_u);
 685	prb_setup_retire_blk_timer(po);
 686	prb_open_block(p1, pbd);
 687}
 688
 689/*  Do NOT update the last_blk_num first.
 690 *  Assumes sk_buff_head lock is held.
 691 */
 692static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 693{
 694	mod_timer(&pkc->retire_blk_timer,
 695			jiffies + pkc->tov_in_jiffies);
 696	pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 697}
 698
 699/*
 700 * Timer logic:
 701 * 1) We refresh the timer only when we open a block.
 702 *    By doing this we don't waste cycles refreshing the timer
 703 *	  on packet-by-packet basis.
 704 *
 705 * With a 1MB block-size, on a 1Gbps line, it will take
 706 * i) ~8 ms to fill a block + ii) memcpy etc.
 707 * In this cut we are not accounting for the memcpy time.
 708 *
 709 * So, if the user sets the 'tmo' to 10ms then the timer
 710 * will never fire while the block is still getting filled
 711 * (which is what we want). However, the user could choose
 712 * to close a block early and that's fine.
 713 *
 714 * But when the timer does fire, we check whether or not to refresh it.
 715 * Since the tmo granularity is in msecs, it is not too expensive
 716 * to refresh the timer, lets say every '8' msecs.
 717 * Either the user can set the 'tmo' or we can derive it based on
 718 * a) line-speed and b) block-size.
 719 * prb_calc_retire_blk_tmo() calculates the tmo.
 720 *
 721 */
 722static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
 723{
 724	struct packet_sock *po =
 725		from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
 726	struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 727	unsigned int frozen;
 728	struct tpacket_block_desc *pbd;
 729
 730	spin_lock(&po->sk.sk_receive_queue.lock);
 731
 732	frozen = prb_queue_frozen(pkc);
 733	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 734
 735	if (unlikely(pkc->delete_blk_timer))
 736		goto out;
 737
 738	/* We only need to plug the race when the block is partially filled.
 739	 * tpacket_rcv:
 740	 *		lock(); increment BLOCK_NUM_PKTS; unlock()
 741	 *		copy_bits() is in progress ...
 742	 *		timer fires on other cpu:
 743	 *		we can't retire the current block because copy_bits
 744	 *		is in progress.
 745	 *
 746	 */
 747	if (BLOCK_NUM_PKTS(pbd)) {
 748		/* Waiting for skb_copy_bits to finish... */
 749		write_lock(&pkc->blk_fill_in_prog_lock);
 750		write_unlock(&pkc->blk_fill_in_prog_lock);
 
 751	}
 752
 753	if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 754		if (!frozen) {
 755			if (!BLOCK_NUM_PKTS(pbd)) {
 756				/* An empty block. Just refresh the timer. */
 757				goto refresh_timer;
 758			}
 759			prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 760			if (!prb_dispatch_next_block(pkc, po))
 761				goto refresh_timer;
 762			else
 763				goto out;
 764		} else {
 765			/* Case 1. Queue was frozen because user-space was
 766			 *	   lagging behind.
 767			 */
 768			if (prb_curr_blk_in_use(pbd)) {
 769				/*
 770				 * Ok, user-space is still behind.
 771				 * So just refresh the timer.
 772				 */
 773				goto refresh_timer;
 774			} else {
 775			       /* Case 2. queue was frozen,user-space caught up,
 776				* now the link went idle && the timer fired.
 777				* We don't have a block to close.So we open this
 778				* block and restart the timer.
 779				* opening a block thaws the queue,restarts timer
 780				* Thawing/timer-refresh is a side effect.
 781				*/
 782				prb_open_block(pkc, pbd);
 783				goto out;
 784			}
 785		}
 786	}
 787
 788refresh_timer:
 789	_prb_refresh_rx_retire_blk_timer(pkc);
 790
 791out:
 792	spin_unlock(&po->sk.sk_receive_queue.lock);
 793}
 794
 795static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 796		struct tpacket_block_desc *pbd1, __u32 status)
 797{
 798	/* Flush everything minus the block header */
 799
 800#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 801	u8 *start, *end;
 802
 803	start = (u8 *)pbd1;
 804
 805	/* Skip the block header(we know header WILL fit in 4K) */
 806	start += PAGE_SIZE;
 807
 808	end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 809	for (; start < end; start += PAGE_SIZE)
 810		flush_dcache_page(pgv_to_page(start));
 811
 812	smp_wmb();
 813#endif
 814
 815	/* Now update the block status. */
 816
 817	BLOCK_STATUS(pbd1) = status;
 818
 819	/* Flush the block header */
 820
 821#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 822	start = (u8 *)pbd1;
 823	flush_dcache_page(pgv_to_page(start));
 824
 825	smp_wmb();
 826#endif
 827}
 828
 829/*
 830 * Side effect:
 831 *
 832 * 1) flush the block
 833 * 2) Increment active_blk_num
 834 *
 835 * Note:We DONT refresh the timer on purpose.
 836 *	Because almost always the next block will be opened.
 837 */
 838static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 839		struct tpacket_block_desc *pbd1,
 840		struct packet_sock *po, unsigned int stat)
 841{
 842	__u32 status = TP_STATUS_USER | stat;
 843
 844	struct tpacket3_hdr *last_pkt;
 845	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 846	struct sock *sk = &po->sk;
 847
 848	if (atomic_read(&po->tp_drops))
 849		status |= TP_STATUS_LOSING;
 850
 851	last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 852	last_pkt->tp_next_offset = 0;
 853
 854	/* Get the ts of the last pkt */
 855	if (BLOCK_NUM_PKTS(pbd1)) {
 856		h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 857		h1->ts_last_pkt.ts_nsec	= last_pkt->tp_nsec;
 858	} else {
 859		/* Ok, we tmo'd - so get the current time.
 860		 *
 861		 * It shouldn't really happen as we don't close empty
 862		 * blocks. See prb_retire_rx_blk_timer_expired().
 863		 */
 864		struct timespec64 ts;
 865		ktime_get_real_ts64(&ts);
 866		h1->ts_last_pkt.ts_sec = ts.tv_sec;
 867		h1->ts_last_pkt.ts_nsec	= ts.tv_nsec;
 868	}
 869
 870	smp_wmb();
 871
 872	/* Flush the block */
 873	prb_flush_block(pkc1, pbd1, status);
 874
 875	sk->sk_data_ready(sk);
 876
 877	pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 878}
 879
 880static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 881{
 882	pkc->reset_pending_on_curr_blk = 0;
 883}
 884
 885/*
 886 * Side effect of opening a block:
 887 *
 888 * 1) prb_queue is thawed.
 889 * 2) retire_blk_timer is refreshed.
 890 *
 891 */
 892static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 893	struct tpacket_block_desc *pbd1)
 894{
 895	struct timespec64 ts;
 896	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 897
 898	smp_rmb();
 899
 900	/* We could have just memset this but we will lose the
 901	 * flexibility of making the priv area sticky
 902	 */
 903
 904	BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 905	BLOCK_NUM_PKTS(pbd1) = 0;
 906	BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 907
 908	ktime_get_real_ts64(&ts);
 909
 910	h1->ts_first_pkt.ts_sec = ts.tv_sec;
 911	h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 912
 913	pkc1->pkblk_start = (char *)pbd1;
 914	pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 915
 916	BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 917	BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 918
 919	pbd1->version = pkc1->version;
 920	pkc1->prev = pkc1->nxt_offset;
 921	pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 922
 923	prb_thaw_queue(pkc1);
 924	_prb_refresh_rx_retire_blk_timer(pkc1);
 925
 926	smp_wmb();
 927}
 928
 929/*
 930 * Queue freeze logic:
 931 * 1) Assume tp_block_nr = 8 blocks.
 932 * 2) At time 't0', user opens Rx ring.
 933 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 934 * 4) user-space is either sleeping or processing block '0'.
 935 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 936 *    it will close block-7,loop around and try to fill block '0'.
 937 *    call-flow:
 938 *    __packet_lookup_frame_in_block
 939 *      prb_retire_current_block()
 940 *      prb_dispatch_next_block()
 941 *        |->(BLOCK_STATUS == USER) evaluates to true
 942 *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 943 * 6) Now there are two cases:
 944 *    6.1) Link goes idle right after the queue is frozen.
 945 *         But remember, the last open_block() refreshed the timer.
 946 *         When this timer expires,it will refresh itself so that we can
 947 *         re-open block-0 in near future.
 948 *    6.2) Link is busy and keeps on receiving packets. This is a simple
 949 *         case and __packet_lookup_frame_in_block will check if block-0
 950 *         is free and can now be re-used.
 951 */
 952static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 953				  struct packet_sock *po)
 954{
 955	pkc->reset_pending_on_curr_blk = 1;
 956	po->stats.stats3.tp_freeze_q_cnt++;
 957}
 958
 959#define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 960
 961/*
 962 * If the next block is free then we will dispatch it
 963 * and return a good offset.
 964 * Else, we will freeze the queue.
 965 * So, caller must check the return value.
 966 */
 967static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 968		struct packet_sock *po)
 969{
 970	struct tpacket_block_desc *pbd;
 971
 972	smp_rmb();
 973
 974	/* 1. Get current block num */
 975	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 976
 977	/* 2. If this block is currently in_use then freeze the queue */
 978	if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 979		prb_freeze_queue(pkc, po);
 980		return NULL;
 981	}
 982
 983	/*
 984	 * 3.
 985	 * open this block and return the offset where the first packet
 986	 * needs to get stored.
 987	 */
 988	prb_open_block(pkc, pbd);
 989	return (void *)pkc->nxt_offset;
 990}
 991
 992static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 993		struct packet_sock *po, unsigned int status)
 994{
 995	struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 996
 997	/* retire/close the current block */
 998	if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 999		/*
1000		 * Plug the case where copy_bits() is in progress on
1001		 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
1002		 * have space to copy the pkt in the current block and
1003		 * called prb_retire_current_block()
1004		 *
1005		 * We don't need to worry about the TMO case because
1006		 * the timer-handler already handled this case.
1007		 */
1008		if (!(status & TP_STATUS_BLK_TMO)) {
1009			/* Waiting for skb_copy_bits to finish... */
1010			write_lock(&pkc->blk_fill_in_prog_lock);
1011			write_unlock(&pkc->blk_fill_in_prog_lock);
 
1012		}
1013		prb_close_block(pkc, pbd, po, status);
1014		return;
1015	}
1016}
1017
1018static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
 
1019{
1020	return TP_STATUS_USER & BLOCK_STATUS(pbd);
1021}
1022
1023static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
1024{
1025	return pkc->reset_pending_on_curr_blk;
1026}
1027
1028static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
1029	__releases(&pkc->blk_fill_in_prog_lock)
1030{
1031	struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1032
1033	read_unlock(&pkc->blk_fill_in_prog_lock);
1034}
1035
1036static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
1037			struct tpacket3_hdr *ppd)
1038{
1039	ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
1040}
1041
1042static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
1043			struct tpacket3_hdr *ppd)
1044{
1045	ppd->hv1.tp_rxhash = 0;
1046}
1047
1048static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1049			struct tpacket3_hdr *ppd)
1050{
1051	struct packet_sock *po = container_of(pkc, struct packet_sock, rx_ring.prb_bdqc);
1052
1053	if (skb_vlan_tag_present(pkc->skb)) {
1054		ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1055		ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1056		ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1057	} else if (unlikely(po->sk.sk_type == SOCK_DGRAM && eth_type_vlan(pkc->skb->protocol))) {
1058		ppd->hv1.tp_vlan_tci = vlan_get_tci(pkc->skb, pkc->skb->dev);
1059		ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->protocol);
1060		ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1061	} else {
1062		ppd->hv1.tp_vlan_tci = 0;
1063		ppd->hv1.tp_vlan_tpid = 0;
1064		ppd->tp_status = TP_STATUS_AVAILABLE;
1065	}
1066}
1067
1068static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1069			struct tpacket3_hdr *ppd)
1070{
1071	ppd->hv1.tp_padding = 0;
1072	prb_fill_vlan_info(pkc, ppd);
1073
1074	if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1075		prb_fill_rxhash(pkc, ppd);
1076	else
1077		prb_clear_rxhash(pkc, ppd);
1078}
1079
1080static void prb_fill_curr_block(char *curr,
1081				struct tpacket_kbdq_core *pkc,
1082				struct tpacket_block_desc *pbd,
1083				unsigned int len)
1084	__acquires(&pkc->blk_fill_in_prog_lock)
1085{
1086	struct tpacket3_hdr *ppd;
1087
1088	ppd  = (struct tpacket3_hdr *)curr;
1089	ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1090	pkc->prev = curr;
1091	pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1092	BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1093	BLOCK_NUM_PKTS(pbd) += 1;
1094	read_lock(&pkc->blk_fill_in_prog_lock);
1095	prb_run_all_ft_ops(pkc, ppd);
1096}
1097
1098/* Assumes caller has the sk->rx_queue.lock */
1099static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1100					    struct sk_buff *skb,
 
1101					    unsigned int len
1102					    )
1103{
1104	struct tpacket_kbdq_core *pkc;
1105	struct tpacket_block_desc *pbd;
1106	char *curr, *end;
1107
1108	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1109	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1110
1111	/* Queue is frozen when user space is lagging behind */
1112	if (prb_queue_frozen(pkc)) {
1113		/*
1114		 * Check if that last block which caused the queue to freeze,
1115		 * is still in_use by user-space.
1116		 */
1117		if (prb_curr_blk_in_use(pbd)) {
1118			/* Can't record this packet */
1119			return NULL;
1120		} else {
1121			/*
1122			 * Ok, the block was released by user-space.
1123			 * Now let's open that block.
1124			 * opening a block also thaws the queue.
1125			 * Thawing is a side effect.
1126			 */
1127			prb_open_block(pkc, pbd);
1128		}
1129	}
1130
1131	smp_mb();
1132	curr = pkc->nxt_offset;
1133	pkc->skb = skb;
1134	end = (char *)pbd + pkc->kblk_size;
1135
1136	/* first try the current block */
1137	if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1138		prb_fill_curr_block(curr, pkc, pbd, len);
1139		return (void *)curr;
1140	}
1141
1142	/* Ok, close the current block */
1143	prb_retire_current_block(pkc, po, 0);
1144
1145	/* Now, try to dispatch the next block */
1146	curr = (char *)prb_dispatch_next_block(pkc, po);
1147	if (curr) {
1148		pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1149		prb_fill_curr_block(curr, pkc, pbd, len);
1150		return (void *)curr;
1151	}
1152
1153	/*
1154	 * No free blocks are available.user_space hasn't caught up yet.
1155	 * Queue was just frozen and now this packet will get dropped.
1156	 */
1157	return NULL;
1158}
1159
1160static void *packet_current_rx_frame(struct packet_sock *po,
1161					    struct sk_buff *skb,
1162					    int status, unsigned int len)
1163{
1164	char *curr = NULL;
1165	switch (po->tp_version) {
1166	case TPACKET_V1:
1167	case TPACKET_V2:
1168		curr = packet_lookup_frame(po, &po->rx_ring,
1169					po->rx_ring.head, status);
1170		return curr;
1171	case TPACKET_V3:
1172		return __packet_lookup_frame_in_block(po, skb, len);
1173	default:
1174		WARN(1, "TPACKET version not supported\n");
1175		BUG();
1176		return NULL;
1177	}
1178}
1179
1180static void *prb_lookup_block(const struct packet_sock *po,
1181			      const struct packet_ring_buffer *rb,
1182			      unsigned int idx,
1183			      int status)
1184{
1185	struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1186	struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1187
1188	if (status != BLOCK_STATUS(pbd))
1189		return NULL;
1190	return pbd;
1191}
1192
1193static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1194{
1195	unsigned int prev;
1196	if (rb->prb_bdqc.kactive_blk_num)
1197		prev = rb->prb_bdqc.kactive_blk_num-1;
1198	else
1199		prev = rb->prb_bdqc.knum_blocks-1;
1200	return prev;
1201}
1202
1203/* Assumes caller has held the rx_queue.lock */
1204static void *__prb_previous_block(struct packet_sock *po,
1205					 struct packet_ring_buffer *rb,
1206					 int status)
1207{
1208	unsigned int previous = prb_previous_blk_num(rb);
1209	return prb_lookup_block(po, rb, previous, status);
1210}
1211
1212static void *packet_previous_rx_frame(struct packet_sock *po,
1213					     struct packet_ring_buffer *rb,
1214					     int status)
1215{
1216	if (po->tp_version <= TPACKET_V2)
1217		return packet_previous_frame(po, rb, status);
1218
1219	return __prb_previous_block(po, rb, status);
1220}
1221
1222static void packet_increment_rx_head(struct packet_sock *po,
1223					    struct packet_ring_buffer *rb)
1224{
1225	switch (po->tp_version) {
1226	case TPACKET_V1:
1227	case TPACKET_V2:
1228		return packet_increment_head(rb);
1229	case TPACKET_V3:
1230	default:
1231		WARN(1, "TPACKET version not supported.\n");
1232		BUG();
1233		return;
1234	}
1235}
1236
1237static void *packet_previous_frame(struct packet_sock *po,
1238		struct packet_ring_buffer *rb,
1239		int status)
1240{
1241	unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1242	return packet_lookup_frame(po, rb, previous, status);
1243}
1244
1245static void packet_increment_head(struct packet_ring_buffer *buff)
1246{
1247	buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1248}
1249
1250static void packet_inc_pending(struct packet_ring_buffer *rb)
1251{
1252	this_cpu_inc(*rb->pending_refcnt);
1253}
1254
1255static void packet_dec_pending(struct packet_ring_buffer *rb)
1256{
1257	this_cpu_dec(*rb->pending_refcnt);
1258}
1259
1260static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1261{
1262	unsigned int refcnt = 0;
1263	int cpu;
1264
1265	/* We don't use pending refcount in rx_ring. */
1266	if (rb->pending_refcnt == NULL)
1267		return 0;
1268
1269	for_each_possible_cpu(cpu)
1270		refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1271
1272	return refcnt;
1273}
1274
1275static int packet_alloc_pending(struct packet_sock *po)
1276{
1277	po->rx_ring.pending_refcnt = NULL;
1278
1279	po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1280	if (unlikely(po->tx_ring.pending_refcnt == NULL))
1281		return -ENOBUFS;
1282
1283	return 0;
1284}
1285
1286static void packet_free_pending(struct packet_sock *po)
1287{
1288	free_percpu(po->tx_ring.pending_refcnt);
1289}
1290
1291#define ROOM_POW_OFF	2
1292#define ROOM_NONE	0x0
1293#define ROOM_LOW	0x1
1294#define ROOM_NORMAL	0x2
1295
1296static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1297{
1298	int idx, len;
1299
1300	len = READ_ONCE(po->rx_ring.frame_max) + 1;
1301	idx = READ_ONCE(po->rx_ring.head);
1302	if (pow_off)
1303		idx += len >> pow_off;
1304	if (idx >= len)
1305		idx -= len;
1306	return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1307}
1308
1309static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1310{
1311	int idx, len;
1312
1313	len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1314	idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1315	if (pow_off)
1316		idx += len >> pow_off;
1317	if (idx >= len)
1318		idx -= len;
1319	return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1320}
1321
1322static int __packet_rcv_has_room(const struct packet_sock *po,
1323				 const struct sk_buff *skb)
1324{
1325	const struct sock *sk = &po->sk;
1326	int ret = ROOM_NONE;
1327
1328	if (po->prot_hook.func != tpacket_rcv) {
1329		int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1330		int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1331				   - (skb ? skb->truesize : 0);
1332
1333		if (avail > (rcvbuf >> ROOM_POW_OFF))
1334			return ROOM_NORMAL;
1335		else if (avail > 0)
1336			return ROOM_LOW;
1337		else
1338			return ROOM_NONE;
1339	}
1340
1341	if (po->tp_version == TPACKET_V3) {
1342		if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1343			ret = ROOM_NORMAL;
1344		else if (__tpacket_v3_has_room(po, 0))
1345			ret = ROOM_LOW;
1346	} else {
1347		if (__tpacket_has_room(po, ROOM_POW_OFF))
1348			ret = ROOM_NORMAL;
1349		else if (__tpacket_has_room(po, 0))
1350			ret = ROOM_LOW;
1351	}
1352
1353	return ret;
1354}
1355
1356static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1357{
1358	bool pressure;
1359	int ret;
 
