xref: /linux/net/packet/af_packet.c (revision 2d7ce0e8a704d1aedc4462817a4eade1c3b00fbe)
1 /*
2  * INET		An implementation of the TCP/IP protocol suite for the LINUX
3  *		operating system.  INET is implemented using the  BSD Socket
4  *		interface as the means of communication with the user level.
5  *
6  *		PACKET - implements raw packet sockets.
7  *
8  * Authors:	Ross Biro
9  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10  *		Alan Cox, <gw4pts@gw4pts.ampr.org>
11  *
12  * Fixes:
13  *		Alan Cox	:	verify_area() now used correctly
14  *		Alan Cox	:	new skbuff lists, look ma no backlogs!
15  *		Alan Cox	:	tidied skbuff lists.
16  *		Alan Cox	:	Now uses generic datagram routines I
17  *					added. Also fixed the peek/read crash
18  *					from all old Linux datagram code.
19  *		Alan Cox	:	Uses the improved datagram code.
20  *		Alan Cox	:	Added NULL's for socket options.
21  *		Alan Cox	:	Re-commented the code.
22  *		Alan Cox	:	Use new kernel side addressing
23  *		Rob Janssen	:	Correct MTU usage.
24  *		Dave Platt	:	Counter leaks caused by incorrect
25  *					interrupt locking and some slightly
26  *					dubious gcc output. Can you read
27  *					compiler: it said _VOLATILE_
28  *	Richard Kooijman	:	Timestamp fixes.
29  *		Alan Cox	:	New buffers. Use sk->mac.raw.
30  *		Alan Cox	:	sendmsg/recvmsg support.
31  *		Alan Cox	:	Protocol setting support
32  *	Alexey Kuznetsov	:	Untied from IPv4 stack.
33  *	Cyrus Durgin		:	Fixed kerneld for kmod.
34  *	Michal Ostrowski        :       Module initialization cleanup.
35  *         Ulises Alonso        :       Frame number limit removal and
36  *                                      packet_set_ring memory leak.
37  *		Eric Biederman	:	Allow for > 8 byte hardware addresses.
38  *					The convention is that longer addresses
39  *					will simply extend the hardware address
40  *					byte arrays at the end of sockaddr_ll
41  *					and packet_mreq.
42  *		Johann Baudy	:	Added TX RING.
43  *		Chetan Loke	:	Implemented TPACKET_V3 block abstraction
44  *					layer.
45  *					Copyright (C) 2011, <lokec@ccs.neu.edu>
46  *
47  *
48  *		This program is free software; you can redistribute it and/or
49  *		modify it under the terms of the GNU General Public License
50  *		as published by the Free Software Foundation; either version
51  *		2 of the License, or (at your option) any later version.
52  *
53  */
54 
55 #include <linux/types.h>
56 #include <linux/mm.h>
57 #include <linux/capability.h>
58 #include <linux/fcntl.h>
59 #include <linux/socket.h>
60 #include <linux/in.h>
61 #include <linux/inet.h>
62 #include <linux/netdevice.h>
63 #include <linux/if_packet.h>
64 #include <linux/wireless.h>
65 #include <linux/kernel.h>
66 #include <linux/kmod.h>
67 #include <linux/slab.h>
68 #include <linux/vmalloc.h>
69 #include <net/net_namespace.h>
70 #include <net/ip.h>
71 #include <net/protocol.h>
72 #include <linux/skbuff.h>
73 #include <net/sock.h>
74 #include <linux/errno.h>
75 #include <linux/timer.h>
76 #include <asm/uaccess.h>
77 #include <asm/ioctls.h>
78 #include <asm/page.h>
79 #include <asm/cacheflush.h>
80 #include <asm/io.h>
81 #include <linux/proc_fs.h>
82 #include <linux/seq_file.h>
83 #include <linux/poll.h>
84 #include <linux/module.h>
85 #include <linux/init.h>
86 #include <linux/mutex.h>
87 #include <linux/if_vlan.h>
88 #include <linux/virtio_net.h>
89 #include <linux/errqueue.h>
90 #include <linux/net_tstamp.h>
91 #include <linux/percpu.h>
92 #ifdef CONFIG_INET
93 #include <net/inet_common.h>
94 #endif
95 
96 #include "internal.h"
97 
98 /*
99    Assumptions:
100    - if device has no dev->hard_header routine, it adds and removes ll header
101      inside itself. In this case ll header is invisible outside of device,
102      but higher levels still should reserve dev->hard_header_len.
103      Some devices are enough clever to reallocate skb, when header
104      will not fit to reserved space (tunnel), another ones are silly
105      (PPP).
106    - packet socket receives packets with pulled ll header,
107      so that SOCK_RAW should push it back.
108 
109 On receive:
110 -----------
111 
112 Incoming, dev->hard_header!=NULL
113    mac_header -> ll header
114    data       -> data
115 
116 Outgoing, dev->hard_header!=NULL
117    mac_header -> ll header
118    data       -> ll header
119 
120 Incoming, dev->hard_header==NULL
121    mac_header -> UNKNOWN position. It is very likely, that it points to ll
122 		 header.  PPP makes it, that is wrong, because introduce
123 		 assymetry between rx and tx paths.
124    data       -> data
125 
126 Outgoing, dev->hard_header==NULL
127    mac_header -> data. ll header is still not built!
128    data       -> data
129 
130 Resume
131   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
132 
133 
134 On transmit:
135 ------------
136 
137 dev->hard_header != NULL
138    mac_header -> ll header
139    data       -> ll header
140 
141 dev->hard_header == NULL (ll header is added by device, we cannot control it)
142    mac_header -> data
143    data       -> data
144 
145    We should set nh.raw on output to correct posistion,
146    packet classifier depends on it.
147  */
148 
149 /* Private packet socket structures. */
150 
151 /* identical to struct packet_mreq except it has
152  * a longer address field.
153  */
154 struct packet_mreq_max {
155 	int		mr_ifindex;
156 	unsigned short	mr_type;
157 	unsigned short	mr_alen;
158 	unsigned char	mr_address[MAX_ADDR_LEN];
159 };
160 
161 union tpacket_uhdr {
162 	struct tpacket_hdr  *h1;
163 	struct tpacket2_hdr *h2;
164 	struct tpacket3_hdr *h3;
165 	void *raw;
166 };
167 
168 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
169 		int closing, int tx_ring);
170 
171 #define V3_ALIGNMENT	(8)
172 
173 #define BLK_HDR_LEN	(ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
174 
175 #define BLK_PLUS_PRIV(sz_of_priv) \
176 	(BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
177 
178 #define PGV_FROM_VMALLOC 1
179 
180 #define BLOCK_STATUS(x)	((x)->hdr.bh1.block_status)
181 #define BLOCK_NUM_PKTS(x)	((x)->hdr.bh1.num_pkts)
182 #define BLOCK_O2FP(x)		((x)->hdr.bh1.offset_to_first_pkt)
183 #define BLOCK_LEN(x)		((x)->hdr.bh1.blk_len)
184 #define BLOCK_SNUM(x)		((x)->hdr.bh1.seq_num)
185 #define BLOCK_O2PRIV(x)	((x)->offset_to_priv)
186 #define BLOCK_PRIV(x)		((void *)((char *)(x) + BLOCK_O2PRIV(x)))
187 
188 struct packet_sock;
189 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
190 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
191 		       struct packet_type *pt, struct net_device *orig_dev);
192 
193 static void *packet_previous_frame(struct packet_sock *po,
194 		struct packet_ring_buffer *rb,
195 		int status);
196 static void packet_increment_head(struct packet_ring_buffer *buff);
197 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
198 			struct tpacket_block_desc *);
199 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
200 			struct packet_sock *);
201 static void prb_retire_current_block(struct tpacket_kbdq_core *,
202 		struct packet_sock *, unsigned int status);
203 static int prb_queue_frozen(struct tpacket_kbdq_core *);
204 static void prb_open_block(struct tpacket_kbdq_core *,
205 		struct tpacket_block_desc *);
206 static void prb_retire_rx_blk_timer_expired(unsigned long);
207 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
208 static void prb_init_blk_timer(struct packet_sock *,
209 		struct tpacket_kbdq_core *,
210 		void (*func) (unsigned long));
211 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
212 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
213 		struct tpacket3_hdr *);
214 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
215 		struct tpacket3_hdr *);
216 static void packet_flush_mclist(struct sock *sk);
217 
218 struct packet_skb_cb {
219 	unsigned int origlen;
220 	union {
221 		struct sockaddr_pkt pkt;
222 		struct sockaddr_ll ll;
223 	} sa;
224 };
225 
226 #define PACKET_SKB_CB(__skb)	((struct packet_skb_cb *)((__skb)->cb))
227 
228 #define GET_PBDQC_FROM_RB(x)	((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
229 #define GET_PBLOCK_DESC(x, bid)	\
230 	((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
231 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)	\
232 	((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
233 #define GET_NEXT_PRB_BLK_NUM(x) \
234 	(((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
235 	((x)->kactive_blk_num+1) : 0)
236 
237 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
238 static void __fanout_link(struct sock *sk, struct packet_sock *po);
239 
240 static int packet_direct_xmit(struct sk_buff *skb)
241 {
242 	struct net_device *dev = skb->dev;
243 	netdev_features_t features;
244 	struct netdev_queue *txq;
245 	int ret = NETDEV_TX_BUSY;
246 
247 	if (unlikely(!netif_running(dev) ||
248 		     !netif_carrier_ok(dev)))
249 		goto drop;
250 
251 	features = netif_skb_features(skb);
252 	if (skb_needs_linearize(skb, features) &&
253 	    __skb_linearize(skb))
254 		goto drop;
255 
256 	txq = skb_get_tx_queue(dev, skb);
257 
258 	local_bh_disable();
259 
260 	HARD_TX_LOCK(dev, txq, smp_processor_id());
261 	if (!netif_xmit_frozen_or_drv_stopped(txq))
262 		ret = netdev_start_xmit(skb, dev, txq, false);
263 	HARD_TX_UNLOCK(dev, txq);
264 
265 	local_bh_enable();
266 
267 	if (!dev_xmit_complete(ret))
268 		kfree_skb(skb);
269 
270 	return ret;
271 drop:
272 	atomic_long_inc(&dev->tx_dropped);
273 	kfree_skb(skb);
274 	return NET_XMIT_DROP;
275 }
276 
277 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
278 {
279 	struct net_device *dev;
280 
281 	rcu_read_lock();
282 	dev = rcu_dereference(po->cached_dev);
283 	if (likely(dev))
284 		dev_hold(dev);
285 	rcu_read_unlock();
286 
287 	return dev;
288 }
289 
290 static void packet_cached_dev_assign(struct packet_sock *po,
291 				     struct net_device *dev)
292 {
293 	rcu_assign_pointer(po->cached_dev, dev);
294 }
295 
296 static void packet_cached_dev_reset(struct packet_sock *po)
297 {
298 	RCU_INIT_POINTER(po->cached_dev, NULL);
299 }
300 
301 static bool packet_use_direct_xmit(const struct packet_sock *po)
302 {
303 	return po->xmit == packet_direct_xmit;
304 }
305 
306 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
307 {
308 	return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
309 }
310 
311 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
312 {
313 	const struct net_device_ops *ops = dev->netdev_ops;
314 	u16 queue_index;
315 
316 	if (ops->ndo_select_queue) {
317 		queue_index = ops->ndo_select_queue(dev, skb, NULL,
318 						    __packet_pick_tx_queue);
319 		queue_index = netdev_cap_txqueue(dev, queue_index);
320 	} else {
321 		queue_index = __packet_pick_tx_queue(dev, skb);
322 	}
323 
324 	skb_set_queue_mapping(skb, queue_index);
325 }
326 
327 /* register_prot_hook must be invoked with the po->bind_lock held,
328  * or from a context in which asynchronous accesses to the packet
329  * socket is not possible (packet_create()).
330  */
331 static void register_prot_hook(struct sock *sk)
332 {
333 	struct packet_sock *po = pkt_sk(sk);
334 
335 	if (!po->running) {
336 		if (po->fanout)
337 			__fanout_link(sk, po);
338 		else
339 			dev_add_pack(&po->prot_hook);
340 
341 		sock_hold(sk);
342 		po->running = 1;
343 	}
344 }
345 
346 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
347  * held.   If the sync parameter is true, we will temporarily drop
348  * the po->bind_lock and do a synchronize_net to make sure no
349  * asynchronous packet processing paths still refer to the elements
350  * of po->prot_hook.  If the sync parameter is false, it is the
351  * callers responsibility to take care of this.
352  */
353 static void __unregister_prot_hook(struct sock *sk, bool sync)
354 {
355 	struct packet_sock *po = pkt_sk(sk);
356 
357 	po->running = 0;
358 
359 	if (po->fanout)
360 		__fanout_unlink(sk, po);
361 	else
362 		__dev_remove_pack(&po->prot_hook);
363 
364 	__sock_put(sk);
365 
366 	if (sync) {
367 		spin_unlock(&po->bind_lock);
368 		synchronize_net();
369 		spin_lock(&po->bind_lock);
370 	}
371 }
372 
373 static void unregister_prot_hook(struct sock *sk, bool sync)
374 {
375 	struct packet_sock *po = pkt_sk(sk);
376 
377 	if (po->running)
378 		__unregister_prot_hook(sk, sync);
379 }
380 
381 static inline struct page * __pure pgv_to_page(void *addr)
382 {
383 	if (is_vmalloc_addr(addr))
384 		return vmalloc_to_page(addr);
385 	return virt_to_page(addr);
386 }
387 
388 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
389 {
390 	union tpacket_uhdr h;
391 
392 	h.raw = frame;
393 	switch (po->tp_version) {
394 	case TPACKET_V1:
395 		h.h1->tp_status = status;
396 		flush_dcache_page(pgv_to_page(&h.h1->tp_status));
397 		break;
398 	case TPACKET_V2:
399 		h.h2->tp_status = status;
400 		flush_dcache_page(pgv_to_page(&h.h2->tp_status));
401 		break;
402 	case TPACKET_V3:
403 	default:
404 		WARN(1, "TPACKET version not supported.\n");
405 		BUG();
406 	}
407 
408 	smp_wmb();
409 }
410 
411 static int __packet_get_status(struct packet_sock *po, void *frame)
412 {
413 	union tpacket_uhdr h;
414 
415 	smp_rmb();
416 
417 	h.raw = frame;
418 	switch (po->tp_version) {
419 	case TPACKET_V1:
420 		flush_dcache_page(pgv_to_page(&h.h1->tp_status));
421 		return h.h1->tp_status;
422 	case TPACKET_V2:
423 		flush_dcache_page(pgv_to_page(&h.h2->tp_status));
424 		return h.h2->tp_status;
425 	case TPACKET_V3:
426 	default:
427 		WARN(1, "TPACKET version not supported.\n");
428 		BUG();
429 		return 0;
430 	}
431 }
432 
433 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
434 				   unsigned int flags)
435 {
436 	struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
437 
438 	if (shhwtstamps &&
439 	    (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
440 	    ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
441 		return TP_STATUS_TS_RAW_HARDWARE;
442 
443 	if (ktime_to_timespec_cond(skb->tstamp, ts))
444 		return TP_STATUS_TS_SOFTWARE;
445 
446 	return 0;
447 }
448 
449 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
450 				    struct sk_buff *skb)
451 {
452 	union tpacket_uhdr h;
453 	struct timespec ts;
454 	__u32 ts_status;
455 
456 	if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
457 		return 0;
458 
459 	h.raw = frame;
460 	switch (po->tp_version) {
461 	case TPACKET_V1:
462 		h.h1->tp_sec = ts.tv_sec;
463 		h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
464 		break;
465 	case TPACKET_V2:
466 		h.h2->tp_sec = ts.tv_sec;
467 		h.h2->tp_nsec = ts.tv_nsec;
468 		break;
469 	case TPACKET_V3:
470 	default:
471 		WARN(1, "TPACKET version not supported.\n");
472 		BUG();
473 	}
474 
475 	/* one flush is safe, as both fields always lie on the same cacheline */
476 	flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
477 	smp_wmb();
478 
479 	return ts_status;
480 }
481 
482 static void *packet_lookup_frame(struct packet_sock *po,
483 		struct packet_ring_buffer *rb,
484 		unsigned int position,
485 		int status)
486 {
487 	unsigned int pg_vec_pos, frame_offset;
488 	union tpacket_uhdr h;
489 
490 	pg_vec_pos = position / rb->frames_per_block;
491 	frame_offset = position % rb->frames_per_block;
492 
493 	h.raw = rb->pg_vec[pg_vec_pos].buffer +
494 		(frame_offset * rb->frame_size);
495 
496 	if (status != __packet_get_status(po, h.raw))
497 		return NULL;
498 
499 	return h.raw;
500 }
501 
502 static void *packet_current_frame(struct packet_sock *po,
503 		struct packet_ring_buffer *rb,
504 		int status)
505 {
506 	return packet_lookup_frame(po, rb, rb->head, status);
507 }
508 
509 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
510 {
511 	del_timer_sync(&pkc->retire_blk_timer);
512 }
513 
514 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
515 		int tx_ring,
516 		struct sk_buff_head *rb_queue)
517 {
518 	struct tpacket_kbdq_core *pkc;
519 
520 	pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
521 			GET_PBDQC_FROM_RB(&po->rx_ring);
522 
523 	spin_lock_bh(&rb_queue->lock);
524 	pkc->delete_blk_timer = 1;
525 	spin_unlock_bh(&rb_queue->lock);
526 
527 	prb_del_retire_blk_timer(pkc);
528 }
529 
530 static void prb_init_blk_timer(struct packet_sock *po,
531 		struct tpacket_kbdq_core *pkc,
532 		void (*func) (unsigned long))
533 {
534 	init_timer(&pkc->retire_blk_timer);
535 	pkc->retire_blk_timer.data = (long)po;
536 	pkc->retire_blk_timer.function = func;
537 	pkc->retire_blk_timer.expires = jiffies;
538 }
539 
540 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
541 {
542 	struct tpacket_kbdq_core *pkc;
543 
544 	if (tx_ring)
545 		BUG();
546 
547 	pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
548 			GET_PBDQC_FROM_RB(&po->rx_ring);
549 	prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
550 }
551 
552 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
553 				int blk_size_in_bytes)
554 {
555 	struct net_device *dev;
556 	unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
557 	struct ethtool_cmd ecmd;
558 	int err;
559 	u32 speed;
560 
561 	rtnl_lock();
562 	dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
563 	if (unlikely(!dev)) {
564 		rtnl_unlock();
565 		return DEFAULT_PRB_RETIRE_TOV;
566 	}
567 	err = __ethtool_get_settings(dev, &ecmd);
568 	speed = ethtool_cmd_speed(&ecmd);
569 	rtnl_unlock();
570 	if (!err) {
571 		/*
572 		 * If the link speed is so slow you don't really
573 		 * need to worry about perf anyways
574 		 */
575 		if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
576 			return DEFAULT_PRB_RETIRE_TOV;
577 		} else {
578 			msec = 1;
579 			div = speed / 1000;
580 		}
581 	}
582 
583 	mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
584 
585 	if (div)
586 		mbits /= div;
587 
588 	tmo = mbits * msec;
589 
590 	if (div)
591 		return tmo+1;
592 	return tmo;
593 }
594 
595 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
596 			union tpacket_req_u *req_u)
597 {
598 	p1->feature_req_word = req_u->req3.tp_feature_req_word;
599 }
600 
601 static void init_prb_bdqc(struct packet_sock *po,
602 			struct packet_ring_buffer *rb,
603 			struct pgv *pg_vec,
604 			union tpacket_req_u *req_u, int tx_ring)
605 {
606 	struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
607 	struct tpacket_block_desc *pbd;
608 
609 	memset(p1, 0x0, sizeof(*p1));
610 
611 	p1->knxt_seq_num = 1;
612 	p1->pkbdq = pg_vec;
613 	pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
614 	p1->pkblk_start	= pg_vec[0].buffer;
615 	p1->kblk_size = req_u->req3.tp_block_size;
616 	p1->knum_blocks	= req_u->req3.tp_block_nr;
617 	p1->hdrlen = po->tp_hdrlen;
618 	p1->version = po->tp_version;
619 	p1->last_kactive_blk_num = 0;
620 	po->stats.stats3.tp_freeze_q_cnt = 0;
621 	if (req_u->req3.tp_retire_blk_tov)
622 		p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
623 	else
624 		p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
625 						req_u->req3.tp_block_size);
626 	p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
627 	p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
628 
629 	p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
630 	prb_init_ft_ops(p1, req_u);
631 	prb_setup_retire_blk_timer(po, tx_ring);
632 	prb_open_block(p1, pbd);
633 }
634 
635 /*  Do NOT update the last_blk_num first.
