Linux Audio

Check our new training course

Loading...
v5.9
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * INET		An implementation of the TCP/IP protocol suite for the LINUX
   4 *		operating system.  INET is implemented using the  BSD Socket
   5 *		interface as the means of communication with the user level.
   6 *
   7 *		PACKET - implements raw packet sockets.
   8 *
   9 * Authors:	Ross Biro
  10 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  11 *		Alan Cox, <gw4pts@gw4pts.ampr.org>
  12 *
  13 * Fixes:
  14 *		Alan Cox	:	verify_area() now used correctly
  15 *		Alan Cox	:	new skbuff lists, look ma no backlogs!
  16 *		Alan Cox	:	tidied skbuff lists.
  17 *		Alan Cox	:	Now uses generic datagram routines I
  18 *					added. Also fixed the peek/read crash
  19 *					from all old Linux datagram code.
  20 *		Alan Cox	:	Uses the improved datagram code.
  21 *		Alan Cox	:	Added NULL's for socket options.
  22 *		Alan Cox	:	Re-commented the code.
  23 *		Alan Cox	:	Use new kernel side addressing
  24 *		Rob Janssen	:	Correct MTU usage.
  25 *		Dave Platt	:	Counter leaks caused by incorrect
  26 *					interrupt locking and some slightly
  27 *					dubious gcc output. Can you read
  28 *					compiler: it said _VOLATILE_
  29 *	Richard Kooijman	:	Timestamp fixes.
  30 *		Alan Cox	:	New buffers. Use sk->mac.raw.
  31 *		Alan Cox	:	sendmsg/recvmsg support.
  32 *		Alan Cox	:	Protocol setting support
  33 *	Alexey Kuznetsov	:	Untied from IPv4 stack.
  34 *	Cyrus Durgin		:	Fixed kerneld for kmod.
  35 *	Michal Ostrowski        :       Module initialization cleanup.
  36 *         Ulises Alonso        :       Frame number limit removal and
  37 *                                      packet_set_ring memory leak.
  38 *		Eric Biederman	:	Allow for > 8 byte hardware addresses.
  39 *					The convention is that longer addresses
  40 *					will simply extend the hardware address
  41 *					byte arrays at the end of sockaddr_ll
  42 *					and packet_mreq.
  43 *		Johann Baudy	:	Added TX RING.
  44 *		Chetan Loke	:	Implemented TPACKET_V3 block abstraction
  45 *					layer.
  46 *					Copyright (C) 2011, <lokec@ccs.neu.edu>
  47 */
  48
 
 
 
 
  49#include <linux/types.h>
  50#include <linux/mm.h>
  51#include <linux/capability.h>
  52#include <linux/fcntl.h>
  53#include <linux/socket.h>
  54#include <linux/in.h>
  55#include <linux/inet.h>
  56#include <linux/netdevice.h>
  57#include <linux/if_packet.h>
  58#include <linux/wireless.h>
  59#include <linux/kernel.h>
  60#include <linux/kmod.h>
  61#include <linux/slab.h>
  62#include <linux/vmalloc.h>
  63#include <net/net_namespace.h>
  64#include <net/ip.h>
  65#include <net/protocol.h>
  66#include <linux/skbuff.h>
  67#include <net/sock.h>
  68#include <linux/errno.h>
  69#include <linux/timer.h>
  70#include <linux/uaccess.h>
  71#include <asm/ioctls.h>
  72#include <asm/page.h>
  73#include <asm/cacheflush.h>
  74#include <asm/io.h>
  75#include <linux/proc_fs.h>
  76#include <linux/seq_file.h>
  77#include <linux/poll.h>
  78#include <linux/module.h>
  79#include <linux/init.h>
  80#include <linux/mutex.h>
  81#include <linux/if_vlan.h>
  82#include <linux/virtio_net.h>
  83#include <linux/errqueue.h>
  84#include <linux/net_tstamp.h>
  85#include <linux/percpu.h>
  86#ifdef CONFIG_INET
  87#include <net/inet_common.h>
  88#endif
  89#include <linux/bpf.h>
  90#include <net/compat.h>
 
  91
  92#include "internal.h"
  93
  94/*
  95   Assumptions:
  96   - if device has no dev->hard_header routine, it adds and removes ll header
  97     inside itself. In this case ll header is invisible outside of device,
  98     but higher levels still should reserve dev->hard_header_len.
  99     Some devices are enough clever to reallocate skb, when header
 100     will not fit to reserved space (tunnel), another ones are silly
 101     (PPP).
 
 
 
 102   - packet socket receives packets with pulled ll header,
 103     so that SOCK_RAW should push it back.
 104
 105On receive:
 106-----------
 107
 108Incoming, dev->hard_header!=NULL
 109   mac_header -> ll header
 110   data       -> data
 111
 112Outgoing, dev->hard_header!=NULL
 113   mac_header -> ll header
 114   data       -> ll header
 115
 116Incoming, dev->hard_header==NULL
 117   mac_header -> UNKNOWN position. It is very likely, that it points to ll
 118		 header.  PPP makes it, that is wrong, because introduce
 119		 assymetry between rx and tx paths.
 120   data       -> data
 121
 122Outgoing, dev->hard_header==NULL
 123   mac_header -> data. ll header is still not built!
 124   data       -> data
 125
 126Resume
 127  If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 
 128
 129
 130On transmit:
 131------------
 132
 133dev->hard_header != NULL
 134   mac_header -> ll header
 135   data       -> ll header
 136
 137dev->hard_header == NULL (ll header is added by device, we cannot control it)
 138   mac_header -> data
 139   data       -> data
 140
 141   We should set nh.raw on output to correct posistion,
 142   packet classifier depends on it.
 143 */
 144
 145/* Private packet socket structures. */
 146
 147/* identical to struct packet_mreq except it has
 148 * a longer address field.
 149 */
 150struct packet_mreq_max {
 151	int		mr_ifindex;
 152	unsigned short	mr_type;
 153	unsigned short	mr_alen;
 154	unsigned char	mr_address[MAX_ADDR_LEN];
 155};
 156
 157union tpacket_uhdr {
 158	struct tpacket_hdr  *h1;
 159	struct tpacket2_hdr *h2;
 160	struct tpacket3_hdr *h3;
 161	void *raw;
 162};
 163
 164static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 165		int closing, int tx_ring);
 166
 167#define V3_ALIGNMENT	(8)
 168
 169#define BLK_HDR_LEN	(ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 170
 171#define BLK_PLUS_PRIV(sz_of_priv) \
 172	(BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 173
 174#define BLOCK_STATUS(x)	((x)->hdr.bh1.block_status)
 175#define BLOCK_NUM_PKTS(x)	((x)->hdr.bh1.num_pkts)
 176#define BLOCK_O2FP(x)		((x)->hdr.bh1.offset_to_first_pkt)
 177#define BLOCK_LEN(x)		((x)->hdr.bh1.blk_len)
 178#define BLOCK_SNUM(x)		((x)->hdr.bh1.seq_num)
 179#define BLOCK_O2PRIV(x)	((x)->offset_to_priv)
 180#define BLOCK_PRIV(x)		((void *)((char *)(x) + BLOCK_O2PRIV(x)))
 181
 182struct packet_sock;
 183static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 184		       struct packet_type *pt, struct net_device *orig_dev);
 185
 186static void *packet_previous_frame(struct packet_sock *po,
 187		struct packet_ring_buffer *rb,
 188		int status);
 189static void packet_increment_head(struct packet_ring_buffer *buff);
 190static int prb_curr_blk_in_use(struct tpacket_block_desc *);
 191static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 192			struct packet_sock *);
 193static void prb_retire_current_block(struct tpacket_kbdq_core *,
 194		struct packet_sock *, unsigned int status);
 195static int prb_queue_frozen(struct tpacket_kbdq_core *);
 196static void prb_open_block(struct tpacket_kbdq_core *,
 197		struct tpacket_block_desc *);
 198static void prb_retire_rx_blk_timer_expired(struct timer_list *);
 199static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 200static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 201static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 202		struct tpacket3_hdr *);
 203static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 204		struct tpacket3_hdr *);
 205static void packet_flush_mclist(struct sock *sk);
 206static u16 packet_pick_tx_queue(struct sk_buff *skb);
 207
 208struct packet_skb_cb {
 209	union {
 210		struct sockaddr_pkt pkt;
 211		union {
 212			/* Trick: alias skb original length with
 213			 * ll.sll_family and ll.protocol in order
 214			 * to save room.
 215			 */
 216			unsigned int origlen;
 217			struct sockaddr_ll ll;
 218		};
 219	} sa;
 220};
 221
 222#define vio_le() virtio_legacy_is_little_endian()
 223
 224#define PACKET_SKB_CB(__skb)	((struct packet_skb_cb *)((__skb)->cb))
 225
 226#define GET_PBDQC_FROM_RB(x)	((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 227#define GET_PBLOCK_DESC(x, bid)	\
 228	((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 229#define GET_CURR_PBLOCK_DESC_FROM_CORE(x)	\
 230	((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 231#define GET_NEXT_PRB_BLK_NUM(x) \
 232	(((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 233	((x)->kactive_blk_num+1) : 0)
 234
 235static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 236static void __fanout_link(struct sock *sk, struct packet_sock *po);
 237
 238static int packet_direct_xmit(struct sk_buff *skb)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 239{
 
 
 
 
 
 
 
 
 
 
 240	return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
 241}
 242
 243static struct net_device *packet_cached_dev_get(struct packet_sock *po)
 244{
 245	struct net_device *dev;
 246
 247	rcu_read_lock();
 248	dev = rcu_dereference(po->cached_dev);
 249	if (likely(dev))
 250		dev_hold(dev);
 251	rcu_read_unlock();
 252
 253	return dev;
 254}
 255
 256static void packet_cached_dev_assign(struct packet_sock *po,
 257				     struct net_device *dev)
 258{
 259	rcu_assign_pointer(po->cached_dev, dev);
 260}
 261
 262static void packet_cached_dev_reset(struct packet_sock *po)
 263{
 264	RCU_INIT_POINTER(po->cached_dev, NULL);
 265}
 266
 267static bool packet_use_direct_xmit(const struct packet_sock *po)
 268{
 269	return po->xmit == packet_direct_xmit;
 270}
 271
 272static u16 packet_pick_tx_queue(struct sk_buff *skb)
 273{
 274	struct net_device *dev = skb->dev;
 275	const struct net_device_ops *ops = dev->netdev_ops;
 276	int cpu = raw_smp_processor_id();
 277	u16 queue_index;
 278
 279#ifdef CONFIG_XPS
 280	skb->sender_cpu = cpu + 1;
 281#endif
 282	skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues);
 283	if (ops->ndo_select_queue) {
 284		queue_index = ops->ndo_select_queue(dev, skb, NULL);
 285		queue_index = netdev_cap_txqueue(dev, queue_index);
 286	} else {
 287		queue_index = netdev_pick_tx(dev, skb, NULL);
 288	}
 289
 290	return queue_index;
 291}
 292
 293/* __register_prot_hook must be invoked through register_prot_hook
 294 * or from a context in which asynchronous accesses to the packet
 295 * socket is not possible (packet_create()).
 296 */
 297static void __register_prot_hook(struct sock *sk)
 298{
 299	struct packet_sock *po = pkt_sk(sk);
 300
 301	if (!po->running) {
 302		if (po->fanout)
 303			__fanout_link(sk, po);
 304		else
 305			dev_add_pack(&po->prot_hook);
 306
 307		sock_hold(sk);
 308		po->running = 1;
 309	}
 310}
 311
 312static void register_prot_hook(struct sock *sk)
 313{
 314	lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
 315	__register_prot_hook(sk);
 316}
 317
 318/* If the sync parameter is true, we will temporarily drop
 319 * the po->bind_lock and do a synchronize_net to make sure no
 320 * asynchronous packet processing paths still refer to the elements
 321 * of po->prot_hook.  If the sync parameter is false, it is the
 322 * callers responsibility to take care of this.
 323 */
 324static void __unregister_prot_hook(struct sock *sk, bool sync)
 325{
 326	struct packet_sock *po = pkt_sk(sk);
 327
 328	lockdep_assert_held_once(&po->bind_lock);
 329
 330	po->running = 0;
 331
 332	if (po->fanout)
 333		__fanout_unlink(sk, po);
 334	else
 335		__dev_remove_pack(&po->prot_hook);
 336
 337	__sock_put(sk);
 338
 339	if (sync) {
 340		spin_unlock(&po->bind_lock);
 341		synchronize_net();
 342		spin_lock(&po->bind_lock);
 343	}
 344}
 345
 346static void unregister_prot_hook(struct sock *sk, bool sync)
 347{
 348	struct packet_sock *po = pkt_sk(sk);
 349
 350	if (po->running)
 351		__unregister_prot_hook(sk, sync);
 352}
 353
 354static inline struct page * __pure pgv_to_page(void *addr)
 355{
 356	if (is_vmalloc_addr(addr))
 357		return vmalloc_to_page(addr);
 358	return virt_to_page(addr);
 359}
 360
 361static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 362{
 363	union tpacket_uhdr h;
 364
 
 
 365	h.raw = frame;
 366	switch (po->tp_version) {
 367	case TPACKET_V1:
 368		h.h1->tp_status = status;
 369		flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 370		break;
 371	case TPACKET_V2:
 372		h.h2->tp_status = status;
 373		flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 374		break;
 375	case TPACKET_V3:
 376		h.h3->tp_status = status;
 377		flush_dcache_page(pgv_to_page(&h.h3->tp_status));
 378		break;
 379	default:
 380		WARN(1, "TPACKET version not supported.\n");
 381		BUG();
 382	}
 383
 384	smp_wmb();
 385}
 386
 387static int __packet_get_status(const struct packet_sock *po, void *frame)
 388{
 389	union tpacket_uhdr h;
 390
 391	smp_rmb();
 392
 
 
 393	h.raw = frame;
 394	switch (po->tp_version) {
 395	case TPACKET_V1:
 396		flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 397		return h.h1->tp_status;
 398	case TPACKET_V2:
 399		flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 400		return h.h2->tp_status;
 401	case TPACKET_V3:
 402		flush_dcache_page(pgv_to_page(&h.h3->tp_status));
 403		return h.h3->tp_status;
 404	default:
 405		WARN(1, "TPACKET version not supported.\n");
 406		BUG();
 407		return 0;
 408	}
 409}
 410
 411static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec64 *ts,
 412				   unsigned int flags)
 413{
 414	struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 415
 416	if (shhwtstamps &&
 417	    (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
 418	    ktime_to_timespec64_cond(shhwtstamps->hwtstamp, ts))
 419		return TP_STATUS_TS_RAW_HARDWARE;
 420
 421	if (ktime_to_timespec64_cond(skb->tstamp, ts))
 
 422		return TP_STATUS_TS_SOFTWARE;
 423
 424	return 0;
 425}
 426
 427static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
 428				    struct sk_buff *skb)
 429{
 430	union tpacket_uhdr h;
 431	struct timespec64 ts;
 432	__u32 ts_status;
 433
 434	if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
 435		return 0;
 436
 437	h.raw = frame;
 438	/*
 439	 * versions 1 through 3 overflow the timestamps in y2106, since they
 440	 * all store the seconds in a 32-bit unsigned integer.
 441	 * If we create a version 4, that should have a 64-bit timestamp,
 442	 * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit
 443	 * nanoseconds.
 444	 */
 445	switch (po->tp_version) {
 446	case TPACKET_V1:
 447		h.h1->tp_sec = ts.tv_sec;
 448		h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
 449		break;
 450	case TPACKET_V2:
 451		h.h2->tp_sec = ts.tv_sec;
 452		h.h2->tp_nsec = ts.tv_nsec;
 453		break;
 454	case TPACKET_V3:
 455		h.h3->tp_sec = ts.tv_sec;
 456		h.h3->tp_nsec = ts.tv_nsec;
 457		break;
 458	default:
 459		WARN(1, "TPACKET version not supported.\n");
 460		BUG();
 461	}
 462
 463	/* one flush is safe, as both fields always lie on the same cacheline */
 464	flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
 465	smp_wmb();
 466
 467	return ts_status;
 468}
 469
 470static void *packet_lookup_frame(const struct packet_sock *po,
 471				 const struct packet_ring_buffer *rb,
 472				 unsigned int position,
 473				 int status)
 474{
 475	unsigned int pg_vec_pos, frame_offset;
 476	union tpacket_uhdr h;
 477
 478	pg_vec_pos = position / rb->frames_per_block;
 479	frame_offset = position % rb->frames_per_block;
 480
 481	h.raw = rb->pg_vec[pg_vec_pos].buffer +
 482		(frame_offset * rb->frame_size);
 483
 484	if (status != __packet_get_status(po, h.raw))
 485		return NULL;
 486
 487	return h.raw;
 488}
 489
 490static void *packet_current_frame(struct packet_sock *po,
 491		struct packet_ring_buffer *rb,
 492		int status)
 493{
 494	return packet_lookup_frame(po, rb, rb->head, status);
 495}
 496
 497static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 498{
 499	del_timer_sync(&pkc->retire_blk_timer);
 500}
 501
 502static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 503		struct sk_buff_head *rb_queue)
 504{
 505	struct tpacket_kbdq_core *pkc;
 506
 507	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 508
 509	spin_lock_bh(&rb_queue->lock);
 510	pkc->delete_blk_timer = 1;
 511	spin_unlock_bh(&rb_queue->lock);
 512
 513	prb_del_retire_blk_timer(pkc);
 514}
 515
 516static void prb_setup_retire_blk_timer(struct packet_sock *po)
 517{
 518	struct tpacket_kbdq_core *pkc;
 519
 520	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 521	timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
 522		    0);
 523	pkc->retire_blk_timer.expires = jiffies;
 524}
 525
 526static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 527				int blk_size_in_bytes)
 528{
 529	struct net_device *dev;
 530	unsigned int mbits, div;
 531	struct ethtool_link_ksettings ecmd;
 532	int err;
 533
 534	rtnl_lock();
 535	dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 536	if (unlikely(!dev)) {
 537		rtnl_unlock();
 538		return DEFAULT_PRB_RETIRE_TOV;
 539	}
 540	err = __ethtool_get_link_ksettings(dev, &ecmd);
 541	rtnl_unlock();
 542	if (err)
 543		return DEFAULT_PRB_RETIRE_TOV;
 544
 545	/* If the link speed is so slow you don't really
 546	 * need to worry about perf anyways
 547	 */
 548	if (ecmd.base.speed < SPEED_1000 ||
 549	    ecmd.base.speed == SPEED_UNKNOWN)
 550		return DEFAULT_PRB_RETIRE_TOV;
 551
 552	div = ecmd.base.speed / 1000;
 553	mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 554
 555	if (div)
 556		mbits /= div;
 557
 558	if (div)
 559		return mbits + 1;
 560	return mbits;
 561}
 562
 563static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 564			union tpacket_req_u *req_u)
 565{
 566	p1->feature_req_word = req_u->req3.tp_feature_req_word;
 567}
 568
 569static void init_prb_bdqc(struct packet_sock *po,
 570			struct packet_ring_buffer *rb,
 571			struct pgv *pg_vec,
 572			union tpacket_req_u *req_u)
 573{
 574	struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
 575	struct tpacket_block_desc *pbd;
 576
 577	memset(p1, 0x0, sizeof(*p1));
 578
 579	p1->knxt_seq_num = 1;
 580	p1->pkbdq = pg_vec;
 581	pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 582	p1->pkblk_start	= pg_vec[0].buffer;
 583	p1->kblk_size = req_u->req3.tp_block_size;
 584	p1->knum_blocks	= req_u->req3.tp_block_nr;
 585	p1->hdrlen = po->tp_hdrlen;
 586	p1->version = po->tp_version;
 587	p1->last_kactive_blk_num = 0;
 588	po->stats.stats3.tp_freeze_q_cnt = 0;
 589	if (req_u->req3.tp_retire_blk_tov)
 590		p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 591	else
 592		p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 593						req_u->req3.tp_block_size);
 594	p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 595	p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 596	rwlock_init(&p1->blk_fill_in_prog_lock);
 597
 598	p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
 599	prb_init_ft_ops(p1, req_u);
 600	prb_setup_retire_blk_timer(po);
 601	prb_open_block(p1, pbd);
 602}
 603
 604/*  Do NOT update the last_blk_num first.
 605 *  Assumes sk_buff_head lock is held.
 606 */
 607static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 608{
 609	mod_timer(&pkc->retire_blk_timer,
 610			jiffies + pkc->tov_in_jiffies);
 611	pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 612}
 613
 614/*
 615 * Timer logic:
 616 * 1) We refresh the timer only when we open a block.
 617 *    By doing this we don't waste cycles refreshing the timer
 618 *	  on packet-by-packet basis.
 619 *
 620 * With a 1MB block-size, on a 1Gbps line, it will take
 621 * i) ~8 ms to fill a block + ii) memcpy etc.
 622 * In this cut we are not accounting for the memcpy time.
 623 *
 624 * So, if the user sets the 'tmo' to 10ms then the timer
 625 * will never fire while the block is still getting filled
 626 * (which is what we want). However, the user could choose
 627 * to close a block early and that's fine.
 628 *
 629 * But when the timer does fire, we check whether or not to refresh it.
 630 * Since the tmo granularity is in msecs, it is not too expensive
 631 * to refresh the timer, lets say every '8' msecs.
 632 * Either the user can set the 'tmo' or we can derive it based on
 633 * a) line-speed and b) block-size.
 634 * prb_calc_retire_blk_tmo() calculates the tmo.
 635 *
 636 */
 637static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
 638{
 639	struct packet_sock *po =
 640		from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
 641	struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 642	unsigned int frozen;
 643	struct tpacket_block_desc *pbd;
 644
 645	spin_lock(&po->sk.sk_receive_queue.lock);
 646
 647	frozen = prb_queue_frozen(pkc);
 648	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 649
 650	if (unlikely(pkc->delete_blk_timer))
 651		goto out;
 652
 653	/* We only need to plug the race when the block is partially filled.
 654	 * tpacket_rcv:
 655	 *		lock(); increment BLOCK_NUM_PKTS; unlock()
 656	 *		copy_bits() is in progress ...
 657	 *		timer fires on other cpu:
 658	 *		we can't retire the current block because copy_bits
 659	 *		is in progress.
 660	 *
 661	 */
 662	if (BLOCK_NUM_PKTS(pbd)) {
 663		/* Waiting for skb_copy_bits to finish... */
 664		write_lock(&pkc->blk_fill_in_prog_lock);
 665		write_unlock(&pkc->blk_fill_in_prog_lock);
 666	}
 667
 668	if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 669		if (!frozen) {
 670			if (!BLOCK_NUM_PKTS(pbd)) {
 671				/* An empty block. Just refresh the timer. */
 672				goto refresh_timer;
 673			}
 674			prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 675			if (!prb_dispatch_next_block(pkc, po))
 676				goto refresh_timer;
 677			else
 678				goto out;
 679		} else {
 680			/* Case 1. Queue was frozen because user-space was
 681			 *	   lagging behind.
 682			 */
 683			if (prb_curr_blk_in_use(pbd)) {
 684				/*
 685				 * Ok, user-space is still behind.
 686				 * So just refresh the timer.
 687				 */
 688				goto refresh_timer;
 689			} else {
 690			       /* Case 2. queue was frozen,user-space caught up,
 691				* now the link went idle && the timer fired.
 692				* We don't have a block to close.So we open this
 693				* block and restart the timer.
 694				* opening a block thaws the queue,restarts timer
 695				* Thawing/timer-refresh is a side effect.
 696				*/
 697				prb_open_block(pkc, pbd);
 698				goto out;
 699			}
 700		}
 701	}
 702
 703refresh_timer:
 704	_prb_refresh_rx_retire_blk_timer(pkc);
 705
 706out:
 707	spin_unlock(&po->sk.sk_receive_queue.lock);
 708}
 709
 710static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 711		struct tpacket_block_desc *pbd1, __u32 status)
 712{
 713	/* Flush everything minus the block header */
 714
 715#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 716	u8 *start, *end;
 717
 718	start = (u8 *)pbd1;
 719
 720	/* Skip the block header(we know header WILL fit in 4K) */
 721	start += PAGE_SIZE;
 722
 723	end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 724	for (; start < end; start += PAGE_SIZE)
 725		flush_dcache_page(pgv_to_page(start));
 726
 727	smp_wmb();
 728#endif
 729
 730	/* Now update the block status. */
 731
 732	BLOCK_STATUS(pbd1) = status;
 733
 734	/* Flush the block header */
 735
 736#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 737	start = (u8 *)pbd1;
 738	flush_dcache_page(pgv_to_page(start));
 739
 740	smp_wmb();
 741#endif
 742}
 743
 744/*
 745 * Side effect:
 746 *
 747 * 1) flush the block
 748 * 2) Increment active_blk_num
 749 *
 750 * Note:We DONT refresh the timer on purpose.
 751 *	Because almost always the next block will be opened.
 752 */
 753static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 754		struct tpacket_block_desc *pbd1,
 755		struct packet_sock *po, unsigned int stat)
 756{
 757	__u32 status = TP_STATUS_USER | stat;
 758
 759	struct tpacket3_hdr *last_pkt;
 760	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 761	struct sock *sk = &po->sk;
 762
 763	if (atomic_read(&po->tp_drops))
 764		status |= TP_STATUS_LOSING;
 765
 766	last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 767	last_pkt->tp_next_offset = 0;
 768
 769	/* Get the ts of the last pkt */
 770	if (BLOCK_NUM_PKTS(pbd1)) {
 771		h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 772		h1->ts_last_pkt.ts_nsec	= last_pkt->tp_nsec;
 773	} else {
 774		/* Ok, we tmo'd - so get the current time.
 775		 *
 776		 * It shouldn't really happen as we don't close empty
 777		 * blocks. See prb_retire_rx_blk_timer_expired().
 778		 */
 779		struct timespec64 ts;
 780		ktime_get_real_ts64(&ts);
 781		h1->ts_last_pkt.ts_sec = ts.tv_sec;
 782		h1->ts_last_pkt.ts_nsec	= ts.tv_nsec;
 783	}
 784
 785	smp_wmb();
 786
 787	/* Flush the block */
 788	prb_flush_block(pkc1, pbd1, status);
 789
 790	sk->sk_data_ready(sk);
 791
 792	pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 793}
 794
 795static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 796{
 797	pkc->reset_pending_on_curr_blk = 0;
 798}
 799
 800/*
 801 * Side effect of opening a block:
 802 *
 803 * 1) prb_queue is thawed.
 804 * 2) retire_blk_timer is refreshed.
 805 *
 806 */
 807static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 808	struct tpacket_block_desc *pbd1)
 809{
 810	struct timespec64 ts;
 811	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 812
 813	smp_rmb();
 814
 815	/* We could have just memset this but we will lose the
 816	 * flexibility of making the priv area sticky
 817	 */
 818
 819	BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 820	BLOCK_NUM_PKTS(pbd1) = 0;
 821	BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 822
 823	ktime_get_real_ts64(&ts);
 824
 825	h1->ts_first_pkt.ts_sec = ts.tv_sec;
 826	h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 827
 828	pkc1->pkblk_start = (char *)pbd1;
 829	pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 830
 831	BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 832	BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 833
 834	pbd1->version = pkc1->version;
 835	pkc1->prev = pkc1->nxt_offset;
 836	pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 837
 838	prb_thaw_queue(pkc1);
 839	_prb_refresh_rx_retire_blk_timer(pkc1);
 840
 841	smp_wmb();
 842}
 843
 844/*
 845 * Queue freeze logic:
 846 * 1) Assume tp_block_nr = 8 blocks.
 847 * 2) At time 't0', user opens Rx ring.
 848 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 849 * 4) user-space is either sleeping or processing block '0'.
 850 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 851 *    it will close block-7,loop around and try to fill block '0'.
 852 *    call-flow:
 853 *    __packet_lookup_frame_in_block
 854 *      prb_retire_current_block()
 855 *      prb_dispatch_next_block()
 856 *        |->(BLOCK_STATUS == USER) evaluates to true
 857 *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 858 * 6) Now there are two cases:
 859 *    6.1) Link goes idle right after the queue is frozen.
 860 *         But remember, the last open_block() refreshed the timer.
 861 *         When this timer expires,it will refresh itself so that we can
 862 *         re-open block-0 in near future.
 863 *    6.2) Link is busy and keeps on receiving packets. This is a simple
 864 *         case and __packet_lookup_frame_in_block will check if block-0
 865 *         is free and can now be re-used.
 866 */
 867static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 868				  struct packet_sock *po)
 869{
 870	pkc->reset_pending_on_curr_blk = 1;
 871	po->stats.stats3.tp_freeze_q_cnt++;
 872}
 873
 874#define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 875
 876/*
 877 * If the next block is free then we will dispatch it
 878 * and return a good offset.
 879 * Else, we will freeze the queue.
 880 * So, caller must check the return value.
 881 */
 882static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 883		struct packet_sock *po)
 884{
 885	struct tpacket_block_desc *pbd;
 886
 887	smp_rmb();
 888
 889	/* 1. Get current block num */
 890	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 891
 892	/* 2. If this block is currently in_use then freeze the queue */
 893	if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 894		prb_freeze_queue(pkc, po);
 895		return NULL;
 896	}
 897
 898	/*
 899	 * 3.
 900	 * open this block and return the offset where the first packet
 901	 * needs to get stored.
 902	 */
 903	prb_open_block(pkc, pbd);
 904	return (void *)pkc->nxt_offset;
 905}
 906
 907static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 908		struct packet_sock *po, unsigned int status)
 909{
 910	struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 911
 912	/* retire/close the current block */
 913	if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 914		/*
 915		 * Plug the case where copy_bits() is in progress on
 916		 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 917		 * have space to copy the pkt in the current block and
 918		 * called prb_retire_current_block()
 919		 *
 920		 * We don't need to worry about the TMO case because
 921		 * the timer-handler already handled this case.
 922		 */
 923		if (!(status & TP_STATUS_BLK_TMO)) {
 924			/* Waiting for skb_copy_bits to finish... */
 925			write_lock(&pkc->blk_fill_in_prog_lock);
 926			write_unlock(&pkc->blk_fill_in_prog_lock);
 927		}
 928		prb_close_block(pkc, pbd, po, status);
 929		return;
 930	}
 931}
 932
 933static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
 934{
 935	return TP_STATUS_USER & BLOCK_STATUS(pbd);
 936}
 937
 938static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 939{
 940	return pkc->reset_pending_on_curr_blk;
 941}
 942
 943static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 944	__releases(&pkc->blk_fill_in_prog_lock)
 945{
 946	struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 947
 948	read_unlock(&pkc->blk_fill_in_prog_lock);
 949}
 950
 951static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 952			struct tpacket3_hdr *ppd)
 953{
 954	ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
 955}
 956
 957static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 958			struct tpacket3_hdr *ppd)
 959{
 960	ppd->hv1.tp_rxhash = 0;
 961}
 962
 963static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
 964			struct tpacket3_hdr *ppd)
 965{
 966	if (skb_vlan_tag_present(pkc->skb)) {
 967		ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
 968		ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
 969		ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
 970	} else {
 971		ppd->hv1.tp_vlan_tci = 0;
 972		ppd->hv1.tp_vlan_tpid = 0;
 973		ppd->tp_status = TP_STATUS_AVAILABLE;
 974	}
 975}
 976
 977static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
 978			struct tpacket3_hdr *ppd)
 979{
 980	ppd->hv1.tp_padding = 0;
 981	prb_fill_vlan_info(pkc, ppd);
 982
 983	if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
 984		prb_fill_rxhash(pkc, ppd);
 985	else
 986		prb_clear_rxhash(pkc, ppd);
 987}
 988
 989static void prb_fill_curr_block(char *curr,
 990				struct tpacket_kbdq_core *pkc,
 991				struct tpacket_block_desc *pbd,
 992				unsigned int len)
 993	__acquires(&pkc->blk_fill_in_prog_lock)
 994{
 995	struct tpacket3_hdr *ppd;
 996
 997	ppd  = (struct tpacket3_hdr *)curr;
 998	ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
 999	pkc->prev = curr;
1000	pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1001	BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1002	BLOCK_NUM_PKTS(pbd) += 1;
1003	read_lock(&pkc->blk_fill_in_prog_lock);
1004	prb_run_all_ft_ops(pkc, ppd);
1005}
1006
1007/* Assumes caller has the sk->rx_queue.lock */
1008static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1009					    struct sk_buff *skb,
1010					    unsigned int len
1011					    )
1012{
1013	struct tpacket_kbdq_core *pkc;
1014	struct tpacket_block_desc *pbd;
1015	char *curr, *end;
1016
1017	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1018	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1019
1020	/* Queue is frozen when user space is lagging behind */
1021	if (prb_queue_frozen(pkc)) {
1022		/*
1023		 * Check if that last block which caused the queue to freeze,
1024		 * is still in_use by user-space.
1025		 */
1026		if (prb_curr_blk_in_use(pbd)) {
1027			/* Can't record this packet */
1028			return NULL;
1029		} else {
1030			/*
1031			 * Ok, the block was released by user-space.
1032			 * Now let's open that block.
1033			 * opening a block also thaws the queue.
1034			 * Thawing is a side effect.
1035			 */
1036			prb_open_block(pkc, pbd);
1037		}
1038	}
1039
1040	smp_mb();
1041	curr = pkc->nxt_offset;
1042	pkc->skb = skb;
1043	end = (char *)pbd + pkc->kblk_size;
1044
1045	/* first try the current block */
1046	if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1047		prb_fill_curr_block(curr, pkc, pbd, len);
1048		return (void *)curr;
1049	}
1050
1051	/* Ok, close the current block */
1052	prb_retire_current_block(pkc, po, 0);
1053
1054	/* Now, try to dispatch the next block */
1055	curr = (char *)prb_dispatch_next_block(pkc, po);
1056	if (curr) {
1057		pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1058		prb_fill_curr_block(curr, pkc, pbd, len);
1059		return (void *)curr;
1060	}
1061
1062	/*
1063	 * No free blocks are available.user_space hasn't caught up yet.
1064	 * Queue was just frozen and now this packet will get dropped.
1065	 */
1066	return NULL;
1067}
1068
1069static void *packet_current_rx_frame(struct packet_sock *po,
1070					    struct sk_buff *skb,
1071					    int status, unsigned int len)
1072{
1073	char *curr = NULL;
1074	switch (po->tp_version) {
1075	case TPACKET_V1:
1076	case TPACKET_V2:
1077		curr = packet_lookup_frame(po, &po->rx_ring,
1078					po->rx_ring.head, status);
1079		return curr;
1080	case TPACKET_V3:
1081		return __packet_lookup_frame_in_block(po, skb, len);
1082	default:
1083		WARN(1, "TPACKET version not supported\n");
1084		BUG();
1085		return NULL;
1086	}
1087}
1088
1089static void *prb_lookup_block(const struct packet_sock *po,
1090			      const struct packet_ring_buffer *rb,
1091			      unsigned int idx,
1092			      int status)
1093{
1094	struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1095	struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1096
1097	if (status != BLOCK_STATUS(pbd))
1098		return NULL;
1099	return pbd;
1100}
1101
1102static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1103{
1104	unsigned int prev;
1105	if (rb->prb_bdqc.kactive_blk_num)
1106		prev = rb->prb_bdqc.kactive_blk_num-1;
1107	else
1108		prev = rb->prb_bdqc.knum_blocks-1;
1109	return prev;
1110}
1111
1112/* Assumes caller has held the rx_queue.lock */
1113static void *__prb_previous_block(struct packet_sock *po,
1114					 struct packet_ring_buffer *rb,
1115					 int status)
1116{
1117	unsigned int previous = prb_previous_blk_num(rb);
1118	return prb_lookup_block(po, rb, previous, status);
1119}
1120
1121static void *packet_previous_rx_frame(struct packet_sock *po,
1122					     struct packet_ring_buffer *rb,
1123					     int status)
1124{
1125	if (po->tp_version <= TPACKET_V2)
1126		return packet_previous_frame(po, rb, status);
1127
1128	return __prb_previous_block(po, rb, status);
1129}
1130
1131static void packet_increment_rx_head(struct packet_sock *po,
1132					    struct packet_ring_buffer *rb)
1133{
1134	switch (po->tp_version) {
1135	case TPACKET_V1:
1136	case TPACKET_V2:
1137		return packet_increment_head(rb);
1138	case TPACKET_V3:
1139	default:
1140		WARN(1, "TPACKET version not supported.\n");
1141		BUG();
1142		return;
1143	}
1144}
1145
1146static void *packet_previous_frame(struct packet_sock *po,
1147		struct packet_ring_buffer *rb,
1148		int status)
1149{
1150	unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1151	return packet_lookup_frame(po, rb, previous, status);
1152}
1153
1154static void packet_increment_head(struct packet_ring_buffer *buff)
1155{
1156	buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1157}
1158
1159static void packet_inc_pending(struct packet_ring_buffer *rb)
1160{
1161	this_cpu_inc(*rb->pending_refcnt);
1162}
1163
1164static void packet_dec_pending(struct packet_ring_buffer *rb)
1165{
1166	this_cpu_dec(*rb->pending_refcnt);
1167}
1168
1169static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1170{
1171	unsigned int refcnt = 0;
1172	int cpu;
1173
1174	/* We don't use pending refcount in rx_ring. */
1175	if (rb->pending_refcnt == NULL)
1176		return 0;
1177
1178	for_each_possible_cpu(cpu)
1179		refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1180
1181	return refcnt;
1182}
1183
1184static int packet_alloc_pending(struct packet_sock *po)
1185{
1186	po->rx_ring.pending_refcnt = NULL;
1187
1188	po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1189	if (unlikely(po->tx_ring.pending_refcnt == NULL))
1190		return -ENOBUFS;
1191
1192	return 0;
1193}
1194
1195static void packet_free_pending(struct packet_sock *po)
1196{
1197	free_percpu(po->tx_ring.pending_refcnt);
1198}
1199
1200#define ROOM_POW_OFF	2
1201#define ROOM_NONE	0x0
1202#define ROOM_LOW	0x1
1203#define ROOM_NORMAL	0x2
1204
1205static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1206{
1207	int idx, len;
1208
1209	len = READ_ONCE(po->rx_ring.frame_max) + 1;
1210	idx = READ_ONCE(po->rx_ring.head);
1211	if (pow_off)
1212		idx += len >> pow_off;
1213	if (idx >= len)
1214		idx -= len;
1215	return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1216}
1217
1218static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1219{
1220	int idx, len;
1221
1222	len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1223	idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1224	if (pow_off)
1225		idx += len >> pow_off;
1226	if (idx >= len)
1227		idx -= len;
1228	return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1229}
1230
1231static int __packet_rcv_has_room(const struct packet_sock *po,
1232				 const struct sk_buff *skb)
1233{
1234	const struct sock *sk = &po->sk;
1235	int ret = ROOM_NONE;
1236
1237	if (po->prot_hook.func != tpacket_rcv) {
1238		int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1239		int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1240				   - (skb ? skb->truesize : 0);
1241
1242		if (avail > (rcvbuf >> ROOM_POW_OFF))
1243			return ROOM_NORMAL;
1244		else if (avail > 0)
1245			return ROOM_LOW;
1246		else
1247			return ROOM_NONE;
1248	}
1249
1250	if (po->tp_version == TPACKET_V3) {
1251		if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1252			ret = ROOM_NORMAL;
1253		else if (__tpacket_v3_has_room(po, 0))
1254			ret = ROOM_LOW;
1255	} else {
1256		if (__tpacket_has_room(po, ROOM_POW_OFF))
1257			ret = ROOM_NORMAL;
1258		else if (__tpacket_has_room(po, 0))
1259			ret = ROOM_LOW;
1260	}
1261
1262	return ret;
1263}
1264
1265static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1266{
1267	int pressure, ret;
 