1360
 
1361	ret = __packet_rcv_has_room(po, skb);
1362	pressure = ret != ROOM_NORMAL;
1363
1364	if (packet_sock_flag(po, PACKET_SOCK_PRESSURE) != pressure)
1365		packet_sock_flag_set(po, PACKET_SOCK_PRESSURE, pressure);
1366
1367	return ret;
1368}
1369
1370static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1371{
1372	if (packet_sock_flag(po, PACKET_SOCK_PRESSURE) &&
1373	    __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1374		packet_sock_flag_set(po, PACKET_SOCK_PRESSURE, false);
1375}
1376
1377static void packet_sock_destruct(struct sock *sk)
1378{
1379	skb_queue_purge(&sk->sk_error_queue);
1380
1381	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1382	WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1383
1384	if (!sock_flag(sk, SOCK_DEAD)) {
1385		pr_err("Attempt to release alive packet socket: %p\n", sk);
1386		return;
1387	}
 
 
1388}
1389
1390static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1391{
1392	u32 *history = po->rollover->history;
1393	u32 victim, rxhash;
1394	int i, count = 0;
1395
1396	rxhash = skb_get_hash(skb);
1397	for (i = 0; i < ROLLOVER_HLEN; i++)
1398		if (READ_ONCE(history[i]) == rxhash)
1399			count++;
1400
1401	victim = get_random_u32_below(ROLLOVER_HLEN);
1402
1403	/* Avoid dirtying the cache line if possible */
1404	if (READ_ONCE(history[victim]) != rxhash)
1405		WRITE_ONCE(history[victim], rxhash);
1406
1407	return count > (ROLLOVER_HLEN >> 1);
1408}
1409
1410static unsigned int fanout_demux_hash(struct packet_fanout *f,
1411				      struct sk_buff *skb,
1412				      unsigned int num)
1413{
1414	return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1415}
1416
1417static unsigned int fanout_demux_lb(struct packet_fanout *f,
1418				    struct sk_buff *skb,
1419				    unsigned int num)
1420{
1421	unsigned int val = atomic_inc_return(&f->rr_cur);
1422
1423	return val % num;
1424}
1425
1426static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1427				     struct sk_buff *skb,
1428				     unsigned int num)
1429{
1430	return smp_processor_id() % num;
1431}
1432
1433static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1434				     struct sk_buff *skb,
1435				     unsigned int num)
1436{
1437	return get_random_u32_below(num);
1438}
1439
1440static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1441					  struct sk_buff *skb,
1442					  unsigned int idx, bool try_self,
1443					  unsigned int num)
1444{
1445	struct packet_sock *po, *po_next, *po_skip = NULL;
1446	unsigned int i, j, room = ROOM_NONE;
1447
1448	po = pkt_sk(rcu_dereference(f->arr[idx]));
1449
1450	if (try_self) {
1451		room = packet_rcv_has_room(po, skb);
1452		if (room == ROOM_NORMAL ||
1453		    (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1454			return idx;
1455		po_skip = po;
1456	}
1457
1458	i = j = min_t(int, po->rollover->sock, num - 1);
1459	do {
1460		po_next = pkt_sk(rcu_dereference(f->arr[i]));
1461		if (po_next != po_skip &&
1462		    !packet_sock_flag(po_next, PACKET_SOCK_PRESSURE) &&
1463		    packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1464			if (i != j)
1465				po->rollover->sock = i;
1466			atomic_long_inc(&po->rollover->num);
1467			if (room == ROOM_LOW)
1468				atomic_long_inc(&po->rollover->num_huge);
1469			return i;
1470		}
1471
1472		if (++i == num)
1473			i = 0;
1474	} while (i != j);
1475
1476	atomic_long_inc(&po->rollover->num_failed);
1477	return idx;
1478}
1479
1480static unsigned int fanout_demux_qm(struct packet_fanout *f,
1481				    struct sk_buff *skb,
1482				    unsigned int num)
1483{
1484	return skb_get_queue_mapping(skb) % num;
1485}
1486
1487static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1488				     struct sk_buff *skb,
1489				     unsigned int num)
1490{
1491	struct bpf_prog *prog;
1492	unsigned int ret = 0;
1493
1494	rcu_read_lock();
1495	prog = rcu_dereference(f->bpf_prog);
1496	if (prog)
1497		ret = bpf_prog_run_clear_cb(prog, skb) % num;
1498	rcu_read_unlock();
1499
1500	return ret;
1501}
1502
1503static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1504{
1505	return f->flags & (flag >> 8);
1506}
1507
1508static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1509			     struct packet_type *pt, struct net_device *orig_dev)
1510{
1511	struct packet_fanout *f = pt->af_packet_priv;
1512	unsigned int num = READ_ONCE(f->num_members);
1513	struct net *net = read_pnet(&f->net);
1514	struct packet_sock *po;
1515	unsigned int idx;
1516
1517	if (!net_eq(dev_net(dev), net) || !num) {
1518		kfree_skb(skb);
1519		return 0;
1520	}
1521
1522	if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1523		skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1524		if (!skb)
1525			return 0;
1526	}
1527	switch (f->type) {
1528	case PACKET_FANOUT_HASH:
1529	default:
1530		idx = fanout_demux_hash(f, skb, num);
1531		break;
1532	case PACKET_FANOUT_LB:
1533		idx = fanout_demux_lb(f, skb, num);
1534		break;
1535	case PACKET_FANOUT_CPU:
1536		idx = fanout_demux_cpu(f, skb, num);
1537		break;
1538	case PACKET_FANOUT_RND:
1539		idx = fanout_demux_rnd(f, skb, num);
1540		break;
1541	case PACKET_FANOUT_QM:
1542		idx = fanout_demux_qm(f, skb, num);
1543		break;
1544	case PACKET_FANOUT_ROLLOVER:
1545		idx = fanout_demux_rollover(f, skb, 0, false, num);
1546		break;
1547	case PACKET_FANOUT_CBPF:
1548	case PACKET_FANOUT_EBPF:
1549		idx = fanout_demux_bpf(f, skb, num);
1550		break;
1551	}
1552
1553	if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1554		idx = fanout_demux_rollover(f, skb, idx, true, num);
1555
1556	po = pkt_sk(rcu_dereference(f->arr[idx]));
1557	return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1558}
1559
1560DEFINE_MUTEX(fanout_mutex);
1561EXPORT_SYMBOL_GPL(fanout_mutex);
1562static LIST_HEAD(fanout_list);
1563static u16 fanout_next_id;
1564
1565static void __fanout_link(struct sock *sk, struct packet_sock *po)
1566{
1567	struct packet_fanout *f = po->fanout;
1568
1569	spin_lock(&f->lock);
1570	rcu_assign_pointer(f->arr[f->num_members], sk);
1571	smp_wmb();
1572	f->num_members++;
1573	if (f->num_members == 1)
1574		dev_add_pack(&f->prot_hook);
1575	spin_unlock(&f->lock);
1576}
1577
1578static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1579{
1580	struct packet_fanout *f = po->fanout;
1581	int i;
1582
1583	spin_lock(&f->lock);
1584	for (i = 0; i < f->num_members; i++) {
1585		if (rcu_dereference_protected(f->arr[i],
1586					      lockdep_is_held(&f->lock)) == sk)
1587			break;
1588	}
1589	BUG_ON(i >= f->num_members);
1590	rcu_assign_pointer(f->arr[i],
1591			   rcu_dereference_protected(f->arr[f->num_members - 1],
1592						     lockdep_is_held(&f->lock)));
1593	f->num_members--;
1594	if (f->num_members == 0)
1595		__dev_remove_pack(&f->prot_hook);
1596	spin_unlock(&f->lock);
1597}
1598
1599static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1600{
1601	if (sk->sk_family != PF_PACKET)
1602		return false;
1603
1604	return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1605}
1606
1607static void fanout_init_data(struct packet_fanout *f)
1608{
1609	switch (f->type) {
1610	case PACKET_FANOUT_LB:
1611		atomic_set(&f->rr_cur, 0);
1612		break;
1613	case PACKET_FANOUT_CBPF:
1614	case PACKET_FANOUT_EBPF:
1615		RCU_INIT_POINTER(f->bpf_prog, NULL);
1616		break;
1617	}
1618}
1619
1620static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1621{
1622	struct bpf_prog *old;
1623
1624	spin_lock(&f->lock);
1625	old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1626	rcu_assign_pointer(f->bpf_prog, new);
1627	spin_unlock(&f->lock);
1628
1629	if (old) {
1630		synchronize_net();
1631		bpf_prog_destroy(old);
1632	}
1633}
1634
1635static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data,
1636				unsigned int len)
1637{
1638	struct bpf_prog *new;
1639	struct sock_fprog fprog;
1640	int ret;
1641
1642	if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1643		return -EPERM;
1644
1645	ret = copy_bpf_fprog_from_user(&fprog, data, len);
1646	if (ret)
1647		return ret;
1648
1649	ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1650	if (ret)
1651		return ret;
1652
1653	__fanout_set_data_bpf(po->fanout, new);
1654	return 0;
1655}
1656
1657static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data,
1658				unsigned int len)
1659{
1660	struct bpf_prog *new;
1661	u32 fd;
1662
1663	if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1664		return -EPERM;
1665	if (len != sizeof(fd))
1666		return -EINVAL;
1667	if (copy_from_sockptr(&fd, data, len))
1668		return -EFAULT;
1669
1670	new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1671	if (IS_ERR(new))
1672		return PTR_ERR(new);
 
 
 
 
1673
1674	__fanout_set_data_bpf(po->fanout, new);
1675	return 0;
1676}
1677
1678static int fanout_set_data(struct packet_sock *po, sockptr_t data,
1679			   unsigned int len)
1680{
1681	switch (po->fanout->type) {
1682	case PACKET_FANOUT_CBPF:
1683		return fanout_set_data_cbpf(po, data, len);
1684	case PACKET_FANOUT_EBPF:
1685		return fanout_set_data_ebpf(po, data, len);
1686	default:
1687		return -EINVAL;
1688	}
1689}
1690
1691static void fanout_release_data(struct packet_fanout *f)
1692{
1693	switch (f->type) {
1694	case PACKET_FANOUT_CBPF:
1695	case PACKET_FANOUT_EBPF:
1696		__fanout_set_data_bpf(f, NULL);
1697	}
1698}
1699
1700static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1701{
1702	struct packet_fanout *f;
1703
1704	list_for_each_entry(f, &fanout_list, list) {
1705		if (f->id == candidate_id &&
1706		    read_pnet(&f->net) == sock_net(sk)) {
1707			return false;
1708		}
1709	}
1710	return true;
1711}
1712
1713static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1714{
1715	u16 id = fanout_next_id;
1716
1717	do {
1718		if (__fanout_id_is_free(sk, id)) {
1719			*new_id = id;
1720			fanout_next_id = id + 1;
1721			return true;
1722		}
1723
1724		id++;
1725	} while (id != fanout_next_id);
1726
1727	return false;
1728}
1729
1730static int fanout_add(struct sock *sk, struct fanout_args *args)
1731{
1732	struct packet_rollover *rollover = NULL;
1733	struct packet_sock *po = pkt_sk(sk);
1734	u16 type_flags = args->type_flags;
1735	struct packet_fanout *f, *match;
1736	u8 type = type_flags & 0xff;
1737	u8 flags = type_flags >> 8;
1738	u16 id = args->id;
1739	int err;
1740
1741	switch (type) {
1742	case PACKET_FANOUT_ROLLOVER:
1743		if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1744			return -EINVAL;
1745		break;
1746	case PACKET_FANOUT_HASH:
1747	case PACKET_FANOUT_LB:
1748	case PACKET_FANOUT_CPU:
1749	case PACKET_FANOUT_RND:
1750	case PACKET_FANOUT_QM:
1751	case PACKET_FANOUT_CBPF:
1752	case PACKET_FANOUT_EBPF:
1753		break;
1754	default:
1755		return -EINVAL;
1756	}
1757
1758	mutex_lock(&fanout_mutex);
 
1759
1760	err = -EALREADY;
1761	if (po->fanout)
1762		goto out;
1763
1764	if (type == PACKET_FANOUT_ROLLOVER ||
1765	    (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1766		err = -ENOMEM;
1767		rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1768		if (!rollover)
1769			goto out;
1770		atomic_long_set(&rollover->num, 0);
1771		atomic_long_set(&rollover->num_huge, 0);
1772		atomic_long_set(&rollover->num_failed, 0);
1773	}
1774
1775	if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1776		if (id != 0) {
1777			err = -EINVAL;
1778			goto out;
1779		}
1780		if (!fanout_find_new_id(sk, &id)) {
1781			err = -ENOMEM;
1782			goto out;
1783		}
1784		/* ephemeral flag for the first socket in the group: drop it */
1785		flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1786	}
1787
 