636  *  Assumes sk_buff_head lock is held.
637  */
638 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
639 {
640 	mod_timer(&pkc->retire_blk_timer,
641 			jiffies + pkc->tov_in_jiffies);
642 	pkc->last_kactive_blk_num = pkc->kactive_blk_num;
643 }
644 
645 /*
646  * Timer logic:
647  * 1) We refresh the timer only when we open a block.
648  *    By doing this we don't waste cycles refreshing the timer
649  *	  on packet-by-packet basis.
650  *
651  * With a 1MB block-size, on a 1Gbps line, it will take
652  * i) ~8 ms to fill a block + ii) memcpy etc.
653  * In this cut we are not accounting for the memcpy time.
654  *
655  * So, if the user sets the 'tmo' to 10ms then the timer
656  * will never fire while the block is still getting filled
657  * (which is what we want). However, the user could choose
658  * to close a block early and that's fine.
659  *
660  * But when the timer does fire, we check whether or not to refresh it.
661  * Since the tmo granularity is in msecs, it is not too expensive
662  * to refresh the timer, lets say every '8' msecs.
663  * Either the user can set the 'tmo' or we can derive it based on
664  * a) line-speed and b) block-size.
665  * prb_calc_retire_blk_tmo() calculates the tmo.
666  *
667  */
668 static void prb_retire_rx_blk_timer_expired(unsigned long data)
669 {
670 	struct packet_sock *po = (struct packet_sock *)data;
671 	struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
672 	unsigned int frozen;
673 	struct tpacket_block_desc *pbd;
674 
675 	spin_lock(&po->sk.sk_receive_queue.lock);
676 
677 	frozen = prb_queue_frozen(pkc);
678 	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
679 
680 	if (unlikely(pkc->delete_blk_timer))
681 		goto out;
682 
683 	/* We only need to plug the race when the block is partially filled.
684 	 * tpacket_rcv:
685 	 *		lock(); increment BLOCK_NUM_PKTS; unlock()
686 	 *		copy_bits() is in progress ...
687 	 *		timer fires on other cpu:
688 	 *		we can't retire the current block because copy_bits
689 	 *		is in progress.
690 	 *
691 	 */
692 	if (BLOCK_NUM_PKTS(pbd)) {
693 		while (atomic_read(&pkc->blk_fill_in_prog)) {
694 			/* Waiting for skb_copy_bits to finish... */
695 			cpu_relax();
696 		}
697 	}
698 
699 	if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
700 		if (!frozen) {
701 			prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
702 			if (!prb_dispatch_next_block(pkc, po))
703 				goto refresh_timer;
704 			else
705 				goto out;
706 		} else {
707 			/* Case 1. Queue was frozen because user-space was
708 			 *	   lagging behind.
709 			 */
710 			if (prb_curr_blk_in_use(pkc, pbd)) {
711 				/*
712 				 * Ok, user-space is still behind.
713 				 * So just refresh the timer.
714 				 */
715 				goto refresh_timer;
716 			} else {
717 			       /* Case 2. queue was frozen,user-space caught up,
718 				* now the link went idle && the timer fired.
719 				* We don't have a block to close.So we open this
720 				* block and restart the timer.
721 				* opening a block thaws the queue,restarts timer
722 				* Thawing/timer-refresh is a side effect.
723 				*/
724 				prb_open_block(pkc, pbd);
725 				goto out;
726 			}
727 		}
728 	}
729 
730 refresh_timer:
731 	_prb_refresh_rx_retire_blk_timer(pkc);
732 
733 out:
734 	spin_unlock(&po->sk.sk_receive_queue.lock);
735 }
736 
737 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
738 		struct tpacket_block_desc *pbd1, __u32 status)
739 {
740 	/* Flush everything minus the block header */
741 
742 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
743 	u8 *start, *end;
744 
745 	start = (u8 *)pbd1;
746 
747 	/* Skip the block header(we know header WILL fit in 4K) */
748 	start += PAGE_SIZE;
749 
750 	end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
751 	for (; start < end; start += PAGE_SIZE)
752 		flush_dcache_page(pgv_to_page(start));
753 
754 	smp_wmb();
755 #endif
756 
757 	/* Now update the block status. */
758 
759 	BLOCK_STATUS(pbd1) = status;
760 
761 	/* Flush the block header */
762 
763 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
764 	start = (u8 *)pbd1;
765 	flush_dcache_page(pgv_to_page(start));
766 
767 	smp_wmb();
768 #endif
769 }
770 
771 /*
772  * Side effect:
773  *
774  * 1) flush the block
775  * 2) Increment active_blk_num
776  *
777  * Note:We DONT refresh the timer on purpose.
778  *	Because almost always the next block will be opened.
779  */
780 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
781 		struct tpacket_block_desc *pbd1,
782 		struct packet_sock *po, unsigned int stat)
783 {
784 	__u32 status = TP_STATUS_USER | stat;
785 
786 	struct tpacket3_hdr *last_pkt;
787 	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
788 
789 	if (po->stats.stats3.tp_drops)
790 		status |= TP_STATUS_LOSING;
791 
792 	last_pkt = (struct tpacket3_hdr *)pkc1->prev;
793 	last_pkt->tp_next_offset = 0;
794 
795 	/* Get the ts of the last pkt */
796 	if (BLOCK_NUM_PKTS(pbd1)) {
797 		h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
798 		h1->ts_last_pkt.ts_nsec	= last_pkt->tp_nsec;
799 	} else {
800 		/* Ok, we tmo'd - so get the current time */
801 		struct timespec ts;
802 		getnstimeofday(&ts);
803 		h1->ts_last_pkt.ts_sec = ts.tv_sec;
804 		h1->ts_last_pkt.ts_nsec	= ts.tv_nsec;
805 	}
806 
807 	smp_wmb();
808 
809 	/* Flush the block */
810 	prb_flush_block(pkc1, pbd1, status);
811 
812 	pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
813 }
814 
815 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
816 {
817 	pkc->reset_pending_on_curr_blk = 0;
818 }
819 
820 /*
821  * Side effect of opening a block:
822  *
823  * 1) prb_queue is thawed.
824  * 2) retire_blk_timer is refreshed.
825  *
826  */
827 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
828 	struct tpacket_block_desc *pbd1)
829 {
830 	struct timespec ts;
831 	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
832 
833 	smp_rmb();
834 
835 	/* We could have just memset this but we will lose the
836 	 * flexibility of making the priv area sticky
837 	 */
838 
839 	BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
840 	BLOCK_NUM_PKTS(pbd1) = 0;
841 	BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
842 
843 	getnstimeofday(&ts);
844 
845 	h1->ts_first_pkt.ts_sec = ts.tv_sec;
846 	h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
847 
848 	pkc1->pkblk_start = (char *)pbd1;
849 	pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
850 
851 	BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
852 	BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
853 
854 	pbd1->version = pkc1->version;
855 	pkc1->prev = pkc1->nxt_offset;
856 	pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
857 
858 	prb_thaw_queue(pkc1);
859 	_prb_refresh_rx_retire_blk_timer(pkc1);
860 
861 	smp_wmb();
862 }
863 
864 /*
865  * Queue freeze logic:
866  * 1) Assume tp_block_nr = 8 blocks.
867  * 2) At time 't0', user opens Rx ring.
868  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
869  * 4) user-space is either sleeping or processing block '0'.
870  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
871  *    it will close block-7,loop around and try to fill block '0'.
872  *    call-flow:
873  *    __packet_lookup_frame_in_block
874  *      prb_retire_current_block()
875  *      prb_dispatch_next_block()
876  *        |->(BLOCK_STATUS == USER) evaluates to true
877  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
878  * 6) Now there are two cases:
879  *    6.1) Link goes idle right after the queue is frozen.
880  *         But remember, the last open_block() refreshed the timer.
881  *         When this timer expires,it will refresh itself so that we can
882  *         re-open block-0 in near future.
883  *    6.2) Link is busy and keeps on receiving packets. This is a simple
884  *         case and __packet_lookup_frame_in_block will check if block-0
885  *         is free and can now be re-used.
886  */
887 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
888 				  struct packet_sock *po)
889 {
890 	pkc->reset_pending_on_curr_blk = 1;
891 	po->stats.stats3.tp_freeze_q_cnt++;
892 }
893 
894 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
895 
896 /*
897  * If the next block is free then we will dispatch it
898  * and return a good offset.
899  * Else, we will freeze the queue.
900  * So, caller must check the return value.
901  */
902 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
903 		struct packet_sock *po)
904 {
905 	struct tpacket_block_desc *pbd;
906 
907 	smp_rmb();
908 
909 	/* 1. Get current block num */
910 	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
911 
912 	/* 2. If this block is currently in_use then freeze the queue */
913 	if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
914 		prb_freeze_queue(pkc, po);
915 		return NULL;
916 	}
917 
918 	/*
919 	 * 3.
920 	 * open this block and return the offset where the first packet
921 	 * needs to get stored.
922 	 */
923 	prb_open_block(pkc, pbd);
924 	return (void *)pkc->nxt_offset;
925 }
926 
927 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
928 		struct packet_sock *po, unsigned int status)
929 {
930 	struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
931 
932 	/* retire/close the current block */
933 	if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
934 		/*
935 		 * Plug the case where copy_bits() is in progress on
936 		 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
937 		 * have space to copy the pkt in the current block and
938 		 * called prb_retire_current_block()
939 		 *
940 		 * We don't need to worry about the TMO case because
941 		 * the timer-handler already handled this case.
942 		 */
943 		if (!(status & TP_STATUS_BLK_TMO)) {
944 			while (atomic_read(&pkc->blk_fill_in_prog)) {
945 				/* Waiting for skb_copy_bits to finish... */
946 				cpu_relax();
947 			}
948 		}
949 		prb_close_block(pkc, pbd, po, status);
950 		return;
951 	}
952 }
953 
954 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
955 				      struct tpacket_block_desc *pbd)
956 {
957 	return TP_STATUS_USER & BLOCK_STATUS(pbd);
958 }
959 
960 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
961 {
962 	return pkc->reset_pending_on_curr_blk;
963 }
964 
965 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
966 {
967 	struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
968 	atomic_dec(&pkc->blk_fill_in_prog);
969 }
970 
971 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
972 			struct tpacket3_hdr *ppd)
973 {
974 	ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
975 }
976 
977 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
978 			struct tpacket3_hdr *ppd)
979 {
980 	ppd->hv1.tp_rxhash = 0;
981 }
982 
983 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
984 			struct tpacket3_hdr *ppd)
985 {
986 	if (vlan_tx_tag_present(pkc->skb)) {
987 		ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
988 		ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
989 		ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
990 	} else {
991 		ppd->hv1.tp_vlan_tci = 0;
992 		ppd->hv1.tp_vlan_tpid = 0;
993 		ppd->tp_status = TP_STATUS_AVAILABLE;
994 	}
995 }
996 
997 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
998 			struct tpacket3_hdr *ppd)
999 {
1000 	ppd->hv1.tp_padding = 0;
1001 	prb_fill_vlan_info(pkc, ppd);
1002 
1003 	if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1004 		prb_fill_rxhash(pkc, ppd);
1005 	else
1006 		prb_clear_rxhash(pkc, ppd);
1007 }
1008 
1009 static void prb_fill_curr_block(char *curr,
1010 				struct tpacket_kbdq_core *pkc,
1011 				struct tpacket_block_desc *pbd,
1012 				unsigned int len)
1013 {
1014 	struct tpacket3_hdr *ppd;
1015 
1016 	ppd  = (struct tpacket3_hdr *)curr;
1017 	ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1018 	pkc->prev = curr;
1019 	pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1020 	BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1021 	BLOCK_NUM_PKTS(pbd) += 1;
1022 	atomic_inc(&pkc->blk_fill_in_prog);
1023 	prb_run_all_ft_ops(pkc, ppd);
1024 }
1025 
1026 /* Assumes caller has the sk->rx_queue.lock */
1027 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1028 					    struct sk_buff *skb,
1029 						int status,
1030 					    unsigned int len
1031 					    )
1032 {
1033 	struct tpacket_kbdq_core *pkc;
1034 	struct tpacket_block_desc *pbd;
1035 	char *curr, *end;
1036 
1037 	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1038 	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1039 
1040 	/* Queue is frozen when user space is lagging behind */
1041 	if (prb_queue_frozen(pkc)) {
1042 		/*
1043 		 * Check if that last block which caused the queue to freeze,
1044 		 * is still in_use by user-space.
1045 		 */
1046 		if (prb_curr_blk_in_use(pkc, pbd)) {
1047 			/* Can't record this packet */
1048 			return NULL;
1049 		} else {
1050 			/*
1051 			 * Ok, the block was released by user-space.
1052 			 * Now let's open that block.
1053 			 * opening a block also thaws the queue.
1054 			 * Thawing is a side effect.