1268
1269	ret = __packet_rcv_has_room(po, skb);
1270	pressure = ret != ROOM_NORMAL;
1271
1272	if (READ_ONCE(po->pressure) != pressure)
1273		WRITE_ONCE(po->pressure, pressure);
1274
1275	return ret;
1276}
1277
1278static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1279{
1280	if (READ_ONCE(po->pressure) &&
1281	    __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1282		WRITE_ONCE(po->pressure,  0);
1283}
1284
1285static void packet_sock_destruct(struct sock *sk)
1286{
1287	skb_queue_purge(&sk->sk_error_queue);
1288
1289	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1290	WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1291
1292	if (!sock_flag(sk, SOCK_DEAD)) {
1293		pr_err("Attempt to release alive packet socket: %p\n", sk);
1294		return;
1295	}
1296
1297	sk_refcnt_debug_dec(sk);
1298}
1299
1300static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1301{
1302	u32 *history = po->rollover->history;
1303	u32 victim, rxhash;
1304	int i, count = 0;
1305
1306	rxhash = skb_get_hash(skb);
1307	for (i = 0; i < ROLLOVER_HLEN; i++)
1308		if (READ_ONCE(history[i]) == rxhash)
1309			count++;
1310
1311	victim = prandom_u32() % ROLLOVER_HLEN;
1312
1313	/* Avoid dirtying the cache line if possible */
1314	if (READ_ONCE(history[victim]) != rxhash)
1315		WRITE_ONCE(history[victim], rxhash);
1316
1317	return count > (ROLLOVER_HLEN >> 1);
1318}
1319
1320static unsigned int fanout_demux_hash(struct packet_fanout *f,
1321				      struct sk_buff *skb,
1322				      unsigned int num)
1323{
1324	return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1325}
1326
1327static unsigned int fanout_demux_lb(struct packet_fanout *f,
1328				    struct sk_buff *skb,
1329				    unsigned int num)
1330{
1331	unsigned int val = atomic_inc_return(&f->rr_cur);
1332
1333	return val % num;
1334}
1335
1336static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1337				     struct sk_buff *skb,
1338				     unsigned int num)
1339{
1340	return smp_processor_id() % num;
1341}
1342
1343static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1344				     struct sk_buff *skb,
1345				     unsigned int num)
1346{
1347	return prandom_u32_max(num);
1348}
1349
1350static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1351					  struct sk_buff *skb,
1352					  unsigned int idx, bool try_self,
1353					  unsigned int num)
1354{
1355	struct packet_sock *po, *po_next, *po_skip = NULL;
1356	unsigned int i, j, room = ROOM_NONE;
1357
1358	po = pkt_sk(f->arr[idx]);
1359
1360	if (try_self) {
1361		room = packet_rcv_has_room(po, skb);
1362		if (room == ROOM_NORMAL ||
1363		    (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1364			return idx;
1365		po_skip = po;
1366	}
1367
1368	i = j = min_t(int, po->rollover->sock, num - 1);
1369	do {
1370		po_next = pkt_sk(f->arr[i]);
1371		if (po_next != po_skip && !READ_ONCE(po_next->pressure) &&
 
1372		    packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1373			if (i != j)
1374				po->rollover->sock = i;
1375			atomic_long_inc(&po->rollover->num);
1376			if (room == ROOM_LOW)
1377				atomic_long_inc(&po->rollover->num_huge);
1378			return i;
1379		}
1380
1381		if (++i == num)
1382			i = 0;
1383	} while (i != j);
1384
1385	atomic_long_inc(&po->rollover->num_failed);
1386	return idx;
1387}
1388
1389static unsigned int fanout_demux_qm(struct packet_fanout *f,
1390				    struct sk_buff *skb,
1391				    unsigned int num)
1392{
1393	return skb_get_queue_mapping(skb) % num;
1394}
1395
1396static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1397				     struct sk_buff *skb,
1398				     unsigned int num)
1399{
1400	struct bpf_prog *prog;
1401	unsigned int ret = 0;
1402
1403	rcu_read_lock();
1404	prog = rcu_dereference(f->bpf_prog);
1405	if (prog)
1406		ret = bpf_prog_run_clear_cb(prog, skb) % num;
1407	rcu_read_unlock();
1408
1409	return ret;
1410}
1411
1412static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1413{
1414	return f->flags & (flag >> 8);
1415}
1416
1417static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1418			     struct packet_type *pt, struct net_device *orig_dev)
1419{
1420	struct packet_fanout *f = pt->af_packet_priv;
1421	unsigned int num = READ_ONCE(f->num_members);
1422	struct net *net = read_pnet(&f->net);
1423	struct packet_sock *po;
1424	unsigned int idx;
1425
1426	if (!net_eq(dev_net(dev), net) || !num) {
1427		kfree_skb(skb);
1428		return 0;
1429	}
1430
1431	if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1432		skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1433		if (!skb)
1434			return 0;
1435	}
1436	switch (f->type) {
1437	case PACKET_FANOUT_HASH:
1438	default:
1439		idx = fanout_demux_hash(f, skb, num);
1440		break;
1441	case PACKET_FANOUT_LB:
1442		idx = fanout_demux_lb(f, skb, num);
1443		break;
1444	case PACKET_FANOUT_CPU:
1445		idx = fanout_demux_cpu(f, skb, num);
1446		break;
1447	case PACKET_FANOUT_RND:
1448		idx = fanout_demux_rnd(f, skb, num);
1449		break;
1450	case PACKET_FANOUT_QM:
1451		idx = fanout_demux_qm(f, skb, num);
1452		break;
1453	case PACKET_FANOUT_ROLLOVER:
1454		idx = fanout_demux_rollover(f, skb, 0, false, num);
1455		break;
1456	case PACKET_FANOUT_CBPF:
1457	case PACKET_FANOUT_EBPF:
1458		idx = fanout_demux_bpf(f, skb, num);
1459		break;
1460	}
1461
1462	if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1463		idx = fanout_demux_rollover(f, skb, idx, true, num);
1464
1465	po = pkt_sk(f->arr[idx]);
1466	return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1467}
1468
1469DEFINE_MUTEX(fanout_mutex);
1470EXPORT_SYMBOL_GPL(fanout_mutex);
1471static LIST_HEAD(fanout_list);
1472static u16 fanout_next_id;
1473
1474static void __fanout_link(struct sock *sk, struct packet_sock *po)
1475{
1476	struct packet_fanout *f = po->fanout;
1477
1478	spin_lock(&f->lock);
1479	f->arr[f->num_members] = sk;
1480	smp_wmb();
1481	f->num_members++;
1482	if (f->num_members == 1)
1483		dev_add_pack(&f->prot_hook);
1484	spin_unlock(&f->lock);
1485}
1486
1487static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1488{
1489	struct packet_fanout *f = po->fanout;
1490	int i;
1491
1492	spin_lock(&f->lock);
1493	for (i = 0; i < f->num_members; i++) {
1494		if (f->arr[i] == sk)
 
1495			break;
1496	}
1497	BUG_ON(i >= f->num_members);
1498	f->arr[i] = f->arr[f->num_members - 1];
 
 
1499	f->num_members--;
1500	if (f->num_members == 0)
1501		__dev_remove_pack(&f->prot_hook);
1502	spin_unlock(&f->lock);
1503}
1504
1505static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1506{
1507	if (sk->sk_family != PF_PACKET)
1508		return false;
1509
1510	return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1511}
1512
1513static void fanout_init_data(struct packet_fanout *f)
1514{
1515	switch (f->type) {
1516	case PACKET_FANOUT_LB:
1517		atomic_set(&f->rr_cur, 0);
1518		break;
1519	case PACKET_FANOUT_CBPF:
1520	case PACKET_FANOUT_EBPF:
1521		RCU_INIT_POINTER(f->bpf_prog, NULL);
1522		break;
1523	}
1524}
1525
1526static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1527{
1528	struct bpf_prog *old;
1529
1530	spin_lock(&f->lock);
1531	old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1532	rcu_assign_pointer(f->bpf_prog, new);
1533	spin_unlock(&f->lock);
1534
1535	if (old) {
1536		synchronize_net();
1537		bpf_prog_destroy(old);
1538	}
1539}
1540
1541static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data,
1542				unsigned int len)
1543{
1544	struct bpf_prog *new;
1545	struct sock_fprog fprog;
1546	int ret;
1547
1548	if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1549		return -EPERM;
1550
1551	ret = copy_bpf_fprog_from_user(&fprog, data, len);
1552	if (ret)
1553		return ret;
1554
1555	ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1556	if (ret)
1557		return ret;
1558
1559	__fanout_set_data_bpf(po->fanout, new);
1560	return 0;
1561}
1562
1563static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data,
1564				unsigned int len)
1565{
1566	struct bpf_prog *new;
1567	u32 fd;
1568
1569	if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1570		return -EPERM;
1571	if (len != sizeof(fd))
1572		return -EINVAL;
1573	if (copy_from_sockptr(&fd, data, len))
1574		return -EFAULT;
1575
1576	new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1577	if (IS_ERR(new))
1578		return PTR_ERR(new);
1579
1580	__fanout_set_data_bpf(po->fanout, new);
1581	return 0;
1582}
1583
1584static int fanout_set_data(struct packet_sock *po, sockptr_t data,
1585			   unsigned int len)
1586{
1587	switch (po->fanout->type) {
1588	case PACKET_FANOUT_CBPF:
1589		return fanout_set_data_cbpf(po, data, len);
1590	case PACKET_FANOUT_EBPF:
1591		return fanout_set_data_ebpf(po, data, len);
1592	default:
1593		return -EINVAL;
1594	}
1595}
1596
1597static void fanout_release_data(struct packet_fanout *f)
1598{
1599	switch (f->type) {
1600	case PACKET_FANOUT_CBPF:
1601	case PACKET_FANOUT_EBPF:
1602		__fanout_set_data_bpf(f, NULL);
1603	}
1604}
1605
1606static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1607{
1608	struct packet_fanout *f;
1609
1610	list_for_each_entry(f, &fanout_list, list) {
1611		if (f->id == candidate_id &&
1612		    read_pnet(&f->net) == sock_net(sk)) {
1613			return false;
1614		}
1615	}
1616	return true;
1617}
1618
1619static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1620{
1621	u16 id = fanout_next_id;
1622
1623	do {
1624		if (__fanout_id_is_free(sk, id)) {
1625			*new_id = id;
1626			fanout_next_id = id + 1;
1627			return true;
1628		}
1629
1630		id++;
1631	} while (id != fanout_next_id);
1632
1633	return false;
1634}
1635
1636static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1637{
1638	struct packet_rollover *rollover = NULL;
1639	struct packet_sock *po = pkt_sk(sk);
 
1640	struct packet_fanout *f, *match;
1641	u8 type = type_flags & 0xff;
1642	u8 flags = type_flags >> 8;
 
1643	int err;
1644
1645	switch (type) {
1646	case PACKET_FANOUT_ROLLOVER:
1647		if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1648			return -EINVAL;
 
1649	case PACKET_FANOUT_HASH:
1650	case PACKET_FANOUT_LB:
1651	case PACKET_FANOUT_CPU:
1652	case PACKET_FANOUT_RND:
1653	case PACKET_FANOUT_QM:
1654	case PACKET_FANOUT_CBPF:
1655	case PACKET_FANOUT_EBPF:
1656		break;
1657	default:
1658		return -EINVAL;
1659	}
1660
1661	mutex_lock(&fanout_mutex);
1662
1663	err = -EALREADY;
1664	if (po->fanout)
1665		goto out;
1666
1667	if (type == PACKET_FANOUT_ROLLOVER ||
1668	    (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1669		err = -ENOMEM;
1670		rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1671		if (!rollover)
1672			goto out;
1673		atomic_long_set(&rollover->num, 0);
1674		atomic_long_set(&rollover->num_huge, 0);
1675		atomic_long_set(&rollover->num_failed, 0);
1676	}
1677
1678	if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1679		if (id != 0) {
1680			err = -EINVAL;
1681			goto out;
1682		}
1683		if (!fanout_find_new_id(sk, &id)) {
1684			err = -ENOMEM;
1685			goto out;
1686		}
1687		/* ephemeral flag for the first socket in the group: drop it */
1688		flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1689	}
1690
1691	match = NULL;
1692	list_for_each_entry(f, &fanout_list, list) {
1693		if (f->id == id &&
1694		    read_pnet(&f->net) == sock_net(sk)) {
1695			match = f;
1696			break;
1697		}
1698	}
1699	err = -EINVAL;
1700	if (match && match->flags != flags)
1701		goto out;
1702	if (!match) {
 
 
 
 
 
 
 
 
 
1703		err = -ENOMEM;
1704		match = kzalloc(sizeof(*match), GFP_KERNEL);
 
1705		if (!match)
1706			goto out;
1707		write_pnet(&match->net, sock_net(sk));
1708		match->id = id;
1709		match->type = type;
1710		match->flags = flags;
1711		INIT_LIST_HEAD(&match->list);
1712		spin_lock_init(&match->lock);
1713		refcount_set(&match->sk_ref, 0);
1714		fanout_init_data(match);
1715		match->prot_hook.type = po->prot_hook.type;
1716		match->prot_hook.dev = po->prot_hook.dev;
1717		match->prot_hook.func = packet_rcv_fanout;
1718		match->prot_hook.af_packet_priv = match;
 
1719		match->prot_hook.id_match = match_fanout_group;
 
 
1720		list_add(&match->list, &fanout_list);
1721	}
1722	err = -EINVAL;
1723
1724	spin_lock(&po->bind_lock);
1725	if (po->running &&
1726	    match->type == type &&
1727	    match->prot_hook.type == po->prot_hook.type &&
1728	    match->prot_hook.dev == po->prot_hook.dev) {
1729		err = -ENOSPC;
1730		if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1731			__dev_remove_pack(&po->prot_hook);
1732			po->fanout = match;
 
 
 
1733			po->rollover = rollover;
1734			rollover = NULL;
1735			refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1736			__fanout_link(sk, po);
1737			err = 0;
1738		}
1739	}
1740	spin_unlock(&po->bind_lock);
1741
1742	if (err && !refcount_read(&match->sk_ref)) {
1743		list_del(&match->list);
1744		kfree(match);
1745	}
1746
1747out:
1748	kfree(rollover);
1749	mutex_unlock(&fanout_mutex);
1750	return err;
1751}
1752
1753/* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1754 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1755 * It is the responsibility of the caller to call fanout_release_data() and
1756 * free the returned packet_fanout (after synchronize_net())
1757 */
1758static struct packet_fanout *fanout_release(struct sock *sk)
1759{
1760	struct packet_sock *po = pkt_sk(sk);
1761	struct packet_fanout *f;
1762
1763	mutex_lock(&fanout_mutex);
1764	f = po->fanout;
1765	if (f) {
1766		po->fanout = NULL;
1767
1768		if (refcount_dec_and_test(&f->sk_ref))
1769			list_del(&f->list);
1770		else
1771			f = NULL;
1772	}
1773	mutex_unlock(&fanout_mutex);
1774
1775	return f;
1776}
1777
1778static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1779					  struct sk_buff *skb)
1780{
1781	/* Earlier code assumed this would be a VLAN pkt, double-check
1782	 * this now that we have the actual packet in hand. We can only
1783	 * do this check on Ethernet devices.
1784	 */
1785	if (unlikely(dev->type != ARPHRD_ETHER))
1786		return false;
1787
1788	skb_reset_mac_header(skb);
1789	return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1790}
1791
1792static const struct proto_ops packet_ops;
1793
1794static const struct proto_ops packet_ops_spkt;
1795
1796static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1797			   struct packet_type *pt, struct net_device *orig_dev)
1798{
1799	struct sock *sk;
1800	struct sockaddr_pkt *spkt;
1801
1802	/*
1803	 *	When we registered the protocol we saved the socket in the data
1804	 *	field for just this event.
1805	 */
1806
1807	sk = pt->af_packet_priv;
1808
1809	/*
1810	 *	Yank back the headers [hope the device set this
1811	 *	right or kerboom...]
1812	 *
1813	 *	Incoming packets have ll header pulled,
1814	 *	push it back.
1815	 *
1816	 *	For outgoing ones skb->data == skb_mac_header(skb)
1817	 *	so that this procedure is noop.
1818	 */
1819
1820	if (skb->pkt_type == PACKET_LOOPBACK)
1821		goto out;
1822
1823	if (!net_eq(dev_net(dev), sock_net(sk)))
1824		goto out;
1825
1826	skb = skb_share_check(skb, GFP_ATOMIC);
1827	if (skb == NULL)
1828		goto oom;
1829
1830	/* drop any routing info */
1831	skb_dst_drop(skb);
1832
1833	/* drop conntrack reference */
1834	nf_reset_ct(skb);
1835
1836	spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1837
1838	skb_push(skb, skb->data - skb_mac_header(skb));
1839
1840	/*
1841	 *	The SOCK_PACKET socket receives _all_ frames.
1842	 */
1843
1844	spkt->spkt_family = dev->type;
1845	strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1846	spkt->spkt_protocol = skb->protocol;
1847
1848	/*
1849	 *	Charge the memory to the socket. This is done specifically
1850	 *	to prevent sockets using all the memory up.
1851	 */
1852
1853	if (sock_queue_rcv_skb(sk, skb) == 0)
1854		return 0;
1855
1856out:
1857	kfree_skb(skb);
1858oom:
1859	return 0;
1860}
1861
1862static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1863{
 
 
1864	if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1865	    sock->type == SOCK_RAW) {
1866		skb_reset_mac_header(skb);
1867		skb->protocol = dev_parse_header_protocol(skb);
1868	}
1869
 
 
 
 
 
 
1870	skb_probe_transport_header(skb);
1871}
1872
1873/*
1874 *	Output a raw packet to a device layer. This bypasses all the other
1875 *	protocol layers and you must therefore supply it with a complete frame
1876 */
1877
1878static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1879			       size_t len)
1880{
1881	struct sock *sk = sock->sk;
1882	DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1883	struct sk_buff *skb = NULL;
1884	struct net_device *dev;
1885	struct sockcm_cookie sockc;
1886	__be16 proto = 0;
1887	int err;
1888	int extra_len = 0;
1889
1890	/*
1891	 *	Get and verify the address.
1892	 */
1893
1894	if (saddr) {
1895		if (msg->msg_namelen < sizeof(struct sockaddr))
1896			return -EINVAL;
1897		if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1898			proto = saddr->spkt_protocol;
1899	} else
1900		return -ENOTCONN;	/* SOCK_PACKET must be sent giving an address */
1901
1902	/*
1903	 *	Find the device first to size check it
1904	 */
1905
1906	saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1907retry:
1908	rcu_read_lock();
1909	dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1910	err = -ENODEV;
1911	if (dev == NULL)
1912		goto out_unlock;
1913
1914	err = -ENETDOWN;
1915	if (!(dev->flags & IFF_UP))
1916		goto out_unlock;
1917
1918	/*
1919	 * You may not queue a frame bigger than the mtu. This is the lowest level
1920	 * raw protocol and you must do your own fragmentation at this level.
1921	 */
1922
1923	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1924		if (!netif_supports_nofcs(dev)) {
1925			err = -EPROTONOSUPPORT;
1926			goto out_unlock;
1927		}
1928		extra_len = 4; /* We're doing our own CRC */
1929	}
1930
1931	err = -EMSGSIZE;
1932	if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1933		goto out_unlock;
1934
1935	if (!skb) {
1936		size_t reserved = LL_RESERVED_SPACE(dev);
1937		int tlen = dev->needed_tailroom;
1938		unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1939
1940		rcu_read_unlock();
1941		skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1942		if (skb == NULL)
1943			return -ENOBUFS;
1944		/* FIXME: Save some space for broken drivers that write a hard
1945		 * header at transmission time by themselves. PPP is the notable
1946		 * one here. This should really be fixed at the driver level.
1947		 */
1948		skb_reserve(skb, reserved);
1949		skb_reset_network_header(skb);
1950
1951		/* Try to align data part correctly */
1952		if (hhlen) {
1953			skb->data -= hhlen;
1954			skb->tail -= hhlen;
1955			if (len < hhlen)
1956				skb_reset_network_header(skb);
1957		}
1958		err = memcpy_from_msg(skb_put(skb, len), msg, len);
1959		if (err)
1960			goto out_free;
1961		goto retry;
1962	}
1963
1964	if (!dev_validate_header(dev, skb->data, len)) {
1965		err = -EINVAL;
1966		goto out_unlock;
1967	}
1968	if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1969	    !packet_extra_vlan_len_allowed(dev, skb)) {
1970		err = -EMSGSIZE;
1971		goto out_unlock;
1972	}
1973
1974	sockcm_init(&sockc, sk);
1975	if (msg->msg_controllen) {
1976		err = sock_cmsg_send(sk, msg, &sockc);
1977		if (unlikely(err))
1978			goto out_unlock;
1979	}
1980
1981	skb->protocol = proto;
1982	skb->dev = dev;
1983	skb->priority = sk->sk_priority;
1984	skb->mark = sk->sk_mark;
1985	skb->tstamp = sockc.transmit_time;
1986
1987	skb_setup_tx_timestamp(skb, sockc.tsflags);
1988
1989	if (unlikely(extra_len == 4))
1990		skb->no_fcs = 1;
1991
1992	packet_parse_headers(skb, sock);
1993
1994	dev_queue_xmit(skb);
1995	rcu_read_unlock();
1996	return len;
1997
1998out_unlock:
1999	rcu_read_unlock();
2000out_free:
2001	kfree_skb(skb);
2002	return err;
2003}
2004
2005static unsigned int run_filter(struct sk_buff *skb,
2006			       const struct sock *sk,
2007			       unsigned int res)
2008{
2009	struct sk_filter *filter;
2010
2011	rcu_read_lock();
2012	filter = rcu_dereference(sk->sk_filter);
2013	if (filter != NULL)
2014		res = bpf_prog_run_clear_cb(filter->prog, skb);
2015	rcu_read_unlock();
2016
2017	return res;
2018}
2019
2020static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2021			   size_t *len)
2022{
2023	struct virtio_net_hdr vnet_hdr;
2024
2025	if (*len < sizeof(vnet_hdr))
2026		return -EINVAL;
2027	*len -= sizeof(vnet_hdr);
2028
2029	if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2030		return -EINVAL;
2031
2032	return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2033}
2034
2035/*
2036 * This function makes lazy skb cloning in hope that most of packets
2037 * are discarded by BPF.
2038 *
2039 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2040 * and skb->cb are mangled. It works because (and until) packets
2041 * falling here are owned by current CPU. Output packets are cloned
2042 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2043 * sequencially, so that if we return skb to original state on exit,
2044 * we will not harm anyone.
2045 */
2046
2047static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2048		      struct packet_type *pt, struct net_device *orig_dev)
2049{
 
2050	struct sock *sk;
2051	struct sockaddr_ll *sll;
2052	struct packet_sock *po;
2053	u8 *skb_head = skb->data;
2054	int skb_len = skb->len;
2055	unsigned int snaplen, res;
2056	bool is_drop_n_account = false;
2057
2058	if (skb->pkt_type == PACKET_LOOPBACK)
2059		goto drop;
2060
2061	sk = pt->af_packet_priv;
2062	po = pkt_sk(sk);
2063
2064	if (!net_eq(dev_net(dev), sock_net(sk)))
2065		goto drop;
2066
2067	skb->dev = dev;
2068
2069	if (dev->header_ops) {
2070		/* The device has an explicit notion of ll header,
2071		 * exported to higher levels.
2072		 *
2073		 * Otherwise, the device hides details of its frame
2074		 * structure, so that corresponding packet head is
2075		 * never delivered to user.
2076		 */
2077		if (sk->sk_type != SOCK_DGRAM)
2078			skb_push(skb, skb->data - skb_mac_header(skb));
2079		else if (skb->pkt_type == PACKET_OUTGOING) {
2080			/* Special case: outgoing packets have ll header at head */
2081			skb_pull(skb, skb_network_offset(skb));
2082		}
2083	}
2084
2085	snaplen = skb->len;
2086
2087	res = run_filter(skb, sk, snaplen);
2088	if (!res)
2089		goto drop_n_restore;
2090	if (snaplen > res)
2091		snaplen = res;
2092
2093	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2094		goto drop_n_acct;
2095
2096	if (skb_shared(skb)) {
2097		struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2098		if (nskb == NULL)
2099			goto drop_n_acct;
2100
2101		if (skb_head != skb->data) {
2102			skb->data = skb_head;
2103			skb->len = skb_len;
2104		}
2105		consume_skb(skb);
2106		skb = nskb;
2107	}
2108
2109	sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2110
2111	sll = &PACKET_SKB_CB(skb)->sa.ll;
2112	sll->sll_hatype = dev->type;
2113	sll->sll_pkttype = skb->pkt_type;
2114	if (unlikely(po->origdev))
2115		sll->sll_ifindex = orig_dev->ifindex;
2116	else
2117		sll->sll_ifindex = dev->ifindex;
2118
2119	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2120
2121	/* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2122	 * Use their space for storing the original skb length.
2123	 */
2124	PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2125
2126	if (pskb_trim(skb, snaplen))
2127		goto drop_n_acct;
2128
2129	skb_set_owner_r(skb, sk);
2130	skb->dev = NULL;
2131	skb_dst_drop(skb);
2132
2133	/* drop conntrack reference */
2134	nf_reset_ct(skb);
2135
2136	spin_lock(&sk->sk_receive_queue.lock);
2137	po->stats.stats1.tp_packets++;
2138	sock_skb_set_dropcount(sk, skb);
 
2139	__skb_queue_tail(&sk->sk_receive_queue, skb);
2140	spin_unlock(&sk->sk_receive_queue.lock);
2141	sk->sk_data_ready(sk);
2142	return 0;
2143
2144drop_n_acct:
2145	is_drop_n_account = true;
2146	atomic_inc(&po->tp_drops);
2147	atomic_inc(&sk->sk_drops);
 
2148
2149drop_n_restore:
2150	if (skb_head != skb->data && skb_shared(skb)) {
2151		skb->data = skb_head;
2152		skb->len = skb_len;
2153	}
2154drop:
2155	if (!is_drop_n_account)
2156		consume_skb(skb);
2157	else
2158		kfree_skb(skb);
2159	return 0;
2160}
2161
2162static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2163		       struct packet_type *pt, struct net_device *orig_dev)
2164{
 
2165	struct sock *sk;
2166	struct packet_sock *po;
2167	struct sockaddr_ll *sll;
2168	union tpacket_uhdr h;
2169	u8 *skb_head = skb->data;
2170	int skb_len = skb->len;
2171	unsigned int snaplen, res;
2172	unsigned long status = TP_STATUS_USER;
2173	unsigned short macoff, hdrlen;
2174	unsigned int netoff;
2175	struct sk_buff *copy_skb = NULL;
2176	struct timespec64 ts;
2177	__u32 ts_status;
2178	bool is_drop_n_account = false;
2179	unsigned int slot_id = 0;
2180	bool do_vnet = false;
2181
2182	/* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2183	 * We may add members to them until current aligned size without forcing
2184	 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2185	 */
2186	BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2187	BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2188
2189	if (skb->pkt_type == PACKET_LOOPBACK)
2190		goto drop;
2191
2192	sk = pt->af_packet_priv;
2193	po = pkt_sk(sk);
2194
2195	if (!net_eq(dev_net(dev), sock_net(sk)))
2196		goto drop;
2197
2198	if (dev->header_ops) {
2199		if (sk->sk_type != SOCK_DGRAM)
2200			skb_push(skb, skb->data - skb_mac_header(skb));
2201		else if (skb->pkt_type == PACKET_OUTGOING) {
2202			/* Special case: outgoing packets have ll header at head */
2203			skb_pull(skb, skb_network_offset(skb));
2204		}
2205	}
2206
2207	snaplen = skb->len;
2208
2209	res = run_filter(skb, sk, snaplen);
2210	if (!res)
2211		goto drop_n_restore;
2212
2213	/* If we are flooded, just give up */
2214	if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2215		atomic_inc(&po->tp_drops);
2216		goto drop_n_restore;
2217	}
2218
2219	if (skb->ip_summed == CHECKSUM_PARTIAL)
2220		status |= TP_STATUS_CSUMNOTREADY;
2221	else if (skb->pkt_type != PACKET_OUTGOING &&
2222		 (skb->ip_summed == CHECKSUM_COMPLETE ||
2223		  skb_csum_unnecessary(skb)))
2224		status |= TP_STATUS_CSUM_VALID;
 
 
2225
2226	if (snaplen > res)
2227		snaplen = res;
2228
2229	if (sk->sk_type == SOCK_DGRAM) {
2230		macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2231				  po->tp_reserve;
2232	} else {
2233		unsigned int maclen = skb_network_offset(skb);
2234		netoff = TPACKET_ALIGN(po->tp_hdrlen +
2235				       (maclen < 16 ? 16 : maclen)) +
2236				       po->tp_reserve;
2237		if (po->has_vnet_hdr) {
2238			netoff += sizeof(struct virtio_net_hdr);
2239			do_vnet = true;
2240		}
2241		macoff = netoff - maclen;
2242	}
2243	if (netoff > USHRT_MAX) {
2244		atomic_inc(&po->tp_drops);
2245		goto drop_n_restore;
2246	}
2247	if (po->tp_version <= TPACKET_V2) {
2248		if (macoff + snaplen > po->rx_ring.frame_size) {
2249			if (po->copy_thresh &&
2250			    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2251				if (skb_shared(skb)) {
2252					copy_skb = skb_clone(skb, GFP_ATOMIC);
2253				} else {
2254					copy_skb = skb_get(skb);
2255					skb_head = skb->data;
2256				}
2257				if (copy_skb)
 