1788	match = NULL;
1789	list_for_each_entry(f, &fanout_list, list) {
1790		if (f->id == id &&
1791		    read_pnet(&f->net) == sock_net(sk)) {
1792			match = f;
1793			break;
1794		}
1795	}
1796	err = -EINVAL;
1797	if (match) {
1798		if (match->flags != flags)
1799			goto out;
1800		if (args->max_num_members &&
1801		    args->max_num_members != match->max_num_members)
1802			goto out;
1803	} else {
1804		if (args->max_num_members > PACKET_FANOUT_MAX)
1805			goto out;
1806		if (!args->max_num_members)
1807			/* legacy PACKET_FANOUT_MAX */
1808			args->max_num_members = 256;
1809		err = -ENOMEM;
1810		match = kvzalloc(struct_size(match, arr, args->max_num_members),
1811				 GFP_KERNEL);
1812		if (!match)
1813			goto out;
1814		write_pnet(&match->net, sock_net(sk));
1815		match->id = id;
1816		match->type = type;
1817		match->flags = flags;
1818		INIT_LIST_HEAD(&match->list);
1819		spin_lock_init(&match->lock);
1820		refcount_set(&match->sk_ref, 0);
1821		fanout_init_data(match);
1822		match->prot_hook.type = po->prot_hook.type;
1823		match->prot_hook.dev = po->prot_hook.dev;
1824		match->prot_hook.func = packet_rcv_fanout;
1825		match->prot_hook.af_packet_priv = match;
1826		match->prot_hook.af_packet_net = read_pnet(&match->net);
1827		match->prot_hook.id_match = match_fanout_group;
1828		match->max_num_members = args->max_num_members;
1829		match->prot_hook.ignore_outgoing = type_flags & PACKET_FANOUT_FLAG_IGNORE_OUTGOING;
1830		list_add(&match->list, &fanout_list);
1831	}
1832	err = -EINVAL;
1833
1834	spin_lock(&po->bind_lock);
1835	if (po->num &&
1836	    match->type == type &&
1837	    match->prot_hook.type == po->prot_hook.type &&
1838	    match->prot_hook.dev == po->prot_hook.dev) {
1839		err = -ENOSPC;
1840		if (refcount_read(&match->sk_ref) < match->max_num_members) {
1841			/* Paired with packet_setsockopt(PACKET_FANOUT_DATA) */
1842			WRITE_ONCE(po->fanout, match);
1843
1844			po->rollover = rollover;
1845			rollover = NULL;
1846			refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1847			if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
1848				__dev_remove_pack(&po->prot_hook);
1849				__fanout_link(sk, po);
1850			}
1851			err = 0;
1852		}
1853	}
1854	spin_unlock(&po->bind_lock);
1855
1856	if (err && !refcount_read(&match->sk_ref)) {
1857		list_del(&match->list);
1858		kvfree(match);
1859	}
1860
1861out:
1862	kfree(rollover);
1863	mutex_unlock(&fanout_mutex);
 
 
 
 
1864	return err;
1865}
1866
1867/* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1868 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1869 * It is the responsibility of the caller to call fanout_release_data() and
1870 * free the returned packet_fanout (after synchronize_net())
1871 */
1872static struct packet_fanout *fanout_release(struct sock *sk)
1873{
1874	struct packet_sock *po = pkt_sk(sk);
1875	struct packet_fanout *f;
1876
 
 
 
 
1877	mutex_lock(&fanout_mutex);
1878	f = po->fanout;
1879	if (f) {
1880		po->fanout = NULL;
1881
1882		if (refcount_dec_and_test(&f->sk_ref))
1883			list_del(&f->list);
1884		else
1885			f = NULL;
 
1886	}
1887	mutex_unlock(&fanout_mutex);
1888
1889	return f;
 
1890}
1891
1892static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1893					  struct sk_buff *skb)
1894{
1895	/* Earlier code assumed this would be a VLAN pkt, double-check
1896	 * this now that we have the actual packet in hand. We can only
1897	 * do this check on Ethernet devices.
1898	 */
1899	if (unlikely(dev->type != ARPHRD_ETHER))
1900		return false;
1901
1902	skb_reset_mac_header(skb);
1903	return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1904}
1905
1906static const struct proto_ops packet_ops;
1907
1908static const struct proto_ops packet_ops_spkt;
1909
1910static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1911			   struct packet_type *pt, struct net_device *orig_dev)
1912{
1913	struct sock *sk;
1914	struct sockaddr_pkt *spkt;
1915
1916	/*
1917	 *	When we registered the protocol we saved the socket in the data
1918	 *	field for just this event.
1919	 */
1920
1921	sk = pt->af_packet_priv;
1922
1923	/*
1924	 *	Yank back the headers [hope the device set this
1925	 *	right or kerboom...]
1926	 *
1927	 *	Incoming packets have ll header pulled,
1928	 *	push it back.
1929	 *
1930	 *	For outgoing ones skb->data == skb_mac_header(skb)
1931	 *	so that this procedure is noop.
1932	 */
1933
1934	if (skb->pkt_type == PACKET_LOOPBACK)
1935		goto out;
1936
1937	if (!net_eq(dev_net(dev), sock_net(sk)))
1938		goto out;
1939
1940	skb = skb_share_check(skb, GFP_ATOMIC);
1941	if (skb == NULL)
1942		goto oom;
1943
1944	/* drop any routing info */
1945	skb_dst_drop(skb);
1946
1947	/* drop conntrack reference */
1948	nf_reset_ct(skb);
1949
1950	spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1951
1952	skb_push(skb, skb->data - skb_mac_header(skb));
1953
1954	/*
1955	 *	The SOCK_PACKET socket receives _all_ frames.
1956	 */
1957
1958	spkt->spkt_family = dev->type;
1959	strscpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1960	spkt->spkt_protocol = skb->protocol;
1961
1962	/*
1963	 *	Charge the memory to the socket. This is done specifically
1964	 *	to prevent sockets using all the memory up.
1965	 */
1966
1967	if (sock_queue_rcv_skb(sk, skb) == 0)
1968		return 0;
1969
1970out:
1971	kfree_skb(skb);
1972oom:
1973	return 0;
1974}
1975
1976static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1977{
1978	int depth;
1979
1980	if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1981	    sock->type == SOCK_RAW) {
1982		skb_reset_mac_header(skb);
1983		skb->protocol = dev_parse_header_protocol(skb);
1984	}
1985
1986	/* Move network header to the right position for VLAN tagged packets */
1987	if (likely(skb->dev->type == ARPHRD_ETHER) &&
1988	    eth_type_vlan(skb->protocol) &&
1989	    vlan_get_protocol_and_depth(skb, skb->protocol, &depth) != 0)
1990		skb_set_network_header(skb, depth);
1991
1992	skb_probe_transport_header(skb);
1993}
1994
1995/*
1996 *	Output a raw packet to a device layer. This bypasses all the other
1997 *	protocol layers and you must therefore supply it with a complete frame
1998 */
1999
2000static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
2001			       size_t len)
2002{
2003	struct sock *sk = sock->sk;
2004	DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
2005	struct sk_buff *skb = NULL;
2006	struct net_device *dev;
2007	struct sockcm_cookie sockc;
2008	__be16 proto = 0;
2009	int err;
2010	int extra_len = 0;
2011
2012	/*
2013	 *	Get and verify the address.
2014	 */
2015
2016	if (saddr) {
2017		if (msg->msg_namelen < sizeof(struct sockaddr))
2018			return -EINVAL;
2019		if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
2020			proto = saddr->spkt_protocol;
2021	} else
2022		return -ENOTCONN;	/* SOCK_PACKET must be sent giving an address */
2023
2024	/*
2025	 *	Find the device first to size check it
2026	 */
2027
2028	saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
2029retry:
2030	rcu_read_lock();
2031	dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
2032	err = -ENODEV;
2033	if (dev == NULL)
2034		goto out_unlock;
2035
2036	err = -ENETDOWN;
2037	if (!(dev->flags & IFF_UP))
2038		goto out_unlock;
2039
2040	/*
2041	 * You may not queue a frame bigger than the mtu. This is the lowest level
2042	 * raw protocol and you must do your own fragmentation at this level.
2043	 */
2044
2045	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2046		if (!netif_supports_nofcs(dev)) {
2047			err = -EPROTONOSUPPORT;
2048			goto out_unlock;
2049		}
2050		extra_len = 4; /* We're doing our own CRC */
2051	}
2052
2053	err = -EMSGSIZE;
2054	if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
2055		goto out_unlock;
2056
2057	if (!skb) {
2058		size_t reserved = LL_RESERVED_SPACE(dev);
2059		int tlen = dev->needed_tailroom;
2060		unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
2061
2062		rcu_read_unlock();
2063		skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
2064		if (skb == NULL)
2065			return -ENOBUFS;
2066		/* FIXME: Save some space for broken drivers that write a hard
2067		 * header at transmission time by themselves. PPP is the notable
2068		 * one here. This should really be fixed at the driver level.
2069		 */
2070		skb_reserve(skb, reserved);
2071		skb_reset_network_header(skb);
2072
2073		/* Try to align data part correctly */
2074		if (hhlen) {
2075			skb->data -= hhlen;
2076			skb->tail -= hhlen;
2077			if (len < hhlen)
2078				skb_reset_network_header(skb);
2079		}
2080		err = memcpy_from_msg(skb_put(skb, len), msg, len);
2081		if (err)
2082			goto out_free;
2083		goto retry;
2084	}
2085
2086	if (!dev_validate_header(dev, skb->data, len) || !skb->len) {
2087		err = -EINVAL;
2088		goto out_unlock;
2089	}
2090	if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
2091	    !packet_extra_vlan_len_allowed(dev, skb)) {
2092		err = -EMSGSIZE;
2093		goto out_unlock;
2094	}
2095
2096	sockcm_init(&sockc, sk);
2097	if (msg->msg_controllen) {
2098		err = sock_cmsg_send(sk, msg, &sockc);
2099		if (unlikely(err))
2100			goto out_unlock;
2101	}
2102
2103	skb->protocol = proto;
2104	skb->dev = dev;
2105	skb->priority = READ_ONCE(sk->sk_priority);
2106	skb->mark = READ_ONCE(sk->sk_mark);
2107	skb_set_delivery_type_by_clockid(skb, sockc.transmit_time, sk->sk_clockid);
2108	skb_setup_tx_timestamp(skb, &sockc);
2109
2110	if (unlikely(extra_len == 4))
2111		skb->no_fcs = 1;
2112
2113	packet_parse_headers(skb, sock);
2114
2115	dev_queue_xmit(skb);
2116	rcu_read_unlock();
2117	return len;
2118
2119out_unlock:
2120	rcu_read_unlock();
2121out_free:
2122	kfree_skb(skb);
2123	return err;
2124}
2125
2126static unsigned int run_filter(struct sk_buff *skb,
2127			       const struct sock *sk,
2128			       unsigned int res)
2129{
2130	struct sk_filter *filter;
2131
2132	rcu_read_lock();
2133	filter = rcu_dereference(sk->sk_filter);
2134	if (filter != NULL)
2135		res = bpf_prog_run_clear_cb(filter->prog, skb);
2136	rcu_read_unlock();
2137
2138	return res;
2139}
2140
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2141static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2142			   size_t *len, int vnet_hdr_sz)
2143{
2144	struct virtio_net_hdr_mrg_rxbuf vnet_hdr = { .num_buffers = 0 };
2145
2146	if (*len < vnet_hdr_sz)
2147		return -EINVAL;
2148	*len -= vnet_hdr_sz;
2149
2150	if (virtio_net_hdr_from_skb(skb, (struct virtio_net_hdr *)&vnet_hdr, vio_le(), true, 0))
2151		return -EINVAL;
2152
2153	return memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_sz);
2154}
2155
2156/*
2157 * This function makes lazy skb cloning in hope that most of packets
2158 * are discarded by BPF.
2159 *
2160 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2161 * and skb->cb are mangled. It works because (and until) packets
2162 * falling here are owned by current CPU. Output packets are cloned
2163 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2164 * sequentially, so that if we return skb to original state on exit,
2165 * we will not harm anyone.
2166 */
2167
2168static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2169		      struct packet_type *pt, struct net_device *orig_dev)
2170{
2171	enum skb_drop_reason drop_reason = SKB_CONSUMED;
2172	struct sock *sk = NULL;
2173	struct sockaddr_ll *sll;
2174	struct packet_sock *po;
2175	u8 *skb_head = skb->data;
2176	int skb_len = skb->len;
2177	unsigned int snaplen, res;
2178
2179	if (skb->pkt_type == PACKET_LOOPBACK)
2180		goto drop;
2181
2182	sk = pt->af_packet_priv;
2183	po = pkt_sk(sk);
2184
2185	if (!net_eq(dev_net(dev), sock_net(sk)))
2186		goto drop;
2187
2188	skb->dev = dev;
2189
2190	if (dev_has_header(dev)) {
2191		/* The device has an explicit notion of ll header,
2192		 * exported to higher levels.
2193		 *
2194		 * Otherwise, the device hides details of its frame
2195		 * structure, so that corresponding packet head is
2196		 * never delivered to user.
2197		 */
2198		if (sk->sk_type != SOCK_DGRAM)
2199			skb_push(skb, skb->data - skb_mac_header(skb));
2200		else if (skb->pkt_type == PACKET_OUTGOING) {
2201			/* Special case: outgoing packets have ll header at head */
2202			skb_pull(skb, skb_network_offset(skb));
2203		}
2204	}
2205
2206	snaplen = skb_frags_readable(skb) ? skb->len : skb_headlen(skb);
2207
2208	res = run_filter(skb, sk, snaplen);
2209	if (!res)
2210		goto drop_n_restore;
2211	if (snaplen > res)
2212		snaplen = res;
2213
2214	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2215		goto drop_n_acct;
2216
2217	if (skb_shared(skb)) {
2218		struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2219		if (nskb == NULL)
2220			goto drop_n_acct;
2221
2222		if (skb_head != skb->data) {
2223			skb->data = skb_head;
2224			skb->len = skb_len;
2225		}
2226		consume_skb(skb);
2227		skb = nskb;
2228	}
2229
2230	sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2231
2232	sll = &PACKET_SKB_CB(skb)->sa.ll;
2233	sll->sll_hatype = dev->type;
2234	sll->sll_pkttype = skb->pkt_type;
2235	if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV)))
2236		sll->sll_ifindex = orig_dev->ifindex;
2237	else
2238		sll->sll_ifindex = dev->ifindex;
2239
2240	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2241
2242	/* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2243	 * Use their space for storing the original skb length.
2244	 */
2245	PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2246
2247	if (pskb_trim(skb, snaplen))
2248		goto drop_n_acct;
2249
2250	skb_set_owner_r(skb, sk);
2251	skb->dev = NULL;
2252	skb_dst_drop(skb);
2253
2254	/* drop conntrack reference */
2255	nf_reset_ct(skb);
2256
2257	spin_lock(&sk->sk_receive_queue.lock);
2258	po->stats.stats1.tp_packets++;
2259	sock_skb_set_dropcount(sk, skb);
2260	skb_clear_delivery_time(skb);
2261	__skb_queue_tail(&sk->sk_receive_queue, skb);
2262	spin_unlock(&sk->sk_receive_queue.lock);
2263	sk->sk_data_ready(sk);
2264	return 0;
2265
2266drop_n_acct:
2267	atomic_inc(&po->tp_drops);
 
2268	atomic_inc(&sk->sk_drops);
2269	drop_reason = SKB_DROP_REASON_PACKET_SOCK_ERROR;
2270
2271drop_n_restore:
2272	if (skb_head != skb->data && skb_shared(skb)) {
2273		skb->data = skb_head;
2274		skb->len = skb_len;
2275	}
2276drop:
2277	sk_skb_reason_drop(sk, skb, drop_reason);
2278	return 0;
2279}
2280
2281static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2282		       struct packet_type *pt, struct net_device *orig_dev)
2283{
2284	enum skb_drop_reason drop_reason = SKB_CONSUMED;
2285	struct sock *sk = NULL;
2286	struct packet_sock *po;
2287	struct sockaddr_ll *sll;
2288	union tpacket_uhdr h;
2289	u8 *skb_head = skb->data;
2290	int skb_len = skb->len;
2291	unsigned int snaplen, res;
2292	unsigned long status = TP_STATUS_USER;
2293	unsigned short macoff, hdrlen;
2294	unsigned int netoff;
2295	struct sk_buff *copy_skb = NULL;
2296	struct timespec64 ts;
2297	__u32 ts_status;
2298	unsigned int slot_id = 0;
2299	int vnet_hdr_sz = 0;
2300
2301	/* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2302	 * We may add members to them until current aligned size without forcing
2303	 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2304	 */
2305	BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2306	BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2307
2308	if (skb->pkt_type == PACKET_LOOPBACK)
2309		goto drop;
2310
2311	sk = pt->af_packet_priv;
2312	po = pkt_sk(sk);
2313
2314	if (!net_eq(dev_net(dev), sock_net(sk)))
2315		goto drop;
2316
2317	if (dev_has_header(dev)) {
2318		if (sk->sk_type != SOCK_DGRAM)
2319			skb_push(skb, skb->data - skb_mac_header(skb));
2320		else if (skb->pkt_type == PACKET_OUTGOING) {
2321			/* Special case: outgoing packets have ll header at head */
2322			skb_pull(skb, skb_network_offset(skb));
2323		}
2324	}
2325
2326	snaplen = skb_frags_readable(skb) ? skb->len : skb_headlen(skb);
2327
2328	res = run_filter(skb, sk, snaplen);
2329	if (!res)
2330		goto drop_n_restore;
2331
2332	/* If we are flooded, just give up */
2333	if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2334		atomic_inc(&po->tp_drops);
2335		goto drop_n_restore;
2336	}
2337
2338	if (skb->ip_summed == CHECKSUM_PARTIAL)
2339		status |= TP_STATUS_CSUMNOTREADY;
2340	else if (skb->pkt_type != PACKET_OUTGOING &&
2341		 skb_csum_unnecessary(skb))
 