1055 			 */
1056 			prb_open_block(pkc, pbd);
1057 		}
1058 	}
1059 
1060 	smp_mb();
1061 	curr = pkc->nxt_offset;
1062 	pkc->skb = skb;
1063 	end = (char *)pbd + pkc->kblk_size;
1064 
1065 	/* first try the current block */
1066 	if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1067 		prb_fill_curr_block(curr, pkc, pbd, len);
1068 		return (void *)curr;
1069 	}
1070 
1071 	/* Ok, close the current block */
1072 	prb_retire_current_block(pkc, po, 0);
1073 
1074 	/* Now, try to dispatch the next block */
1075 	curr = (char *)prb_dispatch_next_block(pkc, po);
1076 	if (curr) {
1077 		pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1078 		prb_fill_curr_block(curr, pkc, pbd, len);
1079 		return (void *)curr;
1080 	}
1081 
1082 	/*
1083 	 * No free blocks are available.user_space hasn't caught up yet.
1084 	 * Queue was just frozen and now this packet will get dropped.
1085 	 */
1086 	return NULL;
1087 }
1088 
1089 static void *packet_current_rx_frame(struct packet_sock *po,
1090 					    struct sk_buff *skb,
1091 					    int status, unsigned int len)
1092 {
1093 	char *curr = NULL;
1094 	switch (po->tp_version) {
1095 	case TPACKET_V1:
1096 	case TPACKET_V2:
1097 		curr = packet_lookup_frame(po, &po->rx_ring,
1098 					po->rx_ring.head, status);
1099 		return curr;
1100 	case TPACKET_V3:
1101 		return __packet_lookup_frame_in_block(po, skb, status, len);
1102 	default:
1103 		WARN(1, "TPACKET version not supported\n");
1104 		BUG();
1105 		return NULL;
1106 	}
1107 }
1108 
1109 static void *prb_lookup_block(struct packet_sock *po,
1110 				     struct packet_ring_buffer *rb,
1111 				     unsigned int idx,
1112 				     int status)
1113 {
1114 	struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1115 	struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1116 
1117 	if (status != BLOCK_STATUS(pbd))
1118 		return NULL;
1119 	return pbd;
1120 }
1121 
1122 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1123 {
1124 	unsigned int prev;
1125 	if (rb->prb_bdqc.kactive_blk_num)
1126 		prev = rb->prb_bdqc.kactive_blk_num-1;
1127 	else
1128 		prev = rb->prb_bdqc.knum_blocks-1;
1129 	return prev;
1130 }
1131 
1132 /* Assumes caller has held the rx_queue.lock */
1133 static void *__prb_previous_block(struct packet_sock *po,
1134 					 struct packet_ring_buffer *rb,
1135 					 int status)
1136 {
1137 	unsigned int previous = prb_previous_blk_num(rb);
1138 	return prb_lookup_block(po, rb, previous, status);
1139 }
1140 
1141 static void *packet_previous_rx_frame(struct packet_sock *po,
1142 					     struct packet_ring_buffer *rb,
1143 					     int status)
1144 {
1145 	if (po->tp_version <= TPACKET_V2)
1146 		return packet_previous_frame(po, rb, status);
1147 
1148 	return __prb_previous_block(po, rb, status);
1149 }
1150 
1151 static void packet_increment_rx_head(struct packet_sock *po,
1152 					    struct packet_ring_buffer *rb)
1153 {
1154 	switch (po->tp_version) {
1155 	case TPACKET_V1:
1156 	case TPACKET_V2:
1157 		return packet_increment_head(rb);
1158 	case TPACKET_V3:
1159 	default:
1160 		WARN(1, "TPACKET version not supported.\n");
1161 		BUG();
1162 		return;
1163 	}
1164 }
1165 
1166 static void *packet_previous_frame(struct packet_sock *po,
1167 		struct packet_ring_buffer *rb,
1168 		int status)
1169 {
1170 	unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1171 	return packet_lookup_frame(po, rb, previous, status);
1172 }
1173 
1174 static void packet_increment_head(struct packet_ring_buffer *buff)
1175 {
1176 	buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1177 }
1178 
1179 static void packet_inc_pending(struct packet_ring_buffer *rb)
1180 {
1181 	this_cpu_inc(*rb->pending_refcnt);
1182 }
1183 
1184 static void packet_dec_pending(struct packet_ring_buffer *rb)
1185 {
1186 	this_cpu_dec(*rb->pending_refcnt);
1187 }
1188 
1189 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1190 {
1191 	unsigned int refcnt = 0;
1192 	int cpu;
1193 
1194 	/* We don't use pending refcount in rx_ring. */
1195 	if (rb->pending_refcnt == NULL)
1196 		return 0;
1197 
1198 	for_each_possible_cpu(cpu)
1199 		refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1200 
1201 	return refcnt;
1202 }
1203 
1204 static int packet_alloc_pending(struct packet_sock *po)
1205 {
1206 	po->rx_ring.pending_refcnt = NULL;
1207 
1208 	po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1209 	if (unlikely(po->tx_ring.pending_refcnt == NULL))
1210 		return -ENOBUFS;
1211 
1212 	return 0;
1213 }
1214 
1215 static void packet_free_pending(struct packet_sock *po)
1216 {
1217 	free_percpu(po->tx_ring.pending_refcnt);
1218 }
1219 
1220 static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1221 {
1222 	struct sock *sk = &po->sk;
1223 	bool has_room;
1224 
1225 	if (po->prot_hook.func != tpacket_rcv)
1226 		return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1227 			<= sk->sk_rcvbuf;
1228 
1229 	spin_lock(&sk->sk_receive_queue.lock);
1230 	if (po->tp_version == TPACKET_V3)
1231 		has_room = prb_lookup_block(po, &po->rx_ring,
1232 					    po->rx_ring.prb_bdqc.kactive_blk_num,
1233 					    TP_STATUS_KERNEL);
1234 	else
1235 		has_room = packet_lookup_frame(po, &po->rx_ring,
1236 					       po->rx_ring.head,
1237 					       TP_STATUS_KERNEL);
1238 	spin_unlock(&sk->sk_receive_queue.lock);
1239 
1240 	return has_room;
1241 }
1242 
1243 static void packet_sock_destruct(struct sock *sk)
1244 {
1245 	skb_queue_purge(&sk->sk_error_queue);
1246 
1247 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1248 	WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1249 
1250 	if (!sock_flag(sk, SOCK_DEAD)) {
1251 		pr_err("Attempt to release alive packet socket: %p\n", sk);
1252 		return;
1253 	}
1254 
1255 	sk_refcnt_debug_dec(sk);
1256 }
1257 
1258 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1259 {
1260 	int x = atomic_read(&f->rr_cur) + 1;
1261 
1262 	if (x >= num)
1263 		x = 0;
1264 
1265 	return x;
1266 }
1267 
1268 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1269 				      struct sk_buff *skb,
1270 				      unsigned int num)
1271 {
1272 	return reciprocal_scale(skb_get_hash(skb), num);
1273 }
1274 
1275 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1276 				    struct sk_buff *skb,
1277 				    unsigned int num)
1278 {
1279 	int cur, old;
1280 
1281 	cur = atomic_read(&f->rr_cur);
1282 	while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1283 				     fanout_rr_next(f, num))) != cur)
1284 		cur = old;
1285 	return cur;
1286 }
1287 
1288 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1289 				     struct sk_buff *skb,
1290 				     unsigned int num)
1291 {
1292 	return smp_processor_id() % num;
1293 }
1294 
1295 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1296 				     struct sk_buff *skb,
1297 				     unsigned int num)
1298 {
1299 	return prandom_u32_max(num);
1300 }
1301 
1302 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1303 					  struct sk_buff *skb,
1304 					  unsigned int idx, unsigned int skip,
1305 					  unsigned int num)
1306 {
1307 	unsigned int i, j;
1308 
1309 	i = j = min_t(int, f->next[idx], num - 1);
1310 	do {
1311 		if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1312 			if (i != j)
1313 				f->next[idx] = i;
1314 			return i;
1315 		}
1316 		if (++i == num)
1317 			i = 0;
1318 	} while (i != j);
1319 
1320 	return idx;
1321 }
1322 
1323 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1324 				    struct sk_buff *skb,
1325 				    unsigned int num)
1326 {
1327 	return skb_get_queue_mapping(skb) % num;
1328 }
1329 
1330 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1331 {
1332 	return f->flags & (flag >> 8);
1333 }
1334 
1335 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1336 			     struct packet_type *pt, struct net_device *orig_dev)
1337 {
1338 	struct packet_fanout *f = pt->af_packet_priv;
1339 	unsigned int num = f->num_members;
1340 	struct packet_sock *po;
1341 	unsigned int idx;
1342 
1343 	if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1344 	    !num) {
1345 		kfree_skb(skb);
1346 		return 0;
1347 	}
1348 
1349 	switch (f->type) {
1350 	case PACKET_FANOUT_HASH:
1351 	default:
1352 		if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1353 			skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1354 			if (!skb)
1355 				return 0;
1356 		}
1357 		idx = fanout_demux_hash(f, skb, num);
1358 		break;
1359 	case PACKET_FANOUT_LB:
1360 		idx = fanout_demux_lb(f, skb, num);
1361 		break;
1362 	case PACKET_FANOUT_CPU:
1363 		idx = fanout_demux_cpu(f, skb, num);
1364 		break;
1365 	case PACKET_FANOUT_RND:
1366 		idx = fanout_demux_rnd(f, skb, num);
1367 		break;
1368 	case PACKET_FANOUT_QM:
1369 		idx = fanout_demux_qm(f, skb, num);
1370 		break;
1371 	case PACKET_FANOUT_ROLLOVER:
1372 		idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1373 		break;
1374 	}
1375 
1376 	po = pkt_sk(f->arr[idx]);
1377 	if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1378 	    unlikely(!packet_rcv_has_room(po, skb))) {
1379 		idx = fanout_demux_rollover(f, skb, idx, idx, num);
1380 		po = pkt_sk(f->arr[idx]);
1381 	}
1382 
1383 	return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1384 }
1385 
1386 DEFINE_MUTEX(fanout_mutex);
1387 EXPORT_SYMBOL_GPL(fanout_mutex);
1388 static LIST_HEAD(fanout_list);
1389 
1390 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1391 {
1392 	struct packet_fanout *f = po->fanout;
1393 
1394 	spin_lock(&f->lock);
1395 	f->arr[f->num_members] = sk;
1396 	smp_wmb();
1397 	f->num_members++;
1398 	spin_unlock(&f->lock);
1399 }
1400 
1401 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1402 {
1403 	struct packet_fanout *f = po->fanout;
1404 	int i;
1405 
1406 	spin_lock(&f->lock);
1407 	for (i = 0; i < f->num_members; i++) {
1408 		if (f->arr[i] == sk)
1409 			break;
1410 	}
1411 	BUG_ON(i >= f->num_members);
1412 	f->arr[i] = f->arr[f->num_members - 1];
1413 	f->num_members--;
1414 	spin_unlock(&f->lock);
1415 }
1416 
1417 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1418 {
1419 	if (ptype->af_packet_priv == (void *)((struct packet_sock *)sk)->fanout)
1420 		return true;
1421 
1422 	return false;
1423 }
1424 
1425 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1426 {
1427 	struct packet_sock *po = pkt_sk(sk);
1428 	struct packet_fanout *f, *match;
1429 	u8 type = type_flags & 0xff;
1430 	u8 flags = type_flags >> 8;
1431 	int err;
1432 
1433 	switch (type) {
1434 	case PACKET_FANOUT_ROLLOVER:
1435 		if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1436 			return -EINVAL;
1437 	case PACKET_FANOUT_HASH:
1438 	case PACKET_FANOUT_LB:
1439 	case PACKET_FANOUT_CPU:
1440 	case PACKET_FANOUT_RND:
1441 	case PACKET_FANOUT_QM:
1442 		break;
1443 	default:
1444 		return -EINVAL;
1445 	}
1446 
1447 	if (!po->running)
1448 		return -EINVAL;
1449 
1450 	if (po->fanout)
1451 		return -EALREADY;
1452 
1453 	mutex_lock(&fanout_mutex);
1454 	match = NULL;
1455 	list_for_each_entry(f, &fanout_list, list) {
1456 		if (f->id == id &&
1457 		    read_pnet(&f->net) == sock_net(sk)) {
1458 			match = f;
1459 			break;
1460 		}
1461 	}
1462 	err = -EINVAL;
1463 	if (match && match->flags != flags)
1464 		goto out;
1465 	if (!match) {
1466 		err = -ENOMEM;
1467 		match = kzalloc(sizeof(*match), GFP_KERNEL);
1468 		if (!match)
1469 			goto out;
1470 		write_pnet(&match->net, sock_net(sk));
1471 		match->id = id;
1472 		match->type = type;
1473 		match->flags = flags;
1474 		atomic_set(&match->rr_cur, 0);
1475 		INIT_LIST_HEAD(&match->list);
1476 		spin_lock_init(&match->lock);
1477 		atomic_set(&match->sk_ref, 0);
1478 		match->prot_hook.type = po->prot_hook.type;
1479 		match->prot_hook.dev = po->prot_hook.dev;
1480 		match->prot_hook.func = packet_rcv_fanout;
1481 		match->prot_hook.af_packet_priv = match;
1482 		match->prot_hook.id_match = match_fanout_group;
1483 		dev_add_pack(&match->prot_hook);
1484 		list_add(&match->list, &fanout_list);
1485 	}
1486 	err = -EINVAL;
1487 	if (match->type == type &&
1488 	    match->prot_hook.type == po->prot_hook.type &&
1489 	    match->prot_hook.dev == po->prot_hook.dev) {
1490 		err = -ENOSPC;
1491 		if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1492 			__dev_remove_pack(&po->prot_hook);
1493 			po->fanout = match;
1494 			atomic_inc(&match->sk_ref);
1495 			__fanout_link(sk, po);
1496 			err = 0;
1497 		}
1498 	}
1499 out:
1500 	mutex_unlock(&fanout_mutex);
1501 	return err;
1502 }
1503 
1504 static void fanout_release(struct sock *sk)
1505 {
1506 	struct packet_sock *po = pkt_sk(sk);
1507 	struct packet_fanout *f;
1508 
1509 	f = po->fanout;
1510 	if (!f)
1511 		return;
1512 
1513 	mutex_lock(&fanout_mutex);
1514 	po->fanout = NULL;
1515 
1516 	if (atomic_dec_and_test(&f->sk_ref)) {
1517 		list_del(&f->list);
1518 		dev_remove_pack(&f->prot_hook);
1519 		kfree(f);
1520 	}
1521 	mutex_unlock(&fanout_mutex);
1522 }
1523 
1524 static const struct proto_ops packet_ops;
1525 
1526 static const struct proto_ops packet_ops_spkt;
1527 
1528 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1529 			   struct packet_type *pt, struct net_device *orig_dev)
1530 {
1531 	struct sock *sk;
1532 	struct sockaddr_pkt *spkt;
1533 
1534 	/*
1535 	 *	When we registered the protocol we saved the socket in the data
1536 	 *	field for just this event.
1537 	 */
1538 
1539 	sk = pt->af_packet_priv;
1540 
1541 	/*
1542 	 *	Yank back the headers [hope the device set this
1543 	 *	right or kerboom...]
1544 	 *
1545 	 *	Incoming packets have ll header pulled,
1546 	 *	push it back.
1547 	 *
1548 	 *	For outgoing ones skb->data == skb_mac_header(skb)
1549 	 *	so that this procedure is noop.
1550 	 */
1551 
1552 	if (skb->pkt_type == PACKET_LOOPBACK)
1553 		goto out;
1554 
1555 	if (!net_eq(dev_net(dev), sock_net(sk)))
1556 		goto out;
1557 
1558 	skb = skb_share_check(skb, GFP_ATOMIC);
1559 	if (skb == NULL)
1560 		goto oom;
1561 
1562 	/* drop any routing info */
1563 	skb_dst_drop(skb);
1564 
1565 	/* drop conntrack reference */
1566 	nf_reset(skb);
1567 
1568 	spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1569 
1570 	skb_push(skb, skb->data - skb_mac_header(skb));
1571 
1572 	/*
1573 	 *	The SOCK_PACKET socket receives _all_ frames.
1574 	 */
1575 
1576 	spkt->spkt_family = dev->type;
1577 	strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1578 	spkt->spkt_protocol = skb->protocol;
1579 
1580 	/*
1581 	 *	Charge the memory to the socket. This is done specifically
1582 	 *	to prevent sockets using all the memory up.