 
2258					skb_set_owner_r(copy_skb, sk);
 
2259			}
2260			snaplen = po->rx_ring.frame_size - macoff;
2261			if ((int)snaplen < 0) {
2262				snaplen = 0;
2263				do_vnet = false;
2264			}
2265		}
2266	} else if (unlikely(macoff + snaplen >
2267			    GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2268		u32 nval;
2269
2270		nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2271		pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2272			    snaplen, nval, macoff);
2273		snaplen = nval;
2274		if (unlikely((int)snaplen < 0)) {
2275			snaplen = 0;
2276			macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2277			do_vnet = false;
2278		}
2279	}
2280	spin_lock(&sk->sk_receive_queue.lock);
2281	h.raw = packet_current_rx_frame(po, skb,
2282					TP_STATUS_KERNEL, (macoff+snaplen));
2283	if (!h.raw)
2284		goto drop_n_account;
2285
2286	if (po->tp_version <= TPACKET_V2) {
2287		slot_id = po->rx_ring.head;
2288		if (test_bit(slot_id, po->rx_ring.rx_owner_map))
2289			goto drop_n_account;
2290		__set_bit(slot_id, po->rx_ring.rx_owner_map);
2291	}
2292
2293	if (do_vnet &&
2294	    virtio_net_hdr_from_skb(skb, h.raw + macoff -
2295				    sizeof(struct virtio_net_hdr),
2296				    vio_le(), true, 0)) {
2297		if (po->tp_version == TPACKET_V3)
2298			prb_clear_blk_fill_status(&po->rx_ring);
2299		goto drop_n_account;
2300	}
2301
2302	if (po->tp_version <= TPACKET_V2) {
2303		packet_increment_rx_head(po, &po->rx_ring);
2304	/*
2305	 * LOSING will be reported till you read the stats,
2306	 * because it's COR - Clear On Read.
2307	 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2308	 * at packet level.
2309	 */
2310		if (atomic_read(&po->tp_drops))
2311			status |= TP_STATUS_LOSING;
2312	}
2313
2314	po->stats.stats1.tp_packets++;
2315	if (copy_skb) {
2316		status |= TP_STATUS_COPY;
 
2317		__skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2318	}
2319	spin_unlock(&sk->sk_receive_queue.lock);
2320
2321	skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2322
2323	if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
 
 
 
 
 
 
2324		ktime_get_real_ts64(&ts);
2325
2326	status |= ts_status;
2327
2328	switch (po->tp_version) {
2329	case TPACKET_V1:
2330		h.h1->tp_len = skb->len;
2331		h.h1->tp_snaplen = snaplen;
2332		h.h1->tp_mac = macoff;
2333		h.h1->tp_net = netoff;
2334		h.h1->tp_sec = ts.tv_sec;
2335		h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2336		hdrlen = sizeof(*h.h1);
2337		break;
2338	case TPACKET_V2:
2339		h.h2->tp_len = skb->len;
2340		h.h2->tp_snaplen = snaplen;
2341		h.h2->tp_mac = macoff;
2342		h.h2->tp_net = netoff;
2343		h.h2->tp_sec = ts.tv_sec;
2344		h.h2->tp_nsec = ts.tv_nsec;
2345		if (skb_vlan_tag_present(skb)) {
2346			h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2347			h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2348			status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2349		} else {
2350			h.h2->tp_vlan_tci = 0;
2351			h.h2->tp_vlan_tpid = 0;
2352		}
2353		memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2354		hdrlen = sizeof(*h.h2);
2355		break;
2356	case TPACKET_V3:
2357		/* tp_nxt_offset,vlan are already populated above.
2358		 * So DONT clear those fields here
2359		 */
2360		h.h3->tp_status |= status;
2361		h.h3->tp_len = skb->len;
2362		h.h3->tp_snaplen = snaplen;
2363		h.h3->tp_mac = macoff;
2364		h.h3->tp_net = netoff;
2365		h.h3->tp_sec  = ts.tv_sec;
2366		h.h3->tp_nsec = ts.tv_nsec;
2367		memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2368		hdrlen = sizeof(*h.h3);
2369		break;
2370	default:
2371		BUG();
2372	}
2373
2374	sll = h.raw + TPACKET_ALIGN(hdrlen);
2375	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2376	sll->sll_family = AF_PACKET;
2377	sll->sll_hatype = dev->type;
2378	sll->sll_protocol = skb->protocol;
2379	sll->sll_pkttype = skb->pkt_type;
2380	if (unlikely(po->origdev))
2381		sll->sll_ifindex = orig_dev->ifindex;
2382	else
2383		sll->sll_ifindex = dev->ifindex;
2384
2385	smp_mb();
2386
2387#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2388	if (po->tp_version <= TPACKET_V2) {
2389		u8 *start, *end;
2390
2391		end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2392					macoff + snaplen);
2393
2394		for (start = h.raw; start < end; start += PAGE_SIZE)
2395			flush_dcache_page(pgv_to_page(start));
2396	}
2397	smp_wmb();
2398#endif
2399
2400	if (po->tp_version <= TPACKET_V2) {
2401		spin_lock(&sk->sk_receive_queue.lock);
2402		__packet_set_status(po, h.raw, status);
2403		__clear_bit(slot_id, po->rx_ring.rx_owner_map);
2404		spin_unlock(&sk->sk_receive_queue.lock);
2405		sk->sk_data_ready(sk);
2406	} else if (po->tp_version == TPACKET_V3) {
2407		prb_clear_blk_fill_status(&po->rx_ring);
2408	}
2409
2410drop_n_restore:
2411	if (skb_head != skb->data && skb_shared(skb)) {
2412		skb->data = skb_head;
2413		skb->len = skb_len;
2414	}
2415drop:
2416	if (!is_drop_n_account)
2417		consume_skb(skb);
2418	else
2419		kfree_skb(skb);
2420	return 0;
2421
2422drop_n_account:
2423	spin_unlock(&sk->sk_receive_queue.lock);
2424	atomic_inc(&po->tp_drops);
2425	is_drop_n_account = true;
2426
2427	sk->sk_data_ready(sk);
2428	kfree_skb(copy_skb);
2429	goto drop_n_restore;
2430}
2431
2432static void tpacket_destruct_skb(struct sk_buff *skb)
2433{
2434	struct packet_sock *po = pkt_sk(skb->sk);
2435
2436	if (likely(po->tx_ring.pg_vec)) {
2437		void *ph;
2438		__u32 ts;
2439
2440		ph = skb_zcopy_get_nouarg(skb);
2441		packet_dec_pending(&po->tx_ring);
2442
2443		ts = __packet_set_timestamp(po, ph, skb);
2444		__packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2445
2446		if (!packet_read_pending(&po->tx_ring))
2447			complete(&po->skb_completion);
2448	}
2449
2450	sock_wfree(skb);
2451}
2452
2453static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2454{
2455	if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2456	    (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2457	     __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2458	      __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2459		vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2460			 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2461			__virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2462
2463	if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2464		return -EINVAL;
2465
2466	return 0;
2467}
2468
2469static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2470				 struct virtio_net_hdr *vnet_hdr)
2471{
2472	if (*len < sizeof(*vnet_hdr))
 
 
2473		return -EINVAL;
2474	*len -= sizeof(*vnet_hdr);
2475
2476	if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2477		return -EFAULT;
2478
2479	return __packet_snd_vnet_parse(vnet_hdr, *len);
 
 
 
 
 
 
 
 
2480}
2481
2482static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2483		void *frame, struct net_device *dev, void *data, int tp_len,
2484		__be16 proto, unsigned char *addr, int hlen, int copylen,
2485		const struct sockcm_cookie *sockc)
2486{
2487	union tpacket_uhdr ph;
2488	int to_write, offset, len, nr_frags, len_max;
2489	struct socket *sock = po->sk.sk_socket;
2490	struct page *page;
2491	int err;
2492
2493	ph.raw = frame;
2494
2495	skb->protocol = proto;
2496	skb->dev = dev;
2497	skb->priority = po->sk.sk_priority;
2498	skb->mark = po->sk.sk_mark;
2499	skb->tstamp = sockc->transmit_time;
2500	skb_setup_tx_timestamp(skb, sockc->tsflags);
2501	skb_zcopy_set_nouarg(skb, ph.raw);
2502
2503	skb_reserve(skb, hlen);
2504	skb_reset_network_header(skb);
2505
2506	to_write = tp_len;
2507
2508	if (sock->type == SOCK_DGRAM) {
2509		err = dev_hard_header(skb, dev, ntohs(proto), addr,
2510				NULL, tp_len);
2511		if (unlikely(err < 0))
2512			return -EINVAL;
2513	} else if (copylen) {
2514		int hdrlen = min_t(int, copylen, tp_len);
2515
2516		skb_push(skb, dev->hard_header_len);
2517		skb_put(skb, copylen - dev->hard_header_len);
2518		err = skb_store_bits(skb, 0, data, hdrlen);
2519		if (unlikely(err))
2520			return err;
2521		if (!dev_validate_header(dev, skb->data, hdrlen))
2522			return -EINVAL;
2523
2524		data += hdrlen;
2525		to_write -= hdrlen;
2526	}
2527
2528	offset = offset_in_page(data);
2529	len_max = PAGE_SIZE - offset;
2530	len = ((to_write > len_max) ? len_max : to_write);
2531
2532	skb->data_len = to_write;
2533	skb->len += to_write;
2534	skb->truesize += to_write;
2535	refcount_add(to_write, &po->sk.sk_wmem_alloc);
2536
2537	while (likely(to_write)) {
2538		nr_frags = skb_shinfo(skb)->nr_frags;
2539
2540		if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2541			pr_err("Packet exceed the number of skb frags(%lu)\n",
2542			       MAX_SKB_FRAGS);
2543			return -EFAULT;
2544		}
2545
2546		page = pgv_to_page(data);
2547		data += len;
2548		flush_dcache_page(page);
2549		get_page(page);
2550		skb_fill_page_desc(skb, nr_frags, page, offset, len);
2551		to_write -= len;
2552		offset = 0;
2553		len_max = PAGE_SIZE;
2554		len = ((to_write > len_max) ? len_max : to_write);
2555	}
2556
2557	packet_parse_headers(skb, sock);
2558
2559	return tp_len;
2560}
2561
2562static int tpacket_parse_header(struct packet_sock *po, void *frame,
2563				int size_max, void **data)
2564{
2565	union tpacket_uhdr ph;
2566	int tp_len, off;
2567
2568	ph.raw = frame;
2569
2570	switch (po->tp_version) {
2571	case TPACKET_V3:
2572		if (ph.h3->tp_next_offset != 0) {
2573			pr_warn_once("variable sized slot not supported");
2574			return -EINVAL;
2575		}
2576		tp_len = ph.h3->tp_len;
2577		break;
2578	case TPACKET_V2:
2579		tp_len = ph.h2->tp_len;
2580		break;
2581	default:
2582		tp_len = ph.h1->tp_len;
2583		break;
2584	}
2585	if (unlikely(tp_len > size_max)) {
2586		pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2587		return -EMSGSIZE;
2588	}
2589
2590	if (unlikely(po->tp_tx_has_off)) {
2591		int off_min, off_max;
2592
2593		off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2594		off_max = po->tx_ring.frame_size - tp_len;
2595		if (po->sk.sk_type == SOCK_DGRAM) {
2596			switch (po->tp_version) {
2597			case TPACKET_V3:
2598				off = ph.h3->tp_net;
2599				break;
2600			case TPACKET_V2:
2601				off = ph.h2->tp_net;
2602				break;
2603			default:
2604				off = ph.h1->tp_net;
2605				break;
2606			}
2607		} else {
2608			switch (po->tp_version) {
2609			case TPACKET_V3:
2610				off = ph.h3->tp_mac;
2611				break;
2612			case TPACKET_V2:
2613				off = ph.h2->tp_mac;
2614				break;
2615			default:
2616				off = ph.h1->tp_mac;
2617				break;
2618			}
2619		}
2620		if (unlikely((off < off_min) || (off_max < off)))
2621			return -EINVAL;
2622	} else {
2623		off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2624	}
2625
2626	*data = frame + off;
2627	return tp_len;
2628}
2629
2630static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2631{
2632	struct sk_buff *skb = NULL;
2633	struct net_device *dev;
2634	struct virtio_net_hdr *vnet_hdr = NULL;
2635	struct sockcm_cookie sockc;
2636	__be16 proto;
2637	int err, reserve = 0;
2638	void *ph;
2639	DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2640	bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
 
2641	unsigned char *addr = NULL;
2642	int tp_len, size_max;
2643	void *data;
2644	int len_sum = 0;
2645	int status = TP_STATUS_AVAILABLE;
2646	int hlen, tlen, copylen = 0;
2647	long timeo = 0;
2648
2649	mutex_lock(&po->pg_vec_lock);
2650
2651	/* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2652	 * we need to confirm it under protection of pg_vec_lock.
2653	 */
2654	if (unlikely(!po->tx_ring.pg_vec)) {
2655		err = -EBUSY;
2656		goto out;
2657	}
2658	if (likely(saddr == NULL)) {
2659		dev	= packet_cached_dev_get(po);
2660		proto	= po->num;
2661	} else {
2662		err = -EINVAL;
2663		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2664			goto out;
2665		if (msg->msg_namelen < (saddr->sll_halen
2666					+ offsetof(struct sockaddr_ll,
2667						sll_addr)))
2668			goto out;
2669		proto	= saddr->sll_protocol;
2670		dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2671		if (po->sk.sk_socket->type == SOCK_DGRAM) {
2672			if (dev && msg->msg_namelen < dev->addr_len +
2673				   offsetof(struct sockaddr_ll, sll_addr))
2674				goto out_put;
2675			addr = saddr->sll_addr;
2676		}
2677	}
2678
2679	err = -ENXIO;
2680	if (unlikely(dev == NULL))
2681		goto out;
2682	err = -ENETDOWN;
2683	if (unlikely(!(dev->flags & IFF_UP)))
2684		goto out_put;
2685
2686	sockcm_init(&sockc, &po->sk);
2687	if (msg->msg_controllen) {
2688		err = sock_cmsg_send(&po->sk, msg, &sockc);
2689		if (unlikely(err))
2690			goto out_put;
2691	}
2692
2693	if (po->sk.sk_socket->type == SOCK_RAW)
2694		reserve = dev->hard_header_len;
2695	size_max = po->tx_ring.frame_size
2696		- (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2697
2698	if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2699		size_max = dev->mtu + reserve + VLAN_HLEN;
2700
2701	reinit_completion(&po->skb_completion);
2702
2703	do {
2704		ph = packet_current_frame(po, &po->tx_ring,
2705					  TP_STATUS_SEND_REQUEST);
2706		if (unlikely(ph == NULL)) {
2707			if (need_wait && skb) {
2708				timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2709				timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2710				if (timeo <= 0) {
2711					err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2712					goto out_put;
2713				}
2714			}
2715			/* check for additional frames */
2716			continue;
2717		}
2718
2719		skb = NULL;
2720		tp_len = tpacket_parse_header(po, ph, size_max, &data);
2721		if (tp_len < 0)
2722			goto tpacket_error;
2723
2724		status = TP_STATUS_SEND_REQUEST;
2725		hlen = LL_RESERVED_SPACE(dev);
2726		tlen = dev->needed_tailroom;
2727		if (po->has_vnet_hdr) {
2728			vnet_hdr = data;
2729			data += sizeof(*vnet_hdr);
2730			tp_len -= sizeof(*vnet_hdr);
2731			if (tp_len < 0 ||
2732			    __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2733				tp_len = -EINVAL;
2734				goto tpacket_error;
2735			}
2736			copylen = __virtio16_to_cpu(vio_le(),
2737						    vnet_hdr->hdr_len);
2738		}
2739		copylen = max_t(int, copylen, dev->hard_header_len);
2740		skb = sock_alloc_send_skb(&po->sk,
2741				hlen + tlen + sizeof(struct sockaddr_ll) +
2742				(copylen - dev->hard_header_len),
2743				!need_wait, &err);
2744
2745		if (unlikely(skb == NULL)) {
2746			/* we assume the socket was initially writeable ... */
2747			if (likely(len_sum > 0))
2748				err = len_sum;
2749			goto out_status;
2750		}
2751		tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2752					  addr, hlen, copylen, &sockc);
2753		if (likely(tp_len >= 0) &&
2754		    tp_len > dev->mtu + reserve &&
2755		    !po->has_vnet_hdr &&
2756		    !packet_extra_vlan_len_allowed(dev, skb))
2757			tp_len = -EMSGSIZE;
2758
2759		if (unlikely(tp_len < 0)) {
2760tpacket_error:
2761			if (po->tp_loss) {
2762				__packet_set_status(po, ph,
2763						TP_STATUS_AVAILABLE);
2764				packet_increment_head(&po->tx_ring);
2765				kfree_skb(skb);
2766				continue;
2767			} else {
2768				status = TP_STATUS_WRONG_FORMAT;
2769				err = tp_len;
2770				goto out_status;
2771			}
2772		}
2773
2774		if (po->has_vnet_hdr) {
2775			if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2776				tp_len = -EINVAL;
2777				goto tpacket_error;
2778			}
2779			virtio_net_hdr_set_proto(skb, vnet_hdr);
2780		}
2781
2782		skb->destructor = tpacket_destruct_skb;
2783		__packet_set_status(po, ph, TP_STATUS_SENDING);
2784		packet_inc_pending(&po->tx_ring);
2785
2786		status = TP_STATUS_SEND_REQUEST;
2787		err = po->xmit(skb);
2788		if (unlikely(err > 0)) {
2789			err = net_xmit_errno(err);
 
2790			if (err && __packet_get_status(po, ph) ==
2791				   TP_STATUS_AVAILABLE) {
2792				/* skb was destructed already */
2793				skb = NULL;
2794				goto out_status;
2795			}
2796			/*
2797			 * skb was dropped but not destructed yet;
2798			 * let's treat it like congestion or err < 0
2799			 */
2800			err = 0;
2801		}
2802		packet_increment_head(&po->tx_ring);
2803		len_sum += tp_len;
2804	} while (likely((ph != NULL) ||
2805		/* Note: packet_read_pending() might be slow if we have
2806		 * to call it as it's per_cpu variable, but in fast-path
2807		 * we already short-circuit the loop with the first
2808		 * condition, and luckily don't have to go that path
2809		 * anyway.
2810		 */
2811		 (need_wait && packet_read_pending(&po->tx_ring))));
2812
2813	err = len_sum;
2814	goto out_put;
2815
2816out_status:
2817	__packet_set_status(po, ph, status);
2818	kfree_skb(skb);
2819out_put:
2820	dev_put(dev);
2821out:
2822	mutex_unlock(&po->pg_vec_lock);
2823	return err;
2824}
2825
2826static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2827				        size_t reserve, size_t len,
2828				        size_t linear, int noblock,
2829				        int *err)
2830{
2831	struct sk_buff *skb;
2832
2833	/* Under a page?  Don't bother with paged skb. */
2834	if (prepad + len < PAGE_SIZE || !linear)
2835		linear = len;
2836
 
 
2837	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2838				   err, 0);
2839	if (!skb)
2840		return NULL;
2841
2842	skb_reserve(skb, reserve);
2843	skb_put(skb, linear);
2844	skb->data_len = len - linear;
2845	skb->len += len - linear;
2846
2847	return skb;
2848}
2849
2850static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2851{
2852	struct sock *sk = sock->sk;
2853	DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2854	struct sk_buff *skb;
2855	struct net_device *dev;
2856	__be16 proto;
2857	unsigned char *addr = NULL;
2858	int err, reserve = 0;
2859	struct sockcm_cookie sockc;
2860	struct virtio_net_hdr vnet_hdr = { 0 };
2861	int offset = 0;
2862	struct packet_sock *po = pkt_sk(sk);
2863	bool has_vnet_hdr = false;
2864	int hlen, tlen, linear;
2865	int extra_len = 0;
2866
2867	/*
2868	 *	Get and verify the address.
2869	 */
2870
2871	if (likely(saddr == NULL)) {
2872		dev	= packet_cached_dev_get(po);
2873		proto	= po->num;
2874	} else {
2875		err = -EINVAL;
2876		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2877			goto out;
2878		if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2879			goto out;
2880		proto	= saddr->sll_protocol;
2881		dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2882		if (sock->type == SOCK_DGRAM) {
2883			if (dev && msg->msg_namelen < dev->addr_len +
2884				   offsetof(struct sockaddr_ll, sll_addr))
2885				goto out_unlock;
2886			addr = saddr->sll_addr;
2887		}
2888	}
2889
2890	err = -ENXIO;
2891	if (unlikely(dev == NULL))
2892		goto out_unlock;
2893	err = -ENETDOWN;
2894	if (unlikely(!(dev->flags & IFF_UP)))
2895		goto out_unlock;
2896
2897	sockcm_init(&sockc, sk);
2898	sockc.mark = sk->sk_mark;
2899	if (msg->msg_controllen) {
2900		err = sock_cmsg_send(sk, msg, &sockc);
2901		if (unlikely(err))
2902			goto out_unlock;
2903	}
2904
2905	if (sock->type == SOCK_RAW)
2906		reserve = dev->hard_header_len;
2907	if (po->has_vnet_hdr) {
2908		err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2909		if (err)
2910			goto out_unlock;
2911		has_vnet_hdr = true;
2912	}
2913
2914	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2915		if (!netif_supports_nofcs(dev)) {
2916			err = -EPROTONOSUPPORT;
2917			goto out_unlock;
2918		}
2919		extra_len = 4; /* We're doing our own CRC */
2920	}
2921
2922	err = -EMSGSIZE;
2923	if (!vnet_hdr.gso_type &&
2924	    (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2925		goto out_unlock;
2926
2927	err = -ENOBUFS;
2928	hlen = LL_RESERVED_SPACE(dev);
2929	tlen = dev->needed_tailroom;
2930	linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2931	linear = max(linear, min_t(int, len, dev->hard_header_len));
2932	skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2933			       msg->msg_flags & MSG_DONTWAIT, &err);
2934	if (skb == NULL)
2935		goto out_unlock;
2936
2937	skb_reset_network_header(skb);
2938
2939	err = -EINVAL;
2940	if (sock->type == SOCK_DGRAM) {
2941		offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2942		if (unlikely(offset < 0))
2943			goto out_free;
2944	} else if (reserve) {
2945		skb_reserve(skb, -reserve);
2946		if (len < reserve + sizeof(struct ipv6hdr) &&
2947		    dev->min_header_len != dev->hard_header_len)
2948			skb_reset_network_header(skb);
2949	}
2950
2951	/* Returns -EFAULT on error */
2952	err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2953	if (err)
2954		goto out_free;
2955
2956	if (sock->type == SOCK_RAW &&
2957	    !dev_validate_header(dev, skb->data, len)) {
2958		err = -EINVAL;
2959		goto out_free;
2960	}
2961
2962	skb_setup_tx_timestamp(skb, sockc.tsflags);
2963
2964	if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2965	    !packet_extra_vlan_len_allowed(dev, skb)) {
2966		err = -EMSGSIZE;
2967		goto out_free;
2968	}
2969
2970	skb->protocol = proto;
2971	skb->dev = dev;
2972	skb->priority = sk->sk_priority;
2973	skb->mark = sockc.mark;
2974	skb->tstamp = sockc.transmit_time;
2975
2976	if (has_vnet_hdr) {
 
 
 
 
 
2977		err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2978		if (err)
2979			goto out_free;
2980		len += sizeof(vnet_hdr);
2981		virtio_net_hdr_set_proto(skb, &vnet_hdr);
2982	}
2983
2984	packet_parse_headers(skb, sock);
2985
2986	if (unlikely(extra_len == 4))
2987		skb->no_fcs = 1;
2988
2989	err = po->xmit(skb);
2990	if (err > 0 && (err = net_xmit_errno(err)) != 0)
2991		goto out_unlock;
 
 
 
2992
2993	dev_put(dev);
2994
2995	return len;
2996
2997out_free:
2998	kfree_skb(skb);
2999out_unlock:
3000	if (dev)
3001		dev_put(dev);
3002out:
3003	return err;
3004}
3005
3006static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
3007{
3008	struct sock *sk = sock->sk;
3009	struct packet_sock *po = pkt_sk(sk);
3010
3011	if (po->tx_ring.pg_vec)
 
 
 
3012		return tpacket_snd(po, msg);
3013	else
3014		return packet_snd(sock, msg, len);
3015}
3016
3017/*
3018 *	Close a PACKET socket. This is fairly simple. We immediately go
3019 *	to 'closed' state and remove our protocol entry in the device list.
3020 */
3021
3022static int packet_release(struct socket *sock)
3023{
3024	struct sock *sk = sock->sk;
3025	struct packet_sock *po;
3026	struct packet_fanout *f;
3027	struct net *net;
3028	union tpacket_req_u req_u;
3029
3030	if (!sk)
3031		return 0;
3032
3033	net = sock_net(sk);
3034	po = pkt_sk(sk);
3035
3036	mutex_lock(&net->packet.sklist_lock);
3037	sk_del_node_init_rcu(sk);
3038	mutex_unlock(&net->packet.sklist_lock);
3039
3040	preempt_disable();
3041	sock_prot_inuse_add(net, sk->sk_prot, -1);
3042	preempt_enable();
3043
3044	spin_lock(&po->bind_lock);
3045	unregister_prot_hook(sk, false);
3046	packet_cached_dev_reset(po);
3047
3048	if (po->prot_hook.dev) {
3049		dev_put(po->prot_hook.dev);
3050		po->prot_hook.dev = NULL;
3051	}
3052	spin_unlock(&po->bind_lock);
3053
3054	packet_flush_mclist(sk);
3055
3056	lock_sock(sk);
3057	if (po->rx_ring.pg_vec) {
3058		memset(&req_u, 0, sizeof(req_u));
3059		packet_set_ring(sk, &req_u, 1, 0);
3060	}
3061
3062	if (po->tx_ring.pg_vec) {
3063		memset(&req_u, 0, sizeof(req_u));
3064		packet_set_ring(sk, &req_u, 1, 1);
3065	}
3066	release_sock(sk);
3067
3068	f = fanout_release(sk);
3069
3070	synchronize_net();
3071
3072	kfree(po->rollover);
3073	if (f) {
3074		fanout_release_data(f);
3075		kfree(f);
3076	}
3077	/*
3078	 *	Now the socket is dead. No more input will appear.
3079	 */
3080	sock_orphan(sk);
3081	sock->sk = NULL;
3082
3083	/* Purge queues */
3084
3085	skb_queue_purge(&sk->sk_receive_queue);
3086	packet_free_pending(po);
3087	sk_refcnt_debug_release(sk);
3088
3089	sock_put(sk);
3090	return 0;
3091}
3092
3093/*
3094 *	Attach a packet hook.
3095 */
3096
3097static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3098			  __be16 proto)
3099{
3100	struct packet_sock *po = pkt_sk(sk);
3101	struct net_device *dev_curr;
3102	__be16 proto_curr;
3103	bool need_rehook;
3104	struct net_device *dev = NULL;
3105	int ret = 0;
3106	bool unlisted = false;
 
 
3107
3108	lock_sock(sk);
3109	spin_lock(&po->bind_lock);
 
 
 
3110	rcu_read_lock();
3111
3112	if (po->fanout) {
3113		ret = -EINVAL;
3114		goto out_unlock;
3115	}
3116
3117	if (name) {
3118		dev = dev_get_by_name_rcu(sock_net(sk), name);
3119		if (!dev) {
3120			ret = -ENODEV;
3121			goto out_unlock;
3122		}
3123	} else if (ifindex) {
3124		dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3125		if (!dev) {
3126			ret = -ENODEV;
3127			goto out_unlock;
3128		}
3129	}
3130
3131	if (dev)
3132		dev_hold(dev);
3133
3134	proto_curr = po->prot_hook.type;
3135	dev_curr = po->prot_hook.dev;
3136
3137	need_rehook = proto_curr != proto || dev_curr != dev;
3138
3139	if (need_rehook) {
3140		if (po->running) {
 
3141			rcu_read_unlock();
3142			/* prevents packet_notifier() from calling
3143			 * register_prot_hook()
3144			 */
3145			po->num = 0;
3146			__unregister_prot_hook(sk, true);
3147			rcu_read_lock();
3148			dev_curr = po->prot_hook.dev;
3149			if (dev)
3150				unlisted = !dev_get_by_index_rcu(sock_net(sk),
3151								 dev->ifindex);
3152		}
3153
3154		BUG_ON(po->running);
3155		po->num = proto;
3156		po->prot_hook.type = proto;
3157
 
 
3158		if (unlikely(unlisted)) {
3159			dev_put(dev);
3160			po->prot_hook.dev = NULL;
3161			po->ifindex = -1;
3162			packet_cached_dev_reset(po);
3163		} else {
 
 
3164			po->prot_hook.dev = dev;
3165			po->ifindex = dev ? dev->ifindex : 0;
3166			packet_cached_dev_assign(po, dev);
3167		}
 
3168	}
3169	if (dev_curr)
3170		dev_put(dev_curr);
3171
3172	if (proto == 0 || !need_rehook)
3173		goto out_unlock;
3174
3175	if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3176		register_prot_hook(sk);
3177	} else {
3178		sk->sk_err = ENETDOWN;
3179		if (!sock_flag(sk, SOCK_DEAD))
3180			sk->sk_error_report(sk);
3181	}
3182
3183out_unlock:
3184	rcu_read_unlock();
3185	spin_unlock(&po->bind_lock);
3186	release_sock(sk);
3187	return ret;
3188}
3189
3190/*
3191 *	Bind a packet socket to a device
3192 */
3193
3194static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3195			    int addr_len)
3196{
3197	struct sock *sk = sock->sk;
3198	char name[sizeof(uaddr->sa_data) + 1];
3199
3200	/*
3201	 *	Check legality
3202	 */
3203
3204	if (addr_len != sizeof(struct sockaddr))
3205		return -EINVAL;
3206	/* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3207	 * zero-terminated.
3208	 */
3209	memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3210	name[sizeof(uaddr->sa_data)] = 0;
3211
3212	return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3213}
3214
3215static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3216{
3217	struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3218	struct sock *sk = sock->sk;
3219
3220	/*
3221	 *	Check legality
3222	 */
3223
3224	if (addr_len < sizeof(struct sockaddr_ll))
3225		return -EINVAL;
3226	if (sll->sll_family != AF_PACKET)
3227		return -EINVAL;
3228
3229	return packet_do_bind(sk, NULL, sll->sll_ifindex,
3230			      sll->sll_protocol ? : pkt_sk(sk)->num);
3231}
3232
3233static struct proto packet_proto = {
3234	.name	  = "PACKET",
3235	.owner	  = THIS_MODULE,
3236	.obj_size = sizeof(struct packet_sock),
3237};
3238
3239/*
3240 *	Create a packet of type SOCK_PACKET.
3241 */
3242
3243static int packet_create(struct net *net, struct socket *sock, int protocol,
3244			 int kern)
3245{
3246	struct sock *sk;
3247	struct packet_sock *po;
3248	__be16 proto = (__force __be16)protocol; /* weird, but documented */
3249	int err;
3250
3251	if (!ns_capable(net->user_ns, CAP_NET_RAW))
3252		return -EPERM;
3253	if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3254	    sock->type != SOCK_PACKET)
3255		return -ESOCKTNOSUPPORT;
3256
3257	sock->state = SS_UNCONNECTED;
3258
3259	err = -ENOBUFS;
3260	sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3261	if (sk == NULL)
3262		goto out;
3263
3264	sock->ops = &packet_ops;
3265	if (sock->type == SOCK_PACKET)
3266		sock->ops = &packet_ops_spkt;
3267
3268	sock_init_data(sock, sk);
3269
3270	po = pkt_sk(sk);
3271	init_completion(&po->skb_completion);
3272	sk->sk_family = PF_PACKET;
3273	po->num = proto;
3274	po->xmit = dev_queue_xmit;
3275
3276	err = packet_alloc_pending(po);
3277	if (err)
3278		goto out2;
3279
3280	packet_cached_dev_reset(po);
3281
3282	sk->sk_destruct = packet_sock_destruct;
3283	sk_refcnt_debug_inc(sk);
3284
3285	/*
3286	 *	Attach a protocol block
3287	 */
3288
3289	spin_lock_init(&po->bind_lock);
3290	mutex_init(&po->pg_vec_lock);
3291	po->rollover = NULL;
3292	po->prot_hook.func = packet_rcv;
3293
3294	if (sock->type == SOCK_PACKET)
3295		po->prot_hook.func = packet_rcv_spkt;
3296
3297	po->prot_hook.af_packet_priv = sk;
 
3298
3299	if (proto) {
3300		po->prot_hook.type = proto;
3301		__register_prot_hook(sk);
3302	}
3303
3304	mutex_lock(&net->packet.sklist_lock);
3305	sk_add_node_tail_rcu(sk, &net->packet.sklist);
3306	mutex_unlock(&net->packet.sklist_lock);
3307
3308	preempt_disable();
3309	sock_prot_inuse_add(net, &packet_proto, 1);
3310	preempt_enable();
3311
3312	return 0;
3313out2:
3314	sk_free(sk);
3315out:
3316	return err;
3317}
3318
3319/*
3320 *	Pull a packet from our receive queue and hand it to the user.
3321 *	If necessary we block.
3322 */
3323
3324static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3325			  int flags)
3326{
3327	struct sock *sk = sock->sk;
3328	struct sk_buff *skb;
3329	int copied, err;
3330	int vnet_hdr_len = 0;
3331	unsigned int origlen = 0;
3332
3333	err = -EINVAL;
3334	if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3335		goto out;
3336
3337#if 0
3338	/* What error should we return now? EUNATTACH? */
3339	if (pkt_sk(sk)->ifindex < 0)
3340		return -ENODEV;
3341#endif
3342
3343	if (flags & MSG_ERRQUEUE) {
3344		err = sock_recv_errqueue(sk, msg, len,
3345					 SOL_PACKET, PACKET_TX_TIMESTAMP);
3346		goto out;
3347	}
3348
3349	/*
3350	 *	Call the generic datagram receiver. This handles all sorts
3351	 *	of horrible races and re-entrancy so we can forget about it
3352	 *	in the protocol layers.
3353	 *
3354	 *	Now it will return ENETDOWN, if device have just gone down,
3355	 *	but then it will block.
3356	 */
3357
3358	skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3359
3360	/*
3361	 *	An error occurred so return it. Because skb_recv_datagram()
3362	 *	handles the blocking we don't see and worry about blocking
3363	 *	retries.
3364	 */
3365
3366	if (skb == NULL)
3367		goto out;
3368
3369	packet_rcv_try_clear_pressure(pkt_sk(sk));
3370
3371	if (pkt_sk(sk)->has_vnet_hdr) {
3372		err = packet_rcv_vnet(msg, skb, &len);
3373		if (err)
3374			goto out_free;
3375		vnet_hdr_len = sizeof(struct virtio_net_hdr);
3376	}
3377
3378	/* You lose any data beyond the buffer you gave. If it worries
3379	 * a user program they can ask the device for its MTU
3380	 * anyway.
3381	 */
3382	copied = skb->len;
3383	if (copied > len) {
3384		copied = len;
3385		msg->msg_flags |= MSG_TRUNC;
3386	}
3387
3388	err = skb_copy_datagram_msg(skb, 0, msg, copied);
3389	if (err)
3390		goto out_free;
3391
3392	if (sock->type != SOCK_PACKET) {
3393		struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3394
3395		/* Original length was stored in sockaddr_ll fields */
3396		origlen = PACKET_SKB_CB(skb)->sa.origlen;
3397		sll->sll_family = AF_PACKET;
3398		sll->sll_protocol = skb->protocol;
3399	}
3400
3401	sock_recv_ts_and_drops(msg, sk, skb);
3402
3403	if (msg->msg_name) {
 