2342		status |= TP_STATUS_CSUM_VALID;
2343	if (skb_is_gso(skb) && skb_is_gso_tcp(skb))
2344		status |= TP_STATUS_GSO_TCP;
2345
2346	if (snaplen > res)
2347		snaplen = res;
2348
2349	if (sk->sk_type == SOCK_DGRAM) {
2350		macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2351				  po->tp_reserve;
2352	} else {
2353		unsigned int maclen = skb_network_offset(skb);
2354		netoff = TPACKET_ALIGN(po->tp_hdrlen +
2355				       (maclen < 16 ? 16 : maclen)) +
2356				       po->tp_reserve;
2357		vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
2358		if (vnet_hdr_sz)
2359			netoff += vnet_hdr_sz;
2360		macoff = netoff - maclen;
2361	}
2362	if (netoff > USHRT_MAX) {
2363		atomic_inc(&po->tp_drops);
2364		goto drop_n_restore;
2365	}
2366	if (po->tp_version <= TPACKET_V2) {
2367		if (macoff + snaplen > po->rx_ring.frame_size) {
2368			if (READ_ONCE(po->copy_thresh) &&
2369			    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2370				if (skb_shared(skb)) {
2371					copy_skb = skb_clone(skb, GFP_ATOMIC);
2372				} else {
2373					copy_skb = skb_get(skb);
2374					skb_head = skb->data;
2375				}
2376				if (copy_skb) {
2377					memset(&PACKET_SKB_CB(copy_skb)->sa.ll, 0,
2378					       sizeof(PACKET_SKB_CB(copy_skb)->sa.ll));
2379					skb_set_owner_r(copy_skb, sk);
2380				}
2381			}
2382			snaplen = po->rx_ring.frame_size - macoff;
2383			if ((int)snaplen < 0) {
2384				snaplen = 0;
2385				vnet_hdr_sz = 0;
2386			}
2387		}
2388	} else if (unlikely(macoff + snaplen >
2389			    GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2390		u32 nval;
2391
2392		nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2393		pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2394			    snaplen, nval, macoff);
2395		snaplen = nval;
2396		if (unlikely((int)snaplen < 0)) {
2397			snaplen = 0;
2398			macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2399			vnet_hdr_sz = 0;
2400		}
2401	}
2402	spin_lock(&sk->sk_receive_queue.lock);
2403	h.raw = packet_current_rx_frame(po, skb,
2404					TP_STATUS_KERNEL, (macoff+snaplen));
2405	if (!h.raw)
2406		goto drop_n_account;
2407
2408	if (po->tp_version <= TPACKET_V2) {
2409		slot_id = po->rx_ring.head;
2410		if (test_bit(slot_id, po->rx_ring.rx_owner_map))
2411			goto drop_n_account;
2412		__set_bit(slot_id, po->rx_ring.rx_owner_map);
2413	}
2414
2415	if (vnet_hdr_sz &&
2416	    virtio_net_hdr_from_skb(skb, h.raw + macoff -
2417				    sizeof(struct virtio_net_hdr),
2418				    vio_le(), true, 0)) {
2419		if (po->tp_version == TPACKET_V3)
2420			prb_clear_blk_fill_status(&po->rx_ring);
2421		goto drop_n_account;
2422	}
2423
2424	if (po->tp_version <= TPACKET_V2) {
2425		packet_increment_rx_head(po, &po->rx_ring);
2426	/*
2427	 * LOSING will be reported till you read the stats,
2428	 * because it's COR - Clear On Read.
2429	 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2430	 * at packet level.
2431	 */
2432		if (atomic_read(&po->tp_drops))
2433			status |= TP_STATUS_LOSING;
2434	}
2435
2436	po->stats.stats1.tp_packets++;
2437	if (copy_skb) {
2438		status |= TP_STATUS_COPY;
2439		skb_clear_delivery_time(copy_skb);
2440		__skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2441	}
2442	spin_unlock(&sk->sk_receive_queue.lock);
2443
 
 
 
 
 
 
 
 
2444	skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2445
2446	/* Always timestamp; prefer an existing software timestamp taken
2447	 * closer to the time of capture.
2448	 */
2449	ts_status = tpacket_get_timestamp(skb, &ts,
2450					  READ_ONCE(po->tp_tstamp) |
2451					  SOF_TIMESTAMPING_SOFTWARE);
2452	if (!ts_status)
2453		ktime_get_real_ts64(&ts);
2454
2455	status |= ts_status;
2456
2457	switch (po->tp_version) {
2458	case TPACKET_V1:
2459		h.h1->tp_len = skb->len;
2460		h.h1->tp_snaplen = snaplen;
2461		h.h1->tp_mac = macoff;
2462		h.h1->tp_net = netoff;
2463		h.h1->tp_sec = ts.tv_sec;
2464		h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2465		hdrlen = sizeof(*h.h1);
2466		break;
2467	case TPACKET_V2:
2468		h.h2->tp_len = skb->len;
2469		h.h2->tp_snaplen = snaplen;
2470		h.h2->tp_mac = macoff;
2471		h.h2->tp_net = netoff;
2472		h.h2->tp_sec = ts.tv_sec;
2473		h.h2->tp_nsec = ts.tv_nsec;
2474		if (skb_vlan_tag_present(skb)) {
2475			h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2476			h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2477			status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2478		} else if (unlikely(sk->sk_type == SOCK_DGRAM && eth_type_vlan(skb->protocol))) {
2479			h.h2->tp_vlan_tci = vlan_get_tci(skb, skb->dev);
2480			h.h2->tp_vlan_tpid = ntohs(skb->protocol);
2481			status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2482		} else {
2483			h.h2->tp_vlan_tci = 0;
2484			h.h2->tp_vlan_tpid = 0;
2485		}
2486		memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2487		hdrlen = sizeof(*h.h2);
2488		break;
2489	case TPACKET_V3:
2490		/* tp_nxt_offset,vlan are already populated above.
2491		 * So DONT clear those fields here
2492		 */
2493		h.h3->tp_status |= status;
2494		h.h3->tp_len = skb->len;
2495		h.h3->tp_snaplen = snaplen;
2496		h.h3->tp_mac = macoff;
2497		h.h3->tp_net = netoff;
2498		h.h3->tp_sec  = ts.tv_sec;
2499		h.h3->tp_nsec = ts.tv_nsec;
2500		memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2501		hdrlen = sizeof(*h.h3);
2502		break;
2503	default:
2504		BUG();
2505	}
2506
2507	sll = h.raw + TPACKET_ALIGN(hdrlen);
2508	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2509	sll->sll_family = AF_PACKET;
2510	sll->sll_hatype = dev->type;
2511	sll->sll_protocol = (sk->sk_type == SOCK_DGRAM) ?
2512		vlan_get_protocol_dgram(skb) : skb->protocol;
2513	sll->sll_pkttype = skb->pkt_type;
2514	if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV)))
2515		sll->sll_ifindex = orig_dev->ifindex;
2516	else
2517		sll->sll_ifindex = dev->ifindex;
2518
2519	smp_mb();
2520
2521#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2522	if (po->tp_version <= TPACKET_V2) {
2523		u8 *start, *end;
2524
2525		end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2526					macoff + snaplen);
2527
2528		for (start = h.raw; start < end; start += PAGE_SIZE)
2529			flush_dcache_page(pgv_to_page(start));
2530	}
2531	smp_wmb();
2532#endif
2533
2534	if (po->tp_version <= TPACKET_V2) {
2535		spin_lock(&sk->sk_receive_queue.lock);
2536		__packet_set_status(po, h.raw, status);
2537		__clear_bit(slot_id, po->rx_ring.rx_owner_map);
2538		spin_unlock(&sk->sk_receive_queue.lock);
2539		sk->sk_data_ready(sk);
2540	} else if (po->tp_version == TPACKET_V3) {
2541		prb_clear_blk_fill_status(&po->rx_ring);
2542	}
2543
2544drop_n_restore:
2545	if (skb_head != skb->data && skb_shared(skb)) {
2546		skb->data = skb_head;
2547		skb->len = skb_len;
2548	}
2549drop:
2550	sk_skb_reason_drop(sk, skb, drop_reason);
2551	return 0;
2552
2553drop_n_account:
 
2554	spin_unlock(&sk->sk_receive_queue.lock);
2555	atomic_inc(&po->tp_drops);
2556	drop_reason = SKB_DROP_REASON_PACKET_SOCK_ERROR;
2557
2558	sk->sk_data_ready(sk);
2559	sk_skb_reason_drop(sk, copy_skb, drop_reason);
2560	goto drop_n_restore;
2561}
2562
2563static void tpacket_destruct_skb(struct sk_buff *skb)
2564{
2565	struct packet_sock *po = pkt_sk(skb->sk);
2566
2567	if (likely(po->tx_ring.pg_vec)) {
2568		void *ph;
2569		__u32 ts;
2570
2571		ph = skb_zcopy_get_nouarg(skb);
2572		packet_dec_pending(&po->tx_ring);
2573
2574		ts = __packet_set_timestamp(po, ph, skb);
2575		__packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2576
2577		complete(&po->skb_completion);
2578	}
2579
2580	sock_wfree(skb);
2581}
2582
 
 
 
 
 
 
 
 
 
2583static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2584{
 
 
2585	if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2586	    (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2587	     __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2588	      __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2589		vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2590			 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2591			__virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2592
2593	if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2594		return -EINVAL;
2595
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2596	return 0;
2597}
2598
2599static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2600				 struct virtio_net_hdr *vnet_hdr, int vnet_hdr_sz)
2601{
2602	int ret;
2603
2604	if (*len < vnet_hdr_sz)
2605		return -EINVAL;
2606	*len -= vnet_hdr_sz;
2607
2608	if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
 
2609		return -EFAULT;
2610
2611	ret = __packet_snd_vnet_parse(vnet_hdr, *len);
2612	if (ret)
2613		return ret;
 
 
 
 
 
 
2614
2615	/* move iter to point to the start of mac header */
2616	if (vnet_hdr_sz != sizeof(struct virtio_net_hdr))
2617		iov_iter_advance(&msg->msg_iter, vnet_hdr_sz - sizeof(struct virtio_net_hdr));
2618
 
 
 
 
 
 
 
2619	return 0;
2620}
2621
2622static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2623		void *frame, struct net_device *dev, void *data, int tp_len,
2624		__be16 proto, unsigned char *addr, int hlen, int copylen,
2625		const struct sockcm_cookie *sockc)
2626{
2627	union tpacket_uhdr ph;
2628	int to_write, offset, len, nr_frags, len_max;
2629	struct socket *sock = po->sk.sk_socket;
2630	struct page *page;
2631	int err;
2632
2633	ph.raw = frame;
2634
2635	skb->protocol = proto;
2636	skb->dev = dev;
2637	skb->priority = READ_ONCE(po->sk.sk_priority);
2638	skb->mark = READ_ONCE(po->sk.sk_mark);
2639	skb_set_delivery_type_by_clockid(skb, sockc->transmit_time, po->sk.sk_clockid);
2640	skb_setup_tx_timestamp(skb, sockc);
2641	skb_zcopy_set_nouarg(skb, ph.raw);
2642
2643	skb_reserve(skb, hlen);
2644	skb_reset_network_header(skb);
2645
2646	to_write = tp_len;
2647
2648	if (sock->type == SOCK_DGRAM) {
2649		err = dev_hard_header(skb, dev, ntohs(proto), addr,
2650				NULL, tp_len);
2651		if (unlikely(err < 0))
2652			return -EINVAL;
2653	} else if (copylen) {
2654		int hdrlen = min_t(int, copylen, tp_len);
2655
2656		skb_push(skb, dev->hard_header_len);
2657		skb_put(skb, copylen - dev->hard_header_len);
2658		err = skb_store_bits(skb, 0, data, hdrlen);
2659		if (unlikely(err))
2660			return err;
2661		if (!dev_validate_header(dev, skb->data, hdrlen))
2662			return -EINVAL;
 
 
2663
2664		data += hdrlen;
2665		to_write -= hdrlen;
2666	}
2667
2668	offset = offset_in_page(data);
2669	len_max = PAGE_SIZE - offset;
2670	len = ((to_write > len_max) ? len_max : to_write);
2671
2672	skb->data_len = to_write;
2673	skb->len += to_write;
2674	skb->truesize += to_write;
2675	refcount_add(to_write, &po->sk.sk_wmem_alloc);
2676
2677	while (likely(to_write)) {
2678		nr_frags = skb_shinfo(skb)->nr_frags;
2679
2680		if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2681			pr_err("Packet exceed the number of skb frags(%u)\n",
2682			       (unsigned int)MAX_SKB_FRAGS);
2683			return -EFAULT;
2684		}
2685
2686		page = pgv_to_page(data);
2687		data += len;
2688		flush_dcache_page(page);
2689		get_page(page);
2690		skb_fill_page_desc(skb, nr_frags, page, offset, len);
2691		to_write -= len;
2692		offset = 0;
2693		len_max = PAGE_SIZE;
2694		len = ((to_write > len_max) ? len_max : to_write);
2695	}
2696
2697	packet_parse_headers(skb, sock);
2698
2699	return tp_len;
2700}
2701
2702static int tpacket_parse_header(struct packet_sock *po, void *frame,
2703				int size_max, void **data)
2704{
2705	union tpacket_uhdr ph;
2706	int tp_len, off;
2707
2708	ph.raw = frame;
2709
2710	switch (po->tp_version) {
2711	case TPACKET_V3:
2712		if (ph.h3->tp_next_offset != 0) {
2713			pr_warn_once("variable sized slot not supported");
2714			return -EINVAL;
2715		}
2716		tp_len = ph.h3->tp_len;
2717		break;
2718	case TPACKET_V2:
2719		tp_len = ph.h2->tp_len;
2720		break;
2721	default:
2722		tp_len = ph.h1->tp_len;
2723		break;
2724	}
2725	if (unlikely(tp_len > size_max)) {
2726		pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2727		return -EMSGSIZE;
2728	}
2729
2730	if (unlikely(packet_sock_flag(po, PACKET_SOCK_TX_HAS_OFF))) {
2731		int off_min, off_max;
2732
2733		off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2734		off_max = po->tx_ring.frame_size - tp_len;
2735		if (po->sk.sk_type == SOCK_DGRAM) {
2736			switch (po->tp_version) {
2737			case TPACKET_V3:
2738				off = ph.h3->tp_net;
2739				break;
2740			case TPACKET_V2:
2741				off = ph.h2->tp_net;
2742				break;
2743			default:
2744				off = ph.h1->tp_net;
2745				break;
2746			}
2747		} else {
2748			switch (po->tp_version) {
2749			case TPACKET_V3:
2750				off = ph.h3->tp_mac;
2751				break;
2752			case TPACKET_V2:
2753				off = ph.h2->tp_mac;
2754				break;
2755			default:
2756				off = ph.h1->tp_mac;
2757				break;
2758			}
2759		}
2760		if (unlikely((off < off_min) || (off_max < off)))
2761			return -EINVAL;
2762	} else {
2763		off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2764	}
2765
2766	*data = frame + off;
2767	return tp_len;
2768}
2769
2770static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2771{
2772	struct sk_buff *skb = NULL;
2773	struct net_device *dev;
2774	struct virtio_net_hdr *vnet_hdr = NULL;
2775	struct sockcm_cookie sockc;
2776	__be16 proto;
2777	int err, reserve = 0;
2778	void *ph;
2779	DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2780	bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2781	int vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
2782	unsigned char *addr = NULL;
2783	int tp_len, size_max;
 
