xref: /linux/net/packet/af_packet.c (revision 2ba9268dd603d23e17643437b2246acb6844953b)
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 			if (!BLOCK_NUM_PKTS(pbd)) {
702 				/* An empty block. Just refresh the timer. */
703 				goto refresh_timer;
704 			}
705 			prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
706 			if (!prb_dispatch_next_block(pkc, po))
707 				goto refresh_timer;
708 			else
709 				goto out;
710 		} else {
711 			/* Case 1. Queue was frozen because user-space was
712 			 *	   lagging behind.
713 			 */
714 			if (prb_curr_blk_in_use(pkc, pbd)) {
715 				/*
716 				 * Ok, user-space is still behind.
717 				 * So just refresh the timer.
718 				 */
719 				goto refresh_timer;
720 			} else {
721 			       /* Case 2. queue was frozen,user-space caught up,
722 				* now the link went idle && the timer fired.
723 				* We don't have a block to close.So we open this
724 				* block and restart the timer.
725 				* opening a block thaws the queue,restarts timer
726 				* Thawing/timer-refresh is a side effect.
727 				*/
728 				prb_open_block(pkc, pbd);
729 				goto out;
730 			}
731 		}
732 	}
733 
734 refresh_timer:
735 	_prb_refresh_rx_retire_blk_timer(pkc);
736 
737 out:
738 	spin_unlock(&po->sk.sk_receive_queue.lock);
739 }
740 
741 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
742 		struct tpacket_block_desc *pbd1, __u32 status)
743 {
744 	/* Flush everything minus the block header */
745 
746 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
747 	u8 *start, *end;
748 
749 	start = (u8 *)pbd1;
750 
751 	/* Skip the block header(we know header WILL fit in 4K) */
752 	start += PAGE_SIZE;
753 
754 	end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
755 	for (; start < end; start += PAGE_SIZE)
756 		flush_dcache_page(pgv_to_page(start));
757 
758 	smp_wmb();
759 #endif
760 
761 	/* Now update the block status. */
762 
763 	BLOCK_STATUS(pbd1) = status;
764 
765 	/* Flush the block header */
766 
767 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
768 	start = (u8 *)pbd1;
769 	flush_dcache_page(pgv_to_page(start));
770 
771 	smp_wmb();
772 #endif
773 }
774 
775 /*
776  * Side effect:
777  *
778  * 1) flush the block
779  * 2) Increment active_blk_num
780  *
781  * Note:We DONT refresh the timer on purpose.
782  *	Because almost always the next block will be opened.
783  */
784 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
785 		struct tpacket_block_desc *pbd1,
786 		struct packet_sock *po, unsigned int stat)
787 {
788 	__u32 status = TP_STATUS_USER | stat;
789 
790 	struct tpacket3_hdr *last_pkt;
791 	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
792 	struct sock *sk = &po->sk;
793 
794 	if (po->stats.stats3.tp_drops)
795 		status |= TP_STATUS_LOSING;
796 
797 	last_pkt = (struct tpacket3_hdr *)pkc1->prev;
798 	last_pkt->tp_next_offset = 0;
799 
800 	/* Get the ts of the last pkt */
801 	if (BLOCK_NUM_PKTS(pbd1)) {
802 		h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
803 		h1->ts_last_pkt.ts_nsec	= last_pkt->tp_nsec;
804 	} else {
805 		/* Ok, we tmo'd - so get the current time.
806 		 *
807 		 * It shouldn't really happen as we don't close empty
808 		 * blocks. See prb_retire_rx_blk_timer_expired().
809 		 */
810 		struct timespec ts;
811 		getnstimeofday(&ts);
812 		h1->ts_last_pkt.ts_sec = ts.tv_sec;
813 		h1->ts_last_pkt.ts_nsec	= ts.tv_nsec;
814 	}
815 
816 	smp_wmb();
817 
818 	/* Flush the block */
819 	prb_flush_block(pkc1, pbd1, status);
820 
821 	sk->sk_data_ready(sk);
822 
823 	pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
824 }
825 
826 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
827 {
828 	pkc->reset_pending_on_curr_blk = 0;
829 }
830 
831 /*
832  * Side effect of opening a block:
833  *
834  * 1) prb_queue is thawed.
835  * 2) retire_blk_timer is refreshed.
836  *
837  */
838 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
839 	struct tpacket_block_desc *pbd1)
840 {
841 	struct timespec ts;
842 	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
843 
844 	smp_rmb();
845 
846 	/* We could have just memset this but we will lose the
847 	 * flexibility of making the priv area sticky
848 	 */
849 
850 	BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
851 	BLOCK_NUM_PKTS(pbd1) = 0;
852 	BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
853 
854 	getnstimeofday(&ts);
855 
856 	h1->ts_first_pkt.ts_sec = ts.tv_sec;
857 	h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
858 
859 	pkc1->pkblk_start = (char *)pbd1;
860 	pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
861 
862 	BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
863 	BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
864 
865 	pbd1->version = pkc1->version;
866 	pkc1->prev = pkc1->nxt_offset;
867 	pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
868 
869 	prb_thaw_queue(pkc1);
870 	_prb_refresh_rx_retire_blk_timer(pkc1);
871 
872 	smp_wmb();
873 }
874 
875 /*
876  * Queue freeze logic:
877  * 1) Assume tp_block_nr = 8 blocks.
878  * 2) At time 't0', user opens Rx ring.