 
3404		int copy_len;
3405
3406		/* If the address length field is there to be filled
3407		 * in, we fill it in now.
3408		 */
3409		if (sock->type == SOCK_PACKET) {
3410			__sockaddr_check_size(sizeof(struct sockaddr_pkt));
3411			msg->msg_namelen = sizeof(struct sockaddr_pkt);
3412			copy_len = msg->msg_namelen;
3413		} else {
3414			struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3415
3416			msg->msg_namelen = sll->sll_halen +
3417				offsetof(struct sockaddr_ll, sll_addr);
3418			copy_len = msg->msg_namelen;
3419			if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3420				memset(msg->msg_name +
3421				       offsetof(struct sockaddr_ll, sll_addr),
3422				       0, sizeof(sll->sll_addr));
3423				msg->msg_namelen = sizeof(struct sockaddr_ll);
3424			}
3425		}
 
 
 
 
3426		memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3427	}
3428
3429	if (pkt_sk(sk)->auxdata) {
3430		struct tpacket_auxdata aux;
3431
3432		aux.tp_status = TP_STATUS_USER;
3433		if (skb->ip_summed == CHECKSUM_PARTIAL)
3434			aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3435		else if (skb->pkt_type != PACKET_OUTGOING &&
3436			 (skb->ip_summed == CHECKSUM_COMPLETE ||
3437			  skb_csum_unnecessary(skb)))
3438			aux.tp_status |= TP_STATUS_CSUM_VALID;
 
 
3439
3440		aux.tp_len = origlen;
3441		aux.tp_snaplen = skb->len;
3442		aux.tp_mac = 0;
3443		aux.tp_net = skb_network_offset(skb);
3444		if (skb_vlan_tag_present(skb)) {
3445			aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3446			aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3447			aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3448		} else {
3449			aux.tp_vlan_tci = 0;
3450			aux.tp_vlan_tpid = 0;
3451		}
3452		put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3453	}
3454
3455	/*
3456	 *	Free or return the buffer as appropriate. Again this
3457	 *	hides all the races and re-entrancy issues from us.
3458	 */
3459	err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3460
3461out_free:
3462	skb_free_datagram(sk, skb);
3463out:
3464	return err;
3465}
3466
3467static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3468			       int peer)
3469{
3470	struct net_device *dev;
3471	struct sock *sk	= sock->sk;
3472
3473	if (peer)
3474		return -EOPNOTSUPP;
3475
3476	uaddr->sa_family = AF_PACKET;
3477	memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3478	rcu_read_lock();
3479	dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3480	if (dev)
3481		strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3482	rcu_read_unlock();
3483
3484	return sizeof(*uaddr);
3485}
3486
3487static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3488			  int peer)
3489{
3490	struct net_device *dev;
3491	struct sock *sk = sock->sk;
3492	struct packet_sock *po = pkt_sk(sk);
3493	DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
 
3494
3495	if (peer)
3496		return -EOPNOTSUPP;
3497
 
3498	sll->sll_family = AF_PACKET;
3499	sll->sll_ifindex = po->ifindex;
3500	sll->sll_protocol = po->num;
3501	sll->sll_pkttype = 0;
3502	rcu_read_lock();
3503	dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3504	if (dev) {
3505		sll->sll_hatype = dev->type;
3506		sll->sll_halen = dev->addr_len;
3507		memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
 
 
 
 
 
3508	} else {
3509		sll->sll_hatype = 0;	/* Bad: we have no ARPHRD_UNSPEC */
3510		sll->sll_halen = 0;
3511	}
3512	rcu_read_unlock();
3513
3514	return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3515}
3516
3517static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3518			 int what)
3519{
3520	switch (i->type) {
3521	case PACKET_MR_MULTICAST:
3522		if (i->alen != dev->addr_len)
3523			return -EINVAL;
3524		if (what > 0)
3525			return dev_mc_add(dev, i->addr);
3526		else
3527			return dev_mc_del(dev, i->addr);
3528		break;
3529	case PACKET_MR_PROMISC:
3530		return dev_set_promiscuity(dev, what);
3531	case PACKET_MR_ALLMULTI:
3532		return dev_set_allmulti(dev, what);
3533	case PACKET_MR_UNICAST:
3534		if (i->alen != dev->addr_len)
3535			return -EINVAL;
3536		if (what > 0)
3537			return dev_uc_add(dev, i->addr);
3538		else
3539			return dev_uc_del(dev, i->addr);
3540		break;
3541	default:
3542		break;
3543	}
3544	return 0;
3545}
3546
3547static void packet_dev_mclist_delete(struct net_device *dev,
3548				     struct packet_mclist **mlp)
3549{
3550	struct packet_mclist *ml;
3551
3552	while ((ml = *mlp) != NULL) {
3553		if (ml->ifindex == dev->ifindex) {
3554			packet_dev_mc(dev, ml, -1);
3555			*mlp = ml->next;
3556			kfree(ml);
3557		} else
3558			mlp = &ml->next;
3559	}
3560}
3561
3562static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3563{
3564	struct packet_sock *po = pkt_sk(sk);
3565	struct packet_mclist *ml, *i;
3566	struct net_device *dev;
3567	int err;
3568
3569	rtnl_lock();
3570
3571	err = -ENODEV;
3572	dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3573	if (!dev)
3574		goto done;
3575
3576	err = -EINVAL;
3577	if (mreq->mr_alen > dev->addr_len)
3578		goto done;
3579
3580	err = -ENOBUFS;
3581	i = kmalloc(sizeof(*i), GFP_KERNEL);
3582	if (i == NULL)
3583		goto done;
3584
3585	err = 0;
3586	for (ml = po->mclist; ml; ml = ml->next) {
3587		if (ml->ifindex == mreq->mr_ifindex &&
3588		    ml->type == mreq->mr_type &&
3589		    ml->alen == mreq->mr_alen &&
3590		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3591			ml->count++;
3592			/* Free the new element ... */
3593			kfree(i);
3594			goto done;
3595		}
3596	}
3597
3598	i->type = mreq->mr_type;
3599	i->ifindex = mreq->mr_ifindex;
3600	i->alen = mreq->mr_alen;
3601	memcpy(i->addr, mreq->mr_address, i->alen);
3602	memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3603	i->count = 1;
3604	i->next = po->mclist;
3605	po->mclist = i;
3606	err = packet_dev_mc(dev, i, 1);
3607	if (err) {
3608		po->mclist = i->next;
3609		kfree(i);
3610	}
3611
3612done:
3613	rtnl_unlock();
3614	return err;
3615}
3616
3617static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3618{
3619	struct packet_mclist *ml, **mlp;
3620
3621	rtnl_lock();
3622
3623	for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3624		if (ml->ifindex == mreq->mr_ifindex &&
3625		    ml->type == mreq->mr_type &&
3626		    ml->alen == mreq->mr_alen &&
3627		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3628			if (--ml->count == 0) {
3629				struct net_device *dev;
3630				*mlp = ml->next;
3631				dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3632				if (dev)
3633					packet_dev_mc(dev, ml, -1);
3634				kfree(ml);
3635			}
3636			break;
3637		}
3638	}
3639	rtnl_unlock();
3640	return 0;
3641}
3642
3643static void packet_flush_mclist(struct sock *sk)
3644{
3645	struct packet_sock *po = pkt_sk(sk);
3646	struct packet_mclist *ml;
3647
3648	if (!po->mclist)
3649		return;
3650
3651	rtnl_lock();
3652	while ((ml = po->mclist) != NULL) {
3653		struct net_device *dev;
3654
3655		po->mclist = ml->next;
3656		dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3657		if (dev != NULL)
3658			packet_dev_mc(dev, ml, -1);
3659		kfree(ml);
3660	}
3661	rtnl_unlock();
3662}
3663
3664static int
3665packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval,
3666		  unsigned int optlen)
3667{
3668	struct sock *sk = sock->sk;
3669	struct packet_sock *po = pkt_sk(sk);
3670	int ret;
3671
3672	if (level != SOL_PACKET)
3673		return -ENOPROTOOPT;
3674
3675	switch (optname) {
3676	case PACKET_ADD_MEMBERSHIP:
3677	case PACKET_DROP_MEMBERSHIP:
3678	{
3679		struct packet_mreq_max mreq;
3680		int len = optlen;
3681		memset(&mreq, 0, sizeof(mreq));
3682		if (len < sizeof(struct packet_mreq))
3683			return -EINVAL;
3684		if (len > sizeof(mreq))
3685			len = sizeof(mreq);
3686		if (copy_from_sockptr(&mreq, optval, len))
3687			return -EFAULT;
3688		if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3689			return -EINVAL;
3690		if (optname == PACKET_ADD_MEMBERSHIP)
3691			ret = packet_mc_add(sk, &mreq);
3692		else
3693			ret = packet_mc_drop(sk, &mreq);
3694		return ret;
3695	}
3696
3697	case PACKET_RX_RING:
3698	case PACKET_TX_RING:
3699	{
3700		union tpacket_req_u req_u;
3701		int len;
3702
3703		lock_sock(sk);
3704		switch (po->tp_version) {
3705		case TPACKET_V1:
3706		case TPACKET_V2:
3707			len = sizeof(req_u.req);
3708			break;
3709		case TPACKET_V3:
3710		default:
3711			len = sizeof(req_u.req3);
3712			break;
3713		}
3714		if (optlen < len) {
3715			ret = -EINVAL;
3716		} else {
3717			if (copy_from_sockptr(&req_u.req, optval, len))
3718				ret = -EFAULT;
3719			else
3720				ret = packet_set_ring(sk, &req_u, 0,
3721						    optname == PACKET_TX_RING);
3722		}
3723		release_sock(sk);
3724		return ret;
3725	}
3726	case PACKET_COPY_THRESH:
3727	{
3728		int val;
3729
3730		if (optlen != sizeof(val))
3731			return -EINVAL;
3732		if (copy_from_sockptr(&val, optval, sizeof(val)))
3733			return -EFAULT;
3734
3735		pkt_sk(sk)->copy_thresh = val;
3736		return 0;
3737	}
3738	case PACKET_VERSION:
3739	{
3740		int val;
3741
3742		if (optlen != sizeof(val))
3743			return -EINVAL;
3744		if (copy_from_sockptr(&val, optval, sizeof(val)))
3745			return -EFAULT;
3746		switch (val) {
3747		case TPACKET_V1:
3748		case TPACKET_V2:
3749		case TPACKET_V3:
3750			break;
3751		default:
3752			return -EINVAL;
3753		}
3754		lock_sock(sk);
3755		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3756			ret = -EBUSY;
3757		} else {
3758			po->tp_version = val;
3759			ret = 0;
3760		}
3761		release_sock(sk);
3762		return ret;
3763	}
3764	case PACKET_RESERVE:
3765	{
3766		unsigned int val;
3767
3768		if (optlen != sizeof(val))
3769			return -EINVAL;
3770		if (copy_from_sockptr(&val, optval, sizeof(val)))
3771			return -EFAULT;
3772		if (val > INT_MAX)
3773			return -EINVAL;
3774		lock_sock(sk);
3775		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3776			ret = -EBUSY;
3777		} else {
3778			po->tp_reserve = val;
3779			ret = 0;
3780		}
3781		release_sock(sk);
3782		return ret;
3783	}
3784	case PACKET_LOSS:
3785	{
3786		unsigned int val;
3787
3788		if (optlen != sizeof(val))
3789			return -EINVAL;
3790		if (copy_from_sockptr(&val, optval, sizeof(val)))
3791			return -EFAULT;
3792
3793		lock_sock(sk);
3794		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3795			ret = -EBUSY;
3796		} else {
3797			po->tp_loss = !!val;
3798			ret = 0;
3799		}
3800		release_sock(sk);
3801		return ret;
3802	}
3803	case PACKET_AUXDATA:
3804	{
3805		int val;
3806
3807		if (optlen < sizeof(val))
3808			return -EINVAL;
3809		if (copy_from_sockptr(&val, optval, sizeof(val)))
3810			return -EFAULT;
3811
3812		lock_sock(sk);
3813		po->auxdata = !!val;
3814		release_sock(sk);
3815		return 0;
3816	}
3817	case PACKET_ORIGDEV:
3818	{
3819		int val;
3820
3821		if (optlen < sizeof(val))
3822			return -EINVAL;
3823		if (copy_from_sockptr(&val, optval, sizeof(val)))
3824			return -EFAULT;
3825
3826		lock_sock(sk);
3827		po->origdev = !!val;
3828		release_sock(sk);
3829		return 0;
3830	}
3831	case PACKET_VNET_HDR:
 
3832	{
3833		int val;
3834
3835		if (sock->type != SOCK_RAW)
3836			return -EINVAL;
3837		if (optlen < sizeof(val))
3838			return -EINVAL;
3839		if (copy_from_sockptr(&val, optval, sizeof(val)))
3840			return -EFAULT;
3841
 
 
 
 
 
 
 
 
3842		lock_sock(sk);
3843		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3844			ret = -EBUSY;
3845		} else {
3846			po->has_vnet_hdr = !!val;
3847			ret = 0;
3848		}
3849		release_sock(sk);
3850		return ret;
3851	}
3852	case PACKET_TIMESTAMP:
3853	{
3854		int val;
3855
3856		if (optlen != sizeof(val))
3857			return -EINVAL;
3858		if (copy_from_sockptr(&val, optval, sizeof(val)))
3859			return -EFAULT;
3860
3861		po->tp_tstamp = val;
3862		return 0;
3863	}
3864	case PACKET_FANOUT:
3865	{
3866		int val;
3867
3868		if (optlen != sizeof(val))
3869			return -EINVAL;
3870		if (copy_from_sockptr(&val, optval, sizeof(val)))
3871			return -EFAULT;
3872
3873		return fanout_add(sk, val & 0xffff, val >> 16);
3874	}
3875	case PACKET_FANOUT_DATA:
3876	{
3877		if (!po->fanout)
 
3878			return -EINVAL;
3879
3880		return fanout_set_data(po, optval, optlen);
3881	}
3882	case PACKET_IGNORE_OUTGOING:
3883	{
3884		int val;
3885
3886		if (optlen != sizeof(val))
3887			return -EINVAL;
3888		if (copy_from_sockptr(&val, optval, sizeof(val)))
3889			return -EFAULT;
3890		if (val < 0 || val > 1)
3891			return -EINVAL;
3892
3893		po->prot_hook.ignore_outgoing = !!val;
3894		return 0;
3895	}
3896	case PACKET_TX_HAS_OFF:
3897	{
3898		unsigned int val;
3899
3900		if (optlen != sizeof(val))
3901			return -EINVAL;
3902		if (copy_from_sockptr(&val, optval, sizeof(val)))
3903			return -EFAULT;
3904
3905		lock_sock(sk);
3906		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3907			ret = -EBUSY;
3908		} else {
3909			po->tp_tx_has_off = !!val;
3910			ret = 0;
3911		}
3912		release_sock(sk);
3913		return 0;
3914	}
3915	case PACKET_QDISC_BYPASS:
3916	{
3917		int val;
3918
3919		if (optlen != sizeof(val))
3920			return -EINVAL;
3921		if (copy_from_sockptr(&val, optval, sizeof(val)))
3922			return -EFAULT;
3923
3924		po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3925		return 0;
3926	}
3927	default:
3928		return -ENOPROTOOPT;
3929	}
3930}
3931
3932static int packet_getsockopt(struct socket *sock, int level, int optname,
3933			     char __user *optval, int __user *optlen)
3934{
3935	int len;
3936	int val, lv = sizeof(val);
3937	struct sock *sk = sock->sk;
3938	struct packet_sock *po = pkt_sk(sk);
3939	void *data = &val;
3940	union tpacket_stats_u st;
3941	struct tpacket_rollover_stats rstats;
3942	int drops;
3943
3944	if (level != SOL_PACKET)
3945		return -ENOPROTOOPT;
3946
3947	if (get_user(len, optlen))
3948		return -EFAULT;
3949
3950	if (len < 0)
3951		return -EINVAL;
3952
3953	switch (optname) {
3954	case PACKET_STATISTICS:
3955		spin_lock_bh(&sk->sk_receive_queue.lock);
3956		memcpy(&st, &po->stats, sizeof(st));
3957		memset(&po->stats, 0, sizeof(po->stats));
3958		spin_unlock_bh(&sk->sk_receive_queue.lock);
3959		drops = atomic_xchg(&po->tp_drops, 0);
3960
3961		if (po->tp_version == TPACKET_V3) {
3962			lv = sizeof(struct tpacket_stats_v3);
3963			st.stats3.tp_drops = drops;
3964			st.stats3.tp_packets += drops;
3965			data = &st.stats3;
3966		} else {
3967			lv = sizeof(struct tpacket_stats);
3968			st.stats1.tp_drops = drops;
3969			st.stats1.tp_packets += drops;
3970			data = &st.stats1;
3971		}
3972
3973		break;
3974	case PACKET_AUXDATA:
3975		val = po->auxdata;
3976		break;
3977	case PACKET_ORIGDEV:
3978		val = po->origdev;
3979		break;
3980	case PACKET_VNET_HDR:
3981		val = po->has_vnet_hdr;
 
 
 
 
 
 
3982		break;
3983	case PACKET_VERSION:
3984		val = po->tp_version;
3985		break;
3986	case PACKET_HDRLEN:
3987		if (len > sizeof(int))
3988			len = sizeof(int);
3989		if (len < sizeof(int))
3990			return -EINVAL;
3991		if (copy_from_user(&val, optval, len))
3992			return -EFAULT;
3993		switch (val) {
3994		case TPACKET_V1:
3995			val = sizeof(struct tpacket_hdr);
3996			break;
3997		case TPACKET_V2:
3998			val = sizeof(struct tpacket2_hdr);
3999			break;
4000		case TPACKET_V3:
4001			val = sizeof(struct tpacket3_hdr);
4002			break;
4003		default:
4004			return -EINVAL;
4005		}
4006		break;
4007	case PACKET_RESERVE:
4008		val = po->tp_reserve;
4009		break;
4010	case PACKET_LOSS:
4011		val = po->tp_loss;
4012		break;
4013	case PACKET_TIMESTAMP:
4014		val = po->tp_tstamp;
4015		break;
4016	case PACKET_FANOUT:
4017		val = (po->fanout ?
4018		       ((u32)po->fanout->id |
4019			((u32)po->fanout->type << 16) |
4020			((u32)po->fanout->flags << 24)) :
4021		       0);
4022		break;
4023	case PACKET_IGNORE_OUTGOING:
4024		val = po->prot_hook.ignore_outgoing;
4025		break;
4026	case PACKET_ROLLOVER_STATS:
4027		if (!po->rollover)
4028			return -EINVAL;
4029		rstats.tp_all = atomic_long_read(&po->rollover->num);
4030		rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4031		rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4032		data = &rstats;
4033		lv = sizeof(rstats);
4034		break;
4035	case PACKET_TX_HAS_OFF:
4036		val = po->tp_tx_has_off;
4037		break;
4038	case PACKET_QDISC_BYPASS:
4039		val = packet_use_direct_xmit(po);
4040		break;
4041	default:
4042		return -ENOPROTOOPT;
4043	}
4044
4045	if (len > lv)
4046		len = lv;
4047	if (put_user(len, optlen))
4048		return -EFAULT;
4049	if (copy_to_user(optval, data, len))
4050		return -EFAULT;
4051	return 0;
4052}
4053
4054static int packet_notifier(struct notifier_block *this,
4055			   unsigned long msg, void *ptr)
4056{
4057	struct sock *sk;
4058	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4059	struct net *net = dev_net(dev);
4060
4061	rcu_read_lock();
4062	sk_for_each_rcu(sk, &net->packet.sklist) {
4063		struct packet_sock *po = pkt_sk(sk);
4064
4065		switch (msg) {
4066		case NETDEV_UNREGISTER:
4067			if (po->mclist)
4068				packet_dev_mclist_delete(dev, &po->mclist);
4069			fallthrough;
4070
4071		case NETDEV_DOWN:
4072			if (dev->ifindex == po->ifindex) {
4073				spin_lock(&po->bind_lock);
4074				if (po->running) {
4075					__unregister_prot_hook(sk, false);
4076					sk->sk_err = ENETDOWN;
4077					if (!sock_flag(sk, SOCK_DEAD))
4078						sk->sk_error_report(sk);
4079				}
4080				if (msg == NETDEV_UNREGISTER) {
4081					packet_cached_dev_reset(po);
4082					po->ifindex = -1;
4083					if (po->prot_hook.dev)
4084						dev_put(po->prot_hook.dev);
4085					po->prot_hook.dev = NULL;
4086				}
4087				spin_unlock(&po->bind_lock);
4088			}
4089			break;
4090		case NETDEV_UP:
4091			if (dev->ifindex == po->ifindex) {
4092				spin_lock(&po->bind_lock);
4093				if (po->num)
4094					register_prot_hook(sk);
4095				spin_unlock(&po->bind_lock);
4096			}
4097			break;
4098		}
4099	}
4100	rcu_read_unlock();
4101	return NOTIFY_DONE;
4102}
4103
4104
4105static int packet_ioctl(struct socket *sock, unsigned int cmd,
4106			unsigned long arg)
4107{
4108	struct sock *sk = sock->sk;
4109
4110	switch (cmd) {
4111	case SIOCOUTQ:
4112	{
4113		int amount = sk_wmem_alloc_get(sk);
4114
4115		return put_user(amount, (int __user *)arg);
4116	}
4117	case SIOCINQ:
4118	{
4119		struct sk_buff *skb;
4120		int amount = 0;
4121
4122		spin_lock_bh(&sk->sk_receive_queue.lock);
4123		skb = skb_peek(&sk->sk_receive_queue);
4124		if (skb)
4125			amount = skb->len;
4126		spin_unlock_bh(&sk->sk_receive_queue.lock);
4127		return put_user(amount, (int __user *)arg);
4128	}
4129#ifdef CONFIG_INET
4130	case SIOCADDRT:
4131	case SIOCDELRT:
4132	case SIOCDARP:
4133	case SIOCGARP:
4134	case SIOCSARP:
4135	case SIOCGIFADDR:
4136	case SIOCSIFADDR:
4137	case SIOCGIFBRDADDR:
4138	case SIOCSIFBRDADDR:
4139	case SIOCGIFNETMASK:
4140	case SIOCSIFNETMASK:
4141	case SIOCGIFDSTADDR:
4142	case SIOCSIFDSTADDR:
4143	case SIOCSIFFLAGS:
4144		return inet_dgram_ops.ioctl(sock, cmd, arg);
4145#endif
4146
4147	default:
4148		return -ENOIOCTLCMD;
4149	}
4150	return 0;
4151}
4152
4153static __poll_t packet_poll(struct file *file, struct socket *sock,
4154				poll_table *wait)
4155{
4156	struct sock *sk = sock->sk;
4157	struct packet_sock *po = pkt_sk(sk);
4158	__poll_t mask = datagram_poll(file, sock, wait);
4159
4160	spin_lock_bh(&sk->sk_receive_queue.lock);
4161	if (po->rx_ring.pg_vec) {
4162		if (!packet_previous_rx_frame(po, &po->rx_ring,
4163			TP_STATUS_KERNEL))
4164			mask |= EPOLLIN | EPOLLRDNORM;
4165	}
4166	packet_rcv_try_clear_pressure(po);
4167	spin_unlock_bh(&sk->sk_receive_queue.lock);
4168	spin_lock_bh(&sk->sk_write_queue.lock);
4169	if (po->tx_ring.pg_vec) {
4170		if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4171			mask |= EPOLLOUT | EPOLLWRNORM;
4172	}
4173	spin_unlock_bh(&sk->sk_write_queue.lock);
4174	return mask;
4175}
4176
4177
4178/* Dirty? Well, I still did not learn better way to account
4179 * for user mmaps.
4180 */
4181
4182static void packet_mm_open(struct vm_area_struct *vma)
4183{
4184	struct file *file = vma->vm_file;
4185	struct socket *sock = file->private_data;
4186	struct sock *sk = sock->sk;
4187
4188	if (sk)
4189		atomic_inc(&pkt_sk(sk)->mapped);
4190}
4191
4192static void packet_mm_close(struct vm_area_struct *vma)
4193{
4194	struct file *file = vma->vm_file;
4195	struct socket *sock = file->private_data;
4196	struct sock *sk = sock->sk;
4197
4198	if (sk)
4199		atomic_dec(&pkt_sk(sk)->mapped);
4200}
4201
4202static const struct vm_operations_struct packet_mmap_ops = {
4203	.open	=	packet_mm_open,
4204	.close	=	packet_mm_close,
4205};
4206
4207static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4208			unsigned int len)
4209{
4210	int i;
4211
4212	for (i = 0; i < len; i++) {
4213		if (likely(pg_vec[i].buffer)) {
4214			if (is_vmalloc_addr(pg_vec[i].buffer))
4215				vfree(pg_vec[i].buffer);
4216			else
4217				free_pages((unsigned long)pg_vec[i].buffer,
4218					   order);
4219			pg_vec[i].buffer = NULL;
4220		}
4221	}
4222	kfree(pg_vec);
4223}
4224
4225static char *alloc_one_pg_vec_page(unsigned long order)
4226{
4227	char *buffer;
4228	gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4229			  __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4230
4231	buffer = (char *) __get_free_pages(gfp_flags, order);
4232	if (buffer)
4233		return buffer;
4234
4235	/* __get_free_pages failed, fall back to vmalloc */
4236	buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4237	if (buffer)
4238		return buffer;
4239
4240	/* vmalloc failed, lets dig into swap here */
4241	gfp_flags &= ~__GFP_NORETRY;
4242	buffer = (char *) __get_free_pages(gfp_flags, order);
4243	if (buffer)
4244		return buffer;
4245
4246	/* complete and utter failure */
4247	return NULL;
4248}
4249
4250static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4251{
4252	unsigned int block_nr = req->tp_block_nr;
4253	struct pgv *pg_vec;
4254	int i;
4255
4256	pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4257	if (unlikely(!pg_vec))
4258		goto out;
4259
4260	for (i = 0; i < block_nr; i++) {
4261		pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4262		if (unlikely(!pg_vec[i].buffer))
4263			goto out_free_pgvec;
4264	}
4265
4266out:
4267	return pg_vec;
4268
4269out_free_pgvec:
4270	free_pg_vec(pg_vec, order, block_nr);
4271	pg_vec = NULL;
4272	goto out;
4273}
4274
4275static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4276		int closing, int tx_ring)
4277{
4278	struct pgv *pg_vec = NULL;
4279	struct packet_sock *po = pkt_sk(sk);
4280	unsigned long *rx_owner_map = NULL;
4281	int was_running, order = 0;
4282	struct packet_ring_buffer *rb;
4283	struct sk_buff_head *rb_queue;
4284	__be16 num;
4285	int err;
4286	/* Added to avoid minimal code churn */
4287	struct tpacket_req *req = &req_u->req;
4288
4289	rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4290	rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4291
4292	err = -EBUSY;
4293	if (!closing) {
4294		if (atomic_read(&po->mapped))
4295			goto out;
4296		if (packet_read_pending(rb))
4297			goto out;
4298	}
4299
4300	if (req->tp_block_nr) {
4301		unsigned int min_frame_size;
4302
4303		/* Sanity tests and some calculations */
4304		err = -EBUSY;
4305		if (unlikely(rb->pg_vec))
4306			goto out;
4307
4308		switch (po->tp_version) {
4309		case TPACKET_V1:
4310			po->tp_hdrlen = TPACKET_HDRLEN;
4311			break;
4312		case TPACKET_V2:
4313			po->tp_hdrlen = TPACKET2_HDRLEN;
4314			break;
4315		case TPACKET_V3:
4316			po->tp_hdrlen = TPACKET3_HDRLEN;
4317			break;
4318		}
4319
4320		err = -EINVAL;
4321		if (unlikely((int)req->tp_block_size <= 0))
4322			goto out;
4323		if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4324			goto out;
4325		min_frame_size = po->tp_hdrlen + po->tp_reserve;
4326		if (po->tp_version >= TPACKET_V3 &&
4327		    req->tp_block_size <
4328		    BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4329			goto out;
4330		if (unlikely(req->tp_frame_size < min_frame_size))
4331			goto out;
4332		if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4333			goto out;
4334
4335		rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4336		if (unlikely(rb->frames_per_block == 0))
4337			goto out;
4338		if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4339			goto out;
4340		if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4341					req->tp_frame_nr))
4342			goto out;
4343
4344		err = -ENOMEM;
4345		order = get_order(req->tp_block_size);
4346		pg_vec = alloc_pg_vec(req, order);
4347		if (unlikely(!pg_vec))
4348			goto out;
4349		switch (po->tp_version) {
4350		case TPACKET_V3:
4351			/* Block transmit is not supported yet */
4352			if (!tx_ring) {
4353				init_prb_bdqc(po, rb, pg_vec, req_u);
4354			} else {
4355				struct tpacket_req3 *req3 = &req_u->req3;
4356
4357				if (req3->tp_retire_blk_tov ||
4358				    req3->tp_sizeof_priv ||
4359				    req3->tp_feature_req_word) {
4360					err = -EINVAL;
4361					goto out_free_pg_vec;
4362				}
4363			}
4364			break;
4365		default:
4366			if (!tx_ring) {
4367				rx_owner_map = bitmap_alloc(req->tp_frame_nr,
4368					GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
4369				if (!rx_owner_map)
4370					goto out_free_pg_vec;
4371			}
4372			break;
4373		}
4374	}
4375	/* Done */
4376	else {
4377		err = -EINVAL;
4378		if (unlikely(req->tp_frame_nr))
4379			goto out;
4380	}
4381
4382
4383	/* Detach socket from network */
4384	spin_lock(&po->bind_lock);
4385	was_running = po->running;
4386	num = po->num;
4387	if (was_running) {
4388		po->num = 0;
4389		__unregister_prot_hook(sk, false);
4390	}
4391	spin_unlock(&po->bind_lock);
4392
4393	synchronize_net();
4394
4395	err = -EBUSY;
4396	mutex_lock(&po->pg_vec_lock);
4397	if (closing || atomic_read(&po->mapped) == 0) {
4398		err = 0;
4399		spin_lock_bh(&rb_queue->lock);
4400		swap(rb->pg_vec, pg_vec);
4401		if (po->tp_version <= TPACKET_V2)
4402			swap(rb->rx_owner_map, rx_owner_map);
4403		rb->frame_max = (req->tp_frame_nr - 1);
4404		rb->head = 0;
4405		rb->frame_size = req->tp_frame_size;
4406		spin_unlock_bh(&rb_queue->lock);
4407
4408		swap(rb->pg_vec_order, order);
4409		swap(rb->pg_vec_len, req->tp_block_nr);
4410
4411		rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4412		po->prot_hook.func = (po->rx_ring.pg_vec) ?
4413						tpacket_rcv : packet_rcv;
4414		skb_queue_purge(rb_queue);
4415		if (atomic_read(&po->mapped))
4416			pr_err("packet_mmap: vma is busy: %d\n",
4417			       atomic_read(&po->mapped));
4418	}
4419	mutex_unlock(&po->pg_vec_lock);
4420
4421	spin_lock(&po->bind_lock);
4422	if (was_running) {
4423		po->num = num;
4424		register_prot_hook(sk);
4425	}
4426	spin_unlock(&po->bind_lock);
4427	if (pg_vec && (po->tp_version > TPACKET_V2)) {
4428		/* Because we don't support block-based V3 on tx-ring */
4429		if (!tx_ring)
4430			prb_shutdown_retire_blk_timer(po, rb_queue);
4431	}
4432
4433out_free_pg_vec:
4434	bitmap_free(rx_owner_map);
4435	if (pg_vec)
4436		free_pg_vec(pg_vec, order, req->tp_block_nr);
 