2784	void *data;
2785	int len_sum = 0;
2786	int status = TP_STATUS_AVAILABLE;
2787	int hlen, tlen, copylen = 0;
2788	long timeo = 0;
2789
2790	mutex_lock(&po->pg_vec_lock);
2791
2792	/* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2793	 * we need to confirm it under protection of pg_vec_lock.
2794	 */
2795	if (unlikely(!po->tx_ring.pg_vec)) {
2796		err = -EBUSY;
2797		goto out;
2798	}
2799	if (likely(saddr == NULL)) {
2800		dev	= packet_cached_dev_get(po);
2801		proto	= READ_ONCE(po->num);
 
2802	} else {
2803		err = -EINVAL;
2804		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2805			goto out;
2806		if (msg->msg_namelen < (saddr->sll_halen
2807					+ offsetof(struct sockaddr_ll,
2808						sll_addr)))
2809			goto out;
2810		proto	= saddr->sll_protocol;
 
2811		dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2812		if (po->sk.sk_socket->type == SOCK_DGRAM) {
2813			if (dev && msg->msg_namelen < dev->addr_len +
2814				   offsetof(struct sockaddr_ll, sll_addr))
2815				goto out_put;
2816			addr = saddr->sll_addr;
2817		}
2818	}
2819
2820	err = -ENXIO;
2821	if (unlikely(dev == NULL))
2822		goto out;
2823	err = -ENETDOWN;
2824	if (unlikely(!(dev->flags & IFF_UP)))
2825		goto out_put;
2826
2827	sockcm_init(&sockc, &po->sk);
2828	if (msg->msg_controllen) {
2829		err = sock_cmsg_send(&po->sk, msg, &sockc);
2830		if (unlikely(err))
2831			goto out_put;
2832	}
2833
2834	if (po->sk.sk_socket->type == SOCK_RAW)
2835		reserve = dev->hard_header_len;
2836	size_max = po->tx_ring.frame_size
2837		- (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2838
2839	if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !vnet_hdr_sz)
2840		size_max = dev->mtu + reserve + VLAN_HLEN;
2841
2842	reinit_completion(&po->skb_completion);
2843
2844	do {
2845		ph = packet_current_frame(po, &po->tx_ring,
2846					  TP_STATUS_SEND_REQUEST);
2847		if (unlikely(ph == NULL)) {
2848			if (need_wait && skb) {
2849				timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2850				timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2851				if (timeo <= 0) {
2852					err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2853					goto out_put;
2854				}
2855			}
2856			/* check for additional frames */
2857			continue;
2858		}
2859
2860		skb = NULL;
2861		tp_len = tpacket_parse_header(po, ph, size_max, &data);
2862		if (tp_len < 0)
2863			goto tpacket_error;
2864
2865		status = TP_STATUS_SEND_REQUEST;
2866		hlen = LL_RESERVED_SPACE(dev);
2867		tlen = dev->needed_tailroom;
2868		if (vnet_hdr_sz) {
2869			vnet_hdr = data;
2870			data += vnet_hdr_sz;
2871			tp_len -= vnet_hdr_sz;
2872			if (tp_len < 0 ||
2873			    __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2874				tp_len = -EINVAL;
2875				goto tpacket_error;
2876			}
2877			copylen = __virtio16_to_cpu(vio_le(),
2878						    vnet_hdr->hdr_len);
2879		}
2880		copylen = max_t(int, copylen, dev->hard_header_len);
2881		skb = sock_alloc_send_skb(&po->sk,
2882				hlen + tlen + sizeof(struct sockaddr_ll) +
2883				(copylen - dev->hard_header_len),
2884				!need_wait, &err);
2885
2886		if (unlikely(skb == NULL)) {
2887			/* we assume the socket was initially writeable ... */
2888			if (likely(len_sum > 0))
2889				err = len_sum;
2890			goto out_status;
2891		}
2892		tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2893					  addr, hlen, copylen, &sockc);
2894		if (likely(tp_len >= 0) &&
2895		    tp_len > dev->mtu + reserve &&
2896		    !vnet_hdr_sz &&
2897		    !packet_extra_vlan_len_allowed(dev, skb))
2898			tp_len = -EMSGSIZE;
2899
2900		if (unlikely(tp_len < 0)) {
2901tpacket_error:
2902			if (packet_sock_flag(po, PACKET_SOCK_TP_LOSS)) {
2903				__packet_set_status(po, ph,
2904						TP_STATUS_AVAILABLE);
2905				packet_increment_head(&po->tx_ring);
2906				kfree_skb(skb);
2907				continue;
2908			} else {
2909				status = TP_STATUS_WRONG_FORMAT;
2910				err = tp_len;
2911				goto out_status;
2912			}
2913		}
2914
2915		if (vnet_hdr_sz) {
2916			if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2917				tp_len = -EINVAL;
2918				goto tpacket_error;
2919			}
2920			virtio_net_hdr_set_proto(skb, vnet_hdr);
2921		}
2922
 
 
2923		skb->destructor = tpacket_destruct_skb;
2924		__packet_set_status(po, ph, TP_STATUS_SENDING);
2925		packet_inc_pending(&po->tx_ring);
2926
2927		status = TP_STATUS_SEND_REQUEST;
2928		err = packet_xmit(po, skb);
2929		if (unlikely(err != 0)) {
2930			if (err > 0)
2931				err = net_xmit_errno(err);
2932			if (err && __packet_get_status(po, ph) ==
2933				   TP_STATUS_AVAILABLE) {
2934				/* skb was destructed already */
2935				skb = NULL;
2936				goto out_status;
2937			}
2938			/*
2939			 * skb was dropped but not destructed yet;
2940			 * let's treat it like congestion or err < 0
2941			 */
2942			err = 0;
2943		}
2944		packet_increment_head(&po->tx_ring);
2945		len_sum += tp_len;
2946	} while (likely((ph != NULL) ||
2947		/* Note: packet_read_pending() might be slow if we have
2948		 * to call it as it's per_cpu variable, but in fast-path
2949		 * we already short-circuit the loop with the first
2950		 * condition, and luckily don't have to go that path
2951		 * anyway.
2952		 */
2953		 (need_wait && packet_read_pending(&po->tx_ring))));
2954
2955	err = len_sum;
2956	goto out_put;
2957
2958out_status:
2959	__packet_set_status(po, ph, status);
2960	kfree_skb(skb);
2961out_put:
2962	dev_put(dev);
2963out:
2964	mutex_unlock(&po->pg_vec_lock);
2965	return err;
2966}
2967
2968static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2969				        size_t reserve, size_t len,
2970				        size_t linear, int noblock,
2971				        int *err)
2972{
2973	struct sk_buff *skb;
2974
2975	/* Under a page?  Don't bother with paged skb. */
2976	if (prepad + len < PAGE_SIZE || !linear)
2977		linear = len;
2978
2979	if (len - linear > MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER))
2980		linear = len - MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER);
2981	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2982				   err, PAGE_ALLOC_COSTLY_ORDER);
2983	if (!skb)
2984		return NULL;
2985
2986	skb_reserve(skb, reserve);
2987	skb_put(skb, linear);
2988	skb->data_len = len - linear;
2989	skb->len += len - linear;
2990
2991	return skb;
2992}
2993
2994static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2995{
2996	struct sock *sk = sock->sk;
2997	DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2998	struct sk_buff *skb;
2999	struct net_device *dev;
3000	__be16 proto;
3001	unsigned char *addr = NULL;
3002	int err, reserve = 0;
3003	struct sockcm_cookie sockc;
3004	struct virtio_net_hdr vnet_hdr = { 0 };
3005	int offset = 0;
3006	struct packet_sock *po = pkt_sk(sk);
3007	int vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
3008	int hlen, tlen, linear;
3009	int extra_len = 0;
3010
3011	/*
3012	 *	Get and verify the address.
3013	 */
3014
3015	if (likely(saddr == NULL)) {
3016		dev	= packet_cached_dev_get(po);
3017		proto	= READ_ONCE(po->num);
 
3018	} else {
3019		err = -EINVAL;
3020		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
3021			goto out;
3022		if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
3023			goto out;
3024		proto	= saddr->sll_protocol;
 
3025		dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
3026		if (sock->type == SOCK_DGRAM) {
3027			if (dev && msg->msg_namelen < dev->addr_len +
3028				   offsetof(struct sockaddr_ll, sll_addr))
3029				goto out_unlock;
3030			addr = saddr->sll_addr;
3031		}
3032	}
3033
3034	err = -ENXIO;
3035	if (unlikely(dev == NULL))
3036		goto out_unlock;
3037	err = -ENETDOWN;
3038	if (unlikely(!(dev->flags & IFF_UP)))
3039		goto out_unlock;
3040
3041	sockcm_init(&sockc, sk);
3042	sockc.mark = READ_ONCE(sk->sk_mark);
3043	if (msg->msg_controllen) {
3044		err = sock_cmsg_send(sk, msg, &sockc);
3045		if (unlikely(err))
3046			goto out_unlock;
3047	}
3048
3049	if (sock->type == SOCK_RAW)
3050		reserve = dev->hard_header_len;
3051	if (vnet_hdr_sz) {
3052		err = packet_snd_vnet_parse(msg, &len, &vnet_hdr, vnet_hdr_sz);
3053		if (err)
3054			goto out_unlock;
3055	}
3056
3057	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
3058		if (!netif_supports_nofcs(dev)) {
3059			err = -EPROTONOSUPPORT;
3060			goto out_unlock;
3061		}
3062		extra_len = 4; /* We're doing our own CRC */
3063	}
3064
3065	err = -EMSGSIZE;
3066	if (!vnet_hdr.gso_type &&
3067	    (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
3068		goto out_unlock;
3069
3070	err = -ENOBUFS;
3071	hlen = LL_RESERVED_SPACE(dev);
3072	tlen = dev->needed_tailroom;
3073	linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
3074	linear = max(linear, min_t(int, len, dev->hard_header_len));
3075	skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
3076			       msg->msg_flags & MSG_DONTWAIT, &err);
3077	if (skb == NULL)
3078		goto out_unlock;
3079
3080	skb_reset_network_header(skb);
3081
3082	err = -EINVAL;
3083	if (sock->type == SOCK_DGRAM) {
3084		offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
3085		if (unlikely(offset < 0))
3086			goto out_free;
3087	} else if (reserve) {
3088		skb_reserve(skb, -reserve);
3089		if (len < reserve + sizeof(struct ipv6hdr) &&
3090		    dev->min_header_len != dev->hard_header_len)
3091			skb_reset_network_header(skb);
3092	}
3093
3094	/* Returns -EFAULT on error */
3095	err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
3096	if (err)
3097		goto out_free;
3098
3099	if ((sock->type == SOCK_RAW &&
3100	     !dev_validate_header(dev, skb->data, len)) || !skb->len) {
3101		err = -EINVAL;
3102		goto out_free;
3103	}
3104
3105	skb_setup_tx_timestamp(skb, &sockc);
3106
3107	if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
3108	    !packet_extra_vlan_len_allowed(dev, skb)) {
3109		err = -EMSGSIZE;
3110		goto out_free;
3111	}
3112
3113	skb->protocol = proto;
3114	skb->dev = dev;
3115	skb->priority = READ_ONCE(sk->sk_priority);
3116	skb->mark = sockc.mark;
3117	skb_set_delivery_type_by_clockid(skb, sockc.transmit_time, sk->sk_clockid);
3118
3119	if (unlikely(extra_len == 4))
3120		skb->no_fcs = 1;
3121
3122	packet_parse_headers(skb, sock);
3123
3124	if (vnet_hdr_sz) {
3125		err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
3126		if (err)
3127			goto out_free;
3128		len += vnet_hdr_sz;
3129		virtio_net_hdr_set_proto(skb, &vnet_hdr);
3130	}
3131
3132	err = packet_xmit(po, skb);
 
 
 
3133
3134	if (unlikely(err != 0)) {
3135		if (err > 0)
3136			err = net_xmit_errno(err);
3137		if (err)
3138			goto out_unlock;
3139	}
3140
3141	dev_put(dev);
3142
3143	return len;
3144
3145out_free:
3146	kfree_skb(skb);
3147out_unlock:
3148	dev_put(dev);
 
3149out:
3150	return err;
3151}
3152
3153static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
3154{
3155	struct sock *sk = sock->sk;
3156	struct packet_sock *po = pkt_sk(sk);
3157
3158	/* Reading tx_ring.pg_vec without holding pg_vec_lock is racy.
3159	 * tpacket_snd() will redo the check safely.
3160	 */
3161	if (data_race(po->tx_ring.pg_vec))
3162		return tpacket_snd(po, msg);
3163
3164	return packet_snd(sock, msg, len);
3165}
3166
3167/*
3168 *	Close a PACKET socket. This is fairly simple. We immediately go
3169 *	to 'closed' state and remove our protocol entry in the device list.
3170 */
3171
3172static int packet_release(struct socket *sock)
3173{
3174	struct sock *sk = sock->sk;
3175	struct packet_sock *po;
3176	struct packet_fanout *f;
3177	struct net *net;
3178	union tpacket_req_u req_u;
3179
3180	if (!sk)
3181		return 0;
3182
3183	net = sock_net(sk);
3184	po = pkt_sk(sk);
3185
3186	mutex_lock(&net->packet.sklist_lock);
3187	sk_del_node_init_rcu(sk);
3188	mutex_unlock(&net->packet.sklist_lock);
3189
 
3190	sock_prot_inuse_add(net, sk->sk_prot, -1);
 
3191
3192	spin_lock(&po->bind_lock);
3193	unregister_prot_hook(sk, false);
3194	packet_cached_dev_reset(po);
3195
3196	if (po->prot_hook.dev) {
3197		netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3198		po->prot_hook.dev = NULL;
3199	}
3200	spin_unlock(&po->bind_lock);
3201
3202	packet_flush_mclist(sk);
3203
3204	lock_sock(sk);
3205	if (po->rx_ring.pg_vec) {
3206		memset(&req_u, 0, sizeof(req_u));
3207		packet_set_ring(sk, &req_u, 1, 0);
3208	}
3209
3210	if (po->tx_ring.pg_vec) {
3211		memset(&req_u, 0, sizeof(req_u));
3212		packet_set_ring(sk, &req_u, 1, 1);
3213	}
3214	release_sock(sk);
3215
3216	f = fanout_release(sk);
3217
3218	synchronize_net();
3219
3220	kfree(po->rollover);
3221	if (f) {
3222		fanout_release_data(f);
3223		kvfree(f);
3224	}
3225	/*
3226	 *	Now the socket is dead. No more input will appear.
3227	 */
3228	sock_orphan(sk);
3229	sock->sk = NULL;
3230
3231	/* Purge queues */
3232
3233	skb_queue_purge(&sk->sk_receive_queue);
3234	packet_free_pending(po);
 
3235
3236	sock_put(sk);
3237	return 0;
3238}
3239
3240/*
3241 *	Attach a packet hook.
3242 */
3243
3244static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3245			  __be16 proto)
3246{
3247	struct packet_sock *po = pkt_sk(sk);
 
 
 
3248	struct net_device *dev = NULL;
 
3249	bool unlisted = false;
3250	bool need_rehook;
3251	int ret = 0;
 
3252
3253	lock_sock(sk);
3254	spin_lock(&po->bind_lock);
3255	if (!proto)
3256		proto = po->num;
3257
3258	rcu_read_lock();
3259
3260	if (po->fanout) {
3261		ret = -EINVAL;
3262		goto out_unlock;
3263	}
3264
3265	if (name) {
3266		dev = dev_get_by_name_rcu(sock_net(sk), name);
3267		if (!dev) {
3268			ret = -ENODEV;
3269			goto out_unlock;
3270		}
3271	} else if (ifindex) {
3272		dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3273		if (!dev) {
3274			ret = -ENODEV;
3275			goto out_unlock;
3276		}
3277	}
3278
3279	need_rehook = po->prot_hook.type != proto || po->prot_hook.dev != dev;
 
 
 
 
 
 
3280
3281	if (need_rehook) {
3282		dev_hold(dev);
3283		if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
3284			rcu_read_unlock();
3285			/* prevents packet_notifier() from calling
3286			 * register_prot_hook()
3287			 */
3288			WRITE_ONCE(po->num, 0);
3289			__unregister_prot_hook(sk, true);
3290			rcu_read_lock();
 
3291			if (dev)
3292				unlisted = !dev_get_by_index_rcu(sock_net(sk),
3293								 dev->ifindex);
3294		}
3295
3296		BUG_ON(packet_sock_flag(po, PACKET_SOCK_RUNNING));
3297		WRITE_ONCE(po->num, proto);
3298		po->prot_hook.type = proto;
3299
3300		netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3301
3302		if (unlikely(unlisted)) {
 