1583 	 */
1584 
1585 	if (sock_queue_rcv_skb(sk, skb) == 0)
1586 		return 0;
1587 
1588 out:
1589 	kfree_skb(skb);
1590 oom:
1591 	return 0;
1592 }
1593 
1594 
1595 /*
1596  *	Output a raw packet to a device layer. This bypasses all the other
1597  *	protocol layers and you must therefore supply it with a complete frame
1598  */
1599 
1600 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1601 			       struct msghdr *msg, size_t len)
1602 {
1603 	struct sock *sk = sock->sk;
1604 	DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1605 	struct sk_buff *skb = NULL;
1606 	struct net_device *dev;
1607 	__be16 proto = 0;
1608 	int err;
1609 	int extra_len = 0;
1610 
1611 	/*
1612 	 *	Get and verify the address.
1613 	 */
1614 
1615 	if (saddr) {
1616 		if (msg->msg_namelen < sizeof(struct sockaddr))
1617 			return -EINVAL;
1618 		if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1619 			proto = saddr->spkt_protocol;
1620 	} else
1621 		return -ENOTCONN;	/* SOCK_PACKET must be sent giving an address */
1622 
1623 	/*
1624 	 *	Find the device first to size check it
1625 	 */
1626 
1627 	saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1628 retry:
1629 	rcu_read_lock();
1630 	dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1631 	err = -ENODEV;
1632 	if (dev == NULL)
1633 		goto out_unlock;
1634 
1635 	err = -ENETDOWN;
1636 	if (!(dev->flags & IFF_UP))
1637 		goto out_unlock;
1638 
1639 	/*
1640 	 * You may not queue a frame bigger than the mtu. This is the lowest level
1641 	 * raw protocol and you must do your own fragmentation at this level.
1642 	 */
1643 
1644 	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1645 		if (!netif_supports_nofcs(dev)) {
1646 			err = -EPROTONOSUPPORT;
1647 			goto out_unlock;
1648 		}
1649 		extra_len = 4; /* We're doing our own CRC */
1650 	}
1651 
1652 	err = -EMSGSIZE;
1653 	if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1654 		goto out_unlock;
1655 
1656 	if (!skb) {
1657 		size_t reserved = LL_RESERVED_SPACE(dev);
1658 		int tlen = dev->needed_tailroom;
1659 		unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1660 
1661 		rcu_read_unlock();
1662 		skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1663 		if (skb == NULL)
1664 			return -ENOBUFS;
1665 		/* FIXME: Save some space for broken drivers that write a hard
1666 		 * header at transmission time by themselves. PPP is the notable
1667 		 * one here. This should really be fixed at the driver level.
1668 		 */
1669 		skb_reserve(skb, reserved);
1670 		skb_reset_network_header(skb);
1671 
1672 		/* Try to align data part correctly */
1673 		if (hhlen) {
1674 			skb->data -= hhlen;
1675 			skb->tail -= hhlen;
1676 			if (len < hhlen)
1677 				skb_reset_network_header(skb);
1678 		}
1679 		err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1680 		if (err)
1681 			goto out_free;
1682 		goto retry;
1683 	}
1684 
1685 	if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1686 		/* Earlier code assumed this would be a VLAN pkt,
1687 		 * double-check this now that we have the actual
1688 		 * packet in hand.
1689 		 */
1690 		struct ethhdr *ehdr;
1691 		skb_reset_mac_header(skb);
1692 		ehdr = eth_hdr(skb);
1693 		if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1694 			err = -EMSGSIZE;
1695 			goto out_unlock;
1696 		}
1697 	}
1698 
1699 	skb->protocol = proto;
1700 	skb->dev = dev;
1701 	skb->priority = sk->sk_priority;
1702 	skb->mark = sk->sk_mark;
1703 
1704 	sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1705 
1706 	if (unlikely(extra_len == 4))
1707 		skb->no_fcs = 1;
1708 
1709 	skb_probe_transport_header(skb, 0);
1710 
1711 	dev_queue_xmit(skb);
1712 	rcu_read_unlock();
1713 	return len;
1714 
1715 out_unlock:
1716 	rcu_read_unlock();
1717 out_free:
1718 	kfree_skb(skb);
1719 	return err;
1720 }
1721 
1722 static unsigned int run_filter(const struct sk_buff *skb,
1723 				      const struct sock *sk,
1724 				      unsigned int res)
1725 {
1726 	struct sk_filter *filter;
1727 
1728 	rcu_read_lock();
1729 	filter = rcu_dereference(sk->sk_filter);
1730 	if (filter != NULL)
1731 		res = SK_RUN_FILTER(filter, skb);
1732 	rcu_read_unlock();
1733 
1734 	return res;
1735 }
1736 
1737 /*
1738  * This function makes lazy skb cloning in hope that most of packets
1739  * are discarded by BPF.
1740  *
1741  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1742  * and skb->cb are mangled. It works because (and until) packets
1743  * falling here are owned by current CPU. Output packets are cloned
1744  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1745  * sequencially, so that if we return skb to original state on exit,
1746  * we will not harm anyone.
1747  */
1748 
1749 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1750 		      struct packet_type *pt, struct net_device *orig_dev)
1751 {
1752 	struct sock *sk;
1753 	struct sockaddr_ll *sll;
1754 	struct packet_sock *po;
1755 	u8 *skb_head = skb->data;
1756 	int skb_len = skb->len;
1757 	unsigned int snaplen, res;
1758 
1759 	if (skb->pkt_type == PACKET_LOOPBACK)
1760 		goto drop;
1761 
1762 	sk = pt->af_packet_priv;
1763 	po = pkt_sk(sk);
1764 
1765 	if (!net_eq(dev_net(dev), sock_net(sk)))
1766 		goto drop;
1767 
1768 	skb->dev = dev;
1769 
1770 	if (dev->header_ops) {
1771 		/* The device has an explicit notion of ll header,
1772 		 * exported to higher levels.
1773 		 *
1774 		 * Otherwise, the device hides details of its frame
1775 		 * structure, so that corresponding packet head is
1776 		 * never delivered to user.
1777 		 */
1778 		if (sk->sk_type != SOCK_DGRAM)
1779 			skb_push(skb, skb->data - skb_mac_header(skb));
1780 		else if (skb->pkt_type == PACKET_OUTGOING) {
1781 			/* Special case: outgoing packets have ll header at head */
1782 			skb_pull(skb, skb_network_offset(skb));
1783 		}
1784 	}
1785 
1786 	snaplen = skb->len;
1787 
1788 	res = run_filter(skb, sk, snaplen);
1789 	if (!res)
1790 		goto drop_n_restore;
1791 	if (snaplen > res)
1792 		snaplen = res;
1793 
1794 	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1795 		goto drop_n_acct;
1796 
1797 	if (skb_shared(skb)) {
1798 		struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1799 		if (nskb == NULL)
1800 			goto drop_n_acct;
1801 
1802 		if (skb_head != skb->data) {
1803 			skb->data = skb_head;
1804 			skb->len = skb_len;
1805 		}
1806 		consume_skb(skb);
1807 		skb = nskb;
1808 	}
1809 
1810 	BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1811 		     sizeof(skb->cb));
1812 
1813 	sll = &PACKET_SKB_CB(skb)->sa.ll;
1814 	sll->sll_family = AF_PACKET;
1815 	sll->sll_hatype = dev->type;
1816 	sll->sll_protocol = skb->protocol;
1817 	sll->sll_pkttype = skb->pkt_type;
1818 	if (unlikely(po->origdev))
1819 		sll->sll_ifindex = orig_dev->ifindex;
1820 	else
1821 		sll->sll_ifindex = dev->ifindex;
1822 
1823 	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1824 
1825 	PACKET_SKB_CB(skb)->origlen = skb->len;
1826 
1827 	if (pskb_trim(skb, snaplen))
1828 		goto drop_n_acct;
1829 
1830 	skb_set_owner_r(skb, sk);
1831 	skb->dev = NULL;
1832 	skb_dst_drop(skb);
1833 
1834 	/* drop conntrack reference */
1835 	nf_reset(skb);
1836 
1837 	spin_lock(&sk->sk_receive_queue.lock);
1838 	po->stats.stats1.tp_packets++;
1839 	skb->dropcount = atomic_read(&sk->sk_drops);
1840 	__skb_queue_tail(&sk->sk_receive_queue, skb);
1841 	spin_unlock(&sk->sk_receive_queue.lock);
1842 	sk->sk_data_ready(sk);
1843 	return 0;
1844 
1845 drop_n_acct:
1846 	spin_lock(&sk->sk_receive_queue.lock);
1847 	po->stats.stats1.tp_drops++;
1848 	atomic_inc(&sk->sk_drops);
1849 	spin_unlock(&sk->sk_receive_queue.lock);
1850 
1851 drop_n_restore:
1852 	if (skb_head != skb->data && skb_shared(skb)) {
1853 		skb->data = skb_head;
1854 		skb->len = skb_len;
1855 	}
1856 drop:
1857 	consume_skb(skb);
1858 	return 0;
1859 }
1860 
1861 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1862 		       struct packet_type *pt, struct net_device *orig_dev)
1863 {
1864 	struct sock *sk;
1865 	struct packet_sock *po;
1866 	struct sockaddr_ll *sll;
1867 	union tpacket_uhdr h;
1868 	u8 *skb_head = skb->data;
1869 	int skb_len = skb->len;
1870 	unsigned int snaplen, res;
1871 	unsigned long status = TP_STATUS_USER;
1872 	unsigned short macoff, netoff, hdrlen;
1873 	struct sk_buff *copy_skb = NULL;
1874 	struct timespec ts;
1875 	__u32 ts_status;
1876 
1877 	/* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
1878 	 * We may add members to them until current aligned size without forcing
1879 	 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
1880 	 */
1881 	BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
1882 	BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
1883 
1884 	if (skb->pkt_type == PACKET_LOOPBACK)
1885 		goto drop;
1886 
1887 	sk = pt->af_packet_priv;
1888 	po = pkt_sk(sk);
1889 
1890 	if (!net_eq(dev_net(dev), sock_net(sk)))
1891 		goto drop;
1892 
1893 	if (dev->header_ops) {
1894 		if (sk->sk_type != SOCK_DGRAM)
1895 			skb_push(skb, skb->data - skb_mac_header(skb));
1896 		else if (skb->pkt_type == PACKET_OUTGOING) {
1897 			/* Special case: outgoing packets have ll header at head */
1898 			skb_pull(skb, skb_network_offset(skb));
1899 		}
1900 	}
1901 
1902 	if (skb->ip_summed == CHECKSUM_PARTIAL)
1903 		status |= TP_STATUS_CSUMNOTREADY;
1904 
1905 	snaplen = skb->len;
1906 
1907 	res = run_filter(skb, sk, snaplen);
1908 	if (!res)
1909 		goto drop_n_restore;
1910 	if (snaplen > res)
1911 		snaplen = res;
1912 
1913 	if (sk->sk_type == SOCK_DGRAM) {
1914 		macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1915 				  po->tp_reserve;
1916 	} else {
1917 		unsigned int maclen = skb_network_offset(skb);
1918 		netoff = TPACKET_ALIGN(po->tp_hdrlen +
1919 				       (maclen < 16 ? 16 : maclen)) +
1920 			po->tp_reserve;
1921 		macoff = netoff - maclen;
1922 	}
1923 	if (po->tp_version <= TPACKET_V2) {
1924 		if (macoff + snaplen > po->rx_ring.frame_size) {
1925 			if (po->copy_thresh &&
1926 			    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1927 				if (skb_shared(skb)) {
1928 					copy_skb = skb_clone(skb, GFP_ATOMIC);
1929 				} else {
1930 					copy_skb = skb_get(skb);
1931 					skb_head = skb->data;
1932 				}
1933 				if (copy_skb)
1934 					skb_set_owner_r(copy_skb, sk);
1935 			}
1936 			snaplen = po->rx_ring.frame_size - macoff;
1937 			if ((int)snaplen < 0)
1938 				snaplen = 0;
1939 		}
1940 	} else if (unlikely(macoff + snaplen >
1941 			    GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
1942 		u32 nval;
1943 
1944 		nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
1945 		pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
1946 			    snaplen, nval, macoff);
1947 		snaplen = nval;
1948 		if (unlikely((int)snaplen < 0)) {
1949 			snaplen = 0;
1950 			macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
1951 		}
1952 	}
1953 	spin_lock(&sk->sk_receive_queue.lock);
1954 	h.raw = packet_current_rx_frame(po, skb,
1955 					TP_STATUS_KERNEL, (macoff+snaplen));
1956 	if (!h.raw)
1957 		goto ring_is_full;
1958 	if (po->tp_version <= TPACKET_V2) {
1959 		packet_increment_rx_head(po, &po->rx_ring);
1960 	/*
1961 	 * LOSING will be reported till you read the stats,
1962 	 * because it's COR - Clear On Read.
1963 	 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1964 	 * at packet level.
1965 	 */
1966 		if (po->stats.stats1.tp_drops)
1967 			status |= TP_STATUS_LOSING;
1968 	}
1969 	po->stats.stats1.tp_packets++;
1970 	if (copy_skb) {
1971 		status |= TP_STATUS_COPY;
1972 		__skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1973 	}
1974 	spin_unlock(&sk->sk_receive_queue.lock);
1975 
1976 	skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1977 
1978 	if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
1979 		getnstimeofday(&ts);
1980 
1981 	status |= ts_status;
1982 
1983 	switch (po->tp_version) {
1984 	case TPACKET_V1:
1985 		h.h1->tp_len = skb->len;
1986 		h.h1->tp_snaplen = snaplen;
1987 		h.h1->tp_mac = macoff;
1988 		h.h1->tp_net = netoff;
1989 		h.h1->tp_sec = ts.tv_sec;
1990 		h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
1991 		hdrlen = sizeof(*h.h1);
1992 		break;
1993 	case TPACKET_V2:
1994 		h.h2->tp_len = skb->len;
1995 		h.h2->tp_snaplen = snaplen;
1996 		h.h2->tp_mac = macoff;
1997 		h.h2->tp_net = netoff;
1998 		h.h2->tp_sec = ts.tv_sec;
1999 		h.h2->tp_nsec = ts.tv_nsec;
2000 		if (vlan_tx_tag_present(skb)) {
2001 			h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
2002 			h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2003 			status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2004 		} else {
2005 			h.h2->tp_vlan_tci = 0;
2006 			h.h2->tp_vlan_tpid = 0;
2007 		}
2008 		memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2009 		hdrlen = sizeof(*h.h2);
2010 		break;
2011 	case TPACKET_V3:
2012 		/* tp_nxt_offset,vlan are already populated above.