4437out:
4438	return err;
4439}
4440
4441static int packet_mmap(struct file *file, struct socket *sock,
4442		struct vm_area_struct *vma)
4443{
4444	struct sock *sk = sock->sk;
4445	struct packet_sock *po = pkt_sk(sk);
4446	unsigned long size, expected_size;
4447	struct packet_ring_buffer *rb;
4448	unsigned long start;
4449	int err = -EINVAL;
4450	int i;
4451
4452	if (vma->vm_pgoff)
4453		return -EINVAL;
4454
4455	mutex_lock(&po->pg_vec_lock);
4456
4457	expected_size = 0;
4458	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4459		if (rb->pg_vec) {
4460			expected_size += rb->pg_vec_len
4461						* rb->pg_vec_pages
4462						* PAGE_SIZE;
4463		}
4464	}
4465
4466	if (expected_size == 0)
4467		goto out;
4468
4469	size = vma->vm_end - vma->vm_start;
4470	if (size != expected_size)
4471		goto out;
4472
4473	start = vma->vm_start;
4474	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4475		if (rb->pg_vec == NULL)
4476			continue;
4477
4478		for (i = 0; i < rb->pg_vec_len; i++) {
4479			struct page *page;
4480			void *kaddr = rb->pg_vec[i].buffer;
4481			int pg_num;
4482
4483			for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4484				page = pgv_to_page(kaddr);
4485				err = vm_insert_page(vma, start, page);
4486				if (unlikely(err))
4487					goto out;
4488				start += PAGE_SIZE;
4489				kaddr += PAGE_SIZE;
4490			}
4491		}
4492	}
4493
4494	atomic_inc(&po->mapped);
4495	vma->vm_ops = &packet_mmap_ops;
4496	err = 0;
4497
4498out:
4499	mutex_unlock(&po->pg_vec_lock);
4500	return err;
4501}
4502
4503static const struct proto_ops packet_ops_spkt = {
4504	.family =	PF_PACKET,
4505	.owner =	THIS_MODULE,
4506	.release =	packet_release,
4507	.bind =		packet_bind_spkt,
4508	.connect =	sock_no_connect,
4509	.socketpair =	sock_no_socketpair,
4510	.accept =	sock_no_accept,
4511	.getname =	packet_getname_spkt,
4512	.poll =		datagram_poll,
4513	.ioctl =	packet_ioctl,
4514	.gettstamp =	sock_gettstamp,
4515	.listen =	sock_no_listen,
4516	.shutdown =	sock_no_shutdown,
4517	.sendmsg =	packet_sendmsg_spkt,
4518	.recvmsg =	packet_recvmsg,
4519	.mmap =		sock_no_mmap,
4520	.sendpage =	sock_no_sendpage,
4521};
4522
4523static const struct proto_ops packet_ops = {
4524	.family =	PF_PACKET,
4525	.owner =	THIS_MODULE,
4526	.release =	packet_release,
4527	.bind =		packet_bind,
4528	.connect =	sock_no_connect,
4529	.socketpair =	sock_no_socketpair,
4530	.accept =	sock_no_accept,
4531	.getname =	packet_getname,
4532	.poll =		packet_poll,
4533	.ioctl =	packet_ioctl,
4534	.gettstamp =	sock_gettstamp,
4535	.listen =	sock_no_listen,
4536	.shutdown =	sock_no_shutdown,
4537	.setsockopt =	packet_setsockopt,
4538	.getsockopt =	packet_getsockopt,
4539	.sendmsg =	packet_sendmsg,
4540	.recvmsg =	packet_recvmsg,
4541	.mmap =		packet_mmap,
4542	.sendpage =	sock_no_sendpage,
4543};
4544
4545static const struct net_proto_family packet_family_ops = {
4546	.family =	PF_PACKET,
4547	.create =	packet_create,
4548	.owner	=	THIS_MODULE,
4549};
4550
4551static struct notifier_block packet_netdev_notifier = {
4552	.notifier_call =	packet_notifier,
4553};
4554
4555#ifdef CONFIG_PROC_FS
4556
4557static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4558	__acquires(RCU)
4559{
4560	struct net *net = seq_file_net(seq);
4561
4562	rcu_read_lock();
4563	return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4564}
4565
4566static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4567{
4568	struct net *net = seq_file_net(seq);
4569	return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4570}
4571
4572static void packet_seq_stop(struct seq_file *seq, void *v)
4573	__releases(RCU)
4574{
4575	rcu_read_unlock();
4576}
4577
4578static int packet_seq_show(struct seq_file *seq, void *v)
4579{
4580	if (v == SEQ_START_TOKEN)
4581		seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
 
 
4582	else {
4583		struct sock *s = sk_entry(v);
4584		const struct packet_sock *po = pkt_sk(s);
4585
4586		seq_printf(seq,
4587			   "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4588			   s,
4589			   refcount_read(&s->sk_refcnt),
4590			   s->sk_type,
4591			   ntohs(po->num),
4592			   po->ifindex,
4593			   po->running,
4594			   atomic_read(&s->sk_rmem_alloc),
4595			   from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4596			   sock_i_ino(s));
4597	}
4598
4599	return 0;
4600}
4601
4602static const struct seq_operations packet_seq_ops = {
4603	.start	= packet_seq_start,
4604	.next	= packet_seq_next,
4605	.stop	= packet_seq_stop,
4606	.show	= packet_seq_show,
4607};
4608#endif
4609
4610static int __net_init packet_net_init(struct net *net)
4611{
4612	mutex_init(&net->packet.sklist_lock);
4613	INIT_HLIST_HEAD(&net->packet.sklist);
4614
 
4615	if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4616			sizeof(struct seq_net_private)))
4617		return -ENOMEM;
 
4618
4619	return 0;
4620}
4621
4622static void __net_exit packet_net_exit(struct net *net)
4623{
4624	remove_proc_entry("packet", net->proc_net);
4625	WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4626}
4627
4628static struct pernet_operations packet_net_ops = {
4629	.init = packet_net_init,
4630	.exit = packet_net_exit,
4631};
4632
4633
4634static void __exit packet_exit(void)
4635{
4636	unregister_netdevice_notifier(&packet_netdev_notifier);
4637	unregister_pernet_subsys(&packet_net_ops);
4638	sock_unregister(PF_PACKET);
4639	proto_unregister(&packet_proto);
 
 
4640}
4641
4642static int __init packet_init(void)
4643{
4644	int rc;
4645
4646	rc = proto_register(&packet_proto, 0);
4647	if (rc)
4648		goto out;
4649	rc = sock_register(&packet_family_ops);
4650	if (rc)
4651		goto out_proto;
4652	rc = register_pernet_subsys(&packet_net_ops);
4653	if (rc)
4654		goto out_sock;
4655	rc = register_netdevice_notifier(&packet_netdev_notifier);
4656	if (rc)
4657		goto out_pernet;
 
 
 
 
 
 
4658
4659	return 0;
4660
4661out_pernet:
4662	unregister_pernet_subsys(&packet_net_ops);
4663out_sock:
4664	sock_unregister(PF_PACKET);
4665out_proto:
4666	proto_unregister(&packet_proto);
 
 
 
 
4667out:
4668	return rc;
4669}
4670
4671module_init(packet_init);
4672module_exit(packet_exit);
 
4673MODULE_LICENSE("GPL");
4674MODULE_ALIAS_NETPROTO(PF_PACKET);
v6.9.4
   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 void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 542{
 543	del_timer_sync(&pkc->retire_blk_timer);
 544}
 545
 546static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 547		struct sk_buff_head *rb_queue)
 548{
 549	struct tpacket_kbdq_core *pkc;
 550
 551	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 552
 553	spin_lock_bh(&rb_queue->lock);
 554	pkc->delete_blk_timer = 1;
 555	spin_unlock_bh(&rb_queue->lock);
 556
 557	prb_del_retire_blk_timer(pkc);
 558}
 559
 560static void prb_setup_retire_blk_timer(struct packet_sock *po)
 561{
 562	struct tpacket_kbdq_core *pkc;
 563
 564	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 565	timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
 566		    0);
 567	pkc->retire_blk_timer.expires = jiffies;
 568}
 569
 570static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 571				int blk_size_in_bytes)
 572{
 573	struct net_device *dev;
 574	unsigned int mbits, div;
 575	struct ethtool_link_ksettings ecmd;
 576	int err;
 577
 578	rtnl_lock();
 579	dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 580	if (unlikely(!dev)) {
 581		rtnl_unlock();
 582		return DEFAULT_PRB_RETIRE_TOV;
 583	}
 584	err = __ethtool_get_link_ksettings(dev, &ecmd);
 585	rtnl_unlock();
 586	if (err)
 587		return DEFAULT_PRB_RETIRE_TOV;
 588
 589	/* If the link speed is so slow you don't really
 590	 * need to worry about perf anyways
 591	 */
 592	if (ecmd.base.speed < SPEED_1000 ||
 593	    ecmd.base.speed == SPEED_UNKNOWN)
 594		return DEFAULT_PRB_RETIRE_TOV;
 595
 596	div = ecmd.base.speed / 1000;
 597	mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 598
 599	if (div)
 600		mbits /= div;
 601
 602	if (div)
 603		return mbits + 1;
 604	return mbits;
 605}
 606
 607static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 608			union tpacket_req_u *req_u)
 609{
 610	p1->feature_req_word = req_u->req3.tp_feature_req_word;
 611}
 612
 613static void init_prb_bdqc(struct packet_sock *po,
 614			struct packet_ring_buffer *rb,
 615			struct pgv *pg_vec,
 616			union tpacket_req_u *req_u)
 617{
 618	struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
 619	struct tpacket_block_desc *pbd;
 620
 621	memset(p1, 0x0, sizeof(*p1));
 622
 623	p1->knxt_seq_num = 1;
 624	p1->pkbdq = pg_vec;
 625	pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 626	p1->pkblk_start	= pg_vec[0].buffer;
 627	p1->kblk_size = req_u->req3.tp_block_size;
 628	p1->knum_blocks	= req_u->req3.tp_block_nr;
 629	p1->hdrlen = po->tp_hdrlen;
 630	p1->version = po->tp_version;
 631	p1->last_kactive_blk_num = 0;
 632	po->stats.stats3.tp_freeze_q_cnt = 0;
 633	if (req_u->req3.tp_retire_blk_tov)
 634		p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 635	else
 636		p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 637						req_u->req3.tp_block_size);
 638	p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 639	p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 640	rwlock_init(&p1->blk_fill_in_prog_lock);
 641
 642	p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
 643	prb_init_ft_ops(p1, req_u);
 644	prb_setup_retire_blk_timer(po);
 645	prb_open_block(p1, pbd);
 646}
 647
 648/*  Do NOT update the last_blk_num first.
 649 *  Assumes sk_buff_head lock is held.
 650 */
 651static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 652{
 653	mod_timer(&pkc->retire_blk_timer,
 654			jiffies + pkc->tov_in_jiffies);
 655	pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 656}
 657
 658/*
 659 * Timer logic:
 660 * 1) We refresh the timer only when we open a block.
 661 *    By doing this we don't waste cycles refreshing the timer
 662 *	  on packet-by-packet basis.
 663 *
 664 * With a 1MB block-size, on a 1Gbps line, it will take
 665 * i) ~8 ms to fill a block + ii) memcpy etc.
 666 * In this cut we are not accounting for the memcpy time.
 667 *
 668 * So, if the user sets the 'tmo' to 10ms then the timer
 669 * will never fire while the block is still getting filled
 670 * (which is what we want). However, the user could choose
 671 * to close a block early and that's fine.
 672 *
 673 * But when the timer does fire, we check whether or not to refresh it.
 674 * Since the tmo granularity is in msecs, it is not too expensive
 675 * to refresh the timer, lets say every '8' msecs.
 676 * Either the user can set the 'tmo' or we can derive it based on
 677 * a) line-speed and b) block-size.
 678 * prb_calc_retire_blk_tmo() calculates the tmo.
 679 *
 680 */
 681static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
 682{
 683	struct packet_sock *po =
 684		from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
 685	struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 686	unsigned int frozen;
 687	struct tpacket_block_desc *pbd;
 688
 689	spin_lock(&po->sk.sk_receive_queue.lock);
 690
 691	frozen = prb_queue_frozen(pkc);
 692	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 693
 694	if (unlikely(pkc->delete_blk_timer))
 695		goto out;
 696
 697	/* We only need to plug the race when the block is partially filled.
 698	 * tpacket_rcv:
 699	 *		lock(); increment BLOCK_NUM_PKTS; unlock()
 700	 *		copy_bits() is in progress ...
 701	 *		timer fires on other cpu:
 702	 *		we can't retire the current block because copy_bits
 703	 *		is in progress.
 704	 *
 705	 */
 706	if (BLOCK_NUM_PKTS(pbd)) {
 707		/* Waiting for skb_copy_bits to finish... */
 708		write_lock(&pkc->blk_fill_in_prog_lock);
 709		write_unlock(&pkc->blk_fill_in_prog_lock);
 710	}
 711
 712	if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 713		if (!frozen) {
 714			if (!BLOCK_NUM_PKTS(pbd)) {
 715				/* An empty block. Just refresh the timer. */
 716				goto refresh_timer;
 717			}
 718			prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 719			if (!prb_dispatch_next_block(pkc, po))
 720				goto refresh_timer;
 721			else
 722				goto out;
 723		} else {
 724			/* Case 1. Queue was frozen because user-space was
 725			 *	   lagging behind.
 726			 */
 727			if (prb_curr_blk_in_use(pbd)) {
 728				/*
 729				 * Ok, user-space is still behind.
 730				 * So just refresh the timer.
 731				 */
 732				goto refresh_timer;
 733			} else {
 734			       /* Case 2. queue was frozen,user-space caught up,
 735				* now the link went idle && the timer fired.
 736				* We don't have a block to close.So we open this
 737				* block and restart the timer.
 738				* opening a block thaws the queue,restarts timer
 739				* Thawing/timer-refresh is a side effect.
 740				*/
 741				prb_open_block(pkc, pbd);
 742				goto out;
 743			}
 744		}
 745	}
 746
 747refresh_timer:
 748	_prb_refresh_rx_retire_blk_timer(pkc);
 749
 750out:
 751	spin_unlock(&po->sk.sk_receive_queue.lock);
 752}
 753
 754static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 755		struct tpacket_block_desc *pbd1, __u32 status)
 756{
 757	/* Flush everything minus the block header */
 758
 759#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 760	u8 *start, *end;
 761
 762	start = (u8 *)pbd1;
 763
 764	/* Skip the block header(we know header WILL fit in 4K) */
 765	start += PAGE_SIZE;
 766
 767	end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 768	for (; start < end; start += PAGE_SIZE)
 769		flush_dcache_page(pgv_to_page(start));
 770
 771	smp_wmb();
 772#endif
 773
 774	/* Now update the block status. */
 775
 776	BLOCK_STATUS(pbd1) = status;
 777
 778	/* Flush the block header */
 779
 780#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 781	start = (u8 *)pbd1;
 782	flush_dcache_page(pgv_to_page(start));
 783
 784	smp_wmb();
 785#endif
 786}
 787
 788/*
 789 * Side effect:
 790 *
 791 * 1) flush the block
 792 * 2) Increment active_blk_num
 793 *
 794 * Note:We DONT refresh the timer on purpose.
 795 *	Because almost always the next block will be opened.
 796 */
 797static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 798		struct tpacket_block_desc *pbd1,
 799		struct packet_sock *po, unsigned int stat)
 800{
 801	__u32 status = TP_STATUS_USER | stat;
 802
 803	struct tpacket3_hdr *last_pkt;
 804	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 805	struct sock *sk = &po->sk;
 806
 807	if (atomic_read(&po->tp_drops))
 808		status |= TP_STATUS_LOSING;
 809
 810	last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 811	last_pkt->tp_next_offset = 0;
 812
 813	/* Get the ts of the last pkt */
 814	if (BLOCK_NUM_PKTS(pbd1)) {
 815		h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 816		h1->ts_last_pkt.ts_nsec	= last_pkt->tp_nsec;
 817	} else {
 818		/* Ok, we tmo'd - so get the current time.
 819		 *
 820		 * It shouldn't really happen as we don't close empty
 821		 * blocks. See prb_retire_rx_blk_timer_expired().
 822		 */
 823		struct timespec64 ts;
 824		ktime_get_real_ts64(&ts);
 825		h1->ts_last_pkt.ts_sec = ts.tv_sec;
 826		h1->ts_last_pkt.ts_nsec	= ts.tv_nsec;
 827	}
 828
 829	smp_wmb();
 830
 831	/* Flush the block */
 832	prb_flush_block(pkc1, pbd1, status);
 833
 834	sk->sk_data_ready(sk);
 835
 836	pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 837}
 838
 839static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 840{
 841	pkc->reset_pending_on_curr_blk = 0;
 842}
 843
 844/*
 845 * Side effect of opening a block:
 846 *
 847 * 1) prb_queue is thawed.
 848 * 2) retire_blk_timer is refreshed.
 849 *
 850 */
 851static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 852	struct tpacket_block_desc *pbd1)
 853{
 854	struct timespec64 ts;
 855	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 856
 857	smp_rmb();
 858
 859	/* We could have just memset this but we will lose the
 860	 * flexibility of making the priv area sticky
 861	 */
 862
 863	BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 864	BLOCK_NUM_PKTS(pbd1) = 0;
 865	BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 866
 867	ktime_get_real_ts64(&ts);
 868
 869	h1->ts_first_pkt.ts_sec = ts.tv_sec;
 870	h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 871
 872	pkc1->pkblk_start = (char *)pbd1;
 873	pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 874
 875	BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 876	BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 877
 878	pbd1->version = pkc1->version;
 879	pkc1->prev = pkc1->nxt_offset;
 880	pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 881
 882	prb_thaw_queue(pkc1);
 883	_prb_refresh_rx_retire_blk_timer(pkc1);
 884
 885	smp_wmb();
 886}
 887
 888/*
 889 * Queue freeze logic:
 890 * 1) Assume tp_block_nr = 8 blocks.
 891 * 2) At time 't0', user opens Rx ring.
 892 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 893 * 4) user-space is either sleeping or processing block '0'.
 894 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 895 *    it will close block-7,loop around and try to fill block '0'.
 896 *    call-flow:
 897 *    __packet_lookup_frame_in_block
 898 *      prb_retire_current_block()
 899 *      prb_dispatch_next_block()
 900 *        |->(BLOCK_STATUS == USER) evaluates to true
 901 *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 902 * 6) Now there are two cases:
 903 *    6.1) Link goes idle right after the queue is frozen.
 904 *         But remember, the last open_block() refreshed the timer.
 905 *         When this timer expires,it will refresh itself so that we can
 906 *         re-open block-0 in near future.
 907 *    6.2) Link is busy and keeps on receiving packets. This is a simple
 908 *         case and __packet_lookup_frame_in_block will check if block-0
 909 *         is free and can now be re-used.
 910 */
 911static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 912				  struct packet_sock *po)
 913{
 914	pkc->reset_pending_on_curr_blk = 1;
 915	po->stats.stats3.tp_freeze_q_cnt++;
 916}
 917
 918#define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 919
 920/*
 921 * If the next block is free then we will dispatch it
 922 * and return a good offset.
 923 * Else, we will freeze the queue.
 924 * So, caller must check the return value.
 925 */
 926static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 927		struct packet_sock *po)
 928{
 929	struct tpacket_block_desc *pbd;
 930
 931	smp_rmb();
 932
 933	/* 1. Get current block num */
 934	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 935
 936	/* 2. If this block is currently in_use then freeze the queue */
 937	if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 938		prb_freeze_queue(pkc, po);
 939		return NULL;
 940	}
 941
 942	/*
 943	 * 3.
 944	 * open this block and return the offset where the first packet
 945	 * needs to get stored.
 946	 */
 947	prb_open_block(pkc, pbd);
 948	return (void *)pkc->nxt_offset;
 949}
 950
 951static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 952		struct packet_sock *po, unsigned int status)
 953{
 954	struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 955
 956	/* retire/close the current block */
 957	if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 958		/*
 959		 * Plug the case where copy_bits() is in progress on
 960		 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 961		 * have space to copy the pkt in the current block and
 962		 * called prb_retire_current_block()
 963		 *
 964		 * We don't need to worry about the TMO case because
 965		 * the timer-handler already handled this case.
 966		 */
 967		if (!(status & TP_STATUS_BLK_TMO)) {
 968			/* Waiting for skb_copy_bits to finish... */
 969			write_lock(&pkc->blk_fill_in_prog_lock);
 970			write_unlock(&pkc->blk_fill_in_prog_lock);
 971		}
 972		prb_close_block(pkc, pbd, po, status);
 973		return;
 974	}
 975}
 976
 977static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
 978{
 979	return TP_STATUS_USER & BLOCK_STATUS(pbd);
 980}
 981
 982static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 983{
 984	return pkc->reset_pending_on_curr_blk;
 985}
 986
 987static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 988	__releases(&pkc->blk_fill_in_prog_lock)
 989{
 990	struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 991
 992	read_unlock(&pkc->blk_fill_in_prog_lock);
 993}
 994
 995static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 996			struct tpacket3_hdr *ppd)
 997{
 998	ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
 999}
1000
1001static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
1002			struct tpacket3_hdr *ppd)
1003{
1004	ppd->hv1.tp_rxhash = 0;
1005}
1006
1007static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1008			struct tpacket3_hdr *ppd)
1009{
1010	if (skb_vlan_tag_present(pkc->skb)) {
1011		ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1012		ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1013		ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1014	} else {
1015		ppd->hv1.tp_vlan_tci = 0;
1016		ppd->hv1.tp_vlan_tpid = 0;
1017		ppd->tp_status = TP_STATUS_AVAILABLE;
1018	}
1019}
1020
1021static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1022			struct tpacket3_hdr *ppd)
1023{
1024	ppd->hv1.tp_padding = 0;
1025	prb_fill_vlan_info(pkc, ppd);
1026
1027	if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1028		prb_fill_rxhash(pkc, ppd);
1029	else
1030		prb_clear_rxhash(pkc, ppd);
1031}
1032
1033static void prb_fill_curr_block(char *curr,
1034				struct tpacket_kbdq_core *pkc,
1035				struct tpacket_block_desc *pbd,
1036				unsigned int len)
1037	__acquires(&pkc->blk_fill_in_prog_lock)
1038{
1039	struct tpacket3_hdr *ppd;
1040
1041	ppd  = (struct tpacket3_hdr *)curr;
1042	ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1043	pkc->prev = curr;
1044	pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1045	BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1046	BLOCK_NUM_PKTS(pbd) += 1;
1047	read_lock(&pkc->blk_fill_in_prog_lock);
1048	prb_run_all_ft_ops(pkc, ppd);
1049}
1050
1051/* Assumes caller has the sk->rx_queue.lock */
1052static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1053					    struct sk_buff *skb,
1054					    unsigned int len
1055					    )
1056{
1057	struct tpacket_kbdq_core *pkc;
1058	struct tpacket_block_desc *pbd;
1059	char *curr, *end;
1060
1061	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1062	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1063
1064	/* Queue is frozen when user space is lagging behind */
1065	if (prb_queue_frozen(pkc)) {
1066		/*
1067		 * Check if that last block which caused the queue to freeze,
1068		 * is still in_use by user-space.
1069		 */
1070		if (prb_curr_blk_in_use(pbd)) {
1071			/* Can't record this packet */
1072			return NULL;
1073		} else {
1074			/*
1075			 * Ok, the block was released by user-space.
1076			 * Now let's open that block.
1077			 * opening a block also thaws the queue.
1078			 * Thawing is a side effect.
1079			 */
1080			prb_open_block(pkc, pbd);
1081		}
1082	}
1083
1084	smp_mb();
1085	curr = pkc->nxt_offset;
1086	pkc->skb = skb;
1087	end = (char *)pbd + pkc->kblk_size;
1088
1089	/* first try the current block */
1090	if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1091		prb_fill_curr_block(curr, pkc, pbd, len);
1092		return (void *)curr;
1093	}
1094
1095	/* Ok, close the current block */
1096	prb_retire_current_block(pkc, po, 0);
1097
1098	/* Now, try to dispatch the next block */
1099	curr = (char *)prb_dispatch_next_block(pkc, po);
1100	if (curr) {
1101		pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1102		prb_fill_curr_block(curr, pkc, pbd, len);
1103		return (void *)curr;
1104	}
1105
1106	/*
1107	 * No free blocks are available.user_space hasn't caught up yet.
1108	 * Queue was just frozen and now this packet will get dropped.
1109	 */
1110	return NULL;
1111}
1112
1113static void *packet_current_rx_frame(struct packet_sock *po,
1114					    struct sk_buff *skb,
1115					    int status, unsigned int len)
1116{
1117	char *curr = NULL;
1118	switch (po->tp_version) {
1119	case TPACKET_V1:
1120	case TPACKET_V2:
1121		curr = packet_lookup_frame(po, &po->rx_ring,
1122					po->rx_ring.head, status);
1123		return curr;
1124	case TPACKET_V3:
1125		return __packet_lookup_frame_in_block(po, skb, len);
1126	default:
1127		WARN(1, "TPACKET version not supported\n");
1128		BUG();
1129		return NULL;
1130	}
1131}
1132
1133static void *prb_lookup_block(const struct packet_sock *po,
1134			      const struct packet_ring_buffer *rb,
1135			      unsigned int idx,
1136			      int status)
1137{
1138	struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1139	struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1140
1141	if (status != BLOCK_STATUS(pbd))
1142		return NULL;
1143	return pbd;
1144}
1145
1146static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1147{
1148	unsigned int prev;
1149	if (rb->prb_bdqc.kactive_blk_num)
1150		prev = rb->prb_bdqc.kactive_blk_num-1;
1151	else
1152		prev = rb->prb_bdqc.knum_blocks-1;
1153	return prev;
1154}
1155
1156/* Assumes caller has held the rx_queue.lock */
1157static void *__prb_previous_block(struct packet_sock *po,
1158					 struct packet_ring_buffer *rb,
1159					 int status)
1160{
1161	unsigned int previous = prb_previous_blk_num(rb);
1162	return prb_lookup_block(po, rb, previous, status);
1163}
1164
1165static void *packet_previous_rx_frame(struct packet_sock *po,
1166					     struct packet_ring_buffer *rb,
1167					     int status)
1168{
1169	if (po->tp_version <= TPACKET_V2)
1170		return packet_previous_frame(po, rb, status);
1171
1172	return __prb_previous_block(po, rb, status);
1173}
1174
1175static void packet_increment_rx_head(struct packet_sock *po,
1176					    struct packet_ring_buffer *rb)
1177{
1178	switch (po->tp_version) {
1179	case TPACKET_V1:
1180	case TPACKET_V2:
1181		return packet_increment_head(rb);
1182	case TPACKET_V3:
1183	default:
1184		WARN(1, "TPACKET version not supported.\n");
1185		BUG();
1186		return;
1187	}
1188}
1189
1190static void *packet_previous_frame(struct packet_sock *po,
1191		struct packet_ring_buffer *rb,
1192		int status)
1193{
1194	unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1195	return packet_lookup_frame(po, rb, previous, status);
1196}
1197
1198static void packet_increment_head(struct packet_ring_buffer *buff)
1199{
1200	buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1201}
1202
1203static void packet_inc_pending(struct packet_ring_buffer *rb)
1204{
1205	this_cpu_inc(*rb->pending_refcnt);
1206}
1207
1208static void packet_dec_pending(struct packet_ring_buffer *rb)
1209{
1210	this_cpu_dec(*rb->pending_refcnt);
1211}
1212
1213static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1214{
1215	unsigned int refcnt = 0;
1216	int cpu;
1217
1218	/* We don't use pending refcount in rx_ring. */
1219	if (rb->pending_refcnt == NULL)
1220		return 0;
1221
1222	for_each_possible_cpu(cpu)
1223		refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1224
1225	return refcnt;
1226}
1227
1228static int packet_alloc_pending(struct packet_sock *po)
1229{
1230	po->rx_ring.pending_refcnt = NULL;
1231
1232	po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1233	if (unlikely(po->tx_ring.pending_refcnt == NULL))
1234		return -ENOBUFS;
1235
1236	return 0;
1237}
1238
1239static void packet_free_pending(struct packet_sock *po)
1240{
1241	free_percpu(po->tx_ring.pending_refcnt);
1242}
1243
1244#define ROOM_POW_OFF	2
1245#define ROOM_NONE	0x0
1246#define ROOM_LOW	0x1
1247#define ROOM_NORMAL	0x2
1248
1249static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1250{
1251	int idx, len;
1252
1253	len = READ_ONCE(po->rx_ring.frame_max) + 1;
1254	idx = READ_ONCE(po->rx_ring.head);
1255	if (pow_off)
1256		idx += len >> pow_off;
1257	if (idx >= len)
1258		idx -= len;
1259	return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1260}
1261
1262static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1263{
1264	int idx, len;
1265
1266	len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1267	idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1268	if (pow_off)
1269		idx += len >> pow_off;
1270	if (idx >= len)
1271		idx -= len;
1272	return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1273}
1274
1275static int __packet_rcv_has_room(const struct packet_sock *po,
1276				 const struct sk_buff *skb)
1277{
1278	const struct sock *sk = &po->sk;
1279	int ret = ROOM_NONE;
1280
1281	if (po->prot_hook.func != tpacket_rcv) {
1282		int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1283		int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1284				   - (skb ? skb->truesize : 0);
1285
1286		if (avail > (rcvbuf >> ROOM_POW_OFF))
1287			return ROOM_NORMAL;
1288		else if (avail > 0)
1289			return ROOM_LOW;
1290		else
1291			return ROOM_NONE;
1292	}
1293
1294	if (po->tp_version == TPACKET_V3) {
1295		if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1296			ret = ROOM_NORMAL;
1297		else if (__tpacket_v3_has_room(po, 0))
1298			ret = ROOM_LOW;
1299	} else {
1300		if (__tpacket_has_room(po, ROOM_POW_OFF))
1301			ret = ROOM_NORMAL;
1302		else if (__tpacket_has_room(po, 0))
1303			ret = ROOM_LOW;
1304	}
1305
1306	return ret;
1307}
1308
1309static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1310{
1311	bool pressure;
1312	int ret;
1313
1314	ret = __packet_rcv_has_room(po, skb);
1315	pressure = ret != ROOM_NORMAL;
1316
1317	if (packet_sock_flag(po, PACKET_SOCK_PRESSURE) != pressure)
1318		packet_sock_flag_set(po, PACKET_SOCK_PRESSURE, pressure);
1319
1320	return ret;
1321}
1322
1323static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1324{
1325	if (packet_sock_flag(po, PACKET_SOCK_PRESSURE) &&
1326	    __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1327		packet_sock_flag_set(po, PACKET_SOCK_PRESSURE, false);
1328}
1329
1330static void packet_sock_destruct(struct sock *sk)
1331{
1332	skb_queue_purge(&sk->sk_error_queue);
1333
1334	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1335	WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1336
1337	if (!sock_flag(sk, SOCK_DEAD)) {
1338		pr_err("Attempt to release alive packet socket: %p\n", sk);
1339		return;
1340	}
 