3303			po->prot_hook.dev = NULL;
3304			WRITE_ONCE(po->ifindex, -1);
3305			packet_cached_dev_reset(po);
3306		} else {
3307			netdev_hold(dev, &po->prot_hook.dev_tracker,
3308				    GFP_ATOMIC);
3309			po->prot_hook.dev = dev;
3310			WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0);
3311			packet_cached_dev_assign(po, dev);
3312		}
3313		dev_put(dev);
3314	}
 
 
3315
3316	if (proto == 0 || !need_rehook)
3317		goto out_unlock;
3318
3319	if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3320		register_prot_hook(sk);
3321	} else {
3322		sk->sk_err = ENETDOWN;
3323		if (!sock_flag(sk, SOCK_DEAD))
3324			sk_error_report(sk);
3325	}
3326
3327out_unlock:
3328	rcu_read_unlock();
3329	spin_unlock(&po->bind_lock);
3330	release_sock(sk);
3331	return ret;
3332}
3333
3334/*
3335 *	Bind a packet socket to a device
3336 */
3337
3338static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3339			    int addr_len)
3340{
3341	struct sock *sk = sock->sk;
3342	char name[sizeof(uaddr->sa_data_min) + 1];
3343
3344	/*
3345	 *	Check legality
3346	 */
3347
3348	if (addr_len != sizeof(struct sockaddr))
3349		return -EINVAL;
3350	/* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3351	 * zero-terminated.
3352	 */
3353	memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data_min));
3354	name[sizeof(uaddr->sa_data_min)] = 0;
3355
3356	return packet_do_bind(sk, name, 0, 0);
3357}
3358
3359static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3360{
3361	struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3362	struct sock *sk = sock->sk;
3363
3364	/*
3365	 *	Check legality
3366	 */
3367
3368	if (addr_len < sizeof(struct sockaddr_ll))
3369		return -EINVAL;
3370	if (sll->sll_family != AF_PACKET)
3371		return -EINVAL;
3372
3373	return packet_do_bind(sk, NULL, sll->sll_ifindex, sll->sll_protocol);
 
3374}
3375
3376static struct proto packet_proto = {
3377	.name	  = "PACKET",
3378	.owner	  = THIS_MODULE,
3379	.obj_size = sizeof(struct packet_sock),
3380};
3381
3382/*
3383 *	Create a packet of type SOCK_PACKET.
3384 */
3385
3386static int packet_create(struct net *net, struct socket *sock, int protocol,
3387			 int kern)
3388{
3389	struct sock *sk;
3390	struct packet_sock *po;
3391	__be16 proto = (__force __be16)protocol; /* weird, but documented */
3392	int err;
3393
3394	if (!ns_capable(net->user_ns, CAP_NET_RAW))
3395		return -EPERM;
3396	if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3397	    sock->type != SOCK_PACKET)
3398		return -ESOCKTNOSUPPORT;
3399
3400	sock->state = SS_UNCONNECTED;
3401
3402	err = -ENOBUFS;
3403	sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3404	if (sk == NULL)
3405		goto out;
3406
3407	sock->ops = &packet_ops;
3408	if (sock->type == SOCK_PACKET)
3409		sock->ops = &packet_ops_spkt;
3410
3411	po = pkt_sk(sk);
3412	err = packet_alloc_pending(po);
3413	if (err)
3414		goto out_sk_free;
3415
3416	sock_init_data(sock, sk);
3417
3418	init_completion(&po->skb_completion);
3419	sk->sk_family = PF_PACKET;
3420	po->num = proto;
 
 
 
 
 
3421
3422	packet_cached_dev_reset(po);
3423
3424	sk->sk_destruct = packet_sock_destruct;
 
3425
3426	/*
3427	 *	Attach a protocol block
3428	 */
3429
3430	spin_lock_init(&po->bind_lock);
3431	mutex_init(&po->pg_vec_lock);
3432	po->rollover = NULL;
3433	po->prot_hook.func = packet_rcv;
3434
3435	if (sock->type == SOCK_PACKET)
3436		po->prot_hook.func = packet_rcv_spkt;
3437
3438	po->prot_hook.af_packet_priv = sk;
3439	po->prot_hook.af_packet_net = sock_net(sk);
3440
3441	if (proto) {
3442		po->prot_hook.type = proto;
3443		__register_prot_hook(sk);
3444	}
3445
3446	mutex_lock(&net->packet.sklist_lock);
3447	sk_add_node_tail_rcu(sk, &net->packet.sklist);
3448	mutex_unlock(&net->packet.sklist_lock);
3449
 
3450	sock_prot_inuse_add(net, &packet_proto, 1);
 
3451
3452	return 0;
3453out_sk_free:
3454	sk_free(sk);
3455out:
3456	return err;
3457}
3458
3459/*
3460 *	Pull a packet from our receive queue and hand it to the user.
3461 *	If necessary we block.
3462 */
3463
3464static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3465			  int flags)
3466{
3467	struct sock *sk = sock->sk;
3468	struct sk_buff *skb;
3469	int copied, err;
3470	int vnet_hdr_len = READ_ONCE(pkt_sk(sk)->vnet_hdr_sz);
3471	unsigned int origlen = 0;
3472
3473	err = -EINVAL;
3474	if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3475		goto out;
3476
3477#if 0
3478	/* What error should we return now? EUNATTACH? */
3479	if (pkt_sk(sk)->ifindex < 0)
3480		return -ENODEV;
3481#endif
3482
3483	if (flags & MSG_ERRQUEUE) {
3484		err = sock_recv_errqueue(sk, msg, len,
3485					 SOL_PACKET, PACKET_TX_TIMESTAMP);
3486		goto out;
3487	}
3488
3489	/*
3490	 *	Call the generic datagram receiver. This handles all sorts
3491	 *	of horrible races and re-entrancy so we can forget about it
3492	 *	in the protocol layers.
3493	 *
3494	 *	Now it will return ENETDOWN, if device have just gone down,
3495	 *	but then it will block.
3496	 */
3497
3498	skb = skb_recv_datagram(sk, flags, &err);
3499
3500	/*
3501	 *	An error occurred so return it. Because skb_recv_datagram()
3502	 *	handles the blocking we don't see and worry about blocking
3503	 *	retries.
3504	 */
3505
3506	if (skb == NULL)
3507		goto out;
3508
3509	packet_rcv_try_clear_pressure(pkt_sk(sk));
 
3510
3511	if (vnet_hdr_len) {
3512		err = packet_rcv_vnet(msg, skb, &len, vnet_hdr_len);
3513		if (err)
3514			goto out_free;
 
3515	}
3516
3517	/* You lose any data beyond the buffer you gave. If it worries
3518	 * a user program they can ask the device for its MTU
3519	 * anyway.
3520	 */
3521	copied = skb->len;
3522	if (copied > len) {
3523		copied = len;
3524		msg->msg_flags |= MSG_TRUNC;
3525	}
3526
3527	err = skb_copy_datagram_msg(skb, 0, msg, copied);
3528	if (err)
3529		goto out_free;
3530
3531	if (sock->type != SOCK_PACKET) {
3532		struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3533
3534		/* Original length was stored in sockaddr_ll fields */
3535		origlen = PACKET_SKB_CB(skb)->sa.origlen;
3536		sll->sll_family = AF_PACKET;
3537		sll->sll_protocol = (sock->type == SOCK_DGRAM) ?
3538			vlan_get_protocol_dgram(skb) : skb->protocol;
3539	}
3540
3541	sock_recv_cmsgs(msg, sk, skb);
3542
3543	if (msg->msg_name) {
3544		const size_t max_len = min(sizeof(skb->cb),
3545					   sizeof(struct sockaddr_storage));
3546		int copy_len;
3547
3548		/* If the address length field is there to be filled
3549		 * in, we fill it in now.
3550		 */
3551		if (sock->type == SOCK_PACKET) {
3552			__sockaddr_check_size(sizeof(struct sockaddr_pkt));
3553			msg->msg_namelen = sizeof(struct sockaddr_pkt);
3554			copy_len = msg->msg_namelen;
3555		} else {
3556			struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3557
3558			msg->msg_namelen = sll->sll_halen +
3559				offsetof(struct sockaddr_ll, sll_addr);
3560			copy_len = msg->msg_namelen;
3561			if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3562				memset(msg->msg_name +
3563				       offsetof(struct sockaddr_ll, sll_addr),
3564				       0, sizeof(sll->sll_addr));
3565				msg->msg_namelen = sizeof(struct sockaddr_ll);
3566			}
3567		}
3568		if (WARN_ON_ONCE(copy_len > max_len)) {
3569			copy_len = max_len;
3570			msg->msg_namelen = copy_len;
3571		}
3572		memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
 
3573	}
3574
3575	if (packet_sock_flag(pkt_sk(sk), PACKET_SOCK_AUXDATA)) {
3576		struct tpacket_auxdata aux;
3577
3578		aux.tp_status = TP_STATUS_USER;
3579		if (skb->ip_summed == CHECKSUM_PARTIAL)
3580			aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3581		else if (skb->pkt_type != PACKET_OUTGOING &&
3582			 skb_csum_unnecessary(skb))
 
3583			aux.tp_status |= TP_STATUS_CSUM_VALID;
3584		if (skb_is_gso(skb) && skb_is_gso_tcp(skb))
3585			aux.tp_status |= TP_STATUS_GSO_TCP;
3586
3587		aux.tp_len = origlen;
3588		aux.tp_snaplen = skb->len;
3589		aux.tp_mac = 0;
3590		aux.tp_net = skb_network_offset(skb);
3591		if (skb_vlan_tag_present(skb)) {
3592			aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3593			aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3594			aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3595		} else if (unlikely(sock->type == SOCK_DGRAM && eth_type_vlan(skb->protocol))) {
3596			struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3597			struct net_device *dev;
3598
3599			rcu_read_lock();
3600			dev = dev_get_by_index_rcu(sock_net(sk), sll->sll_ifindex);
3601			if (dev) {
3602				aux.tp_vlan_tci = vlan_get_tci(skb, dev);
3603				aux.tp_vlan_tpid = ntohs(skb->protocol);
3604				aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3605			} else {
3606				aux.tp_vlan_tci = 0;
3607				aux.tp_vlan_tpid = 0;
3608			}
3609			rcu_read_unlock();
3610		} else {
3611			aux.tp_vlan_tci = 0;
3612			aux.tp_vlan_tpid = 0;
3613		}
3614		put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3615	}
3616
3617	/*
3618	 *	Free or return the buffer as appropriate. Again this
3619	 *	hides all the races and re-entrancy issues from us.
3620	 */
3621	err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3622
3623out_free:
3624	skb_free_datagram(sk, skb);
3625out:
3626	return err;
3627}
3628
3629static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3630			       int peer)
3631{
3632	struct net_device *dev;
3633	struct sock *sk	= sock->sk;
3634
3635	if (peer)
3636		return -EOPNOTSUPP;
3637
3638	uaddr->sa_family = AF_PACKET;
3639	memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data_min));
3640	rcu_read_lock();
3641	dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex));
3642	if (dev)
3643		strscpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data_min));
3644	rcu_read_unlock();
 
3645
3646	return sizeof(*uaddr);
3647}
3648
3649static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3650			  int peer)
3651{
3652	struct net_device *dev;
3653	struct sock *sk = sock->sk;
3654	struct packet_sock *po = pkt_sk(sk);
3655	DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3656	int ifindex;
3657
3658	if (peer)
3659		return -EOPNOTSUPP;
3660
3661	ifindex = READ_ONCE(po->ifindex);
3662	sll->sll_family = AF_PACKET;
3663	sll->sll_ifindex = ifindex;
3664	sll->sll_protocol = READ_ONCE(po->num);
3665	sll->sll_pkttype = 0;
3666	rcu_read_lock();
3667	dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3668	if (dev) {
3669		sll->sll_hatype = dev->type;
3670		sll->sll_halen = dev->addr_len;
3671
3672		/* Let __fortify_memcpy_chk() know the actual buffer size. */
3673		memcpy(((struct sockaddr_storage *)sll)->__data +
3674		       offsetof(struct sockaddr_ll, sll_addr) -
3675		       offsetofend(struct sockaddr_ll, sll_family),
3676		       dev->dev_addr, dev->addr_len);
3677	} else {
3678		sll->sll_hatype = 0;	/* Bad: we have no ARPHRD_UNSPEC */
3679		sll->sll_halen = 0;
3680	}
3681	rcu_read_unlock();
 
3682
3683	return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3684}
3685
3686static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3687			 int what)
3688{
3689	switch (i->type) {
3690	case PACKET_MR_MULTICAST:
3691		if (i->alen != dev->addr_len)
3692			return -EINVAL;
3693		if (what > 0)
3694			return dev_mc_add(dev, i->addr);
3695		else
3696			return dev_mc_del(dev, i->addr);
3697		break;
3698	case PACKET_MR_PROMISC:
3699		return dev_set_promiscuity(dev, what);
3700	case PACKET_MR_ALLMULTI:
3701		return dev_set_allmulti(dev, what);
3702	case PACKET_MR_UNICAST:
3703		if (i->alen != dev->addr_len)
3704			return -EINVAL;
3705		if (what > 0)
3706			return dev_uc_add(dev, i->addr);
3707		else
3708			return dev_uc_del(dev, i->addr);
3709		break;
3710	default:
3711		break;
3712	}
3713	return 0;
3714}
3715
3716static void packet_dev_mclist_delete(struct net_device *dev,
3717				     struct packet_mclist **mlp)
3718{
3719	struct packet_mclist *ml;
3720
3721	while ((ml = *mlp) != NULL) {
3722		if (ml->ifindex == dev->ifindex) {
3723			packet_dev_mc(dev, ml, -1);
3724			*mlp = ml->next;
3725			kfree(ml);
3726		} else
3727			mlp = &ml->next;
3728	}
3729}
3730
3731static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3732{
3733	struct packet_sock *po = pkt_sk(sk);
3734	struct packet_mclist *ml, *i;
3735	struct net_device *dev;
3736	int err;
3737
3738	rtnl_lock();
3739
3740	err = -ENODEV;
3741	dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3742	if (!dev)
3743		goto done;
3744
3745	err = -EINVAL;
3746	if (mreq->mr_alen > dev->addr_len)
3747		goto done;
3748
3749	err = -ENOBUFS;
3750	i = kmalloc(sizeof(*i), GFP_KERNEL);
3751	if (i == NULL)
3752		goto done;
3753
3754	err = 0;
3755	for (ml = po->mclist; ml; ml = ml->next) {
3756		if (ml->ifindex == mreq->mr_ifindex &&
3757		    ml->type == mreq->mr_type &&
3758		    ml->alen == mreq->mr_alen &&
3759		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3760			ml->count++;
3761			/* Free the new element ... */
3762			kfree(i);
3763			goto done;
3764		}
3765	}
3766
3767	i->type = mreq->mr_type;
3768	i->ifindex = mreq->mr_ifindex;
3769	i->alen = mreq->mr_alen;
3770	memcpy(i->addr, mreq->mr_address, i->alen);
3771	memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3772	i->count = 1;
3773	i->next = po->mclist;
3774	po->mclist = i;
3775	err = packet_dev_mc(dev, i, 1);
3776	if (err) {
3777		po->mclist = i->next;
3778		kfree(i);
3779	}
3780
3781done:
3782	rtnl_unlock();
3783	return err;
3784}
3785
3786static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3787{
3788	struct packet_mclist *ml, **mlp;
3789
3790	rtnl_lock();
3791
3792	for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3793		if (ml->ifindex == mreq->mr_ifindex &&
3794		    ml->type == mreq->mr_type &&
3795		    ml->alen == mreq->mr_alen &&
3796		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3797			if (--ml->count == 0) {
3798				struct net_device *dev;
3799				*mlp = ml->next;
3800				dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3801				if (dev)
3802					packet_dev_mc(dev, ml, -1);
3803				kfree(ml);
3804			}
3805			break;
3806		}
3807	}
3808	rtnl_unlock();
3809	return 0;
3810}
3811
3812static void packet_flush_mclist(struct sock *sk)
3813{
3814	struct packet_sock *po = pkt_sk(sk);
3815	struct packet_mclist *ml;
3816
3817	if (!po->mclist)
3818		return;
3819
3820	rtnl_lock();
3821	while ((ml = po->mclist) != NULL) {
3822		struct net_device *dev;
3823
3824		po->mclist = ml->next;
3825		dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3826		if (dev != NULL)
3827			packet_dev_mc(dev, ml, -1);
3828		kfree(ml);
3829	}
3830	rtnl_unlock();
3831}
3832
3833static int
3834packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval,
3835		  unsigned int optlen)
3836{
3837	struct sock *sk = sock->sk;
3838	struct packet_sock *po = pkt_sk(sk);
3839	int ret;
3840
3841	if (level != SOL_PACKET)
3842		return -ENOPROTOOPT;
3843
3844	switch (optname) {
3845	case PACKET_ADD_MEMBERSHIP:
3846	case PACKET_DROP_MEMBERSHIP:
3847	{
3848		struct packet_mreq_max mreq;
3849		int len = optlen;
3850		memset(&mreq, 0, sizeof(mreq));
3851		if (len < sizeof(struct packet_mreq))
3852			return -EINVAL;
3853		if (len > sizeof(mreq))
3854			len = sizeof(mreq);
3855		if (copy_from_sockptr(&mreq, optval, len))
3856			return -EFAULT;
3857		if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3858			return -EINVAL;
3859		if (optname == PACKET_ADD_MEMBERSHIP)
3860			ret = packet_mc_add(sk, &mreq);
3861		else
3862			ret = packet_mc_drop(sk, &mreq);
3863		return ret;
3864	}
3865
3866	case PACKET_RX_RING:
3867	case PACKET_TX_RING:
3868	{
3869		union tpacket_req_u req_u;
 