2013 		 * So DONT clear those fields here
2014 		 */
2015 		h.h3->tp_status |= status;
2016 		h.h3->tp_len = skb->len;
2017 		h.h3->tp_snaplen = snaplen;
2018 		h.h3->tp_mac = macoff;
2019 		h.h3->tp_net = netoff;
2020 		h.h3->tp_sec  = ts.tv_sec;
2021 		h.h3->tp_nsec = ts.tv_nsec;
2022 		memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2023 		hdrlen = sizeof(*h.h3);
2024 		break;
2025 	default:
2026 		BUG();
2027 	}
2028 
2029 	sll = h.raw + TPACKET_ALIGN(hdrlen);
2030 	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2031 	sll->sll_family = AF_PACKET;
2032 	sll->sll_hatype = dev->type;
2033 	sll->sll_protocol = skb->protocol;
2034 	sll->sll_pkttype = skb->pkt_type;
2035 	if (unlikely(po->origdev))
2036 		sll->sll_ifindex = orig_dev->ifindex;
2037 	else
2038 		sll->sll_ifindex = dev->ifindex;
2039 
2040 	smp_mb();
2041 
2042 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2043 	if (po->tp_version <= TPACKET_V2) {
2044 		u8 *start, *end;
2045 
2046 		end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2047 					macoff + snaplen);
2048 
2049 		for (start = h.raw; start < end; start += PAGE_SIZE)
2050 			flush_dcache_page(pgv_to_page(start));
2051 	}
2052 	smp_wmb();
2053 #endif
2054 
2055 	if (po->tp_version <= TPACKET_V2)
2056 		__packet_set_status(po, h.raw, status);
2057 	else
2058 		prb_clear_blk_fill_status(&po->rx_ring);
2059 
2060 	sk->sk_data_ready(sk);
2061 
2062 drop_n_restore:
2063 	if (skb_head != skb->data && skb_shared(skb)) {
2064 		skb->data = skb_head;
2065 		skb->len = skb_len;
2066 	}
2067 drop:
2068 	kfree_skb(skb);
2069 	return 0;
2070 
2071 ring_is_full:
2072 	po->stats.stats1.tp_drops++;
2073 	spin_unlock(&sk->sk_receive_queue.lock);
2074 
2075 	sk->sk_data_ready(sk);
2076 	kfree_skb(copy_skb);
2077 	goto drop_n_restore;
2078 }
2079 
2080 static void tpacket_destruct_skb(struct sk_buff *skb)
2081 {
2082 	struct packet_sock *po = pkt_sk(skb->sk);
2083 
2084 	if (likely(po->tx_ring.pg_vec)) {
2085 		void *ph;
2086 		__u32 ts;
2087 
2088 		ph = skb_shinfo(skb)->destructor_arg;
2089 		packet_dec_pending(&po->tx_ring);
2090 
2091 		ts = __packet_set_timestamp(po, ph, skb);
2092 		__packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2093 	}
2094 
2095 	sock_wfree(skb);
2096 }
2097 
2098 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2099 		void *frame, struct net_device *dev, int size_max,
2100 		__be16 proto, unsigned char *addr, int hlen)
2101 {
2102 	union tpacket_uhdr ph;
2103 	int to_write, offset, len, tp_len, nr_frags, len_max;
2104 	struct socket *sock = po->sk.sk_socket;
2105 	struct page *page;
2106 	void *data;
2107 	int err;
2108 
2109 	ph.raw = frame;
2110 
2111 	skb->protocol = proto;
2112 	skb->dev = dev;
2113 	skb->priority = po->sk.sk_priority;
2114 	skb->mark = po->sk.sk_mark;
2115 	sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2116 	skb_shinfo(skb)->destructor_arg = ph.raw;
2117 
2118 	switch (po->tp_version) {
2119 	case TPACKET_V2:
2120 		tp_len = ph.h2->tp_len;
2121 		break;
2122 	default:
2123 		tp_len = ph.h1->tp_len;
2124 		break;
2125 	}
2126 	if (unlikely(tp_len > size_max)) {
2127 		pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2128 		return -EMSGSIZE;
2129 	}
2130 
2131 	skb_reserve(skb, hlen);
2132 	skb_reset_network_header(skb);
2133 
2134 	if (!packet_use_direct_xmit(po))
2135 		skb_probe_transport_header(skb, 0);
2136 	if (unlikely(po->tp_tx_has_off)) {
2137 		int off_min, off_max, off;
2138 		off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2139 		off_max = po->tx_ring.frame_size - tp_len;
2140 		if (sock->type == SOCK_DGRAM) {
2141 			switch (po->tp_version) {
2142 			case TPACKET_V2:
2143 				off = ph.h2->tp_net;
2144 				break;
2145 			default:
2146 				off = ph.h1->tp_net;
2147 				break;
2148 			}
2149 		} else {
2150 			switch (po->tp_version) {
2151 			case TPACKET_V2:
2152 				off = ph.h2->tp_mac;
2153 				break;
2154 			default:
2155 				off = ph.h1->tp_mac;
2156 				break;
2157 			}
2158 		}
2159 		if (unlikely((off < off_min) || (off_max < off)))
2160 			return -EINVAL;
2161 		data = ph.raw + off;
2162 	} else {
2163 		data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2164 	}
2165 	to_write = tp_len;
2166 
2167 	if (sock->type == SOCK_DGRAM) {
2168 		err = dev_hard_header(skb, dev, ntohs(proto), addr,
2169 				NULL, tp_len);
2170 		if (unlikely(err < 0))
2171 			return -EINVAL;
2172 	} else if (dev->hard_header_len) {
2173 		/* net device doesn't like empty head */
2174 		if (unlikely(tp_len <= dev->hard_header_len)) {
2175 			pr_err("packet size is too short (%d < %d)\n",
2176 			       tp_len, dev->hard_header_len);
2177 			return -EINVAL;
2178 		}
2179 
2180 		skb_push(skb, dev->hard_header_len);
2181 		err = skb_store_bits(skb, 0, data,
2182 				dev->hard_header_len);
2183 		if (unlikely(err))
2184 			return err;
2185 
2186 		data += dev->hard_header_len;
2187 		to_write -= dev->hard_header_len;
2188 	}
2189 
2190 	offset = offset_in_page(data);
2191 	len_max = PAGE_SIZE - offset;
2192 	len = ((to_write > len_max) ? len_max : to_write);
2193 
2194 	skb->data_len = to_write;
2195 	skb->len += to_write;
2196 	skb->truesize += to_write;
2197 	atomic_add(to_write, &po->sk.sk_wmem_alloc);
2198 
2199 	while (likely(to_write)) {
2200 		nr_frags = skb_shinfo(skb)->nr_frags;
2201 
2202 		if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2203 			pr_err("Packet exceed the number of skb frags(%lu)\n",
2204 			       MAX_SKB_FRAGS);
2205 			return -EFAULT;
2206 		}
2207 
2208 		page = pgv_to_page(data);
2209 		data += len;
2210 		flush_dcache_page(page);
2211 		get_page(page);
2212 		skb_fill_page_desc(skb, nr_frags, page, offset, len);
2213 		to_write -= len;
2214 		offset = 0;
2215 		len_max = PAGE_SIZE;
2216 		len = ((to_write > len_max) ? len_max : to_write);
2217 	}
2218 
2219 	return tp_len;
2220 }
2221 
2222 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2223 {
2224 	struct sk_buff *skb;
2225 	struct net_device *dev;
2226 	__be16 proto;
2227 	int err, reserve = 0;
2228 	void *ph;
2229 	DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2230 	bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2231 	int tp_len, size_max;
2232 	unsigned char *addr;
2233 	int len_sum = 0;
2234 	int status = TP_STATUS_AVAILABLE;
2235 	int hlen, tlen;
2236 
2237 	mutex_lock(&po->pg_vec_lock);
2238 
2239 	if (likely(saddr == NULL)) {
2240 		dev	= packet_cached_dev_get(po);
2241 		proto	= po->num;
2242 		addr	= NULL;
2243 	} else {
2244 		err = -EINVAL;
2245 		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2246 			goto out;
2247 		if (msg->msg_namelen < (saddr->sll_halen
2248 					+ offsetof(struct sockaddr_ll,
2249 						sll_addr)))
2250 			goto out;
2251 		proto	= saddr->sll_protocol;
2252 		addr	= saddr->sll_addr;
2253 		dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2254 	}
2255 
2256 	err = -ENXIO;
2257 	if (unlikely(dev == NULL))
2258 		goto out;
2259 	err = -ENETDOWN;
2260 	if (unlikely(!(dev->flags & IFF_UP)))
2261 		goto out_put;
2262 
2263 	reserve = dev->hard_header_len + VLAN_HLEN;
2264 	size_max = po->tx_ring.frame_size
2265 		- (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2266 
2267 	if (size_max > dev->mtu + reserve)
2268 		size_max = dev->mtu + reserve;
2269 
2270 	do {
2271 		ph = packet_current_frame(po, &po->tx_ring,
2272 					  TP_STATUS_SEND_REQUEST);
2273 		if (unlikely(ph == NULL)) {
2274 			if (need_wait && need_resched())
2275 				schedule();
2276 			continue;
2277 		}
2278 
2279 		status = TP_STATUS_SEND_REQUEST;
2280 		hlen = LL_RESERVED_SPACE(dev);
2281 		tlen = dev->needed_tailroom;
2282 		skb = sock_alloc_send_skb(&po->sk,
2283 				hlen + tlen + sizeof(struct sockaddr_ll),
2284 				0, &err);
2285 
2286 		if (unlikely(skb == NULL))
2287 			goto out_status;
2288 
2289 		tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2290 					  addr, hlen);
2291 		if (tp_len > dev->mtu + dev->hard_header_len) {
2292 			struct ethhdr *ehdr;
2293 			/* Earlier code assumed this would be a VLAN pkt,
2294 			 * double-check this now that we have the actual
2295 			 * packet in hand.
2296 			 */
2297 
2298 			skb_reset_mac_header(skb);
2299 			ehdr = eth_hdr(skb);
2300 			if (ehdr->h_proto != htons(ETH_P_8021Q))
2301 				tp_len = -EMSGSIZE;
2302 		}
2303 		if (unlikely(tp_len < 0)) {
2304 			if (po->tp_loss) {
2305 				__packet_set_status(po, ph,
2306 						TP_STATUS_AVAILABLE);
2307 				packet_increment_head(&po->tx_ring);
2308 				kfree_skb(skb);
2309 				continue;
2310 			} else {
2311 				status = TP_STATUS_WRONG_FORMAT;
2312 				err = tp_len;
2313 				goto out_status;
2314 			}
2315 		}
2316 
2317 		packet_pick_tx_queue(dev, skb);
2318 
2319 		skb->destructor = tpacket_destruct_skb;
2320 		__packet_set_status(po, ph, TP_STATUS_SENDING);
2321 		packet_inc_pending(&po->tx_ring);
2322 
2323 		status = TP_STATUS_SEND_REQUEST;
2324 		err = po->xmit(skb);
2325 		if (unlikely(err > 0)) {
2326 			err = net_xmit_errno(err);
2327 			if (err && __packet_get_status(po, ph) ==
2328 				   TP_STATUS_AVAILABLE) {
2329 				/* skb was destructed already */
2330 				skb = NULL;
2331 				goto out_status;
2332 			}
2333 			/*
2334 			 * skb was dropped but not destructed yet;
2335 			 * let's treat it like congestion or err < 0
2336 			 */
2337 			err = 0;
2338 		}
2339 		packet_increment_head(&po->tx_ring);
2340 		len_sum += tp_len;
2341 	} while (likely((ph != NULL) ||
2342 		/* Note: packet_read_pending() might be slow if we have
2343 		 * to call it as it's per_cpu variable, but in fast-path
2344 		 * we already short-circuit the loop with the first
2345 		 * condition, and luckily don't have to go that path
2346 		 * anyway.
2347 		 */
2348 		 (need_wait && packet_read_pending(&po->tx_ring))));
2349 
2350 	err = len_sum;
2351 	goto out_put;
2352 
2353 out_status:
2354 	__packet_set_status(po, ph, status);
2355 	kfree_skb(skb);
2356 out_put:
2357 	dev_put(dev);
2358 out:
2359 	mutex_unlock(&po->pg_vec_lock);
2360 	return err;
2361 }
2362 
2363 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2364 				        size_t reserve, size_t len,
2365 				        size_t linear, int noblock,
2366 				        int *err)
2367 {
2368 	struct sk_buff *skb;
2369 
2370 	/* Under a page?  Don't bother with paged skb. */
2371 	if (prepad + len < PAGE_SIZE || !linear)
2372 		linear = len;
2373 
2374 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2375 				   err, 0);
2376 	if (!skb)
2377 		return NULL;
2378 
2379 	skb_reserve(skb, reserve);
2380 	skb_put(skb, linear);
2381 	skb->data_len = len - linear;
2382 	skb->len += len - linear;
2383 
2384 	return skb;
2385 }
2386 
2387 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2388 {
2389 	struct sock *sk = sock->sk;
2390 	DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2391 	struct sk_buff *skb;
2392 	struct net_device *dev;
2393 	__be16 proto;
2394 	unsigned char *addr;
2395 	int err, reserve = 0;
2396 	struct virtio_net_hdr vnet_hdr = { 0 };
2397 	int offset = 0;
2398 	int vnet_hdr_len;
2399 	struct packet_sock *po = pkt_sk(sk);
2400 	unsigned short gso_type = 0;
2401 	int hlen, tlen;
2402 	int extra_len = 0;
2403 
2404 	/*
2405 	 *	Get and verify the address.
2406 	 */
2407 
2408 	if (likely(saddr == NULL)) {
2409 		dev	= packet_cached_dev_get(po);
2410 		proto	= po->num;
2411 		addr	= NULL;
2412 	} else {
2413 		err = -EINVAL;
2414 		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2415 			goto out;
2416 		if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2417 			goto out;
2418 		proto	= saddr->sll_protocol;
2419 		addr	= saddr->sll_addr;
2420 		dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2421 	}
2422 
2423 	err = -ENXIO;
2424 	if (unlikely(dev == NULL))
2425 		goto out_unlock;
2426 	err = -ENETDOWN;
2427 	if (unlikely(!(dev->flags & IFF_UP)))
2428 		goto out_unlock;
2429 
2430 	if (sock->type == SOCK_RAW)
2431 		reserve = dev->hard_header_len;
2432 	if (po->has_vnet_hdr) {
2433 		vnet_hdr_len = sizeof(vnet_hdr);
2434 
2435 		err = -EINVAL;
2436 		if (len < vnet_hdr_len)
2437 			goto out_unlock;
2438 
2439 		len -= vnet_hdr_len;
2440 
2441 		err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2442 				       vnet_hdr_len);
2443 		if (err < 0)
2444 			goto out_unlock;
2445 
2446 		if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2447 		    (__virtio16_to_cpu(false, vnet_hdr.csum_start) +
2448 		     __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2 >
2449 		      __virtio16_to_cpu(false, vnet_hdr.hdr_len)))
2450 			vnet_hdr.hdr_len = __cpu_to_virtio16(false,
2451 				 __virtio16_to_cpu(false, vnet_hdr.csum_start) +
2452 				__virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2);
2453 
2454 		err = -EINVAL;
2455 		if (__virtio16_to_cpu(false, vnet_hdr.hdr_len) > len)
2456 			goto out_unlock;
2457 
2458 		if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2459 			switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2460 			case VIRTIO_NET_HDR_GSO_TCPV4:
2461 				gso_type = SKB_GSO_TCPV4;
2462 				break;
2463 			case VIRTIO_NET_HDR_GSO_TCPV6:
2464 				gso_type = SKB_GSO_TCPV6;
2465 				break;
2466 			case VIRTIO_NET_HDR_GSO_UDP:
2467 				gso_type = SKB_GSO_UDP;
2468 				break;
2469 			default:
2470 				goto out_unlock;
2471 			}
2472 
2473 			if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2474 				gso_type |= SKB_GSO_TCP_ECN;
2475 
2476 			if (vnet_hdr.gso_size == 0)
2477 				goto out_unlock;
2478 
2479 		}
2480 	}
2481 
2482 	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2483 		if (!netif_supports_nofcs(dev)) {
2484 			err = -EPROTONOSUPPORT;
2485 			goto out_unlock;
2486 		}
2487 		extra_len = 4; /* We're doing our own CRC */
2488 	}
2489 
2490 	err = -EMSGSIZE;
2491 	if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2492 		goto out_unlock;
2493 
2494 	err = -ENOBUFS;
2495 	hlen = LL_RESERVED_SPACE(dev);
2496 	tlen = dev->needed_tailroom;
2497 	skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
2498 			       __virtio16_to_cpu(false, vnet_hdr.hdr_len),
2499 			       msg->msg_flags & MSG_DONTWAIT, &err);
2500 	if (skb == NULL)
2501 		goto out_unlock;
2502 
2503 	skb_set_network_header(skb, reserve);
2504 
2505 	err = -EINVAL;
2506 	if (sock->type == SOCK_DGRAM &&
2507 	    (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2508 		goto out_free;
2509 
2510 	/* Returns -EFAULT on error */
2511 	err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2512 	if (err)
2513 		goto out_free;
2514 
2515 	sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2516 
2517 	if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2518 		/* Earlier code assumed this would be a VLAN pkt,
2519 		 * double-check this now that we have the actual
2520 		 * packet in hand.
2521 		 */
2522 		struct ethhdr *ehdr;
2523 		skb_reset_mac_header(skb);
2524 		ehdr = eth_hdr(skb);
2525 		if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2526 			err = -EMSGSIZE;
2527 			goto out_free;
2528 		}
2529 	}
2530 
2531 	skb->protocol = proto;
2532 	skb->dev = dev;
2533 	skb->priority = sk->sk_priority;
2534 	skb->mark = sk->sk_mark;
2535 
2536 	packet_pick_tx_queue(dev, skb);
2537 
2538 	if (po->has_vnet_hdr) {
2539 		if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2540 			u16 s = __virtio16_to_cpu(false, vnet_hdr.csum_start);
2541 			u16 o = __virtio16_to_cpu(false, vnet_hdr.csum_offset);
2542 			if (!skb_partial_csum_set(skb, s, o)) {
2543 				err = -EINVAL;
2544 				goto out_free;
2545 			}
2546 		}
2547 
2548 		skb_shinfo(skb)->gso_size =
2549 			__virtio16_to_cpu(false, vnet_hdr.gso_size);
2550 		skb_shinfo(skb)->gso_type = gso_type;
2551 
2552 		/* Header must be checked, and gso_segs computed. */
2553 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2554 		skb_shinfo(skb)->gso_segs = 0;
2555 
2556 		len += vnet_hdr_len;
2557 	}
2558 
2559 	if (!packet_use_direct_xmit(po))
2560 		skb_probe_transport_header(skb, reserve);
2561 	if (unlikely(extra_len == 4))
2562 		skb->no_fcs = 1;
2563 
2564 	err = po->xmit(skb);
2565 	if (err > 0 && (err = net_xmit_errno(err)) != 0)
2566 		goto out_unlock;
2567 
2568 	dev_put(dev);
2569 
2570 	return len;
2571 
2572 out_free:
2573 	kfree_skb(skb);
2574 out_unlock:
2575 	if (dev)
2576 		dev_put(dev);
2577 out:
2578 	return err;
2579 }
2580 
2581 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2582 		struct msghdr *msg, size_t len)
2583 {
2584 	struct sock *sk = sock->sk;
2585 	struct packet_sock *po = pkt_sk(sk);
2586 
2587 	if (po->tx_ring.pg_vec)
2588 		return tpacket_snd(po, msg);
2589 	else
2590 		return packet_snd(sock, msg, len);
2591 }
2592 
2593 /*
2594  *	Close a PACKET socket. This is fairly simple. We immediately go
2595  *	to 'closed' state and remove our protocol entry in the device list.