 
1341}
1342
1343static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1344{
1345	u32 *history = po->rollover->history;
1346	u32 victim, rxhash;
1347	int i, count = 0;
1348
1349	rxhash = skb_get_hash(skb);
1350	for (i = 0; i < ROLLOVER_HLEN; i++)
1351		if (READ_ONCE(history[i]) == rxhash)
1352			count++;
1353
1354	victim = get_random_u32_below(ROLLOVER_HLEN);
1355
1356	/* Avoid dirtying the cache line if possible */
1357	if (READ_ONCE(history[victim]) != rxhash)
1358		WRITE_ONCE(history[victim], rxhash);
1359
1360	return count > (ROLLOVER_HLEN >> 1);
1361}
1362
1363static unsigned int fanout_demux_hash(struct packet_fanout *f,
1364				      struct sk_buff *skb,
1365				      unsigned int num)
1366{
1367	return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1368}
1369
1370static unsigned int fanout_demux_lb(struct packet_fanout *f,
1371				    struct sk_buff *skb,
1372				    unsigned int num)
1373{
1374	unsigned int val = atomic_inc_return(&f->rr_cur);
1375
1376	return val % num;
1377}
1378
1379static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1380				     struct sk_buff *skb,
1381				     unsigned int num)
1382{
1383	return smp_processor_id() % num;
1384}
1385
1386static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1387				     struct sk_buff *skb,
1388				     unsigned int num)
1389{
1390	return get_random_u32_below(num);
1391}
1392
1393static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1394					  struct sk_buff *skb,
1395					  unsigned int idx, bool try_self,
1396					  unsigned int num)
1397{
1398	struct packet_sock *po, *po_next, *po_skip = NULL;
1399	unsigned int i, j, room = ROOM_NONE;
1400
1401	po = pkt_sk(rcu_dereference(f->arr[idx]));
1402
1403	if (try_self) {
1404		room = packet_rcv_has_room(po, skb);
1405		if (room == ROOM_NORMAL ||
1406		    (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1407			return idx;
1408		po_skip = po;
1409	}
1410
1411	i = j = min_t(int, po->rollover->sock, num - 1);
1412	do {
1413		po_next = pkt_sk(rcu_dereference(f->arr[i]));
1414		if (po_next != po_skip &&
1415		    !packet_sock_flag(po_next, PACKET_SOCK_PRESSURE) &&
1416		    packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1417			if (i != j)
1418				po->rollover->sock = i;
1419			atomic_long_inc(&po->rollover->num);
1420			if (room == ROOM_LOW)
1421				atomic_long_inc(&po->rollover->num_huge);
1422			return i;
1423		}
1424
1425		if (++i == num)
1426			i = 0;
1427	} while (i != j);
1428
1429	atomic_long_inc(&po->rollover->num_failed);
1430	return idx;
1431}
1432
1433static unsigned int fanout_demux_qm(struct packet_fanout *f,
1434				    struct sk_buff *skb,
1435				    unsigned int num)
1436{
1437	return skb_get_queue_mapping(skb) % num;
1438}
1439
1440static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1441				     struct sk_buff *skb,
1442				     unsigned int num)
1443{
1444	struct bpf_prog *prog;
1445	unsigned int ret = 0;
1446
1447	rcu_read_lock();
1448	prog = rcu_dereference(f->bpf_prog);
1449	if (prog)
1450		ret = bpf_prog_run_clear_cb(prog, skb) % num;
1451	rcu_read_unlock();
1452
1453	return ret;
1454}
1455
1456static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1457{
1458	return f->flags & (flag >> 8);
1459}
1460
1461static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1462			     struct packet_type *pt, struct net_device *orig_dev)
1463{
1464	struct packet_fanout *f = pt->af_packet_priv;
1465	unsigned int num = READ_ONCE(f->num_members);
1466	struct net *net = read_pnet(&f->net);
1467	struct packet_sock *po;
1468	unsigned int idx;
1469
1470	if (!net_eq(dev_net(dev), net) || !num) {
1471		kfree_skb(skb);
1472		return 0;
1473	}
1474
1475	if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1476		skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1477		if (!skb)
1478			return 0;
1479	}
1480	switch (f->type) {
1481	case PACKET_FANOUT_HASH:
1482	default:
1483		idx = fanout_demux_hash(f, skb, num);
1484		break;
1485	case PACKET_FANOUT_LB:
1486		idx = fanout_demux_lb(f, skb, num);
1487		break;
1488	case PACKET_FANOUT_CPU:
1489		idx = fanout_demux_cpu(f, skb, num);
1490		break;
1491	case PACKET_FANOUT_RND:
1492		idx = fanout_demux_rnd(f, skb, num);
1493		break;
1494	case PACKET_FANOUT_QM:
1495		idx = fanout_demux_qm(f, skb, num);
1496		break;
1497	case PACKET_FANOUT_ROLLOVER:
1498		idx = fanout_demux_rollover(f, skb, 0, false, num);
1499		break;
1500	case PACKET_FANOUT_CBPF:
1501	case PACKET_FANOUT_EBPF:
1502		idx = fanout_demux_bpf(f, skb, num);
1503		break;
1504	}
1505
1506	if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1507		idx = fanout_demux_rollover(f, skb, idx, true, num);
1508
1509	po = pkt_sk(rcu_dereference(f->arr[idx]));
1510	return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1511}
1512
1513DEFINE_MUTEX(fanout_mutex);
1514EXPORT_SYMBOL_GPL(fanout_mutex);
1515static LIST_HEAD(fanout_list);
1516static u16 fanout_next_id;
1517
1518static void __fanout_link(struct sock *sk, struct packet_sock *po)
1519{
1520	struct packet_fanout *f = po->fanout;
1521
1522	spin_lock(&f->lock);
1523	rcu_assign_pointer(f->arr[f->num_members], sk);
1524	smp_wmb();
1525	f->num_members++;
1526	if (f->num_members == 1)
1527		dev_add_pack(&f->prot_hook);
1528	spin_unlock(&f->lock);
1529}
1530
1531static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1532{
1533	struct packet_fanout *f = po->fanout;
1534	int i;
1535
1536	spin_lock(&f->lock);
1537	for (i = 0; i < f->num_members; i++) {
1538		if (rcu_dereference_protected(f->arr[i],
1539					      lockdep_is_held(&f->lock)) == sk)
1540			break;
1541	}
1542	BUG_ON(i >= f->num_members);
1543	rcu_assign_pointer(f->arr[i],
1544			   rcu_dereference_protected(f->arr[f->num_members - 1],
1545						     lockdep_is_held(&f->lock)));
1546	f->num_members--;
1547	if (f->num_members == 0)
1548		__dev_remove_pack(&f->prot_hook);
1549	spin_unlock(&f->lock);
1550}
1551
1552static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1553{
1554	if (sk->sk_family != PF_PACKET)
1555		return false;
1556
1557	return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1558}
1559
1560static void fanout_init_data(struct packet_fanout *f)
1561{
1562	switch (f->type) {
1563	case PACKET_FANOUT_LB:
1564		atomic_set(&f->rr_cur, 0);
1565		break;
1566	case PACKET_FANOUT_CBPF:
1567	case PACKET_FANOUT_EBPF:
1568		RCU_INIT_POINTER(f->bpf_prog, NULL);
1569		break;
1570	}
1571}
1572
1573static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1574{
1575	struct bpf_prog *old;
1576
1577	spin_lock(&f->lock);
1578	old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1579	rcu_assign_pointer(f->bpf_prog, new);
1580	spin_unlock(&f->lock);
1581
1582	if (old) {
1583		synchronize_net();
1584		bpf_prog_destroy(old);
1585	}
1586}
1587
1588static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data,
1589				unsigned int len)
1590{
1591	struct bpf_prog *new;
1592	struct sock_fprog fprog;
1593	int ret;
1594
1595	if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1596		return -EPERM;
1597
1598	ret = copy_bpf_fprog_from_user(&fprog, data, len);
1599	if (ret)
1600		return ret;
1601
1602	ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1603	if (ret)
1604		return ret;
1605
1606	__fanout_set_data_bpf(po->fanout, new);
1607	return 0;
1608}
1609
1610static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data,
1611				unsigned int len)
1612{
1613	struct bpf_prog *new;
1614	u32 fd;
1615
1616	if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1617		return -EPERM;
1618	if (len != sizeof(fd))
1619		return -EINVAL;
1620	if (copy_from_sockptr(&fd, data, len))
1621		return -EFAULT;
1622
1623	new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1624	if (IS_ERR(new))
1625		return PTR_ERR(new);
1626
1627	__fanout_set_data_bpf(po->fanout, new);
1628	return 0;
1629}
1630
1631static int fanout_set_data(struct packet_sock *po, sockptr_t data,
1632			   unsigned int len)
1633{
1634	switch (po->fanout->type) {
1635	case PACKET_FANOUT_CBPF:
1636		return fanout_set_data_cbpf(po, data, len);
1637	case PACKET_FANOUT_EBPF:
1638		return fanout_set_data_ebpf(po, data, len);
1639	default:
1640		return -EINVAL;
1641	}
1642}
1643
1644static void fanout_release_data(struct packet_fanout *f)
1645{
1646	switch (f->type) {
1647	case PACKET_FANOUT_CBPF:
1648	case PACKET_FANOUT_EBPF:
1649		__fanout_set_data_bpf(f, NULL);
1650	}
1651}
1652
1653static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1654{
1655	struct packet_fanout *f;
1656
1657	list_for_each_entry(f, &fanout_list, list) {
1658		if (f->id == candidate_id &&
1659		    read_pnet(&f->net) == sock_net(sk)) {
1660			return false;
1661		}
1662	}
1663	return true;
1664}
1665
1666static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1667{
1668	u16 id = fanout_next_id;
1669
1670	do {
1671		if (__fanout_id_is_free(sk, id)) {
1672			*new_id = id;
1673			fanout_next_id = id + 1;
1674			return true;
1675		}
1676
1677		id++;
1678	} while (id != fanout_next_id);
1679
1680	return false;
1681}
1682
1683static int fanout_add(struct sock *sk, struct fanout_args *args)
1684{
1685	struct packet_rollover *rollover = NULL;
1686	struct packet_sock *po = pkt_sk(sk);
1687	u16 type_flags = args->type_flags;
1688	struct packet_fanout *f, *match;
1689	u8 type = type_flags & 0xff;
1690	u8 flags = type_flags >> 8;
1691	u16 id = args->id;
1692	int err;
1693
1694	switch (type) {
1695	case PACKET_FANOUT_ROLLOVER:
1696		if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1697			return -EINVAL;
1698		break;
1699	case PACKET_FANOUT_HASH:
1700	case PACKET_FANOUT_LB:
1701	case PACKET_FANOUT_CPU:
1702	case PACKET_FANOUT_RND:
1703	case PACKET_FANOUT_QM:
1704	case PACKET_FANOUT_CBPF:
1705	case PACKET_FANOUT_EBPF:
1706		break;
1707	default:
1708		return -EINVAL;
1709	}
1710
1711	mutex_lock(&fanout_mutex);
1712
1713	err = -EALREADY;
1714	if (po->fanout)
1715		goto out;
1716
1717	if (type == PACKET_FANOUT_ROLLOVER ||
1718	    (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1719		err = -ENOMEM;
1720		rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1721		if (!rollover)
1722			goto out;
1723		atomic_long_set(&rollover->num, 0);
1724		atomic_long_set(&rollover->num_huge, 0);
1725		atomic_long_set(&rollover->num_failed, 0);
1726	}
1727
1728	if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1729		if (id != 0) {
1730			err = -EINVAL;
1731			goto out;
1732		}
1733		if (!fanout_find_new_id(sk, &id)) {
1734			err = -ENOMEM;
1735			goto out;
1736		}
1737		/* ephemeral flag for the first socket in the group: drop it */
1738		flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1739	}
1740
1741	match = NULL;
1742	list_for_each_entry(f, &fanout_list, list) {
1743		if (f->id == id &&
1744		    read_pnet(&f->net) == sock_net(sk)) {
1745			match = f;
1746			break;
1747		}
1748	}
1749	err = -EINVAL;
1750	if (match) {
1751		if (match->flags != flags)
1752			goto out;
1753		if (args->max_num_members &&
1754		    args->max_num_members != match->max_num_members)
1755			goto out;
1756	} else {
1757		if (args->max_num_members > PACKET_FANOUT_MAX)
1758			goto out;
1759		if (!args->max_num_members)
1760			/* legacy PACKET_FANOUT_MAX */
1761			args->max_num_members = 256;
1762		err = -ENOMEM;
1763		match = kvzalloc(struct_size(match, arr, args->max_num_members),
1764				 GFP_KERNEL);
1765		if (!match)
1766			goto out;
1767		write_pnet(&match->net, sock_net(sk));
1768		match->id = id;
1769		match->type = type;
1770		match->flags = flags;
1771		INIT_LIST_HEAD(&match->list);
1772		spin_lock_init(&match->lock);
1773		refcount_set(&match->sk_ref, 0);
1774		fanout_init_data(match);
1775		match->prot_hook.type = po->prot_hook.type;
1776		match->prot_hook.dev = po->prot_hook.dev;
1777		match->prot_hook.func = packet_rcv_fanout;
1778		match->prot_hook.af_packet_priv = match;
1779		match->prot_hook.af_packet_net = read_pnet(&match->net);
1780		match->prot_hook.id_match = match_fanout_group;
1781		match->max_num_members = args->max_num_members;
1782		match->prot_hook.ignore_outgoing = type_flags & PACKET_FANOUT_FLAG_IGNORE_OUTGOING;
1783		list_add(&match->list, &fanout_list);
1784	}
1785	err = -EINVAL;
1786
1787	spin_lock(&po->bind_lock);
1788	if (packet_sock_flag(po, PACKET_SOCK_RUNNING) &&
1789	    match->type == type &&
1790	    match->prot_hook.type == po->prot_hook.type &&
1791	    match->prot_hook.dev == po->prot_hook.dev) {
1792		err = -ENOSPC;
1793		if (refcount_read(&match->sk_ref) < match->max_num_members) {
1794			__dev_remove_pack(&po->prot_hook);
1795
1796			/* Paired with packet_setsockopt(PACKET_FANOUT_DATA) */
1797			WRITE_ONCE(po->fanout, match);
1798
1799			po->rollover = rollover;
1800			rollover = NULL;
1801			refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1802			__fanout_link(sk, po);
1803			err = 0;
1804		}
1805	}
1806	spin_unlock(&po->bind_lock);
1807
1808	if (err && !refcount_read(&match->sk_ref)) {
1809		list_del(&match->list);
1810		kvfree(match);
1811	}
1812
1813out:
1814	kfree(rollover);
1815	mutex_unlock(&fanout_mutex);
1816	return err;
1817}
1818
1819/* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1820 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1821 * It is the responsibility of the caller to call fanout_release_data() and
1822 * free the returned packet_fanout (after synchronize_net())
1823 */
1824static struct packet_fanout *fanout_release(struct sock *sk)
1825{
1826	struct packet_sock *po = pkt_sk(sk);
1827	struct packet_fanout *f;
1828
1829	mutex_lock(&fanout_mutex);
1830	f = po->fanout;
1831	if (f) {
1832		po->fanout = NULL;
1833
1834		if (refcount_dec_and_test(&f->sk_ref))
1835			list_del(&f->list);
1836		else
1837			f = NULL;
1838	}
1839	mutex_unlock(&fanout_mutex);
1840
1841	return f;
1842}
1843
1844static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1845					  struct sk_buff *skb)
1846{
1847	/* Earlier code assumed this would be a VLAN pkt, double-check
1848	 * this now that we have the actual packet in hand. We can only
1849	 * do this check on Ethernet devices.
1850	 */
1851	if (unlikely(dev->type != ARPHRD_ETHER))
1852		return false;
1853
1854	skb_reset_mac_header(skb);
1855	return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1856}
1857
1858static const struct proto_ops packet_ops;
1859
1860static const struct proto_ops packet_ops_spkt;
1861
1862static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1863			   struct packet_type *pt, struct net_device *orig_dev)
1864{
1865	struct sock *sk;
1866	struct sockaddr_pkt *spkt;
1867
1868	/*
1869	 *	When we registered the protocol we saved the socket in the data
1870	 *	field for just this event.
1871	 */
1872
1873	sk = pt->af_packet_priv;
1874
1875	/*
1876	 *	Yank back the headers [hope the device set this
1877	 *	right or kerboom...]
1878	 *
1879	 *	Incoming packets have ll header pulled,
1880	 *	push it back.
1881	 *
1882	 *	For outgoing ones skb->data == skb_mac_header(skb)
1883	 *	so that this procedure is noop.
1884	 */
1885
1886	if (skb->pkt_type == PACKET_LOOPBACK)
1887		goto out;
1888
1889	if (!net_eq(dev_net(dev), sock_net(sk)))
1890		goto out;
1891
1892	skb = skb_share_check(skb, GFP_ATOMIC);
1893	if (skb == NULL)
1894		goto oom;
1895
1896	/* drop any routing info */
1897	skb_dst_drop(skb);
1898
1899	/* drop conntrack reference */
1900	nf_reset_ct(skb);
1901
1902	spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1903
1904	skb_push(skb, skb->data - skb_mac_header(skb));
1905
1906	/*
1907	 *	The SOCK_PACKET socket receives _all_ frames.
1908	 */
1909
1910	spkt->spkt_family = dev->type;
1911	strscpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1912	spkt->spkt_protocol = skb->protocol;
1913
1914	/*
1915	 *	Charge the memory to the socket. This is done specifically
1916	 *	to prevent sockets using all the memory up.
1917	 */
1918
1919	if (sock_queue_rcv_skb(sk, skb) == 0)
1920		return 0;
1921
1922out:
1923	kfree_skb(skb);
1924oom:
1925	return 0;
1926}
1927
1928static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1929{
1930	int depth;
1931
1932	if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1933	    sock->type == SOCK_RAW) {
1934		skb_reset_mac_header(skb);
1935		skb->protocol = dev_parse_header_protocol(skb);
1936	}
1937
1938	/* Move network header to the right position for VLAN tagged packets */
1939	if (likely(skb->dev->type == ARPHRD_ETHER) &&
1940	    eth_type_vlan(skb->protocol) &&
1941	    vlan_get_protocol_and_depth(skb, skb->protocol, &depth) != 0)
1942		skb_set_network_header(skb, depth);
1943
1944	skb_probe_transport_header(skb);
1945}
1946
1947/*
1948 *	Output a raw packet to a device layer. This bypasses all the other
1949 *	protocol layers and you must therefore supply it with a complete frame
1950 */
1951
1952static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1953			       size_t len)
1954{
1955	struct sock *sk = sock->sk;
1956	DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1957	struct sk_buff *skb = NULL;
1958	struct net_device *dev;
1959	struct sockcm_cookie sockc;
1960	__be16 proto = 0;
1961	int err;
1962	int extra_len = 0;
1963
1964	/*
1965	 *	Get and verify the address.
1966	 */
1967
1968	if (saddr) {
1969		if (msg->msg_namelen < sizeof(struct sockaddr))
1970			return -EINVAL;
1971		if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1972			proto = saddr->spkt_protocol;
1973	} else
1974		return -ENOTCONN;	/* SOCK_PACKET must be sent giving an address */
1975
1976	/*
1977	 *	Find the device first to size check it
1978	 */
1979
1980	saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1981retry:
1982	rcu_read_lock();
1983	dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1984	err = -ENODEV;
1985	if (dev == NULL)
1986		goto out_unlock;
1987
1988	err = -ENETDOWN;
1989	if (!(dev->flags & IFF_UP))
1990		goto out_unlock;
1991
1992	/*
1993	 * You may not queue a frame bigger than the mtu. This is the lowest level
1994	 * raw protocol and you must do your own fragmentation at this level.
1995	 */
1996
1997	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1998		if (!netif_supports_nofcs(dev)) {
1999			err = -EPROTONOSUPPORT;
2000			goto out_unlock;
2001		}
2002		extra_len = 4; /* We're doing our own CRC */
2003	}
2004
2005	err = -EMSGSIZE;
2006	if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
2007		goto out_unlock;
2008
2009	if (!skb) {
2010		size_t reserved = LL_RESERVED_SPACE(dev);
2011		int tlen = dev->needed_tailroom;
2012		unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
2013
2014		rcu_read_unlock();
2015		skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
2016		if (skb == NULL)
2017			return -ENOBUFS;
2018		/* FIXME: Save some space for broken drivers that write a hard
2019		 * header at transmission time by themselves. PPP is the notable
2020		 * one here. This should really be fixed at the driver level.
2021		 */
2022		skb_reserve(skb, reserved);
2023		skb_reset_network_header(skb);
2024
2025		/* Try to align data part correctly */
2026		if (hhlen) {
2027			skb->data -= hhlen;
2028			skb->tail -= hhlen;
2029			if (len < hhlen)
2030				skb_reset_network_header(skb);
2031		}
2032		err = memcpy_from_msg(skb_put(skb, len), msg, len);
2033		if (err)
2034			goto out_free;
2035		goto retry;
2036	}
2037
2038	if (!dev_validate_header(dev, skb->data, len) || !skb->len) {
2039		err = -EINVAL;
2040		goto out_unlock;
2041	}
2042	if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
2043	    !packet_extra_vlan_len_allowed(dev, skb)) {
2044		err = -EMSGSIZE;
2045		goto out_unlock;
2046	}
2047
2048	sockcm_init(&sockc, sk);
2049	if (msg->msg_controllen) {
2050		err = sock_cmsg_send(sk, msg, &sockc);
2051		if (unlikely(err))
2052			goto out_unlock;
2053	}
2054
2055	skb->protocol = proto;
2056	skb->dev = dev;
2057	skb->priority = READ_ONCE(sk->sk_priority);
2058	skb->mark = READ_ONCE(sk->sk_mark);
2059	skb->tstamp = sockc.transmit_time;
2060
2061	skb_setup_tx_timestamp(skb, sockc.tsflags);
2062
2063	if (unlikely(extra_len == 4))
2064		skb->no_fcs = 1;
2065
2066	packet_parse_headers(skb, sock);
2067
2068	dev_queue_xmit(skb);
2069	rcu_read_unlock();
2070	return len;
2071
2072out_unlock:
2073	rcu_read_unlock();
2074out_free:
2075	kfree_skb(skb);
2076	return err;
2077}
2078
2079static unsigned int run_filter(struct sk_buff *skb,
2080			       const struct sock *sk,
2081			       unsigned int res)
2082{
2083	struct sk_filter *filter;
2084
2085	rcu_read_lock();
2086	filter = rcu_dereference(sk->sk_filter);
2087	if (filter != NULL)
2088		res = bpf_prog_run_clear_cb(filter->prog, skb);
2089	rcu_read_unlock();
2090
2091	return res;
2092}
2093
2094static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2095			   size_t *len, int vnet_hdr_sz)
2096{
2097	struct virtio_net_hdr_mrg_rxbuf vnet_hdr = { .num_buffers = 0 };
2098
2099	if (*len < vnet_hdr_sz)
2100		return -EINVAL;
2101	*len -= vnet_hdr_sz;
2102
2103	if (virtio_net_hdr_from_skb(skb, (struct virtio_net_hdr *)&vnet_hdr, vio_le(), true, 0))
2104		return -EINVAL;
2105
2106	return memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_sz);
2107}
2108
2109/*
2110 * This function makes lazy skb cloning in hope that most of packets
2111 * are discarded by BPF.
2112 *
2113 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2114 * and skb->cb are mangled. It works because (and until) packets
2115 * falling here are owned by current CPU. Output packets are cloned
2116 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2117 * sequentially, so that if we return skb to original state on exit,
2118 * we will not harm anyone.
2119 */
2120
2121static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2122		      struct packet_type *pt, struct net_device *orig_dev)
2123{
2124	enum skb_drop_reason drop_reason = SKB_CONSUMED;
2125	struct sock *sk;
2126	struct sockaddr_ll *sll;
2127	struct packet_sock *po;
2128	u8 *skb_head = skb->data;
2129	int skb_len = skb->len;
2130	unsigned int snaplen, res;
 
2131
2132	if (skb->pkt_type == PACKET_LOOPBACK)
2133		goto drop;
2134
2135	sk = pt->af_packet_priv;
2136	po = pkt_sk(sk);
2137
2138	if (!net_eq(dev_net(dev), sock_net(sk)))
2139		goto drop;
2140
2141	skb->dev = dev;
2142
2143	if (dev_has_header(dev)) {
2144		/* The device has an explicit notion of ll header,
2145		 * exported to higher levels.
2146		 *
2147		 * Otherwise, the device hides details of its frame
2148		 * structure, so that corresponding packet head is
2149		 * never delivered to user.
2150		 */
2151		if (sk->sk_type != SOCK_DGRAM)
2152			skb_push(skb, skb->data - skb_mac_header(skb));
2153		else if (skb->pkt_type == PACKET_OUTGOING) {
2154			/* Special case: outgoing packets have ll header at head */
2155			skb_pull(skb, skb_network_offset(skb));
2156		}
2157	}
2158
2159	snaplen = skb->len;
2160
2161	res = run_filter(skb, sk, snaplen);
2162	if (!res)
2163		goto drop_n_restore;
2164	if (snaplen > res)
2165		snaplen = res;
2166
2167	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2168		goto drop_n_acct;
2169
2170	if (skb_shared(skb)) {
2171		struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2172		if (nskb == NULL)
2173			goto drop_n_acct;
2174
2175		if (skb_head != skb->data) {
2176			skb->data = skb_head;
2177			skb->len = skb_len;
2178		}
2179		consume_skb(skb);
2180		skb = nskb;
2181	}
2182
2183	sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2184
2185	sll = &PACKET_SKB_CB(skb)->sa.ll;
2186	sll->sll_hatype = dev->type;
2187	sll->sll_pkttype = skb->pkt_type;
2188	if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV)))
2189		sll->sll_ifindex = orig_dev->ifindex;
2190	else
2191		sll->sll_ifindex = dev->ifindex;
2192
2193	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2194
2195	/* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2196	 * Use their space for storing the original skb length.
2197	 */
2198	PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2199
2200	if (pskb_trim(skb, snaplen))
2201		goto drop_n_acct;
2202
2203	skb_set_owner_r(skb, sk);
2204	skb->dev = NULL;
2205	skb_dst_drop(skb);
2206
2207	/* drop conntrack reference */
2208	nf_reset_ct(skb);
2209
2210	spin_lock(&sk->sk_receive_queue.lock);
2211	po->stats.stats1.tp_packets++;
2212	sock_skb_set_dropcount(sk, skb);
2213	skb_clear_delivery_time(skb);
2214	__skb_queue_tail(&sk->sk_receive_queue, skb);
2215	spin_unlock(&sk->sk_receive_queue.lock);
2216	sk->sk_data_ready(sk);
2217	return 0;
2218
2219drop_n_acct:
 
2220	atomic_inc(&po->tp_drops);
2221	atomic_inc(&sk->sk_drops);
2222	drop_reason = SKB_DROP_REASON_PACKET_SOCK_ERROR;
2223
2224drop_n_restore:
2225	if (skb_head != skb->data && skb_shared(skb)) {
2226		skb->data = skb_head;
2227		skb->len = skb_len;
2228	}
2229drop:
2230	kfree_skb_reason(skb, drop_reason);
 
 
 
2231	return 0;
2232}
2233
2234static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2235		       struct packet_type *pt, struct net_device *orig_dev)
2236{
2237	enum skb_drop_reason drop_reason = SKB_CONSUMED;
2238	struct sock *sk;
2239	struct packet_sock *po;
2240	struct sockaddr_ll *sll;
2241	union tpacket_uhdr h;
2242	u8 *skb_head = skb->data;
2243	int skb_len = skb->len;
2244	unsigned int snaplen, res;
2245	unsigned long status = TP_STATUS_USER;
2246	unsigned short macoff, hdrlen;
2247	unsigned int netoff;
2248	struct sk_buff *copy_skb = NULL;
2249	struct timespec64 ts;
2250	__u32 ts_status;
 
2251	unsigned int slot_id = 0;
2252	int vnet_hdr_sz = 0;
2253
2254	/* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2255	 * We may add members to them until current aligned size without forcing
2256	 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2257	 */
2258	BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2259	BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2260
2261	if (skb->pkt_type == PACKET_LOOPBACK)
2262		goto drop;
2263
2264	sk = pt->af_packet_priv;
2265	po = pkt_sk(sk);
2266
2267	if (!net_eq(dev_net(dev), sock_net(sk)))
2268		goto drop;
2269
2270	if (dev_has_header(dev)) {
2271		if (sk->sk_type != SOCK_DGRAM)
2272			skb_push(skb, skb->data - skb_mac_header(skb));
2273		else if (skb->pkt_type == PACKET_OUTGOING) {
2274			/* Special case: outgoing packets have ll header at head */
2275			skb_pull(skb, skb_network_offset(skb));
2276		}
2277	}
2278
2279	snaplen = skb->len;
2280
2281	res = run_filter(skb, sk, snaplen);
2282	if (!res)
2283		goto drop_n_restore;
2284
2285	/* If we are flooded, just give up */
2286	if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2287		atomic_inc(&po->tp_drops);
2288		goto drop_n_restore;
2289	}
2290
2291	if (skb->ip_summed == CHECKSUM_PARTIAL)
2292		status |= TP_STATUS_CSUMNOTREADY;
2293	else if (skb->pkt_type != PACKET_OUTGOING &&
2294		 skb_csum_unnecessary(skb))
 
2295		status |= TP_STATUS_CSUM_VALID;
2296	if (skb_is_gso(skb) && skb_is_gso_tcp(skb))
2297		status |= TP_STATUS_GSO_TCP;
2298
2299	if (snaplen > res)
2300		snaplen = res;
2301
2302	if (sk->sk_type == SOCK_DGRAM) {
2303		macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2304				  po->tp_reserve;
2305	} else {
2306		unsigned int maclen = skb_network_offset(skb);
2307		netoff = TPACKET_ALIGN(po->tp_hdrlen +
2308				       (maclen < 16 ? 16 : maclen)) +
2309				       po->tp_reserve;
2310		vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
2311		if (vnet_hdr_sz)
2312			netoff += vnet_hdr_sz;
 
2313		macoff = netoff - maclen;
2314	}
2315	if (netoff > USHRT_MAX) {
2316		atomic_inc(&po->tp_drops);
2317		goto drop_n_restore;
2318	}
2319	if (po->tp_version <= TPACKET_V2) {
2320		if (macoff + snaplen > po->rx_ring.frame_size) {
2321			if (READ_ONCE(po->copy_thresh) &&
2322			    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2323				if (skb_shared(skb)) {
2324					copy_skb = skb_clone(skb, GFP_ATOMIC);
2325				} else {
2326					copy_skb = skb_get(skb);
2327					skb_head = skb->data;
2328				}
2329				if (copy_skb) {
2330					memset(&PACKET_SKB_CB(copy_skb)->sa.ll, 0,
2331					       sizeof(PACKET_SKB_CB(copy_skb)->sa.ll));
2332					skb_set_owner_r(copy_skb, sk);
2333				}
2334			}
2335			snaplen = po->rx_ring.frame_size - macoff;
2336			if ((int)snaplen < 0) {
2337				snaplen = 0;
2338				vnet_hdr_sz = 0;
2339			}
2340		}
2341	} else if (unlikely(macoff + snaplen >
2342			    GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2343		u32 nval;
2344
2345		nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2346		pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2347			    snaplen, nval, macoff);
2348		snaplen = nval;
2349		if (unlikely((int)snaplen < 0)) {
2350			snaplen = 0;
2351			macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2352			vnet_hdr_sz = 0;
2353		}
2354	}
2355	spin_lock(&sk->sk_receive_queue.lock);
2356	h.raw = packet_current_rx_frame(po, skb,
2357					TP_STATUS_KERNEL, (macoff+snaplen));
2358	if (!h.raw)
2359		goto drop_n_account;
2360
2361	if (po->tp_version <= TPACKET_V2) {
2362		slot_id = po->rx_ring.head;
2363		if (test_bit(slot_id, po->rx_ring.rx_owner_map))
2364			goto drop_n_account;
2365		__set_bit(slot_id, po->rx_ring.rx_owner_map);
2366	}
2367
2368	if (vnet_hdr_sz &&
2369	    virtio_net_hdr_from_skb(skb, h.raw + macoff -
2370				    sizeof(struct virtio_net_hdr),
2371				    vio_le(), true, 0)) {
2372		if (po->tp_version == TPACKET_V3)
2373			prb_clear_blk_fill_status(&po->rx_ring);
2374		goto drop_n_account;
2375	}
2376
2377	if (po->tp_version <= TPACKET_V2) {
2378		packet_increment_rx_head(po, &po->rx_ring);
2379	/*
2380	 * LOSING will be reported till you read the stats,
2381	 * because it's COR - Clear On Read.
2382	 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2383	 * at packet level.
2384	 */
2385		if (atomic_read(&po->tp_drops))
2386			status |= TP_STATUS_LOSING;
2387	}
2388
2389	po->stats.stats1.tp_packets++;
2390	if (copy_skb) {
2391		status |= TP_STATUS_COPY;
2392		skb_clear_delivery_time(copy_skb);
2393		__skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2394	}
2395	spin_unlock(&sk->sk_receive_queue.lock);
2396
2397	skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2398
2399	/* Always timestamp; prefer an existing software timestamp taken
2400	 * closer to the time of capture.
2401	 */
2402	ts_status = tpacket_get_timestamp(skb, &ts,
2403					  READ_ONCE(po->tp_tstamp) |
2404					  SOF_TIMESTAMPING_SOFTWARE);
2405	if (!ts_status)
2406		ktime_get_real_ts64(&ts);
2407
2408	status |= ts_status;
2409
2410	switch (po->tp_version) {
2411	case TPACKET_V1:
2412		h.h1->tp_len = skb->len;
2413		h.h1->tp_snaplen = snaplen;
2414		h.h1->tp_mac = macoff;
2415		h.h1->tp_net = netoff;
2416		h.h1->tp_sec = ts.tv_sec;
2417		h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2418		hdrlen = sizeof(*h.h1);
2419		break;
2420	case TPACKET_V2:
2421		h.h2->tp_len = skb->len;
2422		h.h2->tp_snaplen = snaplen;
2423		h.h2->tp_mac = macoff;
2424		h.h2->tp_net = netoff;
2425		h.h2->tp_sec = ts.tv_sec;
2426		h.h2->tp_nsec = ts.tv_nsec;
2427		if (skb_vlan_tag_present(skb)) {
2428			h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2429			h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2430			status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2431		} else {
2432			h.h2->tp_vlan_tci = 0;
2433			h.h2->tp_vlan_tpid = 0;
2434		}
2435		memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2436		hdrlen = sizeof(*h.h2);
2437		break;
2438	case TPACKET_V3:
2439		/* tp_nxt_offset,vlan are already populated above.
2440		 * So DONT clear those fields here
2441		 */
2442		h.h3->tp_status |= status;
2443		h.h3->tp_len = skb->len;
2444		h.h3->tp_snaplen = snaplen;
2445		h.h3->tp_mac = macoff;
2446		h.h3->tp_net = netoff;
2447		h.h3->tp_sec  = ts.tv_sec;
2448		h.h3->tp_nsec = ts.tv_nsec;
2449		memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2450		hdrlen = sizeof(*h.h3);
2451		break;
2452	default:
2453		BUG();
2454	}
2455
2456	sll = h.raw + TPACKET_ALIGN(hdrlen);
2457	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2458	sll->sll_family = AF_PACKET;
2459	sll->sll_hatype = dev->type;
2460	sll->sll_protocol = skb->protocol;
2461	sll->sll_pkttype = skb->pkt_type;
2462	if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV)))
2463		sll->sll_ifindex = orig_dev->ifindex;
2464	else
2465		sll->sll_ifindex = dev->ifindex;
2466
2467	smp_mb();
2468
2469#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2470	if (po->tp_version <= TPACKET_V2) {
2471		u8 *start, *end;
2472
2473		end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2474					macoff + snaplen);
2475
2476		for (start = h.raw; start < end; start += PAGE_SIZE)
2477			flush_dcache_page(pgv_to_page(start));
2478	}
2479	smp_wmb();
2480#endif
2481
2482	if (po->tp_version <= TPACKET_V2) {
2483		spin_lock(&sk->sk_receive_queue.lock);
2484		__packet_set_status(po, h.raw, status);
2485		__clear_bit(slot_id, po->rx_ring.rx_owner_map);
2486		spin_unlock(&sk->sk_receive_queue.lock);
2487		sk->sk_data_ready(sk);
2488	} else if (po->tp_version == TPACKET_V3) {
2489		prb_clear_blk_fill_status(&po->rx_ring);
2490	}
2491
2492drop_n_restore:
2493	if (skb_head != skb->data && skb_shared(skb)) {
2494		skb->data = skb_head;
2495		skb->len = skb_len;
2496	}
2497drop:
2498	kfree_skb_reason(skb, drop_reason);
 
 
 
2499	return 0;
2500
2501drop_n_account:
2502	spin_unlock(&sk->sk_receive_queue.lock);
2503	atomic_inc(&po->tp_drops);
2504	drop_reason = SKB_DROP_REASON_PACKET_SOCK_ERROR;
2505
2506	sk->sk_data_ready(sk);
2507	kfree_skb_reason(copy_skb, drop_reason);
2508	goto drop_n_restore;
2509}
2510
2511static void tpacket_destruct_skb(struct sk_buff *skb)
2512{
2513	struct packet_sock *po = pkt_sk(skb->sk);
2514
2515	if (likely(po->tx_ring.pg_vec)) {
2516		void *ph;
2517		__u32 ts;
2518
2519		ph = skb_zcopy_get_nouarg(skb);
2520		packet_dec_pending(&po->tx_ring);
2521
2522		ts = __packet_set_timestamp(po, ph, skb);
2523		__packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2524
2525		complete(&po->skb_completion);
 