3870
3871		ret = -EINVAL;
3872		lock_sock(sk);
3873		switch (po->tp_version) {
3874		case TPACKET_V1:
3875		case TPACKET_V2:
3876			if (optlen < sizeof(req_u.req))
3877				break;
3878			ret = copy_from_sockptr(&req_u.req, optval,
3879						sizeof(req_u.req)) ?
3880						-EINVAL : 0;
3881			break;
3882		case TPACKET_V3:
3883		default:
3884			if (optlen < sizeof(req_u.req3))
3885				break;
3886			ret = copy_from_sockptr(&req_u.req3, optval,
3887						sizeof(req_u.req3)) ?
3888						-EINVAL : 0;
3889			break;
3890		}
3891		if (!ret)
3892			ret = packet_set_ring(sk, &req_u, 0,
3893					      optname == PACKET_TX_RING);
3894		release_sock(sk);
3895		return ret;
 
3896	}
3897	case PACKET_COPY_THRESH:
3898	{
3899		int val;
3900
3901		if (optlen != sizeof(val))
3902			return -EINVAL;
3903		if (copy_from_sockptr(&val, optval, sizeof(val)))
3904			return -EFAULT;
3905
3906		WRITE_ONCE(pkt_sk(sk)->copy_thresh, val);
3907		return 0;
3908	}
3909	case PACKET_VERSION:
3910	{
3911		int val;
3912
3913		if (optlen != sizeof(val))
3914			return -EINVAL;
3915		if (copy_from_sockptr(&val, optval, sizeof(val)))
 
 
3916			return -EFAULT;
3917		switch (val) {
3918		case TPACKET_V1:
3919		case TPACKET_V2:
3920		case TPACKET_V3:
3921			break;
 
3922		default:
3923			return -EINVAL;
3924		}
3925		lock_sock(sk);
3926		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3927			ret = -EBUSY;
3928		} else {
3929			po->tp_version = val;
3930			ret = 0;
3931		}
3932		release_sock(sk);
3933		return ret;
3934	}
3935	case PACKET_RESERVE:
3936	{
3937		unsigned int val;
3938
3939		if (optlen != sizeof(val))
3940			return -EINVAL;
3941		if (copy_from_sockptr(&val, optval, sizeof(val)))
 
 
3942			return -EFAULT;
3943		if (val > INT_MAX)
3944			return -EINVAL;
3945		lock_sock(sk);
3946		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3947			ret = -EBUSY;
3948		} else {
3949			po->tp_reserve = val;
3950			ret = 0;
3951		}
3952		release_sock(sk);
3953		return ret;
3954	}
3955	case PACKET_LOSS:
3956	{
3957		unsigned int val;
3958
3959		if (optlen != sizeof(val))
3960			return -EINVAL;
3961		if (copy_from_sockptr(&val, optval, sizeof(val)))
 
 
3962			return -EFAULT;
3963
3964		lock_sock(sk);
3965		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3966			ret = -EBUSY;
3967		} else {
3968			packet_sock_flag_set(po, PACKET_SOCK_TP_LOSS, val);
3969			ret = 0;
3970		}
3971		release_sock(sk);
3972		return ret;
3973	}
3974	case PACKET_AUXDATA:
3975	{
3976		int val;
3977
3978		if (optlen < sizeof(val))
3979			return -EINVAL;
3980		if (copy_from_sockptr(&val, optval, sizeof(val)))
3981			return -EFAULT;
3982
3983		packet_sock_flag_set(po, PACKET_SOCK_AUXDATA, val);
3984		return 0;
3985	}
3986	case PACKET_ORIGDEV:
3987	{
3988		int val;
3989
3990		if (optlen < sizeof(val))
3991			return -EINVAL;
3992		if (copy_from_sockptr(&val, optval, sizeof(val)))
3993			return -EFAULT;
3994
3995		packet_sock_flag_set(po, PACKET_SOCK_ORIGDEV, val);
3996		return 0;
3997	}
3998	case PACKET_VNET_HDR:
3999	case PACKET_VNET_HDR_SZ:
4000	{
4001		int val, hdr_len;
4002
4003		if (sock->type != SOCK_RAW)
4004			return -EINVAL;
 
 
4005		if (optlen < sizeof(val))
4006			return -EINVAL;
4007		if (copy_from_sockptr(&val, optval, sizeof(val)))
4008			return -EFAULT;
4009
4010		if (optname == PACKET_VNET_HDR_SZ) {
4011			if (val && val != sizeof(struct virtio_net_hdr) &&
4012			    val != sizeof(struct virtio_net_hdr_mrg_rxbuf))
4013				return -EINVAL;
4014			hdr_len = val;
4015		} else {
4016			hdr_len = val ? sizeof(struct virtio_net_hdr) : 0;
4017		}
4018		lock_sock(sk);
4019		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
4020			ret = -EBUSY;
4021		} else {
4022			WRITE_ONCE(po->vnet_hdr_sz, hdr_len);
4023			ret = 0;
4024		}
4025		release_sock(sk);
4026		return ret;
4027	}
4028	case PACKET_TIMESTAMP:
4029	{
4030		int val;
4031
4032		if (optlen != sizeof(val))
4033			return -EINVAL;
4034		if (copy_from_sockptr(&val, optval, sizeof(val)))
4035			return -EFAULT;
4036
4037		WRITE_ONCE(po->tp_tstamp, val);
4038		return 0;
4039	}
4040	case PACKET_FANOUT:
4041	{
4042		struct fanout_args args = { 0 };
4043
4044		if (optlen != sizeof(int) && optlen != sizeof(args))
4045			return -EINVAL;
4046		if (copy_from_sockptr(&args, optval, optlen))
4047			return -EFAULT;
4048
4049		return fanout_add(sk, &args);
4050	}
4051	case PACKET_FANOUT_DATA:
4052	{
4053		/* Paired with the WRITE_ONCE() in fanout_add() */
4054		if (!READ_ONCE(po->fanout))
4055			return -EINVAL;
4056
4057		return fanout_set_data(po, optval, optlen);
4058	}
4059	case PACKET_IGNORE_OUTGOING:
4060	{
4061		int val;
4062
4063		if (optlen != sizeof(val))
4064			return -EINVAL;
4065		if (copy_from_sockptr(&val, optval, sizeof(val)))
4066			return -EFAULT;
4067		if (val < 0 || val > 1)
4068			return -EINVAL;
4069
4070		WRITE_ONCE(po->prot_hook.ignore_outgoing, !!val);
4071		return 0;
4072	}
4073	case PACKET_TX_HAS_OFF:
4074	{
4075		unsigned int val;
4076
4077		if (optlen != sizeof(val))
4078			return -EINVAL;
4079		if (copy_from_sockptr(&val, optval, sizeof(val)))
 
 
4080			return -EFAULT;
4081
4082		lock_sock(sk);
4083		if (!po->rx_ring.pg_vec && !po->tx_ring.pg_vec)
4084			packet_sock_flag_set(po, PACKET_SOCK_TX_HAS_OFF, val);
4085
4086		release_sock(sk);
4087		return 0;
4088	}
4089	case PACKET_QDISC_BYPASS:
4090	{
4091		int val;
4092
4093		if (optlen != sizeof(val))
4094			return -EINVAL;
4095		if (copy_from_sockptr(&val, optval, sizeof(val)))
4096			return -EFAULT;
4097
4098		packet_sock_flag_set(po, PACKET_SOCK_QDISC_BYPASS, val);
4099		return 0;
4100	}
4101	default:
4102		return -ENOPROTOOPT;
4103	}
4104}
4105
4106static int packet_getsockopt(struct socket *sock, int level, int optname,
4107			     char __user *optval, int __user *optlen)
4108{
4109	int len;
4110	int val, lv = sizeof(val);
4111	struct sock *sk = sock->sk;
4112	struct packet_sock *po = pkt_sk(sk);
4113	void *data = &val;
4114	union tpacket_stats_u st;
4115	struct tpacket_rollover_stats rstats;
4116	int drops;
4117
4118	if (level != SOL_PACKET)
4119		return -ENOPROTOOPT;
4120
4121	if (get_user(len, optlen))
4122		return -EFAULT;
4123
4124	if (len < 0)
4125		return -EINVAL;
4126
4127	switch (optname) {
4128	case PACKET_STATISTICS:
4129		spin_lock_bh(&sk->sk_receive_queue.lock);
4130		memcpy(&st, &po->stats, sizeof(st));
4131		memset(&po->stats, 0, sizeof(po->stats));
4132		spin_unlock_bh(&sk->sk_receive_queue.lock);
4133		drops = atomic_xchg(&po->tp_drops, 0);
4134
4135		if (po->tp_version == TPACKET_V3) {
4136			lv = sizeof(struct tpacket_stats_v3);
4137			st.stats3.tp_drops = drops;
4138			st.stats3.tp_packets += drops;
4139			data = &st.stats3;
4140		} else {
4141			lv = sizeof(struct tpacket_stats);
4142			st.stats1.tp_drops = drops;
4143			st.stats1.tp_packets += drops;
4144			data = &st.stats1;
4145		}
4146
4147		break;
4148	case PACKET_AUXDATA:
4149		val = packet_sock_flag(po, PACKET_SOCK_AUXDATA);
4150		break;
4151	case PACKET_ORIGDEV:
4152		val = packet_sock_flag(po, PACKET_SOCK_ORIGDEV);
4153		break;
4154	case PACKET_VNET_HDR:
4155		val = !!READ_ONCE(po->vnet_hdr_sz);
4156		break;
4157	case PACKET_VNET_HDR_SZ:
4158		val = READ_ONCE(po->vnet_hdr_sz);
4159		break;
4160	case PACKET_COPY_THRESH:
4161		val = READ_ONCE(pkt_sk(sk)->copy_thresh);
4162		break;
4163	case PACKET_VERSION:
4164		val = po->tp_version;
4165		break;
4166	case PACKET_HDRLEN:
4167		if (len > sizeof(int))
4168			len = sizeof(int);
4169		if (len < sizeof(int))
4170			return -EINVAL;
4171		if (copy_from_user(&val, optval, len))
4172			return -EFAULT;
4173		switch (val) {
4174		case TPACKET_V1:
4175			val = sizeof(struct tpacket_hdr);
4176			break;
4177		case TPACKET_V2:
4178			val = sizeof(struct tpacket2_hdr);
4179			break;
4180		case TPACKET_V3:
4181			val = sizeof(struct tpacket3_hdr);
4182			break;
4183		default:
4184			return -EINVAL;
4185		}
4186		break;
4187	case PACKET_RESERVE:
4188		val = po->tp_reserve;
4189		break;
4190	case PACKET_LOSS:
4191		val = packet_sock_flag(po, PACKET_SOCK_TP_LOSS);
4192		break;
4193	case PACKET_TIMESTAMP:
4194		val = READ_ONCE(po->tp_tstamp);
4195		break;
4196	case PACKET_FANOUT:
4197		val = (po->fanout ?
4198		       ((u32)po->fanout->id |
4199			((u32)po->fanout->type << 16) |
4200			((u32)po->fanout->flags << 24)) :
4201		       0);
4202		break;
4203	case PACKET_IGNORE_OUTGOING:
4204		val = READ_ONCE(po->prot_hook.ignore_outgoing);
4205		break;
4206	case PACKET_ROLLOVER_STATS:
4207		if (!po->rollover)
4208			return -EINVAL;
4209		rstats.tp_all = atomic_long_read(&po->rollover->num);
4210		rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4211		rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4212		data = &rstats;
4213		lv = sizeof(rstats);
4214		break;
4215	case PACKET_TX_HAS_OFF:
4216		val = packet_sock_flag(po, PACKET_SOCK_TX_HAS_OFF);
4217		break;
4218	case PACKET_QDISC_BYPASS:
4219		val = packet_sock_flag(po, PACKET_SOCK_QDISC_BYPASS);
4220		break;
4221	default:
4222		return -ENOPROTOOPT;
4223	}
4224
4225	if (len > lv)
4226		len = lv;
4227	if (put_user(len, optlen))
4228		return -EFAULT;
4229	if (copy_to_user(optval, data, len))
4230		return -EFAULT;
4231	return 0;
4232}
4233
 
4234static int packet_notifier(struct notifier_block *this,
4235			   unsigned long msg, void *ptr)
4236{
4237	struct sock *sk;
4238	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4239	struct net *net = dev_net(dev);
4240
4241	rcu_read_lock();
4242	sk_for_each_rcu(sk, &net->packet.sklist) {
4243		struct packet_sock *po = pkt_sk(sk);
4244
4245		switch (msg) {
4246		case NETDEV_UNREGISTER:
4247			if (po->mclist)
4248				packet_dev_mclist_delete(dev, &po->mclist);
4249			fallthrough;
4250
4251		case NETDEV_DOWN:
4252			if (dev->ifindex == po->ifindex) {
4253				spin_lock(&po->bind_lock);
4254				if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
4255					__unregister_prot_hook(sk, false);
4256					sk->sk_err = ENETDOWN;
4257					if (!sock_flag(sk, SOCK_DEAD))
4258						sk_error_report(sk);
4259				}
4260				if (msg == NETDEV_UNREGISTER) {
4261					packet_cached_dev_reset(po);
4262					WRITE_ONCE(po->ifindex, -1);
4263					netdev_put(po->prot_hook.dev,
4264						   &po->prot_hook.dev_tracker);
4265					po->prot_hook.dev = NULL;
4266				}
4267				spin_unlock(&po->bind_lock);
4268			}
4269			break;
4270		case NETDEV_UP:
4271			if (dev->ifindex == po->ifindex) {
4272				spin_lock(&po->bind_lock);
4273				if (po->num)
4274					register_prot_hook(sk);
4275				spin_unlock(&po->bind_lock);
4276			}
4277			break;
4278		}
4279	}
4280	rcu_read_unlock();
4281	return NOTIFY_DONE;
4282}
4283
4284
4285static int packet_ioctl(struct socket *sock, unsigned int cmd,
4286			unsigned long arg)
4287{
4288	struct sock *sk = sock->sk;
4289
4290	switch (cmd) {
4291	case SIOCOUTQ:
4292	{
4293		int amount = sk_wmem_alloc_get(sk);
4294
4295		return put_user(amount, (int __user *)arg);
4296	}
4297	case SIOCINQ:
4298	{
4299		struct sk_buff *skb;
4300		int amount = 0;
4301
4302		spin_lock_bh(&sk->sk_receive_queue.lock);
4303		skb = skb_peek(&sk->sk_receive_queue);
4304		if (skb)
4305			amount = skb->len;
4306		spin_unlock_bh(&sk->sk_receive_queue.lock);
4307		return put_user(amount, (int __user *)arg);
4308	}
 
 
 
 
 
4309#ifdef CONFIG_INET
4310	case SIOCADDRT:
4311	case SIOCDELRT:
4312	case SIOCDARP:
4313	case SIOCGARP:
4314	case SIOCSARP:
4315	case SIOCGIFADDR:
4316	case SIOCSIFADDR:
4317	case SIOCGIFBRDADDR:
4318	case SIOCSIFBRDADDR:
4319	case SIOCGIFNETMASK:
4320	case SIOCSIFNETMASK:
4321	case SIOCGIFDSTADDR:
4322	case SIOCSIFDSTADDR:
4323	case SIOCSIFFLAGS:
4324		return inet_dgram_ops.ioctl(sock, cmd, arg);
4325#endif
4326
4327	default:
4328		return -ENOIOCTLCMD;
4329	}
4330	return 0;
4331}
4332
4333static __poll_t packet_poll(struct file *file, struct socket *sock,
4334				poll_table *wait)
4335{
4336	struct sock *sk = sock->sk;
4337	struct packet_sock *po = pkt_sk(sk);
4338	__poll_t mask = datagram_poll(file, sock, wait);
4339
4340	spin_lock_bh(&sk->sk_receive_queue.lock);
4341	if (po->rx_ring.pg_vec) {
4342		if (!packet_previous_rx_frame(po, &po->rx_ring,
4343			TP_STATUS_KERNEL))
4344			mask |= EPOLLIN | EPOLLRDNORM;
4345	}
4346	packet_rcv_try_clear_pressure(po);
 