2596  */
2597 
2598 static int packet_release(struct socket *sock)
2599 {
2600 	struct sock *sk = sock->sk;
2601 	struct packet_sock *po;
2602 	struct net *net;
2603 	union tpacket_req_u req_u;
2604 
2605 	if (!sk)
2606 		return 0;
2607 
2608 	net = sock_net(sk);
2609 	po = pkt_sk(sk);
2610 
2611 	mutex_lock(&net->packet.sklist_lock);
2612 	sk_del_node_init_rcu(sk);
2613 	mutex_unlock(&net->packet.sklist_lock);
2614 
2615 	preempt_disable();
2616 	sock_prot_inuse_add(net, sk->sk_prot, -1);
2617 	preempt_enable();
2618 
2619 	spin_lock(&po->bind_lock);
2620 	unregister_prot_hook(sk, false);
2621 	packet_cached_dev_reset(po);
2622 
2623 	if (po->prot_hook.dev) {
2624 		dev_put(po->prot_hook.dev);
2625 		po->prot_hook.dev = NULL;
2626 	}
2627 	spin_unlock(&po->bind_lock);
2628 
2629 	packet_flush_mclist(sk);
2630 
2631 	if (po->rx_ring.pg_vec) {
2632 		memset(&req_u, 0, sizeof(req_u));
2633 		packet_set_ring(sk, &req_u, 1, 0);
2634 	}
2635 
2636 	if (po->tx_ring.pg_vec) {
2637 		memset(&req_u, 0, sizeof(req_u));
2638 		packet_set_ring(sk, &req_u, 1, 1);
2639 	}
2640 
2641 	fanout_release(sk);
2642 
2643 	synchronize_net();
2644 	/*
2645 	 *	Now the socket is dead. No more input will appear.
2646 	 */
2647 	sock_orphan(sk);
2648 	sock->sk = NULL;
2649 
2650 	/* Purge queues */
2651 
2652 	skb_queue_purge(&sk->sk_receive_queue);
2653 	packet_free_pending(po);
2654 	sk_refcnt_debug_release(sk);
2655 
2656 	sock_put(sk);
2657 	return 0;
2658 }
2659 
2660 /*
2661  *	Attach a packet hook.
2662  */
2663 
2664 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 proto)
2665 {
2666 	struct packet_sock *po = pkt_sk(sk);
2667 	const struct net_device *dev_curr;
2668 	__be16 proto_curr;
2669 	bool need_rehook;
2670 
2671 	if (po->fanout) {
2672 		if (dev)
2673 			dev_put(dev);
2674 
2675 		return -EINVAL;
2676 	}
2677 
2678 	lock_sock(sk);
2679 	spin_lock(&po->bind_lock);
2680 
2681 	proto_curr = po->prot_hook.type;
2682 	dev_curr = po->prot_hook.dev;
2683 
2684 	need_rehook = proto_curr != proto || dev_curr != dev;
2685 
2686 	if (need_rehook) {
2687 		unregister_prot_hook(sk, true);
2688 
2689 		po->num = proto;
2690 		po->prot_hook.type = proto;
2691 
2692 		if (po->prot_hook.dev)
2693 			dev_put(po->prot_hook.dev);
2694 
2695 		po->prot_hook.dev = dev;
2696 
2697 		po->ifindex = dev ? dev->ifindex : 0;
2698 		packet_cached_dev_assign(po, dev);
2699 	}
2700 
2701 	if (proto == 0 || !need_rehook)
2702 		goto out_unlock;
2703 
2704 	if (!dev || (dev->flags & IFF_UP)) {
2705 		register_prot_hook(sk);
2706 	} else {
2707 		sk->sk_err = ENETDOWN;
2708 		if (!sock_flag(sk, SOCK_DEAD))
2709 			sk->sk_error_report(sk);
2710 	}
2711 
2712 out_unlock:
2713 	spin_unlock(&po->bind_lock);
2714 	release_sock(sk);
2715 	return 0;
2716 }
2717 
2718 /*
2719  *	Bind a packet socket to a device
2720  */
2721 
2722 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2723 			    int addr_len)
2724 {
2725 	struct sock *sk = sock->sk;
2726 	char name[15];
2727 	struct net_device *dev;
2728 	int err = -ENODEV;
2729 
2730 	/*
2731 	 *	Check legality
2732 	 */
2733 
2734 	if (addr_len != sizeof(struct sockaddr))
2735 		return -EINVAL;
2736 	strlcpy(name, uaddr->sa_data, sizeof(name));
2737 
2738 	dev = dev_get_by_name(sock_net(sk), name);
2739 	if (dev)
2740 		err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2741 	return err;
2742 }
2743 
2744 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2745 {
2746 	struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2747 	struct sock *sk = sock->sk;
2748 	struct net_device *dev = NULL;
2749 	int err;
2750 
2751 
2752 	/*
2753 	 *	Check legality
2754 	 */
2755 
2756 	if (addr_len < sizeof(struct sockaddr_ll))
2757 		return -EINVAL;
2758 	if (sll->sll_family != AF_PACKET)
2759 		return -EINVAL;
2760 
2761 	if (sll->sll_ifindex) {
2762 		err = -ENODEV;
2763 		dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2764 		if (dev == NULL)
2765 			goto out;
2766 	}
2767 	err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2768 
2769 out:
2770 	return err;
2771 }
2772 
2773 static struct proto packet_proto = {
2774 	.name	  = "PACKET",
2775 	.owner	  = THIS_MODULE,
2776 	.obj_size = sizeof(struct packet_sock),
2777 };
2778 
2779 /*
2780  *	Create a packet of type SOCK_PACKET.
2781  */
2782 
2783 static int packet_create(struct net *net, struct socket *sock, int protocol,
2784 			 int kern)
2785 {
2786 	struct sock *sk;
2787 	struct packet_sock *po;
2788 	__be16 proto = (__force __be16)protocol; /* weird, but documented */
2789 	int err;
2790 
2791 	if (!ns_capable(net->user_ns, CAP_NET_RAW))
2792 		return -EPERM;
2793 	if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2794 	    sock->type != SOCK_PACKET)
2795 		return -ESOCKTNOSUPPORT;
2796 
2797 	sock->state = SS_UNCONNECTED;
2798 
2799 	err = -ENOBUFS;
2800 	sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2801 	if (sk == NULL)
2802 		goto out;
2803 
2804 	sock->ops = &packet_ops;
2805 	if (sock->type == SOCK_PACKET)
2806 		sock->ops = &packet_ops_spkt;
2807 
2808 	sock_init_data(sock, sk);
2809 
2810 	po = pkt_sk(sk);
2811 	sk->sk_family = PF_PACKET;
2812 	po->num = proto;
2813 	po->xmit = dev_queue_xmit;
2814 
2815 	err = packet_alloc_pending(po);
2816 	if (err)
2817 		goto out2;
2818 
2819 	packet_cached_dev_reset(po);
2820 
2821 	sk->sk_destruct = packet_sock_destruct;
2822 	sk_refcnt_debug_inc(sk);
2823 
2824 	/*
2825 	 *	Attach a protocol block
2826 	 */
2827 
2828 	spin_lock_init(&po->bind_lock);
2829 	mutex_init(&po->pg_vec_lock);
2830 	po->prot_hook.func = packet_rcv;
2831 
2832 	if (sock->type == SOCK_PACKET)
2833 		po->prot_hook.func = packet_rcv_spkt;
2834 
2835 	po->prot_hook.af_packet_priv = sk;
2836 
2837 	if (proto) {
2838 		po->prot_hook.type = proto;
2839 		register_prot_hook(sk);
2840 	}
2841 
2842 	mutex_lock(&net->packet.sklist_lock);
2843 	sk_add_node_rcu(sk, &net->packet.sklist);
2844 	mutex_unlock(&net->packet.sklist_lock);
2845 
2846 	preempt_disable();
2847 	sock_prot_inuse_add(net, &packet_proto, 1);
2848 	preempt_enable();
2849 
2850 	return 0;
2851 out2:
2852 	sk_free(sk);
2853 out:
2854 	return err;
2855 }
2856 
2857 /*
2858  *	Pull a packet from our receive queue and hand it to the user.
2859  *	If necessary we block.
2860  */
2861 
2862 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2863 			  struct msghdr *msg, size_t len, int flags)
2864 {
2865 	struct sock *sk = sock->sk;
2866 	struct sk_buff *skb;
2867 	int copied, err;
2868 	int vnet_hdr_len = 0;
2869 
2870 	err = -EINVAL;
2871 	if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2872 		goto out;
2873 
2874 #if 0
2875 	/* What error should we return now? EUNATTACH? */
2876 	if (pkt_sk(sk)->ifindex < 0)
2877 		return -ENODEV;
2878 #endif
2879 
2880 	if (flags & MSG_ERRQUEUE) {
2881 		err = sock_recv_errqueue(sk, msg, len,
2882 					 SOL_PACKET, PACKET_TX_TIMESTAMP);
2883 		goto out;
2884 	}
2885 
2886 	/*
2887 	 *	Call the generic datagram receiver. This handles all sorts
2888 	 *	of horrible races and re-entrancy so we can forget about it
2889 	 *	in the protocol layers.
2890 	 *
2891 	 *	Now it will return ENETDOWN, if device have just gone down,
2892 	 *	but then it will block.
2893 	 */
2894 
2895 	skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2896 
2897 	/*
2898 	 *	An error occurred so return it. Because skb_recv_datagram()
2899 	 *	handles the blocking we don't see and worry about blocking
2900 	 *	retries.
2901 	 */
2902 
2903 	if (skb == NULL)
2904 		goto out;
2905 
2906 	if (pkt_sk(sk)->has_vnet_hdr) {
2907 		struct virtio_net_hdr vnet_hdr = { 0 };
2908 
2909 		err = -EINVAL;
2910 		vnet_hdr_len = sizeof(vnet_hdr);
2911 		if (len < vnet_hdr_len)
2912 			goto out_free;
2913 
2914 		len -= vnet_hdr_len;
2915 
2916 		if (skb_is_gso(skb)) {
2917 			struct skb_shared_info *sinfo = skb_shinfo(skb);
2918 
2919 			/* This is a hint as to how much should be linear. */
2920 			vnet_hdr.hdr_len =
2921 				__cpu_to_virtio16(false, skb_headlen(skb));
2922 			vnet_hdr.gso_size =
2923 				__cpu_to_virtio16(false, sinfo->gso_size);
2924 			if (sinfo->gso_type & SKB_GSO_TCPV4)
2925 				vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2926 			else if (sinfo->gso_type & SKB_GSO_TCPV6)
2927 				vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2928 			else if (sinfo->gso_type & SKB_GSO_UDP)
2929 				vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2930 			else if (sinfo->gso_type & SKB_GSO_FCOE)
2931 				goto out_free;
2932 			else
2933 				BUG();
2934 			if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2935 				vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2936 		} else
2937 			vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2938 
2939 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
2940 			vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2941 			vnet_hdr.csum_start = __cpu_to_virtio16(false,
2942 					  skb_checksum_start_offset(skb));
2943 			vnet_hdr.csum_offset = __cpu_to_virtio16(false,
2944 							 skb->csum_offset);
2945 		} else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2946 			vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2947 		} /* else everything is zero */
2948 
2949 		err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2950 				     vnet_hdr_len);
2951 		if (err < 0)
2952 			goto out_free;
2953 	}
2954 
2955 	/* You lose any data beyond the buffer you gave. If it worries
2956 	 * a user program they can ask the device for its MTU
2957 	 * anyway.
2958 	 */
2959 	copied = skb->len;
2960 	if (copied > len) {
2961 		copied = len;
2962 		msg->msg_flags |= MSG_TRUNC;
2963 	}
2964 
2965 	err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2966 	if (err)
2967 		goto out_free;
2968 
2969 	sock_recv_ts_and_drops(msg, sk, skb);
2970 
2971 	if (msg->msg_name) {
2972 		/* If the address length field is there to be filled
2973 		 * in, we fill it in now.
2974 		 */
2975 		if (sock->type == SOCK_PACKET) {
2976 			__sockaddr_check_size(sizeof(struct sockaddr_pkt));
2977 			msg->msg_namelen = sizeof(struct sockaddr_pkt);
2978 		} else {
2979 			struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2980 			msg->msg_namelen = sll->sll_halen +
2981 				offsetof(struct sockaddr_ll, sll_addr);
2982 		}
2983 		memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2984 		       msg->msg_namelen);
2985 	}
2986 
2987 	if (pkt_sk(sk)->auxdata) {
2988 		struct tpacket_auxdata aux;
2989 
2990 		aux.tp_status = TP_STATUS_USER;
2991 		if (skb->ip_summed == CHECKSUM_PARTIAL)
2992 			aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2993 		aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2994 		aux.tp_snaplen = skb->len;
2995 		aux.tp_mac = 0;
2996 		aux.tp_net = skb_network_offset(skb);
2997 		if (vlan_tx_tag_present(skb)) {
2998 			aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2999 			aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3000 			aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3001 		} else {
3002 			aux.tp_vlan_tci = 0;
3003 			aux.tp_vlan_tpid = 0;
3004 		}
3005 		put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3006 	}
3007 
3008 	/*
3009 	 *	Free or return the buffer as appropriate. Again this
3010 	 *	hides all the races and re-entrancy issues from us.