2526	}
2527
2528	sock_wfree(skb);
2529}
2530
2531static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2532{
2533	if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2534	    (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2535	     __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2536	      __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2537		vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2538			 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2539			__virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2540
2541	if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2542		return -EINVAL;
2543
2544	return 0;
2545}
2546
2547static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2548				 struct virtio_net_hdr *vnet_hdr, int vnet_hdr_sz)
2549{
2550	int ret;
2551
2552	if (*len < vnet_hdr_sz)
2553		return -EINVAL;
2554	*len -= vnet_hdr_sz;
2555
2556	if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2557		return -EFAULT;
2558
2559	ret = __packet_snd_vnet_parse(vnet_hdr, *len);
2560	if (ret)
2561		return ret;
2562
2563	/* move iter to point to the start of mac header */
2564	if (vnet_hdr_sz != sizeof(struct virtio_net_hdr))
2565		iov_iter_advance(&msg->msg_iter, vnet_hdr_sz - sizeof(struct virtio_net_hdr));
2566
2567	return 0;
2568}
2569
2570static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2571		void *frame, struct net_device *dev, void *data, int tp_len,
2572		__be16 proto, unsigned char *addr, int hlen, int copylen,
2573		const struct sockcm_cookie *sockc)
2574{
2575	union tpacket_uhdr ph;
2576	int to_write, offset, len, nr_frags, len_max;
2577	struct socket *sock = po->sk.sk_socket;
2578	struct page *page;
2579	int err;
2580
2581	ph.raw = frame;
2582
2583	skb->protocol = proto;
2584	skb->dev = dev;
2585	skb->priority = READ_ONCE(po->sk.sk_priority);
2586	skb->mark = READ_ONCE(po->sk.sk_mark);
2587	skb->tstamp = sockc->transmit_time;
2588	skb_setup_tx_timestamp(skb, sockc->tsflags);
2589	skb_zcopy_set_nouarg(skb, ph.raw);
2590
2591	skb_reserve(skb, hlen);
2592	skb_reset_network_header(skb);
2593
2594	to_write = tp_len;
2595
2596	if (sock->type == SOCK_DGRAM) {
2597		err = dev_hard_header(skb, dev, ntohs(proto), addr,
2598				NULL, tp_len);
2599		if (unlikely(err < 0))
2600			return -EINVAL;
2601	} else if (copylen) {
2602		int hdrlen = min_t(int, copylen, tp_len);
2603
2604		skb_push(skb, dev->hard_header_len);
2605		skb_put(skb, copylen - dev->hard_header_len);
2606		err = skb_store_bits(skb, 0, data, hdrlen);
2607		if (unlikely(err))
2608			return err;
2609		if (!dev_validate_header(dev, skb->data, hdrlen))
2610			return -EINVAL;
2611
2612		data += hdrlen;
2613		to_write -= hdrlen;
2614	}
2615
2616	offset = offset_in_page(data);
2617	len_max = PAGE_SIZE - offset;
2618	len = ((to_write > len_max) ? len_max : to_write);
2619
2620	skb->data_len = to_write;
2621	skb->len += to_write;
2622	skb->truesize += to_write;
2623	refcount_add(to_write, &po->sk.sk_wmem_alloc);
2624
2625	while (likely(to_write)) {
2626		nr_frags = skb_shinfo(skb)->nr_frags;
2627
2628		if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2629			pr_err("Packet exceed the number of skb frags(%u)\n",
2630			       (unsigned int)MAX_SKB_FRAGS);
2631			return -EFAULT;
2632		}
2633
2634		page = pgv_to_page(data);
2635		data += len;
2636		flush_dcache_page(page);
2637		get_page(page);
2638		skb_fill_page_desc(skb, nr_frags, page, offset, len);
2639		to_write -= len;
2640		offset = 0;
2641		len_max = PAGE_SIZE;
2642		len = ((to_write > len_max) ? len_max : to_write);
2643	}
2644
2645	packet_parse_headers(skb, sock);
2646
2647	return tp_len;
2648}
2649
2650static int tpacket_parse_header(struct packet_sock *po, void *frame,
2651				int size_max, void **data)
2652{
2653	union tpacket_uhdr ph;
2654	int tp_len, off;
2655
2656	ph.raw = frame;
2657
2658	switch (po->tp_version) {
2659	case TPACKET_V3:
2660		if (ph.h3->tp_next_offset != 0) {
2661			pr_warn_once("variable sized slot not supported");
2662			return -EINVAL;
2663		}
2664		tp_len = ph.h3->tp_len;
2665		break;
2666	case TPACKET_V2:
2667		tp_len = ph.h2->tp_len;
2668		break;
2669	default:
2670		tp_len = ph.h1->tp_len;
2671		break;
2672	}
2673	if (unlikely(tp_len > size_max)) {
2674		pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2675		return -EMSGSIZE;
2676	}
2677
2678	if (unlikely(packet_sock_flag(po, PACKET_SOCK_TX_HAS_OFF))) {
2679		int off_min, off_max;
2680
2681		off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2682		off_max = po->tx_ring.frame_size - tp_len;
2683		if (po->sk.sk_type == SOCK_DGRAM) {
2684			switch (po->tp_version) {
2685			case TPACKET_V3:
2686				off = ph.h3->tp_net;
2687				break;
2688			case TPACKET_V2:
2689				off = ph.h2->tp_net;
2690				break;
2691			default:
2692				off = ph.h1->tp_net;
2693				break;
2694			}
2695		} else {
2696			switch (po->tp_version) {
2697			case TPACKET_V3:
2698				off = ph.h3->tp_mac;
2699				break;
2700			case TPACKET_V2:
2701				off = ph.h2->tp_mac;
2702				break;
2703			default:
2704				off = ph.h1->tp_mac;
2705				break;
2706			}
2707		}
2708		if (unlikely((off < off_min) || (off_max < off)))
2709			return -EINVAL;
2710	} else {
2711		off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2712	}
2713
2714	*data = frame + off;
2715	return tp_len;
2716}
2717
2718static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2719{
2720	struct sk_buff *skb = NULL;
2721	struct net_device *dev;
2722	struct virtio_net_hdr *vnet_hdr = NULL;
2723	struct sockcm_cookie sockc;
2724	__be16 proto;
2725	int err, reserve = 0;
2726	void *ph;
2727	DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2728	bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2729	int vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
2730	unsigned char *addr = NULL;
2731	int tp_len, size_max;
2732	void *data;
2733	int len_sum = 0;
2734	int status = TP_STATUS_AVAILABLE;
2735	int hlen, tlen, copylen = 0;
2736	long timeo = 0;
2737
2738	mutex_lock(&po->pg_vec_lock);
2739
2740	/* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2741	 * we need to confirm it under protection of pg_vec_lock.
2742	 */
2743	if (unlikely(!po->tx_ring.pg_vec)) {
2744		err = -EBUSY;
2745		goto out;
2746	}
2747	if (likely(saddr == NULL)) {
2748		dev	= packet_cached_dev_get(po);
2749		proto	= READ_ONCE(po->num);
2750	} else {
2751		err = -EINVAL;
2752		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2753			goto out;
2754		if (msg->msg_namelen < (saddr->sll_halen
2755					+ offsetof(struct sockaddr_ll,
2756						sll_addr)))
2757			goto out;
2758		proto	= saddr->sll_protocol;
2759		dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2760		if (po->sk.sk_socket->type == SOCK_DGRAM) {
2761			if (dev && msg->msg_namelen < dev->addr_len +
2762				   offsetof(struct sockaddr_ll, sll_addr))
2763				goto out_put;
2764			addr = saddr->sll_addr;
2765		}
2766	}
2767
2768	err = -ENXIO;
2769	if (unlikely(dev == NULL))
2770		goto out;
2771	err = -ENETDOWN;
2772	if (unlikely(!(dev->flags & IFF_UP)))
2773		goto out_put;
2774
2775	sockcm_init(&sockc, &po->sk);
2776	if (msg->msg_controllen) {
2777		err = sock_cmsg_send(&po->sk, msg, &sockc);
2778		if (unlikely(err))
2779			goto out_put;
2780	}
2781
2782	if (po->sk.sk_socket->type == SOCK_RAW)
2783		reserve = dev->hard_header_len;
2784	size_max = po->tx_ring.frame_size
2785		- (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2786
2787	if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !vnet_hdr_sz)
2788		size_max = dev->mtu + reserve + VLAN_HLEN;
2789
2790	reinit_completion(&po->skb_completion);
2791
2792	do {
2793		ph = packet_current_frame(po, &po->tx_ring,
2794					  TP_STATUS_SEND_REQUEST);
2795		if (unlikely(ph == NULL)) {
2796			if (need_wait && skb) {
2797				timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2798				timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2799				if (timeo <= 0) {
2800					err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2801					goto out_put;
2802				}
2803			}
2804			/* check for additional frames */
2805			continue;
2806		}
2807
2808		skb = NULL;
2809		tp_len = tpacket_parse_header(po, ph, size_max, &data);
2810		if (tp_len < 0)
2811			goto tpacket_error;
2812
2813		status = TP_STATUS_SEND_REQUEST;
2814		hlen = LL_RESERVED_SPACE(dev);
2815		tlen = dev->needed_tailroom;
2816		if (vnet_hdr_sz) {
2817			vnet_hdr = data;
2818			data += vnet_hdr_sz;
2819			tp_len -= vnet_hdr_sz;
2820			if (tp_len < 0 ||
2821			    __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2822				tp_len = -EINVAL;
2823				goto tpacket_error;
2824			}
2825			copylen = __virtio16_to_cpu(vio_le(),
2826						    vnet_hdr->hdr_len);
2827		}
2828		copylen = max_t(int, copylen, dev->hard_header_len);
2829		skb = sock_alloc_send_skb(&po->sk,
2830				hlen + tlen + sizeof(struct sockaddr_ll) +
2831				(copylen - dev->hard_header_len),
2832				!need_wait, &err);
2833
2834		if (unlikely(skb == NULL)) {
2835			/* we assume the socket was initially writeable ... */
2836			if (likely(len_sum > 0))
2837				err = len_sum;
2838			goto out_status;
2839		}
2840		tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2841					  addr, hlen, copylen, &sockc);
2842		if (likely(tp_len >= 0) &&
2843		    tp_len > dev->mtu + reserve &&
2844		    !vnet_hdr_sz &&
2845		    !packet_extra_vlan_len_allowed(dev, skb))
2846			tp_len = -EMSGSIZE;
2847
2848		if (unlikely(tp_len < 0)) {
2849tpacket_error:
2850			if (packet_sock_flag(po, PACKET_SOCK_TP_LOSS)) {
2851				__packet_set_status(po, ph,
2852						TP_STATUS_AVAILABLE);
2853				packet_increment_head(&po->tx_ring);
2854				kfree_skb(skb);
2855				continue;
2856			} else {
2857				status = TP_STATUS_WRONG_FORMAT;
2858				err = tp_len;
2859				goto out_status;
2860			}
2861		}
2862
2863		if (vnet_hdr_sz) {
2864			if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2865				tp_len = -EINVAL;
2866				goto tpacket_error;
2867			}
2868			virtio_net_hdr_set_proto(skb, vnet_hdr);
2869		}
2870
2871		skb->destructor = tpacket_destruct_skb;
2872		__packet_set_status(po, ph, TP_STATUS_SENDING);
2873		packet_inc_pending(&po->tx_ring);
2874
2875		status = TP_STATUS_SEND_REQUEST;
2876		err = packet_xmit(po, skb);
2877		if (unlikely(err != 0)) {
2878			if (err > 0)
2879				err = net_xmit_errno(err);
2880			if (err && __packet_get_status(po, ph) ==
2881				   TP_STATUS_AVAILABLE) {
2882				/* skb was destructed already */
2883				skb = NULL;
2884				goto out_status;
2885			}
2886			/*
2887			 * skb was dropped but not destructed yet;
2888			 * let's treat it like congestion or err < 0
2889			 */
2890			err = 0;
2891		}
2892		packet_increment_head(&po->tx_ring);
2893		len_sum += tp_len;
2894	} while (likely((ph != NULL) ||
2895		/* Note: packet_read_pending() might be slow if we have
2896		 * to call it as it's per_cpu variable, but in fast-path
2897		 * we already short-circuit the loop with the first
2898		 * condition, and luckily don't have to go that path
2899		 * anyway.
2900		 */
2901		 (need_wait && packet_read_pending(&po->tx_ring))));
2902
2903	err = len_sum;
2904	goto out_put;
2905
2906out_status:
2907	__packet_set_status(po, ph, status);
2908	kfree_skb(skb);
2909out_put:
2910	dev_put(dev);
2911out:
2912	mutex_unlock(&po->pg_vec_lock);
2913	return err;
2914}
2915
2916static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2917				        size_t reserve, size_t len,
2918				        size_t linear, int noblock,
2919				        int *err)
2920{
2921	struct sk_buff *skb;
2922
2923	/* Under a page?  Don't bother with paged skb. */
2924	if (prepad + len < PAGE_SIZE || !linear)
2925		linear = len;
2926
2927	if (len - linear > MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER))
2928		linear = len - MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER);
2929	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2930				   err, PAGE_ALLOC_COSTLY_ORDER);
2931	if (!skb)
2932		return NULL;
2933
2934	skb_reserve(skb, reserve);
2935	skb_put(skb, linear);
2936	skb->data_len = len - linear;
2937	skb->len += len - linear;
2938
2939	return skb;
2940}
2941
2942static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2943{
2944	struct sock *sk = sock->sk;
2945	DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2946	struct sk_buff *skb;
2947	struct net_device *dev;
2948	__be16 proto;
2949	unsigned char *addr = NULL;
2950	int err, reserve = 0;
2951	struct sockcm_cookie sockc;
2952	struct virtio_net_hdr vnet_hdr = { 0 };
2953	int offset = 0;
2954	struct packet_sock *po = pkt_sk(sk);
2955	int vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
2956	int hlen, tlen, linear;
2957	int extra_len = 0;
2958
2959	/*
2960	 *	Get and verify the address.
2961	 */
2962
2963	if (likely(saddr == NULL)) {
2964		dev	= packet_cached_dev_get(po);
2965		proto	= READ_ONCE(po->num);
2966	} else {
2967		err = -EINVAL;
2968		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2969			goto out;
2970		if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2971			goto out;
2972		proto	= saddr->sll_protocol;
2973		dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2974		if (sock->type == SOCK_DGRAM) {
2975			if (dev && msg->msg_namelen < dev->addr_len +
2976				   offsetof(struct sockaddr_ll, sll_addr))
2977				goto out_unlock;
2978			addr = saddr->sll_addr;
2979		}
2980	}
2981
2982	err = -ENXIO;
2983	if (unlikely(dev == NULL))
2984		goto out_unlock;
2985	err = -ENETDOWN;
2986	if (unlikely(!(dev->flags & IFF_UP)))
2987		goto out_unlock;
2988
2989	sockcm_init(&sockc, sk);
2990	sockc.mark = READ_ONCE(sk->sk_mark);
2991	if (msg->msg_controllen) {
2992		err = sock_cmsg_send(sk, msg, &sockc);
2993		if (unlikely(err))
2994			goto out_unlock;
2995	}
2996
2997	if (sock->type == SOCK_RAW)
2998		reserve = dev->hard_header_len;
2999	if (vnet_hdr_sz) {
3000		err = packet_snd_vnet_parse(msg, &len, &vnet_hdr, vnet_hdr_sz);
3001		if (err)
3002			goto out_unlock;
 
3003	}
3004
3005	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
3006		if (!netif_supports_nofcs(dev)) {
3007			err = -EPROTONOSUPPORT;
3008			goto out_unlock;
3009		}
3010		extra_len = 4; /* We're doing our own CRC */
3011	}
3012
3013	err = -EMSGSIZE;
3014	if (!vnet_hdr.gso_type &&
3015	    (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
3016		goto out_unlock;
3017
3018	err = -ENOBUFS;
3019	hlen = LL_RESERVED_SPACE(dev);
3020	tlen = dev->needed_tailroom;
3021	linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
3022	linear = max(linear, min_t(int, len, dev->hard_header_len));
3023	skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
3024			       msg->msg_flags & MSG_DONTWAIT, &err);
3025	if (skb == NULL)
3026		goto out_unlock;
3027
3028	skb_reset_network_header(skb);
3029
3030	err = -EINVAL;
3031	if (sock->type == SOCK_DGRAM) {
3032		offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
3033		if (unlikely(offset < 0))
3034			goto out_free;
3035	} else if (reserve) {
3036		skb_reserve(skb, -reserve);
3037		if (len < reserve + sizeof(struct ipv6hdr) &&
3038		    dev->min_header_len != dev->hard_header_len)
3039			skb_reset_network_header(skb);
3040	}
3041
3042	/* Returns -EFAULT on error */
3043	err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
3044	if (err)
3045		goto out_free;
3046
3047	if ((sock->type == SOCK_RAW &&
3048	     !dev_validate_header(dev, skb->data, len)) || !skb->len) {
3049		err = -EINVAL;
3050		goto out_free;
3051	}
3052
3053	skb_setup_tx_timestamp(skb, sockc.tsflags);
3054
3055	if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
3056	    !packet_extra_vlan_len_allowed(dev, skb)) {
3057		err = -EMSGSIZE;
3058		goto out_free;
3059	}
3060
3061	skb->protocol = proto;
3062	skb->dev = dev;
3063	skb->priority = READ_ONCE(sk->sk_priority);
3064	skb->mark = sockc.mark;
3065	skb->tstamp = sockc.transmit_time;
3066
3067	if (unlikely(extra_len == 4))
3068		skb->no_fcs = 1;
3069
3070	packet_parse_headers(skb, sock);
3071
3072	if (vnet_hdr_sz) {
3073		err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
3074		if (err)
3075			goto out_free;
3076		len += vnet_hdr_sz;
3077		virtio_net_hdr_set_proto(skb, &vnet_hdr);
3078	}
3079
3080	err = packet_xmit(po, skb);
 
 
 
3081
3082	if (unlikely(err != 0)) {
3083		if (err > 0)
3084			err = net_xmit_errno(err);
3085		if (err)
3086			goto out_unlock;
3087	}
3088
3089	dev_put(dev);
3090
3091	return len;
3092
3093out_free:
3094	kfree_skb(skb);
3095out_unlock:
3096	dev_put(dev);
 
3097out:
3098	return err;
3099}
3100
3101static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
3102{
3103	struct sock *sk = sock->sk;
3104	struct packet_sock *po = pkt_sk(sk);
3105
3106	/* Reading tx_ring.pg_vec without holding pg_vec_lock is racy.
3107	 * tpacket_snd() will redo the check safely.
3108	 */
3109	if (data_race(po->tx_ring.pg_vec))
3110		return tpacket_snd(po, msg);
3111
3112	return packet_snd(sock, msg, len);
3113}
3114
3115/*
3116 *	Close a PACKET socket. This is fairly simple. We immediately go
3117 *	to 'closed' state and remove our protocol entry in the device list.
3118 */
3119
3120static int packet_release(struct socket *sock)
3121{
3122	struct sock *sk = sock->sk;
3123	struct packet_sock *po;
3124	struct packet_fanout *f;
3125	struct net *net;
3126	union tpacket_req_u req_u;
3127
3128	if (!sk)
3129		return 0;
3130
3131	net = sock_net(sk);
3132	po = pkt_sk(sk);
3133
3134	mutex_lock(&net->packet.sklist_lock);
3135	sk_del_node_init_rcu(sk);
3136	mutex_unlock(&net->packet.sklist_lock);
3137
 
3138	sock_prot_inuse_add(net, sk->sk_prot, -1);
 
3139
3140	spin_lock(&po->bind_lock);
3141	unregister_prot_hook(sk, false);
3142	packet_cached_dev_reset(po);
3143
3144	if (po->prot_hook.dev) {
3145		netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3146		po->prot_hook.dev = NULL;
3147	}
3148	spin_unlock(&po->bind_lock);
3149
3150	packet_flush_mclist(sk);
3151
3152	lock_sock(sk);
3153	if (po->rx_ring.pg_vec) {
3154		memset(&req_u, 0, sizeof(req_u));
3155		packet_set_ring(sk, &req_u, 1, 0);
3156	}
3157
3158	if (po->tx_ring.pg_vec) {
3159		memset(&req_u, 0, sizeof(req_u));
3160		packet_set_ring(sk, &req_u, 1, 1);
3161	}
3162	release_sock(sk);
3163
3164	f = fanout_release(sk);
3165
3166	synchronize_net();
3167
3168	kfree(po->rollover);
3169	if (f) {
3170		fanout_release_data(f);
3171		kvfree(f);
3172	}
3173	/*
3174	 *	Now the socket is dead. No more input will appear.
3175	 */
3176	sock_orphan(sk);
3177	sock->sk = NULL;
3178
3179	/* Purge queues */
3180
3181	skb_queue_purge(&sk->sk_receive_queue);
3182	packet_free_pending(po);
 
3183
3184	sock_put(sk);
3185	return 0;
3186}
3187
3188/*
3189 *	Attach a packet hook.
3190 */
3191
3192static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3193			  __be16 proto)
3194{
3195	struct packet_sock *po = pkt_sk(sk);
 
 
 
3196	struct net_device *dev = NULL;
 
3197	bool unlisted = false;
3198	bool need_rehook;
3199	int ret = 0;
3200
3201	lock_sock(sk);
3202	spin_lock(&po->bind_lock);
3203	if (!proto)
3204		proto = po->num;
3205
3206	rcu_read_lock();
3207
3208	if (po->fanout) {
3209		ret = -EINVAL;
3210		goto out_unlock;
3211	}
3212
3213	if (name) {
3214		dev = dev_get_by_name_rcu(sock_net(sk), name);
3215		if (!dev) {
3216			ret = -ENODEV;
3217			goto out_unlock;
3218		}
3219	} else if (ifindex) {
3220		dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3221		if (!dev) {
3222			ret = -ENODEV;
3223			goto out_unlock;
3224		}
3225	}
3226
3227	need_rehook = po->prot_hook.type != proto || po->prot_hook.dev != dev;
 
 
 
 
 
 
3228
3229	if (need_rehook) {
3230		dev_hold(dev);
3231		if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
3232			rcu_read_unlock();
3233			/* prevents packet_notifier() from calling
3234			 * register_prot_hook()
3235			 */
3236			WRITE_ONCE(po->num, 0);
3237			__unregister_prot_hook(sk, true);
3238			rcu_read_lock();
 
3239			if (dev)
3240				unlisted = !dev_get_by_index_rcu(sock_net(sk),
3241								 dev->ifindex);
3242		}
3243
3244		BUG_ON(packet_sock_flag(po, PACKET_SOCK_RUNNING));
3245		WRITE_ONCE(po->num, proto);
3246		po->prot_hook.type = proto;
3247
3248		netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3249
3250		if (unlikely(unlisted)) {
 
3251			po->prot_hook.dev = NULL;
3252			WRITE_ONCE(po->ifindex, -1);
3253			packet_cached_dev_reset(po);
3254		} else {
3255			netdev_hold(dev, &po->prot_hook.dev_tracker,
3256				    GFP_ATOMIC);
3257			po->prot_hook.dev = dev;
3258			WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0);
3259			packet_cached_dev_assign(po, dev);
3260		}
3261		dev_put(dev);
3262	}
 
 
3263
3264	if (proto == 0 || !need_rehook)
3265		goto out_unlock;
3266
3267	if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3268		register_prot_hook(sk);
3269	} else {
3270		sk->sk_err = ENETDOWN;
3271		if (!sock_flag(sk, SOCK_DEAD))
3272			sk_error_report(sk);
3273	}
3274
3275out_unlock:
3276	rcu_read_unlock();
3277	spin_unlock(&po->bind_lock);
3278	release_sock(sk);
3279	return ret;
3280}
3281
3282/*
3283 *	Bind a packet socket to a device
3284 */
3285
3286static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3287			    int addr_len)
3288{
3289	struct sock *sk = sock->sk;
3290	char name[sizeof(uaddr->sa_data_min) + 1];
3291
3292	/*
3293	 *	Check legality
3294	 */
3295
3296	if (addr_len != sizeof(struct sockaddr))
3297		return -EINVAL;
3298	/* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3299	 * zero-terminated.
3300	 */
3301	memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data_min));
3302	name[sizeof(uaddr->sa_data_min)] = 0;
3303
3304	return packet_do_bind(sk, name, 0, 0);
3305}
3306
3307static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3308{
3309	struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3310	struct sock *sk = sock->sk;
3311
3312	/*
3313	 *	Check legality
3314	 */
3315
3316	if (addr_len < sizeof(struct sockaddr_ll))
3317		return -EINVAL;
3318	if (sll->sll_family != AF_PACKET)
3319		return -EINVAL;
3320
3321	return packet_do_bind(sk, NULL, sll->sll_ifindex, sll->sll_protocol);
 
3322}
3323
3324static struct proto packet_proto = {
3325	.name	  = "PACKET",
3326	.owner	  = THIS_MODULE,
3327	.obj_size = sizeof(struct packet_sock),
3328};
3329
3330/*
3331 *	Create a packet of type SOCK_PACKET.
3332 */
3333
3334static int packet_create(struct net *net, struct socket *sock, int protocol,
3335			 int kern)
3336{
3337	struct sock *sk;
3338	struct packet_sock *po;
3339	__be16 proto = (__force __be16)protocol; /* weird, but documented */
3340	int err;
3341
3342	if (!ns_capable(net->user_ns, CAP_NET_RAW))
3343		return -EPERM;
3344	if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3345	    sock->type != SOCK_PACKET)
3346		return -ESOCKTNOSUPPORT;
3347
3348	sock->state = SS_UNCONNECTED;
3349
3350	err = -ENOBUFS;
3351	sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3352	if (sk == NULL)
3353		goto out;
3354
3355	sock->ops = &packet_ops;
3356	if (sock->type == SOCK_PACKET)
3357		sock->ops = &packet_ops_spkt;
3358
3359	sock_init_data(sock, sk);
3360
3361	po = pkt_sk(sk);
3362	init_completion(&po->skb_completion);
3363	sk->sk_family = PF_PACKET;
3364	po->num = proto;
 
3365
3366	err = packet_alloc_pending(po);
3367	if (err)
3368		goto out2;
3369
3370	packet_cached_dev_reset(po);
3371
3372	sk->sk_destruct = packet_sock_destruct;
 
3373
3374	/*
3375	 *	Attach a protocol block
3376	 */
3377
3378	spin_lock_init(&po->bind_lock);
3379	mutex_init(&po->pg_vec_lock);
3380	po->rollover = NULL;
3381	po->prot_hook.func = packet_rcv;
3382
3383	if (sock->type == SOCK_PACKET)
3384		po->prot_hook.func = packet_rcv_spkt;
3385
3386	po->prot_hook.af_packet_priv = sk;
3387	po->prot_hook.af_packet_net = sock_net(sk);
3388
3389	if (proto) {
3390		po->prot_hook.type = proto;
3391		__register_prot_hook(sk);
3392	}
3393
3394	mutex_lock(&net->packet.sklist_lock);
3395	sk_add_node_tail_rcu(sk, &net->packet.sklist);
3396	mutex_unlock(&net->packet.sklist_lock);
3397
 
3398	sock_prot_inuse_add(net, &packet_proto, 1);
 
3399
3400	return 0;
3401out2:
3402	sk_free(sk);
3403out:
3404	return err;
3405}
3406
3407/*
3408 *	Pull a packet from our receive queue and hand it to the user.
3409 *	If necessary we block.
3410 */
3411
3412static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3413			  int flags)
3414{
3415	struct sock *sk = sock->sk;
3416	struct sk_buff *skb;
3417	int copied, err;
3418	int vnet_hdr_len = READ_ONCE(pkt_sk(sk)->vnet_hdr_sz);
3419	unsigned int origlen = 0;
3420
3421	err = -EINVAL;
3422	if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3423		goto out;
3424
3425#if 0
3426	/* What error should we return now? EUNATTACH? */
3427	if (pkt_sk(sk)->ifindex < 0)
3428		return -ENODEV;
3429#endif
3430
3431	if (flags & MSG_ERRQUEUE) {
3432		err = sock_recv_errqueue(sk, msg, len,
3433					 SOL_PACKET, PACKET_TX_TIMESTAMP);
3434		goto out;
3435	}
3436
3437	/*
3438	 *	Call the generic datagram receiver. This handles all sorts
3439	 *	of horrible races and re-entrancy so we can forget about it
3440	 *	in the protocol layers.
3441	 *
3442	 *	Now it will return ENETDOWN, if device have just gone down,
3443	 *	but then it will block.
3444	 */
3445
3446	skb = skb_recv_datagram(sk, flags, &err);
3447
3448	/*
3449	 *	An error occurred so return it. Because skb_recv_datagram()
3450	 *	handles the blocking we don't see and worry about blocking
3451	 *	retries.
3452	 */
3453
3454	if (skb == NULL)
3455		goto out;
3456
3457	packet_rcv_try_clear_pressure(pkt_sk(sk));
3458
3459	if (vnet_hdr_len) {
3460		err = packet_rcv_vnet(msg, skb, &len, vnet_hdr_len);
3461		if (err)
3462			goto out_free;
 
3463	}
3464
3465	/* You lose any data beyond the buffer you gave. If it worries
3466	 * a user program they can ask the device for its MTU
3467	 * anyway.
3468	 */
3469	copied = skb->len;
3470	if (copied > len) {
3471		copied = len;
3472		msg->msg_flags |= MSG_TRUNC;
3473	}
3474
3475	err = skb_copy_datagram_msg(skb, 0, msg, copied);
3476	if (err)
3477		goto out_free;
3478
3479	if (sock->type != SOCK_PACKET) {
3480		struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3481
3482		/* Original length was stored in sockaddr_ll fields */
3483		origlen = PACKET_SKB_CB(skb)->sa.origlen;
3484		sll->sll_family = AF_PACKET;
3485		sll->sll_protocol = skb->protocol;
3486	}
3487
3488	sock_recv_cmsgs(msg, sk, skb);
3489
3490	if (msg->msg_name) {
3491		const size_t max_len = min(sizeof(skb->cb),
3492					   sizeof(struct sockaddr_storage));
3493		int copy_len;
3494
3495		/* If the address length field is there to be filled
3496		 * in, we fill it in now.
3497		 */
3498		if (sock->type == SOCK_PACKET) {
3499			__sockaddr_check_size(sizeof(struct sockaddr_pkt));
3500			msg->msg_namelen = sizeof(struct sockaddr_pkt);
3501			copy_len = msg->msg_namelen;
3502		} else {
3503			struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3504
3505			msg->msg_namelen = sll->sll_halen +
3506				offsetof(struct sockaddr_ll, sll_addr);
3507			copy_len = msg->msg_namelen;
3508			if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3509				memset(msg->msg_name +
3510				       offsetof(struct sockaddr_ll, sll_addr),
3511				       0, sizeof(sll->sll_addr));
3512				msg->msg_namelen = sizeof(struct sockaddr_ll);
3513			}
3514		}
3515		if (WARN_ON_ONCE(copy_len > max_len)) {
3516			copy_len = max_len;
3517			msg->msg_namelen = copy_len;
3518		}
3519		memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3520	}
3521
3522	if (packet_sock_flag(pkt_sk(sk), PACKET_SOCK_AUXDATA)) {
3523		struct tpacket_auxdata aux;
3524
3525		aux.tp_status = TP_STATUS_USER;
3526		if (skb->ip_summed == CHECKSUM_PARTIAL)
3527			aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3528		else if (skb->pkt_type != PACKET_OUTGOING &&
3529			 skb_csum_unnecessary(skb))
 