4347	spin_unlock_bh(&sk->sk_receive_queue.lock);
4348	spin_lock_bh(&sk->sk_write_queue.lock);
4349	if (po->tx_ring.pg_vec) {
4350		if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4351			mask |= EPOLLOUT | EPOLLWRNORM;
4352	}
4353	spin_unlock_bh(&sk->sk_write_queue.lock);
4354	return mask;
4355}
4356
4357
4358/* Dirty? Well, I still did not learn better way to account
4359 * for user mmaps.
4360 */
4361
4362static void packet_mm_open(struct vm_area_struct *vma)
4363{
4364	struct file *file = vma->vm_file;
4365	struct socket *sock = file->private_data;
4366	struct sock *sk = sock->sk;
4367
4368	if (sk)
4369		atomic_long_inc(&pkt_sk(sk)->mapped);
4370}
4371
4372static void packet_mm_close(struct vm_area_struct *vma)
4373{
4374	struct file *file = vma->vm_file;
4375	struct socket *sock = file->private_data;
4376	struct sock *sk = sock->sk;
4377
4378	if (sk)
4379		atomic_long_dec(&pkt_sk(sk)->mapped);
4380}
4381
4382static const struct vm_operations_struct packet_mmap_ops = {
4383	.open	=	packet_mm_open,
4384	.close	=	packet_mm_close,
4385};
4386
4387static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4388			unsigned int len)
4389{
4390	int i;
4391
4392	for (i = 0; i < len; i++) {
4393		if (likely(pg_vec[i].buffer)) {
4394			if (is_vmalloc_addr(pg_vec[i].buffer))
4395				vfree(pg_vec[i].buffer);
4396			else
4397				free_pages((unsigned long)pg_vec[i].buffer,
4398					   order);
4399			pg_vec[i].buffer = NULL;
4400		}
4401	}
4402	kfree(pg_vec);
4403}
4404
4405static char *alloc_one_pg_vec_page(unsigned long order)
4406{
4407	char *buffer;
4408	gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4409			  __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4410
4411	buffer = (char *) __get_free_pages(gfp_flags, order);
4412	if (buffer)
4413		return buffer;
4414
4415	/* __get_free_pages failed, fall back to vmalloc */
4416	buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4417	if (buffer)
4418		return buffer;
4419
4420	/* vmalloc failed, lets dig into swap here */
4421	gfp_flags &= ~__GFP_NORETRY;
4422	buffer = (char *) __get_free_pages(gfp_flags, order);
4423	if (buffer)
4424		return buffer;
4425
4426	/* complete and utter failure */
4427	return NULL;
4428}
4429
4430static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4431{
4432	unsigned int block_nr = req->tp_block_nr;
4433	struct pgv *pg_vec;
4434	int i;
4435
4436	pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4437	if (unlikely(!pg_vec))
4438		goto out;
4439
4440	for (i = 0; i < block_nr; i++) {
4441		pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4442		if (unlikely(!pg_vec[i].buffer))
4443			goto out_free_pgvec;
4444	}
4445
4446out:
4447	return pg_vec;
4448
4449out_free_pgvec:
4450	free_pg_vec(pg_vec, order, block_nr);
4451	pg_vec = NULL;
4452	goto out;
4453}
4454
4455static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4456		int closing, int tx_ring)
4457{
4458	struct pgv *pg_vec = NULL;
4459	struct packet_sock *po = pkt_sk(sk);
4460	unsigned long *rx_owner_map = NULL;
4461	int was_running, order = 0;
4462	struct packet_ring_buffer *rb;
4463	struct sk_buff_head *rb_queue;
4464	__be16 num;
4465	int err;
4466	/* Added to avoid minimal code churn */
4467	struct tpacket_req *req = &req_u->req;
4468
 
 
 
 
 
 
4469	rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4470	rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4471
4472	err = -EBUSY;
4473	if (!closing) {
4474		if (atomic_long_read(&po->mapped))
4475			goto out;
4476		if (packet_read_pending(rb))
4477			goto out;
4478	}
4479
4480	if (req->tp_block_nr) {
4481		unsigned int min_frame_size;
4482
4483		/* Sanity tests and some calculations */
4484		err = -EBUSY;
4485		if (unlikely(rb->pg_vec))
4486			goto out;
4487
4488		switch (po->tp_version) {
4489		case TPACKET_V1:
4490			po->tp_hdrlen = TPACKET_HDRLEN;
4491			break;
4492		case TPACKET_V2:
4493			po->tp_hdrlen = TPACKET2_HDRLEN;
4494			break;
4495		case TPACKET_V3:
4496			po->tp_hdrlen = TPACKET3_HDRLEN;
4497			break;
4498		}
4499
4500		err = -EINVAL;
4501		if (unlikely((int)req->tp_block_size <= 0))
4502			goto out;
4503		if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4504			goto out;
4505		min_frame_size = po->tp_hdrlen + po->tp_reserve;
4506		if (po->tp_version >= TPACKET_V3 &&
4507		    req->tp_block_size <
4508		    BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4509			goto out;
4510		if (unlikely(req->tp_frame_size < min_frame_size))
 
4511			goto out;
4512		if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4513			goto out;
4514
4515		rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4516		if (unlikely(rb->frames_per_block == 0))
4517			goto out;
4518		if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4519			goto out;
4520		if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4521					req->tp_frame_nr))
4522			goto out;
4523
4524		err = -ENOMEM;
4525		order = get_order(req->tp_block_size);
4526		pg_vec = alloc_pg_vec(req, order);
4527		if (unlikely(!pg_vec))
4528			goto out;
4529		switch (po->tp_version) {
4530		case TPACKET_V3:
4531			/* Block transmit is not supported yet */
4532			if (!tx_ring) {
 
 
4533				init_prb_bdqc(po, rb, pg_vec, req_u);
4534			} else {
4535				struct tpacket_req3 *req3 = &req_u->req3;
4536
4537				if (req3->tp_retire_blk_tov ||
4538				    req3->tp_sizeof_priv ||
4539				    req3->tp_feature_req_word) {
4540					err = -EINVAL;
4541					goto out_free_pg_vec;
4542				}
4543			}
4544			break;
4545		default:
4546			if (!tx_ring) {
4547				rx_owner_map = bitmap_alloc(req->tp_frame_nr,
4548					GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
4549				if (!rx_owner_map)
4550					goto out_free_pg_vec;
4551			}
4552			break;
4553		}
4554	}
4555	/* Done */
4556	else {
4557		err = -EINVAL;
4558		if (unlikely(req->tp_frame_nr))
4559			goto out;
4560	}
4561
 
4562
4563	/* Detach socket from network */
4564	spin_lock(&po->bind_lock);
4565	was_running = packet_sock_flag(po, PACKET_SOCK_RUNNING);
4566	num = po->num;
4567	if (was_running) {
4568		WRITE_ONCE(po->num, 0);
4569		__unregister_prot_hook(sk, false);
4570	}
4571	spin_unlock(&po->bind_lock);
4572
4573	synchronize_net();
4574
4575	err = -EBUSY;
4576	mutex_lock(&po->pg_vec_lock);
4577	if (closing || atomic_long_read(&po->mapped) == 0) {
4578		err = 0;
4579		spin_lock_bh(&rb_queue->lock);
4580		swap(rb->pg_vec, pg_vec);
4581		if (po->tp_version <= TPACKET_V2)
4582			swap(rb->rx_owner_map, rx_owner_map);
4583		rb->frame_max = (req->tp_frame_nr - 1);
4584		rb->head = 0;
4585		rb->frame_size = req->tp_frame_size;
4586		spin_unlock_bh(&rb_queue->lock);
4587
4588		swap(rb->pg_vec_order, order);
4589		swap(rb->pg_vec_len, req->tp_block_nr);
4590
4591		rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4592		po->prot_hook.func = (po->rx_ring.pg_vec) ?
4593						tpacket_rcv : packet_rcv;
4594		skb_queue_purge(rb_queue);
4595		if (atomic_long_read(&po->mapped))
4596			pr_err("packet_mmap: vma is busy: %ld\n",
4597			       atomic_long_read(&po->mapped));
4598	}
4599	mutex_unlock(&po->pg_vec_lock);
4600
4601	spin_lock(&po->bind_lock);
4602	if (was_running) {
4603		WRITE_ONCE(po->num, num);
4604		register_prot_hook(sk);
4605	}
4606	spin_unlock(&po->bind_lock);
4607	if (pg_vec && (po->tp_version > TPACKET_V2)) {
4608		/* Because we don't support block-based V3 on tx-ring */
4609		if (!tx_ring)
4610			prb_shutdown_retire_blk_timer(po, rb_queue);
4611	}
 
4612
4613out_free_pg_vec:
4614	if (pg_vec) {
4615		bitmap_free(rx_owner_map);
4616		free_pg_vec(pg_vec, order, req->tp_block_nr);
4617	}
4618out:
4619	return err;
4620}
4621
4622static int packet_mmap(struct file *file, struct socket *sock,
4623		struct vm_area_struct *vma)
4624{
4625	struct sock *sk = sock->sk;
4626	struct packet_sock *po = pkt_sk(sk);
4627	unsigned long size, expected_size;
4628	struct packet_ring_buffer *rb;
4629	unsigned long start;
4630	int err = -EINVAL;
4631	int i;
4632
4633	if (vma->vm_pgoff)
4634		return -EINVAL;
4635
4636	mutex_lock(&po->pg_vec_lock);
4637
4638	expected_size = 0;
4639	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4640		if (rb->pg_vec) {
4641			expected_size += rb->pg_vec_len
4642						* rb->pg_vec_pages
4643						* PAGE_SIZE;
4644		}
4645	}
4646
4647	if (expected_size == 0)
4648		goto out;
4649
4650	size = vma->vm_end - vma->vm_start;
4651	if (size != expected_size)
4652		goto out;
4653
4654	start = vma->vm_start;
4655	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4656		if (rb->pg_vec == NULL)
4657			continue;
4658
4659		for (i = 0; i < rb->pg_vec_len; i++) {
4660			struct page *page;
4661			void *kaddr = rb->pg_vec[i].buffer;
4662			int pg_num;
4663
4664			for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4665				page = pgv_to_page(kaddr);
4666				err = vm_insert_page(vma, start, page);
4667				if (unlikely(err))
4668					goto out;
4669				start += PAGE_SIZE;
4670				kaddr += PAGE_SIZE;
4671			}
4672		}
4673	}
4674
4675	atomic_long_inc(&po->mapped);
4676	vma->vm_ops = &packet_mmap_ops;
4677	err = 0;
4678
4679out:
4680	mutex_unlock(&po->pg_vec_lock);
4681	return err;
4682}
4683
4684static const struct proto_ops packet_ops_spkt = {
4685	.family =	PF_PACKET,
4686	.owner =	THIS_MODULE,
4687	.release =	packet_release,
4688	.bind =		packet_bind_spkt,
4689	.connect =	sock_no_connect,
4690	.socketpair =	sock_no_socketpair,
4691	.accept =	sock_no_accept,
4692	.getname =	packet_getname_spkt,
4693	.poll =		datagram_poll,
4694	.ioctl =	packet_ioctl,
4695	.gettstamp =	sock_gettstamp,
4696	.listen =	sock_no_listen,
4697	.shutdown =	sock_no_shutdown,
 
 
4698	.sendmsg =	packet_sendmsg_spkt,
4699	.recvmsg =	packet_recvmsg,
4700	.mmap =		sock_no_mmap,
 
4701};
4702
4703static const struct proto_ops packet_ops = {
4704	.family =	PF_PACKET,
4705	.owner =	THIS_MODULE,
4706	.release =	packet_release,
4707	.bind =		packet_bind,
4708	.connect =	sock_no_connect,
4709	.socketpair =	sock_no_socketpair,
4710	.accept =	sock_no_accept,
4711	.getname =	packet_getname,
4712	.poll =		packet_poll,
4713	.ioctl =	packet_ioctl,
4714	.gettstamp =	sock_gettstamp,
4715	.listen =	sock_no_listen,
4716	.shutdown =	sock_no_shutdown,
4717	.setsockopt =	packet_setsockopt,
4718	.getsockopt =	packet_getsockopt,
4719	.sendmsg =	packet_sendmsg,
4720	.recvmsg =	packet_recvmsg,
4721	.mmap =		packet_mmap,
 
4722};
4723
4724static const struct net_proto_family packet_family_ops = {
4725	.family =	PF_PACKET,
4726	.create =	packet_create,
4727	.owner	=	THIS_MODULE,
4728};
4729
4730static struct notifier_block packet_netdev_notifier = {
4731	.notifier_call =	packet_notifier,
4732};
4733
4734#ifdef CONFIG_PROC_FS
4735
4736static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4737	__acquires(RCU)
4738{
4739	struct net *net = seq_file_net(seq);
4740
4741	rcu_read_lock();
4742	return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4743}
4744
4745static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4746{
4747	struct net *net = seq_file_net(seq);
4748	return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4749}
4750
4751static void packet_seq_stop(struct seq_file *seq, void *v)
4752	__releases(RCU)
4753{
4754	rcu_read_unlock();
4755}
4756
4757static int packet_seq_show(struct seq_file *seq, void *v)
4758{
4759	if (v == SEQ_START_TOKEN)
4760		seq_printf(seq,
4761			   "%*sRefCnt Type Proto  Iface R Rmem   User   Inode\n",
4762			   IS_ENABLED(CONFIG_64BIT) ? -17 : -9, "sk");
4763	else {
4764		struct sock *s = sk_entry(v);
4765		const struct packet_sock *po = pkt_sk(s);
4766
4767		seq_printf(seq,
4768			   "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4769			   s,
4770			   refcount_read(&s->sk_refcnt),
4771			   s->sk_type,
4772			   ntohs(READ_ONCE(po->num)),
4773			   READ_ONCE(po->ifindex),
4774			   packet_sock_flag(po, PACKET_SOCK_RUNNING),
4775			   atomic_read(&s->sk_rmem_alloc),
4776			   from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4777			   sock_i_ino(s));
4778	}
4779
4780	return 0;
4781}
4782
4783static const struct seq_operations packet_seq_ops = {
4784	.start	= packet_seq_start,
4785	.next	= packet_seq_next,
4786	.stop	= packet_seq_stop,
4787	.show	= packet_seq_show,
4788};
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4789#endif
4790
4791static int __net_init packet_net_init(struct net *net)
4792{
4793	mutex_init(&net->packet.sklist_lock);
4794	INIT_HLIST_HEAD(&net->packet.sklist);
4795
4796#ifdef CONFIG_PROC_FS
4797	if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4798			sizeof(struct seq_net_private)))
4799		return -ENOMEM;
4800#endif /* CONFIG_PROC_FS */
4801
4802	return 0;
4803}
4804
4805static void __net_exit packet_net_exit(struct net *net)
4806{
4807	remove_proc_entry("packet", net->proc_net);
4808	WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4809}
4810
4811static struct pernet_operations packet_net_ops = {
4812	.init = packet_net_init,
4813	.exit = packet_net_exit,
4814};
4815
4816
4817static void __exit packet_exit(void)
4818{
 
 
4819	sock_unregister(PF_PACKET);
4820	proto_unregister(&packet_proto);
4821	unregister_netdevice_notifier(&packet_netdev_notifier);
4822	unregister_pernet_subsys(&packet_net_ops);
4823}
4824
4825static int __init packet_init(void)
4826{
4827	int rc;
4828
4829	rc = register_pernet_subsys(&packet_net_ops);
4830	if (rc)
4831		goto out;
4832	rc = register_netdevice_notifier(&packet_netdev_notifier);
4833	if (rc)
4834		goto out_pernet;
4835	rc = proto_register(&packet_proto, 0);
4836	if (rc)
4837		goto out_notifier;
4838	rc = sock_register(&packet_family_ops);
4839	if (rc)
4840		goto out_proto;
4841
4842	return 0;
4843
4844out_proto:
4845	proto_unregister(&packet_proto);
4846out_notifier:
4847	unregister_netdevice_notifier(&packet_netdev_notifier);
4848out_pernet:
4849	unregister_pernet_subsys(&packet_net_ops);
4850out:
4851	return rc;
4852}
4853
4854module_init(packet_init);
4855module_exit(packet_exit);
4856MODULE_DESCRIPTION("Packet socket support (AF_PACKET)");
4857MODULE_LICENSE("GPL");
4858MODULE_ALIAS_NETPROTO(PF_PACKET);