3011 	 */
3012 	err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3013 
3014 out_free:
3015 	skb_free_datagram(sk, skb);
3016 out:
3017 	return err;
3018 }
3019 
3020 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3021 			       int *uaddr_len, int peer)
3022 {
3023 	struct net_device *dev;
3024 	struct sock *sk	= sock->sk;
3025 
3026 	if (peer)
3027 		return -EOPNOTSUPP;
3028 
3029 	uaddr->sa_family = AF_PACKET;
3030 	memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3031 	rcu_read_lock();
3032 	dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3033 	if (dev)
3034 		strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3035 	rcu_read_unlock();
3036 	*uaddr_len = sizeof(*uaddr);
3037 
3038 	return 0;
3039 }
3040 
3041 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3042 			  int *uaddr_len, int peer)
3043 {
3044 	struct net_device *dev;
3045 	struct sock *sk = sock->sk;
3046 	struct packet_sock *po = pkt_sk(sk);
3047 	DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3048 
3049 	if (peer)
3050 		return -EOPNOTSUPP;
3051 
3052 	sll->sll_family = AF_PACKET;
3053 	sll->sll_ifindex = po->ifindex;
3054 	sll->sll_protocol = po->num;
3055 	sll->sll_pkttype = 0;
3056 	rcu_read_lock();
3057 	dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3058 	if (dev) {
3059 		sll->sll_hatype = dev->type;
3060 		sll->sll_halen = dev->addr_len;
3061 		memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3062 	} else {
3063 		sll->sll_hatype = 0;	/* Bad: we have no ARPHRD_UNSPEC */
3064 		sll->sll_halen = 0;
3065 	}
3066 	rcu_read_unlock();
3067 	*uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3068 
3069 	return 0;
3070 }
3071 
3072 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3073 			 int what)
3074 {
3075 	switch (i->type) {
3076 	case PACKET_MR_MULTICAST:
3077 		if (i->alen != dev->addr_len)
3078 			return -EINVAL;
3079 		if (what > 0)
3080 			return dev_mc_add(dev, i->addr);
3081 		else
3082 			return dev_mc_del(dev, i->addr);
3083 		break;
3084 	case PACKET_MR_PROMISC:
3085 		return dev_set_promiscuity(dev, what);
3086 	case PACKET_MR_ALLMULTI:
3087 		return dev_set_allmulti(dev, what);
3088 	case PACKET_MR_UNICAST:
3089 		if (i->alen != dev->addr_len)
3090 			return -EINVAL;
3091 		if (what > 0)
3092 			return dev_uc_add(dev, i->addr);
3093 		else
3094 			return dev_uc_del(dev, i->addr);
3095 		break;
3096 	default:
3097 		break;
3098 	}
3099 	return 0;
3100 }
3101 
3102 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
3103 {
3104 	for ( ; i; i = i->next) {
3105 		if (i->ifindex == dev->ifindex)
3106 			packet_dev_mc(dev, i, what);
3107 	}
3108 }
3109 
3110 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3111 {
3112 	struct packet_sock *po = pkt_sk(sk);
3113 	struct packet_mclist *ml, *i;
3114 	struct net_device *dev;
3115 	int err;
3116 
3117 	rtnl_lock();
3118 
3119 	err = -ENODEV;
3120 	dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3121 	if (!dev)
3122 		goto done;
3123 
3124 	err = -EINVAL;
3125 	if (mreq->mr_alen > dev->addr_len)
3126 		goto done;
3127 
3128 	err = -ENOBUFS;
3129 	i = kmalloc(sizeof(*i), GFP_KERNEL);
3130 	if (i == NULL)
3131 		goto done;
3132 
3133 	err = 0;
3134 	for (ml = po->mclist; ml; ml = ml->next) {
3135 		if (ml->ifindex == mreq->mr_ifindex &&
3136 		    ml->type == mreq->mr_type &&
3137 		    ml->alen == mreq->mr_alen &&
3138 		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3139 			ml->count++;
3140 			/* Free the new element ... */
3141 			kfree(i);
3142 			goto done;
3143 		}
3144 	}
3145 
3146 	i->type = mreq->mr_type;
3147 	i->ifindex = mreq->mr_ifindex;
3148 	i->alen = mreq->mr_alen;
3149 	memcpy(i->addr, mreq->mr_address, i->alen);
3150 	i->count = 1;
3151 	i->next = po->mclist;
3152 	po->mclist = i;
3153 	err = packet_dev_mc(dev, i, 1);
3154 	if (err) {
3155 		po->mclist = i->next;
3156 		kfree(i);
3157 	}
3158 
3159 done:
3160 	rtnl_unlock();
3161 	return err;
3162 }
3163 
3164 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3165 {
3166 	struct packet_mclist *ml, **mlp;
3167 
3168 	rtnl_lock();
3169 
3170 	for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3171 		if (ml->ifindex == mreq->mr_ifindex &&
3172 		    ml->type == mreq->mr_type &&
3173 		    ml->alen == mreq->mr_alen &&
3174 		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3175 			if (--ml->count == 0) {
3176 				struct net_device *dev;
3177 				*mlp = ml->next;
3178 				dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3179 				if (dev)
3180 					packet_dev_mc(dev, ml, -1);
3181 				kfree(ml);
3182 			}
3183 			rtnl_unlock();
3184 			return 0;
3185 		}
3186 	}
3187 	rtnl_unlock();
3188 	return -EADDRNOTAVAIL;
3189 }
3190 
3191 static void packet_flush_mclist(struct sock *sk)
3192 {
3193 	struct packet_sock *po = pkt_sk(sk);
3194 	struct packet_mclist *ml;
3195 
3196 	if (!po->mclist)
3197 		return;
3198 
3199 	rtnl_lock();
3200 	while ((ml = po->mclist) != NULL) {
3201 		struct net_device *dev;
3202 
3203 		po->mclist = ml->next;
3204 		dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3205 		if (dev != NULL)
3206 			packet_dev_mc(dev, ml, -1);
3207 		kfree(ml);
3208 	}
3209 	rtnl_unlock();
3210 }
3211 
3212 static int
3213 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3214 {
3215 	struct sock *sk = sock->sk;
3216 	struct packet_sock *po = pkt_sk(sk);
3217 	int ret;
3218 
3219 	if (level != SOL_PACKET)
3220 		return -ENOPROTOOPT;
3221 
3222 	switch (optname) {
3223 	case PACKET_ADD_MEMBERSHIP:
3224 	case PACKET_DROP_MEMBERSHIP:
3225 	{
3226 		struct packet_mreq_max mreq;
3227 		int len = optlen;
3228 		memset(&mreq, 0, sizeof(mreq));
3229 		if (len < sizeof(struct packet_mreq))
3230 			return -EINVAL;
3231 		if (len > sizeof(mreq))
3232 			len = sizeof(mreq);
3233 		if (copy_from_user(&mreq, optval, len))
3234 			return -EFAULT;
3235 		if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3236 			return -EINVAL;
3237 		if (optname == PACKET_ADD_MEMBERSHIP)
3238 			ret = packet_mc_add(sk, &mreq);
3239 		else
3240 			ret = packet_mc_drop(sk, &mreq);
3241 		return ret;
3242 	}
3243 
3244 	case PACKET_RX_RING:
3245 	case PACKET_TX_RING:
3246 	{
3247 		union tpacket_req_u req_u;
3248 		int len;
3249 
3250 		switch (po->tp_version) {
3251 		case TPACKET_V1:
3252 		case TPACKET_V2:
3253 			len = sizeof(req_u.req);
3254 			break;
3255 		case TPACKET_V3:
3256 		default:
3257 			len = sizeof(req_u.req3);
3258 			break;
3259 		}
3260 		if (optlen < len)
3261 			return -EINVAL;
3262 		if (pkt_sk(sk)->has_vnet_hdr)
3263 			return -EINVAL;
3264 		if (copy_from_user(&req_u.req, optval, len))
3265 			return -EFAULT;
3266 		return packet_set_ring(sk, &req_u, 0,
3267 			optname == PACKET_TX_RING);
3268 	}
3269 	case PACKET_COPY_THRESH:
3270 	{
3271 		int val;
3272 
3273 		if (optlen != sizeof(val))
3274 			return -EINVAL;
3275 		if (copy_from_user(&val, optval, sizeof(val)))
3276 			return -EFAULT;
3277 
3278 		pkt_sk(sk)->copy_thresh = val;
3279 		return 0;
3280 	}
3281 	case PACKET_VERSION:
3282 	{
3283 		int val;
3284 
3285 		if (optlen != sizeof(val))
3286 			return -EINVAL;
3287 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3288 			return -EBUSY;
3289 		if (copy_from_user(&val, optval, sizeof(val)))
3290 			return -EFAULT;
3291 		switch (val) {
3292 		case TPACKET_V1:
3293 		case TPACKET_V2:
3294 		case TPACKET_V3:
3295 			po->tp_version = val;
3296 			return 0;
3297 		default:
3298 			return -EINVAL;
3299 		}
3300 	}
3301 	case PACKET_RESERVE:
3302 	{
3303 		unsigned int val;
3304 
3305 		if (optlen != sizeof(val))
3306 			return -EINVAL;
3307 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3308 			return -EBUSY;
3309 		if (copy_from_user(&val, optval, sizeof(val)))
3310 			return -EFAULT;
3311 		po->tp_reserve = val;
3312 		return 0;
3313 	}
3314 	case PACKET_LOSS:
3315 	{
3316 		unsigned int val;
3317 
3318 		if (optlen != sizeof(val))
3319 			return -EINVAL;
3320 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3321 			return -EBUSY;
3322 		if (copy_from_user(&val, optval, sizeof(val)))
3323 			return -EFAULT;
3324 		po->tp_loss = !!val;
3325 		return 0;
3326 	}
3327 	case PACKET_AUXDATA:
3328 	{
3329 		int val;
3330 
3331 		if (optlen < sizeof(val))
3332 			return -EINVAL;
3333 		if (copy_from_user(&val, optval, sizeof(val)))
3334 			return -EFAULT;
3335 
3336 		po->auxdata = !!val;
3337 		return 0;
3338 	}
3339 	case PACKET_ORIGDEV:
3340 	{
3341 		int val;
3342 
3343 		if (optlen < sizeof(val))
3344 			return -EINVAL;
3345 		if (copy_from_user(&val, optval, sizeof(val)))
3346 			return -EFAULT;
3347 
3348 		po->origdev = !!val;
3349 		return 0;
3350 	}
3351 	case PACKET_VNET_HDR:
3352 	{
3353 		int val;
3354 
3355 		if (sock->type != SOCK_RAW)
3356 			return -EINVAL;
3357 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3358 			return -EBUSY;
3359 		if (optlen < sizeof(val))
3360 			return -EINVAL;
3361 		if (copy_from_user(&val, optval, sizeof(val)))
3362 			return -EFAULT;
3363 
3364 		po->has_vnet_hdr = !!val;
3365 		return 0;
3366 	}
3367 	case PACKET_TIMESTAMP:
3368 	{
3369 		int val;
3370 
3371 		if (optlen != sizeof(val))
3372 			return -EINVAL;
3373 		if (copy_from_user(&val, optval, sizeof(val)))
3374 			return -EFAULT;
3375 
3376 		po->tp_tstamp = val;
3377 		return 0;
3378 	}
3379 	case PACKET_FANOUT:
3380 	{
3381 		int val;
3382 
3383 		if (optlen != sizeof(val))
3384 			return -EINVAL;
3385 		if (copy_from_user(&val, optval, sizeof(val)))
3386 			return -EFAULT;
3387 
3388 		return fanout_add(sk, val & 0xffff, val >> 16);
3389 	}
3390 	case PACKET_TX_HAS_OFF:
3391 	{
3392 		unsigned int val;
3393 
3394 		if (optlen != sizeof(val))
3395 			return -EINVAL;
3396 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3397 			return -EBUSY;
3398 		if (copy_from_user(&val, optval, sizeof(val)))
3399 			return -EFAULT;
3400 		po->tp_tx_has_off = !!val;
3401 		return 0;
3402 	}
3403 	case PACKET_QDISC_BYPASS:
3404 	{
3405 		int val;
3406 
3407 		if (optlen != sizeof(val))
3408 			return -EINVAL;
3409 		if (copy_from_user(&val, optval, sizeof(val)))
3410 			return -EFAULT;
3411 
3412 		po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3413 		return 0;
3414 	}
3415 	default:
3416 		return -ENOPROTOOPT;
3417 	}
3418 }
3419 
3420 static int packet_getsockopt(struct socket *sock, int level, int optname,
3421 			     char __user *optval, int __user *optlen)
3422 {
3423 	int len;
3424 	int val, lv = sizeof(val);
3425 	struct sock *sk = sock->sk;
3426 	struct packet_sock *po = pkt_sk(sk);
3427 	void *data = &val;
3428 	union tpacket_stats_u st;
3429 
3430 	if (level != SOL_PACKET)
3431 		return -ENOPROTOOPT;
3432 
3433 	if (get_user(len, optlen))
3434 		return -EFAULT;
3435 
3436 	if (len < 0)
3437 		return -EINVAL;
3438 
3439 	switch (optname) {
3440 	case PACKET_STATISTICS:
3441 		spin_lock_bh(&sk->sk_receive_queue.lock);
3442 		memcpy(&st, &po->stats, sizeof(st));
3443 		memset(&po->stats, 0, sizeof(po->stats));
3444 		spin_unlock_bh(&sk->sk_receive_queue.lock);
3445 
3446 		if (po->tp_version == TPACKET_V3) {
3447 			lv = sizeof(struct tpacket_stats_v3);
3448 			st.stats3.tp_packets += st.stats3.tp_drops;
3449 			data = &st.stats3;
3450 		} else {
3451 			lv = sizeof(struct tpacket_stats);
3452 			st.stats1.tp_packets += st.stats1.tp_drops;
3453 			data = &st.stats1;
3454 		}
3455 
3456 		break;
3457 	case PACKET_AUXDATA:
3458 		val = po->auxdata;
3459 		break;
3460 	case PACKET_ORIGDEV:
3461 		val = po->origdev;
3462 		break;
3463 	case PACKET_VNET_HDR:
3464 		val = po->has_vnet_hdr;
3465 		break;
3466 	case PACKET_VERSION:
3467 		val = po->tp_version;
3468 		break;
3469 	case PACKET_HDRLEN:
3470 		if (len > sizeof(int))
3471 			len = sizeof(int);
3472 		if (copy_from_user(&val, optval, len))
3473 			return -EFAULT;
3474 		switch (val) {
3475 		case TPACKET_V1:
3476 			val = sizeof(struct tpacket_hdr);
3477 			break;
3478 		case TPACKET_V2:
3479 			val = sizeof(struct tpacket2_hdr);
3480 			break;
3481 		case TPACKET_V3:
3482 			val = sizeof(struct tpacket3_hdr);
3483 			break;
3484 		default:
3485 			return -EINVAL;
3486 		}
3487 		break;
3488 	case PACKET_RESERVE:
3489 		val = po->tp_reserve;
3490 		break;
3491 	case PACKET_LOSS:
3492 		val = po->tp_loss;
3493 		break;
3494 	case PACKET_TIMESTAMP:
3495 		val = po->tp_tstamp;
3496 		break;
3497 	case PACKET_FANOUT:
3498 		val = (po->fanout ?
3499 		       ((u32)po->fanout->id |
3500 			((u32)po->fanout->type << 16) |
3501 			((u32)po->fanout->flags << 24)) :
3502 		       0);
3503 		break;
3504 	case PACKET_TX_HAS_OFF:
3505 		val = po->tp_tx_has_off;
3506 		break;
3507 	case PACKET_QDISC_BYPASS:
3508 		val = packet_use_direct_xmit(po);
3509 		break;
3510 	default:
3511 		return -ENOPROTOOPT;
3512 	}
3513 
3514 	if (len > lv)
3515 		len = lv;
3516 	if (put_user(len, optlen))
3517 		return -EFAULT;
3518 	if (copy_to_user(optval, data, len))
3519 		return -EFAULT;
3520 	return 0;
3521 }
3522 
3523 
3524 static int packet_notifier(struct notifier_block *this,
3525 			   unsigned long msg, void *ptr)
3526 {
3527 	struct sock *sk;
3528 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3529 	struct net *net = dev_net(dev);
3530 
3531 	rcu_read_lock();
3532 	sk_for_each_rcu(sk, &net->packet.sklist) {
3533 		struct packet_sock *po = pkt_sk(sk);
3534 
3535 		switch (msg) {
3536 		case NETDEV_UNREGISTER:
3537 			if (po->mclist)
3538 				packet_dev_mclist(dev, po->mclist, -1);
3539 			/* fallthrough */
3540 
3541 		case NETDEV_DOWN:
3542 			if (dev->ifindex == po->ifindex) {
3543 				spin_lock(&po->bind_lock);
3544 				if (po->running) {
3545 					__unregister_prot_hook(sk, false);
3546 					sk->sk_err = ENETDOWN;
3547 					if (!sock_flag(sk, SOCK_DEAD))
3548 						sk->sk_error_report(sk);
3549 				}
3550 				if (msg == NETDEV_UNREGISTER) {
3551 					packet_cached_dev_reset(po);
3552 					po->ifindex = -1;
3553 					if (po->prot_hook.dev)
3554 						dev_put(po->prot_hook.dev);
3555 					po->prot_hook.dev = NULL;
3556 				}
3557 				spin_unlock(&po->bind_lock);
3558 			}
3559 			break;
3560 		case NETDEV_UP:
3561 			if (dev->ifindex == po->ifindex) {
3562 				spin_lock(&po->bind_lock);
3563 				if (po->num)
3564 					register_prot_hook(sk);
3565 				spin_unlock(&po->bind_lock);
3566 			}
3567 			break;
3568 		}
3569 	}
3570 	rcu_read_unlock();
3571 	return NOTIFY_DONE;
3572 }
3573 
3574 
3575 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3576 			unsigned long arg)
3577 {
3578 	struct sock *sk = sock->sk;
3579 
3580 	switch (cmd) {
3581 	case SIOCOUTQ:
3582 	{
3583 		int amount = sk_wmem_alloc_get(sk);
3584 
3585 		return put_user(amount, (int __user *)arg);
3586 	}
3587 	case SIOCINQ:
3588 	{
3589 		struct sk_buff *skb;
3590 		int amount = 0;
3591 
3592 		spin_lock_bh(&sk->sk_receive_queue.lock);
3593 		skb = skb_peek(&sk->sk_receive_queue);
3594 		if (skb)
3595 			amount = skb->len;
3596 		spin_unlock_bh(&sk->sk_receive_queue.lock);
3597 		return put_user(amount, (int __user *)arg);
3598 	}
3599 	case SIOCGSTAMP:
3600 		return sock_get_timestamp(sk, (struct timeval __user *)arg);
3601 	case SIOCGSTAMPNS:
3602 		return sock_get_timestampns(sk, (struct timespec __user *)arg);
3603 
3604 #ifdef CONFIG_INET
3605 	case SIOCADDRT:
3606 	case SIOCDELRT:
3607 	case SIOCDARP:
3608 	case SIOCGARP:
3609 	case SIOCSARP:
3610 	case SIOCGIFADDR:
3611 	case SIOCSIFADDR:
3612 	case SIOCGIFBRDADDR:
3613 	case SIOCSIFBRDADDR:
3614 	case SIOCGIFNETMASK:
3615 	case SIOCSIFNETMASK:
3616 	case SIOCGIFDSTADDR:
3617 	case SIOCSIFDSTADDR:
3618 	case SIOCSIFFLAGS:
3619 		return inet_dgram_ops.ioctl(sock, cmd, arg);
3620 #endif
3621 
3622 	default:
3623 		return -ENOIOCTLCMD;
3624 	}
3625 	return 0;
3626 }
3627 
3628 static unsigned int packet_poll(struct file *file, struct socket *sock,
3629 				poll_table *wait)
3630 {
3631 	struct sock *sk = sock->sk;
3632 	struct packet_sock *po = pkt_sk(sk);
3633 	unsigned int mask = datagram_poll(file, sock, wait);
3634 
3635 	spin_lock_bh(&sk->sk_receive_queue.lock);
3636 	if (po->rx_ring.pg_vec) {
3637 		if (!packet_previous_rx_frame(po, &po->rx_ring,
3638 			TP_STATUS_KERNEL))
3639 			mask |= POLLIN | POLLRDNORM;
3640 	}
3641 	spin_unlock_bh(&sk->sk_receive_queue.lock);
3642 	spin_lock_bh(&sk->sk_write_queue.lock);
3643 	if (po->tx_ring.pg_vec) {
3644 		if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3645 			mask |= POLLOUT | POLLWRNORM;
3646 	}
3647 	spin_unlock_bh(&sk->sk_write_queue.lock);
3648 	return mask;
3649 }
3650 
3651 
3652 /* Dirty? Well, I still did not learn better way to account
3653  * for user mmaps.