3530			aux.tp_status |= TP_STATUS_CSUM_VALID;
3531		if (skb_is_gso(skb) && skb_is_gso_tcp(skb))
3532			aux.tp_status |= TP_STATUS_GSO_TCP;
3533
3534		aux.tp_len = origlen;
3535		aux.tp_snaplen = skb->len;
3536		aux.tp_mac = 0;
3537		aux.tp_net = skb_network_offset(skb);
3538		if (skb_vlan_tag_present(skb)) {
3539			aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3540			aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3541			aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3542		} else {
3543			aux.tp_vlan_tci = 0;
3544			aux.tp_vlan_tpid = 0;
3545		}
3546		put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3547	}
3548
3549	/*
3550	 *	Free or return the buffer as appropriate. Again this
3551	 *	hides all the races and re-entrancy issues from us.
3552	 */
3553	err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3554
3555out_free:
3556	skb_free_datagram(sk, skb);
3557out:
3558	return err;
3559}
3560
3561static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3562			       int peer)
3563{
3564	struct net_device *dev;
3565	struct sock *sk	= sock->sk;
3566
3567	if (peer)
3568		return -EOPNOTSUPP;
3569
3570	uaddr->sa_family = AF_PACKET;
3571	memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data_min));
3572	rcu_read_lock();
3573	dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex));
3574	if (dev)
3575		strscpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data_min));
3576	rcu_read_unlock();
3577
3578	return sizeof(*uaddr);
3579}
3580
3581static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3582			  int peer)
3583{
3584	struct net_device *dev;
3585	struct sock *sk = sock->sk;
3586	struct packet_sock *po = pkt_sk(sk);
3587	DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3588	int ifindex;
3589
3590	if (peer)
3591		return -EOPNOTSUPP;
3592
3593	ifindex = READ_ONCE(po->ifindex);
3594	sll->sll_family = AF_PACKET;
3595	sll->sll_ifindex = ifindex;
3596	sll->sll_protocol = READ_ONCE(po->num);
3597	sll->sll_pkttype = 0;
3598	rcu_read_lock();
3599	dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3600	if (dev) {
3601		sll->sll_hatype = dev->type;
3602		sll->sll_halen = dev->addr_len;
3603
3604		/* Let __fortify_memcpy_chk() know the actual buffer size. */
3605		memcpy(((struct sockaddr_storage *)sll)->__data +
3606		       offsetof(struct sockaddr_ll, sll_addr) -
3607		       offsetofend(struct sockaddr_ll, sll_family),
3608		       dev->dev_addr, dev->addr_len);
3609	} else {
3610		sll->sll_hatype = 0;	/* Bad: we have no ARPHRD_UNSPEC */
3611		sll->sll_halen = 0;
3612	}
3613	rcu_read_unlock();
3614
3615	return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3616}
3617
3618static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3619			 int what)
3620{
3621	switch (i->type) {
3622	case PACKET_MR_MULTICAST:
3623		if (i->alen != dev->addr_len)
3624			return -EINVAL;
3625		if (what > 0)
3626			return dev_mc_add(dev, i->addr);
3627		else
3628			return dev_mc_del(dev, i->addr);
3629		break;
3630	case PACKET_MR_PROMISC:
3631		return dev_set_promiscuity(dev, what);
3632	case PACKET_MR_ALLMULTI:
3633		return dev_set_allmulti(dev, what);
3634	case PACKET_MR_UNICAST:
3635		if (i->alen != dev->addr_len)
3636			return -EINVAL;
3637		if (what > 0)
3638			return dev_uc_add(dev, i->addr);
3639		else
3640			return dev_uc_del(dev, i->addr);
3641		break;
3642	default:
3643		break;
3644	}
3645	return 0;
3646}
3647
3648static void packet_dev_mclist_delete(struct net_device *dev,
3649				     struct packet_mclist **mlp)
3650{
3651	struct packet_mclist *ml;
3652
3653	while ((ml = *mlp) != NULL) {
3654		if (ml->ifindex == dev->ifindex) {
3655			packet_dev_mc(dev, ml, -1);
3656			*mlp = ml->next;
3657			kfree(ml);
3658		} else
3659			mlp = &ml->next;
3660	}
3661}
3662
3663static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3664{
3665	struct packet_sock *po = pkt_sk(sk);
3666	struct packet_mclist *ml, *i;
3667	struct net_device *dev;
3668	int err;
3669
3670	rtnl_lock();
3671
3672	err = -ENODEV;
3673	dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3674	if (!dev)
3675		goto done;
3676
3677	err = -EINVAL;
3678	if (mreq->mr_alen > dev->addr_len)
3679		goto done;
3680
3681	err = -ENOBUFS;
3682	i = kmalloc(sizeof(*i), GFP_KERNEL);
3683	if (i == NULL)
3684		goto done;
3685
3686	err = 0;
3687	for (ml = po->mclist; ml; ml = ml->next) {
3688		if (ml->ifindex == mreq->mr_ifindex &&
3689		    ml->type == mreq->mr_type &&
3690		    ml->alen == mreq->mr_alen &&
3691		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3692			ml->count++;
3693			/* Free the new element ... */
3694			kfree(i);
3695			goto done;
3696		}
3697	}
3698
3699	i->type = mreq->mr_type;
3700	i->ifindex = mreq->mr_ifindex;
3701	i->alen = mreq->mr_alen;
3702	memcpy(i->addr, mreq->mr_address, i->alen);
3703	memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3704	i->count = 1;
3705	i->next = po->mclist;
3706	po->mclist = i;
3707	err = packet_dev_mc(dev, i, 1);
3708	if (err) {
3709		po->mclist = i->next;
3710		kfree(i);
3711	}
3712
3713done:
3714	rtnl_unlock();
3715	return err;
3716}
3717
3718static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3719{
3720	struct packet_mclist *ml, **mlp;
3721
3722	rtnl_lock();
3723
3724	for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3725		if (ml->ifindex == mreq->mr_ifindex &&
3726		    ml->type == mreq->mr_type &&
3727		    ml->alen == mreq->mr_alen &&
3728		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3729			if (--ml->count == 0) {
3730				struct net_device *dev;
3731				*mlp = ml->next;
3732				dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3733				if (dev)
3734					packet_dev_mc(dev, ml, -1);
3735				kfree(ml);
3736			}
3737			break;
3738		}
3739	}
3740	rtnl_unlock();
3741	return 0;
3742}
3743
3744static void packet_flush_mclist(struct sock *sk)
3745{
3746	struct packet_sock *po = pkt_sk(sk);
3747	struct packet_mclist *ml;
3748
3749	if (!po->mclist)
3750		return;
3751
3752	rtnl_lock();
3753	while ((ml = po->mclist) != NULL) {
3754		struct net_device *dev;
3755
3756		po->mclist = ml->next;
3757		dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3758		if (dev != NULL)
3759			packet_dev_mc(dev, ml, -1);
3760		kfree(ml);
3761	}
3762	rtnl_unlock();
3763}
3764
3765static int
3766packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval,
3767		  unsigned int optlen)
3768{
3769	struct sock *sk = sock->sk;
3770	struct packet_sock *po = pkt_sk(sk);
3771	int ret;
3772
3773	if (level != SOL_PACKET)
3774		return -ENOPROTOOPT;
3775
3776	switch (optname) {
3777	case PACKET_ADD_MEMBERSHIP:
3778	case PACKET_DROP_MEMBERSHIP:
3779	{
3780		struct packet_mreq_max mreq;
3781		int len = optlen;
3782		memset(&mreq, 0, sizeof(mreq));
3783		if (len < sizeof(struct packet_mreq))
3784			return -EINVAL;
3785		if (len > sizeof(mreq))
3786			len = sizeof(mreq);
3787		if (copy_from_sockptr(&mreq, optval, len))
3788			return -EFAULT;
3789		if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3790			return -EINVAL;
3791		if (optname == PACKET_ADD_MEMBERSHIP)
3792			ret = packet_mc_add(sk, &mreq);
3793		else
3794			ret = packet_mc_drop(sk, &mreq);
3795		return ret;
3796	}
3797
3798	case PACKET_RX_RING:
3799	case PACKET_TX_RING:
3800	{
3801		union tpacket_req_u req_u;
3802		int len;
3803
3804		lock_sock(sk);
3805		switch (po->tp_version) {
3806		case TPACKET_V1:
3807		case TPACKET_V2:
3808			len = sizeof(req_u.req);
3809			break;
3810		case TPACKET_V3:
3811		default:
3812			len = sizeof(req_u.req3);
3813			break;
3814		}
3815		if (optlen < len) {
3816			ret = -EINVAL;
3817		} else {
3818			if (copy_from_sockptr(&req_u.req, optval, len))
3819				ret = -EFAULT;
3820			else
3821				ret = packet_set_ring(sk, &req_u, 0,
3822						    optname == PACKET_TX_RING);
3823		}
3824		release_sock(sk);
3825		return ret;
3826	}
3827	case PACKET_COPY_THRESH:
3828	{
3829		int val;
3830
3831		if (optlen != sizeof(val))
3832			return -EINVAL;
3833		if (copy_from_sockptr(&val, optval, sizeof(val)))
3834			return -EFAULT;
3835
3836		WRITE_ONCE(pkt_sk(sk)->copy_thresh, val);
3837		return 0;
3838	}
3839	case PACKET_VERSION:
3840	{
3841		int val;
3842
3843		if (optlen != sizeof(val))
3844			return -EINVAL;
3845		if (copy_from_sockptr(&val, optval, sizeof(val)))
3846			return -EFAULT;
3847		switch (val) {
3848		case TPACKET_V1:
3849		case TPACKET_V2:
3850		case TPACKET_V3:
3851			break;
3852		default:
3853			return -EINVAL;
3854		}
3855		lock_sock(sk);
3856		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3857			ret = -EBUSY;
3858		} else {
3859			po->tp_version = val;
3860			ret = 0;
3861		}
3862		release_sock(sk);
3863		return ret;
3864	}
3865	case PACKET_RESERVE:
3866	{
3867		unsigned int val;
3868
3869		if (optlen != sizeof(val))
3870			return -EINVAL;
3871		if (copy_from_sockptr(&val, optval, sizeof(val)))
3872			return -EFAULT;
3873		if (val > INT_MAX)
3874			return -EINVAL;
3875		lock_sock(sk);
3876		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3877			ret = -EBUSY;
3878		} else {
3879			po->tp_reserve = val;
3880			ret = 0;
3881		}
3882		release_sock(sk);
3883		return ret;
3884	}
3885	case PACKET_LOSS:
3886	{
3887		unsigned int val;
3888
3889		if (optlen != sizeof(val))
3890			return -EINVAL;
3891		if (copy_from_sockptr(&val, optval, sizeof(val)))
3892			return -EFAULT;
3893
3894		lock_sock(sk);
3895		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3896			ret = -EBUSY;
3897		} else {
3898			packet_sock_flag_set(po, PACKET_SOCK_TP_LOSS, val);
3899			ret = 0;
3900		}
3901		release_sock(sk);
3902		return ret;
3903	}
3904	case PACKET_AUXDATA:
3905	{
3906		int val;
3907
3908		if (optlen < sizeof(val))
3909			return -EINVAL;
3910		if (copy_from_sockptr(&val, optval, sizeof(val)))
3911			return -EFAULT;
3912
3913		packet_sock_flag_set(po, PACKET_SOCK_AUXDATA, val);
 
 
3914		return 0;
3915	}
3916	case PACKET_ORIGDEV:
3917	{
3918		int val;
3919
3920		if (optlen < sizeof(val))
3921			return -EINVAL;
3922		if (copy_from_sockptr(&val, optval, sizeof(val)))
3923			return -EFAULT;
3924
3925		packet_sock_flag_set(po, PACKET_SOCK_ORIGDEV, val);
 
 
3926		return 0;
3927	}
3928	case PACKET_VNET_HDR:
3929	case PACKET_VNET_HDR_SZ:
3930	{
3931		int val, hdr_len;
3932
3933		if (sock->type != SOCK_RAW)
3934			return -EINVAL;
3935		if (optlen < sizeof(val))
3936			return -EINVAL;
3937		if (copy_from_sockptr(&val, optval, sizeof(val)))
3938			return -EFAULT;
3939
3940		if (optname == PACKET_VNET_HDR_SZ) {
3941			if (val && val != sizeof(struct virtio_net_hdr) &&
3942			    val != sizeof(struct virtio_net_hdr_mrg_rxbuf))
3943				return -EINVAL;
3944			hdr_len = val;
3945		} else {
3946			hdr_len = val ? sizeof(struct virtio_net_hdr) : 0;
3947		}
3948		lock_sock(sk);
3949		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3950			ret = -EBUSY;
3951		} else {
3952			WRITE_ONCE(po->vnet_hdr_sz, hdr_len);
3953			ret = 0;
3954		}
3955		release_sock(sk);
3956		return ret;
3957	}
3958	case PACKET_TIMESTAMP:
3959	{
3960		int val;
3961
3962		if (optlen != sizeof(val))
3963			return -EINVAL;
3964		if (copy_from_sockptr(&val, optval, sizeof(val)))
3965			return -EFAULT;
3966
3967		WRITE_ONCE(po->tp_tstamp, val);
3968		return 0;
3969	}
3970	case PACKET_FANOUT:
3971	{
3972		struct fanout_args args = { 0 };
3973
3974		if (optlen != sizeof(int) && optlen != sizeof(args))
3975			return -EINVAL;
3976		if (copy_from_sockptr(&args, optval, optlen))
3977			return -EFAULT;
3978
3979		return fanout_add(sk, &args);
3980	}
3981	case PACKET_FANOUT_DATA:
3982	{
3983		/* Paired with the WRITE_ONCE() in fanout_add() */
3984		if (!READ_ONCE(po->fanout))
3985			return -EINVAL;
3986
3987		return fanout_set_data(po, optval, optlen);
3988	}
3989	case PACKET_IGNORE_OUTGOING:
3990	{
3991		int val;
3992
3993		if (optlen != sizeof(val))
3994			return -EINVAL;
3995		if (copy_from_sockptr(&val, optval, sizeof(val)))
3996			return -EFAULT;
3997		if (val < 0 || val > 1)
3998			return -EINVAL;
3999
4000		WRITE_ONCE(po->prot_hook.ignore_outgoing, !!val);
4001		return 0;
4002	}
4003	case PACKET_TX_HAS_OFF:
4004	{
4005		unsigned int val;
4006
4007		if (optlen != sizeof(val))
4008			return -EINVAL;
4009		if (copy_from_sockptr(&val, optval, sizeof(val)))
4010			return -EFAULT;
4011
4012		lock_sock(sk);
4013		if (!po->rx_ring.pg_vec && !po->tx_ring.pg_vec)
4014			packet_sock_flag_set(po, PACKET_SOCK_TX_HAS_OFF, val);
4015
 
 
 
4016		release_sock(sk);
4017		return 0;
4018	}
4019	case PACKET_QDISC_BYPASS:
4020	{
4021		int val;
4022
4023		if (optlen != sizeof(val))
4024			return -EINVAL;
4025		if (copy_from_sockptr(&val, optval, sizeof(val)))
4026			return -EFAULT;
4027
4028		packet_sock_flag_set(po, PACKET_SOCK_QDISC_BYPASS, val);
4029		return 0;
4030	}
4031	default:
4032		return -ENOPROTOOPT;
4033	}
4034}
4035
4036static int packet_getsockopt(struct socket *sock, int level, int optname,
4037			     char __user *optval, int __user *optlen)
4038{
4039	int len;
4040	int val, lv = sizeof(val);
4041	struct sock *sk = sock->sk;
4042	struct packet_sock *po = pkt_sk(sk);
4043	void *data = &val;
4044	union tpacket_stats_u st;
4045	struct tpacket_rollover_stats rstats;
4046	int drops;
4047
4048	if (level != SOL_PACKET)
4049		return -ENOPROTOOPT;
4050
4051	if (get_user(len, optlen))
4052		return -EFAULT;
4053
4054	if (len < 0)
4055		return -EINVAL;
4056
4057	switch (optname) {
4058	case PACKET_STATISTICS:
4059		spin_lock_bh(&sk->sk_receive_queue.lock);
4060		memcpy(&st, &po->stats, sizeof(st));
4061		memset(&po->stats, 0, sizeof(po->stats));
4062		spin_unlock_bh(&sk->sk_receive_queue.lock);
4063		drops = atomic_xchg(&po->tp_drops, 0);
4064
4065		if (po->tp_version == TPACKET_V3) {
4066			lv = sizeof(struct tpacket_stats_v3);
4067			st.stats3.tp_drops = drops;
4068			st.stats3.tp_packets += drops;
4069			data = &st.stats3;
4070		} else {
4071			lv = sizeof(struct tpacket_stats);
4072			st.stats1.tp_drops = drops;
4073			st.stats1.tp_packets += drops;
4074			data = &st.stats1;
4075		}
4076
4077		break;
4078	case PACKET_AUXDATA:
4079		val = packet_sock_flag(po, PACKET_SOCK_AUXDATA);
4080		break;
4081	case PACKET_ORIGDEV:
4082		val = packet_sock_flag(po, PACKET_SOCK_ORIGDEV);
4083		break;
4084	case PACKET_VNET_HDR:
4085		val = !!READ_ONCE(po->vnet_hdr_sz);
4086		break;
4087	case PACKET_VNET_HDR_SZ:
4088		val = READ_ONCE(po->vnet_hdr_sz);
4089		break;
4090	case PACKET_COPY_THRESH:
4091		val = READ_ONCE(pkt_sk(sk)->copy_thresh);
4092		break;
4093	case PACKET_VERSION:
4094		val = po->tp_version;
4095		break;
4096	case PACKET_HDRLEN:
4097		if (len > sizeof(int))
4098			len = sizeof(int);
4099		if (len < sizeof(int))
4100			return -EINVAL;
4101		if (copy_from_user(&val, optval, len))
4102			return -EFAULT;
4103		switch (val) {
4104		case TPACKET_V1:
4105			val = sizeof(struct tpacket_hdr);
4106			break;
4107		case TPACKET_V2:
4108			val = sizeof(struct tpacket2_hdr);
4109			break;
4110		case TPACKET_V3:
4111			val = sizeof(struct tpacket3_hdr);
4112			break;
4113		default:
4114			return -EINVAL;
4115		}
4116		break;
4117	case PACKET_RESERVE:
4118		val = po->tp_reserve;
4119		break;
4120	case PACKET_LOSS:
4121		val = packet_sock_flag(po, PACKET_SOCK_TP_LOSS);
4122		break;
4123	case PACKET_TIMESTAMP:
4124		val = READ_ONCE(po->tp_tstamp);
4125		break;
4126	case PACKET_FANOUT:
4127		val = (po->fanout ?
4128		       ((u32)po->fanout->id |
4129			((u32)po->fanout->type << 16) |
4130			((u32)po->fanout->flags << 24)) :
4131		       0);
4132		break;
4133	case PACKET_IGNORE_OUTGOING:
4134		val = READ_ONCE(po->prot_hook.ignore_outgoing);
4135		break;
4136	case PACKET_ROLLOVER_STATS:
4137		if (!po->rollover)
4138			return -EINVAL;
4139		rstats.tp_all = atomic_long_read(&po->rollover->num);
4140		rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4141		rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4142		data = &rstats;
4143		lv = sizeof(rstats);
4144		break;
4145	case PACKET_TX_HAS_OFF:
4146		val = packet_sock_flag(po, PACKET_SOCK_TX_HAS_OFF);
4147		break;
4148	case PACKET_QDISC_BYPASS:
4149		val = packet_sock_flag(po, PACKET_SOCK_QDISC_BYPASS);
4150		break;
4151	default:
4152		return -ENOPROTOOPT;
4153	}
4154
4155	if (len > lv)
4156		len = lv;
4157	if (put_user(len, optlen))
4158		return -EFAULT;
4159	if (copy_to_user(optval, data, len))
4160		return -EFAULT;
4161	return 0;
4162}
4163
4164static int packet_notifier(struct notifier_block *this,
4165			   unsigned long msg, void *ptr)
4166{
4167	struct sock *sk;
4168	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4169	struct net *net = dev_net(dev);
4170
4171	rcu_read_lock();
4172	sk_for_each_rcu(sk, &net->packet.sklist) {
4173		struct packet_sock *po = pkt_sk(sk);
4174
4175		switch (msg) {
4176		case NETDEV_UNREGISTER:
4177			if (po->mclist)
4178				packet_dev_mclist_delete(dev, &po->mclist);
4179			fallthrough;
4180
4181		case NETDEV_DOWN:
4182			if (dev->ifindex == po->ifindex) {
4183				spin_lock(&po->bind_lock);
4184				if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
4185					__unregister_prot_hook(sk, false);
4186					sk->sk_err = ENETDOWN;
4187					if (!sock_flag(sk, SOCK_DEAD))
4188						sk_error_report(sk);
4189				}
4190				if (msg == NETDEV_UNREGISTER) {
4191					packet_cached_dev_reset(po);
4192					WRITE_ONCE(po->ifindex, -1);
4193					netdev_put(po->prot_hook.dev,
4194						   &po->prot_hook.dev_tracker);
4195					po->prot_hook.dev = NULL;
4196				}
4197				spin_unlock(&po->bind_lock);
4198			}
4199			break;
4200		case NETDEV_UP:
4201			if (dev->ifindex == po->ifindex) {
4202				spin_lock(&po->bind_lock);
4203				if (po->num)
4204					register_prot_hook(sk);
4205				spin_unlock(&po->bind_lock);
4206			}
4207			break;
4208		}
4209	}
4210	rcu_read_unlock();
4211	return NOTIFY_DONE;
4212}
4213
4214
4215static int packet_ioctl(struct socket *sock, unsigned int cmd,
4216			unsigned long arg)
4217{
4218	struct sock *sk = sock->sk;
4219
4220	switch (cmd) {
4221	case SIOCOUTQ:
4222	{
4223		int amount = sk_wmem_alloc_get(sk);
4224
4225		return put_user(amount, (int __user *)arg);
4226	}
4227	case SIOCINQ:
4228	{
4229		struct sk_buff *skb;
4230		int amount = 0;
4231
4232		spin_lock_bh(&sk->sk_receive_queue.lock);
4233		skb = skb_peek(&sk->sk_receive_queue);
4234		if (skb)
4235			amount = skb->len;
4236		spin_unlock_bh(&sk->sk_receive_queue.lock);
4237		return put_user(amount, (int __user *)arg);
4238	}
4239#ifdef CONFIG_INET
4240	case SIOCADDRT:
4241	case SIOCDELRT:
4242	case SIOCDARP:
4243	case SIOCGARP:
4244	case SIOCSARP:
4245	case SIOCGIFADDR:
4246	case SIOCSIFADDR:
4247	case SIOCGIFBRDADDR:
4248	case SIOCSIFBRDADDR:
4249	case SIOCGIFNETMASK:
4250	case SIOCSIFNETMASK:
4251	case SIOCGIFDSTADDR:
4252	case SIOCSIFDSTADDR:
4253	case SIOCSIFFLAGS:
4254		return inet_dgram_ops.ioctl(sock, cmd, arg);
4255#endif
4256
4257	default:
4258		return -ENOIOCTLCMD;
4259	}
4260	return 0;
4261}
4262
4263static __poll_t packet_poll(struct file *file, struct socket *sock,
4264				poll_table *wait)
4265{
4266	struct sock *sk = sock->sk;
4267	struct packet_sock *po = pkt_sk(sk);
4268	__poll_t mask = datagram_poll(file, sock, wait);
4269
4270	spin_lock_bh(&sk->sk_receive_queue.lock);
4271	if (po->rx_ring.pg_vec) {
4272		if (!packet_previous_rx_frame(po, &po->rx_ring,
4273			TP_STATUS_KERNEL))
4274			mask |= EPOLLIN | EPOLLRDNORM;
4275	}
4276	packet_rcv_try_clear_pressure(po);
4277	spin_unlock_bh(&sk->sk_receive_queue.lock);
4278	spin_lock_bh(&sk->sk_write_queue.lock);
4279	if (po->tx_ring.pg_vec) {
4280		if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4281			mask |= EPOLLOUT | EPOLLWRNORM;
4282	}
4283	spin_unlock_bh(&sk->sk_write_queue.lock);
4284	return mask;
4285}
4286
4287
4288/* Dirty? Well, I still did not learn better way to account
4289 * for user mmaps.
4290 */
4291
4292static void packet_mm_open(struct vm_area_struct *vma)
4293{
4294	struct file *file = vma->vm_file;
4295	struct socket *sock = file->private_data;
4296	struct sock *sk = sock->sk;
4297
4298	if (sk)
4299		atomic_long_inc(&pkt_sk(sk)->mapped);
4300}
4301
4302static void packet_mm_close(struct vm_area_struct *vma)
4303{
4304	struct file *file = vma->vm_file;
4305	struct socket *sock = file->private_data;
4306	struct sock *sk = sock->sk;
4307
4308	if (sk)
4309		atomic_long_dec(&pkt_sk(sk)->mapped);
4310}
4311
4312static const struct vm_operations_struct packet_mmap_ops = {
4313	.open	=	packet_mm_open,
4314	.close	=	packet_mm_close,
4315};
4316
4317static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4318			unsigned int len)
4319{
4320	int i;
4321
4322	for (i = 0; i < len; i++) {
4323		if (likely(pg_vec[i].buffer)) {
4324			if (is_vmalloc_addr(pg_vec[i].buffer))
4325				vfree(pg_vec[i].buffer);
4326			else
4327				free_pages((unsigned long)pg_vec[i].buffer,
4328					   order);
4329			pg_vec[i].buffer = NULL;
4330		}
4331	}
4332	kfree(pg_vec);
4333}
4334
4335static char *alloc_one_pg_vec_page(unsigned long order)
4336{
4337	char *buffer;
4338	gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4339			  __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4340
4341	buffer = (char *) __get_free_pages(gfp_flags, order);
4342	if (buffer)
4343		return buffer;
4344
4345	/* __get_free_pages failed, fall back to vmalloc */
4346	buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4347	if (buffer)
4348		return buffer;
4349
4350	/* vmalloc failed, lets dig into swap here */
4351	gfp_flags &= ~__GFP_NORETRY;
4352	buffer = (char *) __get_free_pages(gfp_flags, order);
4353	if (buffer)
4354		return buffer;
4355
4356	/* complete and utter failure */
4357	return NULL;
4358}
4359
4360static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4361{
4362	unsigned int block_nr = req->tp_block_nr;
4363	struct pgv *pg_vec;
4364	int i;
4365
4366	pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4367	if (unlikely(!pg_vec))
4368		goto out;
4369
4370	for (i = 0; i < block_nr; i++) {
4371		pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4372		if (unlikely(!pg_vec[i].buffer))
4373			goto out_free_pgvec;
4374	}
4375
4376out:
4377	return pg_vec;
4378
4379out_free_pgvec:
4380	free_pg_vec(pg_vec, order, block_nr);
4381	pg_vec = NULL;
4382	goto out;
4383}
4384
4385static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4386		int closing, int tx_ring)
4387{
4388	struct pgv *pg_vec = NULL;
4389	struct packet_sock *po = pkt_sk(sk);
4390	unsigned long *rx_owner_map = NULL;
4391	int was_running, order = 0;
4392	struct packet_ring_buffer *rb;
4393	struct sk_buff_head *rb_queue;
4394	__be16 num;
4395	int err;
4396	/* Added to avoid minimal code churn */
4397	struct tpacket_req *req = &req_u->req;
4398
4399	rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4400	rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4401
4402	err = -EBUSY;
4403	if (!closing) {
4404		if (atomic_long_read(&po->mapped))
4405			goto out;
4406		if (packet_read_pending(rb))
4407			goto out;
4408	}
4409
4410	if (req->tp_block_nr) {
4411		unsigned int min_frame_size;
4412
4413		/* Sanity tests and some calculations */
4414		err = -EBUSY;
4415		if (unlikely(rb->pg_vec))
4416			goto out;
4417
4418		switch (po->tp_version) {
4419		case TPACKET_V1:
4420			po->tp_hdrlen = TPACKET_HDRLEN;
4421			break;
4422		case TPACKET_V2:
4423			po->tp_hdrlen = TPACKET2_HDRLEN;
4424			break;
4425		case TPACKET_V3:
4426			po->tp_hdrlen = TPACKET3_HDRLEN;
4427			break;
4428		}
4429
4430		err = -EINVAL;
4431		if (unlikely((int)req->tp_block_size <= 0))
4432			goto out;
4433		if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4434			goto out;
4435		min_frame_size = po->tp_hdrlen + po->tp_reserve;
4436		if (po->tp_version >= TPACKET_V3 &&
4437		    req->tp_block_size <
4438		    BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4439			goto out;
4440		if (unlikely(req->tp_frame_size < min_frame_size))
4441			goto out;
4442		if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4443			goto out;
4444
4445		rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4446		if (unlikely(rb->frames_per_block == 0))
4447			goto out;
4448		if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4449			goto out;
4450		if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4451					req->tp_frame_nr))
4452			goto out;
4453
4454		err = -ENOMEM;
4455		order = get_order(req->tp_block_size);
4456		pg_vec = alloc_pg_vec(req, order);
4457		if (unlikely(!pg_vec))
4458			goto out;
4459		switch (po->tp_version) {
4460		case TPACKET_V3:
4461			/* Block transmit is not supported yet */
4462			if (!tx_ring) {
4463				init_prb_bdqc(po, rb, pg_vec, req_u);
4464			} else {
4465				struct tpacket_req3 *req3 = &req_u->req3;
4466
4467				if (req3->tp_retire_blk_tov ||
4468				    req3->tp_sizeof_priv ||
4469				    req3->tp_feature_req_word) {
4470					err = -EINVAL;
4471					goto out_free_pg_vec;
4472				}
4473			}
4474			break;
4475		default:
4476			if (!tx_ring) {
4477				rx_owner_map = bitmap_alloc(req->tp_frame_nr,
4478					GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
4479				if (!rx_owner_map)
4480					goto out_free_pg_vec;
4481			}
4482			break;
4483		}
4484	}
4485	/* Done */
4486	else {
4487		err = -EINVAL;
4488		if (unlikely(req->tp_frame_nr))
4489			goto out;
4490	}
4491
4492
4493	/* Detach socket from network */
4494	spin_lock(&po->bind_lock);
4495	was_running = packet_sock_flag(po, PACKET_SOCK_RUNNING);
4496	num = po->num;
4497	if (was_running) {
4498		WRITE_ONCE(po->num, 0);
4499		__unregister_prot_hook(sk, false);
4500	}
4501	spin_unlock(&po->bind_lock);
4502
4503	synchronize_net();
4504
4505	err = -EBUSY;
4506	mutex_lock(&po->pg_vec_lock);
4507	if (closing || atomic_long_read(&po->mapped) == 0) {
4508		err = 0;
4509		spin_lock_bh(&rb_queue->lock);
4510		swap(rb->pg_vec, pg_vec);
4511		if (po->tp_version <= TPACKET_V2)
4512			swap(rb->rx_owner_map, rx_owner_map);
4513		rb->frame_max = (req->tp_frame_nr - 1);
4514		rb->head = 0;
4515		rb->frame_size = req->tp_frame_size;
4516		spin_unlock_bh(&rb_queue->lock);
4517
4518		swap(rb->pg_vec_order, order);
4519		swap(rb->pg_vec_len, req->tp_block_nr);
4520
4521		rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4522		po->prot_hook.func = (po->rx_ring.pg_vec) ?
4523						tpacket_rcv : packet_rcv;
4524		skb_queue_purge(rb_queue);
4525		if (atomic_long_read(&po->mapped))
4526			pr_err("packet_mmap: vma is busy: %ld\n",
4527			       atomic_long_read(&po->mapped));
4528	}
4529	mutex_unlock(&po->pg_vec_lock);
4530
4531	spin_lock(&po->bind_lock);
4532	if (was_running) {
4533		WRITE_ONCE(po->num, num);
4534		register_prot_hook(sk);
4535	}
4536	spin_unlock(&po->bind_lock);
4537	if (pg_vec && (po->tp_version > TPACKET_V2)) {
4538		/* Because we don't support block-based V3 on tx-ring */
4539		if (!tx_ring)
4540			prb_shutdown_retire_blk_timer(po, rb_queue);
4541	}
4542
4543out_free_pg_vec:
4544	if (pg_vec) {
4545		bitmap_free(rx_owner_map);
4546		free_pg_vec(pg_vec, order, req->tp_block_nr);
4547	}
4548out:
4549	return err;
4550}
4551
4552static int packet_mmap(struct file *file, struct socket *sock,
4553		struct vm_area_struct *vma)
4554{
4555	struct sock *sk = sock->sk;
4556	struct packet_sock *po = pkt_sk(sk);
4557	unsigned long size, expected_size;
4558	struct packet_ring_buffer *rb;
4559	unsigned long start;
4560	int err = -EINVAL;
4561	int i;
4562
4563	if (vma->vm_pgoff)
4564		return -EINVAL;
4565
4566	mutex_lock(&po->pg_vec_lock);
4567
4568	expected_size = 0;
4569	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4570		if (rb->pg_vec) {
4571			expected_size += rb->pg_vec_len
4572						* rb->pg_vec_pages
4573						* PAGE_SIZE;
4574		}
4575	}
4576
4577	if (expected_size == 0)
4578		goto out;
4579
4580	size = vma->vm_end - vma->vm_start;
4581	if (size != expected_size)
4582		goto out;
4583
4584	start = vma->vm_start;
4585	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4586		if (rb->pg_vec == NULL)
4587			continue;
4588
4589		for (i = 0; i < rb->pg_vec_len; i++) {
4590			struct page *page;
4591			void *kaddr = rb->pg_vec[i].buffer;
4592			int pg_num;
4593
4594			for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4595				page = pgv_to_page(kaddr);
4596				err = vm_insert_page(vma, start, page);
4597				if (unlikely(err))
4598					goto out;
4599				start += PAGE_SIZE;
4600				kaddr += PAGE_SIZE;
4601			}
4602		}
4603	}
4604
4605	atomic_long_inc(&po->mapped);
4606	vma->vm_ops = &packet_mmap_ops;
4607	err = 0;
4608
4609out:
4610	mutex_unlock(&po->pg_vec_lock);
4611	return err;
4612}
4613
4614static const struct proto_ops packet_ops_spkt = {
4615	.family =	PF_PACKET,
4616	.owner =	THIS_MODULE,
4617	.release =	packet_release,
4618	.bind =		packet_bind_spkt,
4619	.connect =	sock_no_connect,
4620	.socketpair =	sock_no_socketpair,
4621	.accept =	sock_no_accept,
4622	.getname =	packet_getname_spkt,
4623	.poll =		datagram_poll,
4624	.ioctl =	packet_ioctl,
4625	.gettstamp =	sock_gettstamp,
4626	.listen =	sock_no_listen,
4627	.shutdown =	sock_no_shutdown,
4628	.sendmsg =	packet_sendmsg_spkt,
4629	.recvmsg =	packet_recvmsg,
4630	.mmap =		sock_no_mmap,
 
4631};
4632
4633static const struct proto_ops packet_ops = {
4634	.family =	PF_PACKET,
4635	.owner =	THIS_MODULE,
4636	.release =	packet_release,
4637	.bind =		packet_bind,
4638	.connect =	sock_no_connect,
4639	.socketpair =	sock_no_socketpair,
4640	.accept =	sock_no_accept,
4641	.getname =	packet_getname,
4642	.poll =		packet_poll,
4643	.ioctl =	packet_ioctl,
4644	.gettstamp =	sock_gettstamp,
4645	.listen =	sock_no_listen,
4646	.shutdown =	sock_no_shutdown,
4647	.setsockopt =	packet_setsockopt,
4648	.getsockopt =	packet_getsockopt,
4649	.sendmsg =	packet_sendmsg,
4650	.recvmsg =	packet_recvmsg,
4651	.mmap =		packet_mmap,
 
4652};
4653
4654static const struct net_proto_family packet_family_ops = {
4655	.family =	PF_PACKET,
4656	.create =	packet_create,
4657	.owner	=	THIS_MODULE,
4658};
4659
4660static struct notifier_block packet_netdev_notifier = {
4661	.notifier_call =	packet_notifier,
4662};
4663
4664#ifdef CONFIG_PROC_FS
4665
4666static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4667	__acquires(RCU)
4668{
4669	struct net *net = seq_file_net(seq);
4670
4671	rcu_read_lock();
4672	return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4673}
4674
4675static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4676{
4677	struct net *net = seq_file_net(seq);
4678	return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4679}
4680
4681static void packet_seq_stop(struct seq_file *seq, void *v)
4682	__releases(RCU)
4683{
4684	rcu_read_unlock();
4685}
4686
4687static int packet_seq_show(struct seq_file *seq, void *v)
4688{
4689	if (v == SEQ_START_TOKEN)
4690		seq_printf(seq,
4691			   "%*sRefCnt Type Proto  Iface R Rmem   User   Inode\n",
4692			   IS_ENABLED(CONFIG_64BIT) ? -17 : -9, "sk");
4693	else {
4694		struct sock *s = sk_entry(v);
4695		const struct packet_sock *po = pkt_sk(s);
4696
4697		seq_printf(seq,
4698			   "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4699			   s,
4700			   refcount_read(&s->sk_refcnt),
4701			   s->sk_type,
4702			   ntohs(READ_ONCE(po->num)),
4703			   READ_ONCE(po->ifindex),
4704			   packet_sock_flag(po, PACKET_SOCK_RUNNING),
4705			   atomic_read(&s->sk_rmem_alloc),
4706			   from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4707			   sock_i_ino(s));
4708	}
4709
4710	return 0;
4711}
4712
4713static const struct seq_operations packet_seq_ops = {
4714	.start	= packet_seq_start,
4715	.next	= packet_seq_next,
4716	.stop	= packet_seq_stop,
4717	.show	= packet_seq_show,
4718};
4719#endif
4720
4721static int __net_init packet_net_init(struct net *net)
4722{
4723	mutex_init(&net->packet.sklist_lock);
4724	INIT_HLIST_HEAD(&net->packet.sklist);
4725
4726#ifdef CONFIG_PROC_FS
4727	if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4728			sizeof(struct seq_net_private)))
4729		return -ENOMEM;
4730#endif /* CONFIG_PROC_FS */
4731
4732	return 0;
4733}
4734
4735static void __net_exit packet_net_exit(struct net *net)
4736{
4737	remove_proc_entry("packet", net->proc_net);
4738	WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4739}
4740
4741static struct pernet_operations packet_net_ops = {
4742	.init = packet_net_init,
4743	.exit = packet_net_exit,
4744};
4745
4746
4747static void __exit packet_exit(void)
4748{
 
 
4749	sock_unregister(PF_PACKET);
4750	proto_unregister(&packet_proto);
4751	unregister_netdevice_notifier(&packet_netdev_notifier);
4752	unregister_pernet_subsys(&packet_net_ops);
4753}
4754
4755static int __init packet_init(void)
4756{
4757	int rc;
4758
 
 
 
 
 
 
4759	rc = register_pernet_subsys(&packet_net_ops);
4760	if (rc)
4761		goto out;
4762	rc = register_netdevice_notifier(&packet_netdev_notifier);
4763	if (rc)
4764		goto out_pernet;
4765	rc = proto_register(&packet_proto, 0);
4766	if (rc)
4767		goto out_notifier;
4768	rc = sock_register(&packet_family_ops);
4769	if (rc)
4770		goto out_proto;
4771
4772	return 0;
4773
 
 
 
 
4774out_proto:
4775	proto_unregister(&packet_proto);
4776out_notifier:
4777	unregister_netdevice_notifier(&packet_netdev_notifier);
4778out_pernet:
4779	unregister_pernet_subsys(&packet_net_ops);
4780out:
4781	return rc;
4782}
4783
4784module_init(packet_init);
4785module_exit(packet_exit);
4786MODULE_DESCRIPTION("Packet socket support (AF_PACKET)");
4787MODULE_LICENSE("GPL");
4788MODULE_ALIAS_NETPROTO(PF_PACKET);