3654  */
3655 
3656 static void packet_mm_open(struct vm_area_struct *vma)
3657 {
3658 	struct file *file = vma->vm_file;
3659 	struct socket *sock = file->private_data;
3660 	struct sock *sk = sock->sk;
3661 
3662 	if (sk)
3663 		atomic_inc(&pkt_sk(sk)->mapped);
3664 }
3665 
3666 static void packet_mm_close(struct vm_area_struct *vma)
3667 {
3668 	struct file *file = vma->vm_file;
3669 	struct socket *sock = file->private_data;
3670 	struct sock *sk = sock->sk;
3671 
3672 	if (sk)
3673 		atomic_dec(&pkt_sk(sk)->mapped);
3674 }
3675 
3676 static const struct vm_operations_struct packet_mmap_ops = {
3677 	.open	=	packet_mm_open,
3678 	.close	=	packet_mm_close,
3679 };
3680 
3681 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3682 			unsigned int len)
3683 {
3684 	int i;
3685 
3686 	for (i = 0; i < len; i++) {
3687 		if (likely(pg_vec[i].buffer)) {
3688 			if (is_vmalloc_addr(pg_vec[i].buffer))
3689 				vfree(pg_vec[i].buffer);
3690 			else
3691 				free_pages((unsigned long)pg_vec[i].buffer,
3692 					   order);
3693 			pg_vec[i].buffer = NULL;
3694 		}
3695 	}
3696 	kfree(pg_vec);
3697 }
3698 
3699 static char *alloc_one_pg_vec_page(unsigned long order)
3700 {
3701 	char *buffer;
3702 	gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3703 			  __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3704 
3705 	buffer = (char *) __get_free_pages(gfp_flags, order);
3706 	if (buffer)
3707 		return buffer;
3708 
3709 	/* __get_free_pages failed, fall back to vmalloc */
3710 	buffer = vzalloc((1 << order) * PAGE_SIZE);
3711 	if (buffer)
3712 		return buffer;
3713 
3714 	/* vmalloc failed, lets dig into swap here */
3715 	gfp_flags &= ~__GFP_NORETRY;
3716 	buffer = (char *) __get_free_pages(gfp_flags, order);
3717 	if (buffer)
3718 		return buffer;
3719 
3720 	/* complete and utter failure */
3721 	return NULL;
3722 }
3723 
3724 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3725 {
3726 	unsigned int block_nr = req->tp_block_nr;
3727 	struct pgv *pg_vec;
3728 	int i;
3729 
3730 	pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3731 	if (unlikely(!pg_vec))
3732 		goto out;
3733 
3734 	for (i = 0; i < block_nr; i++) {
3735 		pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3736 		if (unlikely(!pg_vec[i].buffer))
3737 			goto out_free_pgvec;
3738 	}
3739 
3740 out:
3741 	return pg_vec;
3742 
3743 out_free_pgvec:
3744 	free_pg_vec(pg_vec, order, block_nr);
3745 	pg_vec = NULL;
3746 	goto out;
3747 }
3748 
3749 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3750 		int closing, int tx_ring)
3751 {
3752 	struct pgv *pg_vec = NULL;
3753 	struct packet_sock *po = pkt_sk(sk);
3754 	int was_running, order = 0;
3755 	struct packet_ring_buffer *rb;
3756 	struct sk_buff_head *rb_queue;
3757 	__be16 num;
3758 	int err = -EINVAL;
3759 	/* Added to avoid minimal code churn */
3760 	struct tpacket_req *req = &req_u->req;
3761 
3762 	/* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3763 	if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3764 		WARN(1, "Tx-ring is not supported.\n");
3765 		goto out;
3766 	}
3767 
3768 	rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3769 	rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3770 
3771 	err = -EBUSY;
3772 	if (!closing) {
3773 		if (atomic_read(&po->mapped))
3774 			goto out;
3775 		if (packet_read_pending(rb))
3776 			goto out;
3777 	}
3778 
3779 	if (req->tp_block_nr) {
3780 		/* Sanity tests and some calculations */
3781 		err = -EBUSY;
3782 		if (unlikely(rb->pg_vec))
3783 			goto out;
3784 
3785 		switch (po->tp_version) {
3786 		case TPACKET_V1:
3787 			po->tp_hdrlen = TPACKET_HDRLEN;
3788 			break;
3789 		case TPACKET_V2:
3790 			po->tp_hdrlen = TPACKET2_HDRLEN;
3791 			break;
3792 		case TPACKET_V3:
3793 			po->tp_hdrlen = TPACKET3_HDRLEN;
3794 			break;
3795 		}
3796 
3797 		err = -EINVAL;
3798 		if (unlikely((int)req->tp_block_size <= 0))
3799 			goto out;
3800 		if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3801 			goto out;
3802 		if (po->tp_version >= TPACKET_V3 &&
3803 		    (int)(req->tp_block_size -
3804 			  BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
3805 			goto out;
3806 		if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3807 					po->tp_reserve))
3808 			goto out;
3809 		if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3810 			goto out;
3811 
3812 		rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3813 		if (unlikely(rb->frames_per_block <= 0))
3814 			goto out;
3815 		if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3816 					req->tp_frame_nr))
3817 			goto out;
3818 
3819 		err = -ENOMEM;
3820 		order = get_order(req->tp_block_size);
3821 		pg_vec = alloc_pg_vec(req, order);
3822 		if (unlikely(!pg_vec))
3823 			goto out;
3824 		switch (po->tp_version) {
3825 		case TPACKET_V3:
3826 		/* Transmit path is not supported. We checked
3827 		 * it above but just being paranoid
3828 		 */
3829 			if (!tx_ring)
3830 				init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3831 			break;
3832 		default:
3833 			break;
3834 		}
3835 	}
3836 	/* Done */
3837 	else {
3838 		err = -EINVAL;
3839 		if (unlikely(req->tp_frame_nr))
3840 			goto out;
3841 	}
3842 
3843 	lock_sock(sk);
3844 
3845 	/* Detach socket from network */
3846 	spin_lock(&po->bind_lock);
3847 	was_running = po->running;
3848 	num = po->num;
3849 	if (was_running) {
3850 		po->num = 0;
3851 		__unregister_prot_hook(sk, false);
3852 	}
3853 	spin_unlock(&po->bind_lock);
3854 
3855 	synchronize_net();
3856 
3857 	err = -EBUSY;
3858 	mutex_lock(&po->pg_vec_lock);
3859 	if (closing || atomic_read(&po->mapped) == 0) {
3860 		err = 0;
3861 		spin_lock_bh(&rb_queue->lock);
3862 		swap(rb->pg_vec, pg_vec);
3863 		rb->frame_max = (req->tp_frame_nr - 1);
3864 		rb->head = 0;
3865 		rb->frame_size = req->tp_frame_size;
3866 		spin_unlock_bh(&rb_queue->lock);
3867 
3868 		swap(rb->pg_vec_order, order);
3869 		swap(rb->pg_vec_len, req->tp_block_nr);
3870 
3871 		rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3872 		po->prot_hook.func = (po->rx_ring.pg_vec) ?
3873 						tpacket_rcv : packet_rcv;
3874 		skb_queue_purge(rb_queue);
3875 		if (atomic_read(&po->mapped))
3876 			pr_err("packet_mmap: vma is busy: %d\n",
3877 			       atomic_read(&po->mapped));
3878 	}
3879 	mutex_unlock(&po->pg_vec_lock);
3880 
3881 	spin_lock(&po->bind_lock);
3882 	if (was_running) {
3883 		po->num = num;
3884 		register_prot_hook(sk);
3885 	}
3886 	spin_unlock(&po->bind_lock);
3887 	if (closing && (po->tp_version > TPACKET_V2)) {
3888 		/* Because we don't support block-based V3 on tx-ring */
3889 		if (!tx_ring)
3890 			prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3891 	}
3892 	release_sock(sk);
3893 
3894 	if (pg_vec)
3895 		free_pg_vec(pg_vec, order, req->tp_block_nr);
3896 out:
3897 	return err;
3898 }
3899 
3900 static int packet_mmap(struct file *file, struct socket *sock,
3901 		struct vm_area_struct *vma)
3902 {
3903 	struct sock *sk = sock->sk;
3904 	struct packet_sock *po = pkt_sk(sk);
3905 	unsigned long size, expected_size;
3906 	struct packet_ring_buffer *rb;
3907 	unsigned long start;
3908 	int err = -EINVAL;
3909 	int i;
3910 
3911 	if (vma->vm_pgoff)
3912 		return -EINVAL;
3913 
3914 	mutex_lock(&po->pg_vec_lock);
3915 
3916 	expected_size = 0;
3917 	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3918 		if (rb->pg_vec) {
3919 			expected_size += rb->pg_vec_len
3920 						* rb->pg_vec_pages
3921 						* PAGE_SIZE;
3922 		}
3923 	}
3924 
3925 	if (expected_size == 0)
3926 		goto out;
3927 
3928 	size = vma->vm_end - vma->vm_start;
3929 	if (size != expected_size)
3930 		goto out;
3931 
3932 	start = vma->vm_start;
3933 	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3934 		if (rb->pg_vec == NULL)
3935 			continue;
3936 
3937 		for (i = 0; i < rb->pg_vec_len; i++) {
3938 			struct page *page;
3939 			void *kaddr = rb->pg_vec[i].buffer;
3940 			int pg_num;
3941 
3942 			for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3943 				page = pgv_to_page(kaddr);
3944 				err = vm_insert_page(vma, start, page);
3945 				if (unlikely(err))
3946 					goto out;
3947 				start += PAGE_SIZE;
3948 				kaddr += PAGE_SIZE;
3949 			}
3950 		}
3951 	}
3952 
3953 	atomic_inc(&po->mapped);
3954 	vma->vm_ops = &packet_mmap_ops;
3955 	err = 0;
3956 
3957 out:
3958 	mutex_unlock(&po->pg_vec_lock);
3959 	return err;
3960 }
3961 
3962 static const struct proto_ops packet_ops_spkt = {
3963 	.family =	PF_PACKET,
3964 	.owner =	THIS_MODULE,
3965 	.release =	packet_release,
3966 	.bind =		packet_bind_spkt,
3967 	.connect =	sock_no_connect,
3968 	.socketpair =	sock_no_socketpair,
3969 	.accept =	sock_no_accept,
3970 	.getname =	packet_getname_spkt,
3971 	.poll =		datagram_poll,
3972 	.ioctl =	packet_ioctl,
3973 	.listen =	sock_no_listen,
3974 	.shutdown =	sock_no_shutdown,
3975 	.setsockopt =	sock_no_setsockopt,
3976 	.getsockopt =	sock_no_getsockopt,
3977 	.sendmsg =	packet_sendmsg_spkt,
3978 	.recvmsg =	packet_recvmsg,
3979 	.mmap =		sock_no_mmap,
3980 	.sendpage =	sock_no_sendpage,
3981 };
3982 
3983 static const struct proto_ops packet_ops = {
3984 	.family =	PF_PACKET,
3985 	.owner =	THIS_MODULE,
3986 	.release =	packet_release,
3987 	.bind =		packet_bind,
3988 	.connect =	sock_no_connect,
3989 	.socketpair =	sock_no_socketpair,
3990 	.accept =	sock_no_accept,
3991 	.getname =	packet_getname,
3992 	.poll =		packet_poll,
3993 	.ioctl =	packet_ioctl,
3994 	.listen =	sock_no_listen,
3995 	.shutdown =	sock_no_shutdown,
3996 	.setsockopt =	packet_setsockopt,
3997 	.getsockopt =	packet_getsockopt,
3998 	.sendmsg =	packet_sendmsg,
3999 	.recvmsg =	packet_recvmsg,
4000 	.mmap =		packet_mmap,
4001 	.sendpage =	sock_no_sendpage,
4002 };
4003 
4004 static const struct net_proto_family packet_family_ops = {
4005 	.family =	PF_PACKET,
4006 	.create =	packet_create,
4007 	.owner	=	THIS_MODULE,
4008 };
4009 
4010 static struct notifier_block packet_netdev_notifier = {
4011 	.notifier_call =	packet_notifier,
4012 };
4013 
4014 #ifdef CONFIG_PROC_FS
4015 
4016 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4017 	__acquires(RCU)
4018 {
4019 	struct net *net = seq_file_net(seq);
4020 
4021 	rcu_read_lock();
4022 	return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4023 }
4024 
4025 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4026 {
4027 	struct net *net = seq_file_net(seq);
4028 	return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4029 }
4030 
4031 static void packet_seq_stop(struct seq_file *seq, void *v)
4032 	__releases(RCU)
4033 {
4034 	rcu_read_unlock();
4035 }
4036 
4037 static int packet_seq_show(struct seq_file *seq, void *v)
4038 {
4039 	if (v == SEQ_START_TOKEN)
4040 		seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
4041 	else {
4042 		struct sock *s = sk_entry(v);
4043 		const struct packet_sock *po = pkt_sk(s);
4044 
4045 		seq_printf(seq,
4046 			   "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4047 			   s,
4048 			   atomic_read(&s->sk_refcnt),
4049 			   s->sk_type,
4050 			   ntohs(po->num),
4051 			   po->ifindex,
4052 			   po->running,
4053 			   atomic_read(&s->sk_rmem_alloc),
4054 			   from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4055 			   sock_i_ino(s));
4056 	}
4057 
4058 	return 0;
4059 }
4060 
4061 static const struct seq_operations packet_seq_ops = {
4062 	.start	= packet_seq_start,
4063 	.next	= packet_seq_next,
4064 	.stop	= packet_seq_stop,
4065 	.show	= packet_seq_show,
4066 };
4067 
4068 static int packet_seq_open(struct inode *inode, struct file *file)
4069 {
4070 	return seq_open_net(inode, file, &packet_seq_ops,
4071 			    sizeof(struct seq_net_private));
4072 }
4073 
4074 static const struct file_operations packet_seq_fops = {
4075 	.owner		= THIS_MODULE,
4076 	.open		= packet_seq_open,
4077 	.read		= seq_read,
4078 	.llseek		= seq_lseek,
4079 	.release	= seq_release_net,
4080 };
4081 
4082 #endif
4083 
4084 static int __net_init packet_net_init(struct net *net)
4085 {
4086 	mutex_init(&net->packet.sklist_lock);
4087 	INIT_HLIST_HEAD(&net->packet.sklist);
4088 
4089 	if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4090 		return -ENOMEM;
4091 
4092 	return 0;
4093 }
4094 
4095 static void __net_exit packet_net_exit(struct net *net)
4096 {
4097 	remove_proc_entry("packet", net->proc_net);
4098 }
4099 
4100 static struct pernet_operations packet_net_ops = {
4101 	.init = packet_net_init,
4102 	.exit = packet_net_exit,
4103 };
4104 
4105 
4106 static void __exit packet_exit(void)
4107 {
4108 	unregister_netdevice_notifier(&packet_netdev_notifier);
4109 	unregister_pernet_subsys(&packet_net_ops);
4110 	sock_unregister(PF_PACKET);
4111 	proto_unregister(&packet_proto);
4112 }
4113 
4114 static int __init packet_init(void)
4115 {
4116 	int rc = proto_register(&packet_proto, 0);
4117 
4118 	if (rc != 0)
4119 		goto out;
4120 
4121 	sock_register(&packet_family_ops);
4122 	register_pernet_subsys(&packet_net_ops);
4123 	register_netdevice_notifier(&packet_netdev_notifier);
4124 out:
4125 	return rc;
4126 }
4127 
4128 module_init(packet_init);
4129 module_exit(packet_exit);
4130 MODULE_LICENSE("GPL");
4131 MODULE_ALIAS_NETPROTO(PF_PACKET);
4132