879  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
880  * 4) user-space is either sleeping or processing block '0'.
881  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
882  *    it will close block-7,loop around and try to fill block '0'.
883  *    call-flow:
884  *    __packet_lookup_frame_in_block
885  *      prb_retire_current_block()
886  *      prb_dispatch_next_block()
887  *        |->(BLOCK_STATUS == USER) evaluates to true
888  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
889  * 6) Now there are two cases:
890  *    6.1) Link goes idle right after the queue is frozen.
891  *         But remember, the last open_block() refreshed the timer.
892  *         When this timer expires,it will refresh itself so that we can
893  *         re-open block-0 in near future.
894  *    6.2) Link is busy and keeps on receiving packets. This is a simple
895  *         case and __packet_lookup_frame_in_block will check if block-0
896  *         is free and can now be re-used.
897  */
898 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
899 				  struct packet_sock *po)
900 {
901 	pkc->reset_pending_on_curr_blk = 1;
902 	po->stats.stats3.tp_freeze_q_cnt++;
903 }
904 
905 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
906 
907 /*
908  * If the next block is free then we will dispatch it
909  * and return a good offset.
910  * Else, we will freeze the queue.
911  * So, caller must check the return value.
912  */
913 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
914 		struct packet_sock *po)
915 {
916 	struct tpacket_block_desc *pbd;
917 
918 	smp_rmb();
919 
920 	/* 1. Get current block num */
921 	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
922 
923 	/* 2. If this block is currently in_use then freeze the queue */
924 	if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
925 		prb_freeze_queue(pkc, po);
926 		return NULL;
927 	}
928 
929 	/*
930 	 * 3.
931 	 * open this block and return the offset where the first packet
932 	 * needs to get stored.
933 	 */
934 	prb_open_block(pkc, pbd);
935 	return (void *)pkc->nxt_offset;
936 }
937 
938 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
939 		struct packet_sock *po, unsigned int status)
940 {
941 	struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
942 
943 	/* retire/close the current block */
944 	if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
945 		/*
946 		 * Plug the case where copy_bits() is in progress on
947 		 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
948 		 * have space to copy the pkt in the current block and
949 		 * called prb_retire_current_block()
950 		 *
951 		 * We don't need to worry about the TMO case because
952 		 * the timer-handler already handled this case.
953 		 */
954 		if (!(status & TP_STATUS_BLK_TMO)) {
955 			while (atomic_read(&pkc->blk_fill_in_prog)) {
956 				/* Waiting for skb_copy_bits to finish... */
957 				cpu_relax();
958 			}
959 		}
960 		prb_close_block(pkc, pbd, po, status);
961 		return;
962 	}
963 }
964 
965 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
966 				      struct tpacket_block_desc *pbd)
967 {
968 	return TP_STATUS_USER & BLOCK_STATUS(pbd);
969 }
970 
971 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
972 {
973 	return pkc->reset_pending_on_curr_blk;
974 }
975 
976 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
977 {
978 	struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
979 	atomic_dec(&pkc->blk_fill_in_prog);
980 }
981 
982 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
983 			struct tpacket3_hdr *ppd)
984 {
985 	ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
986 }
987 
988 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
989 			struct tpacket3_hdr *ppd)
990 {
991 	ppd->hv1.tp_rxhash = 0;
992 }
993 
994 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
995 			struct tpacket3_hdr *ppd)
996 {
997 	if (skb_vlan_tag_present(pkc->skb)) {
998 		ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
999 		ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1000 		ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1001 	} else {
1002 		ppd->hv1.tp_vlan_tci = 0;
1003 		ppd->hv1.tp_vlan_tpid = 0;
1004 		ppd->tp_status = TP_STATUS_AVAILABLE;
1005 	}
1006 }
1007 
1008 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1009 			struct tpacket3_hdr *ppd)
1010 {
1011 	ppd->hv1.tp_padding = 0;
1012 	prb_fill_vlan_info(pkc, ppd);
1013 
1014 	if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1015 		prb_fill_rxhash(pkc, ppd);
1016 	else
1017 		prb_clear_rxhash(pkc, ppd);
1018 }
1019 
1020 static void prb_fill_curr_block(char *curr,
1021 				struct tpacket_kbdq_core *pkc,
1022 				struct tpacket_block_desc *pbd,
1023 				unsigned int len)
1024 {
1025 	struct tpacket3_hdr *ppd;
1026 
1027 	ppd  = (struct tpacket3_hdr *)curr;
1028 	ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1029 	pkc->prev = curr;
1030 	pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1031 	BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1032 	BLOCK_NUM_PKTS(pbd) += 1;
1033 	atomic_inc(&pkc->blk_fill_in_prog);
1034 	prb_run_all_ft_ops(pkc, ppd);
1035 }
1036 
1037 /* Assumes caller has the sk->rx_queue.lock */
1038 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1039 					    struct sk_buff *skb,
1040 						int status,
1041 					    unsigned int len
1042 					    )
1043 {
1044 	struct tpacket_kbdq_core *pkc;
1045 	struct tpacket_block_desc *pbd;
1046 	char *curr, *end;
1047 
1048 	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1049 	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1050 
1051 	/* Queue is frozen when user space is lagging behind */
1052 	if (prb_queue_frozen(pkc)) {
1053 		/*
1054 		 * Check if that last block which caused the queue to freeze,
1055 		 * is still in_use by user-space.
1056 		 */
1057 		if (prb_curr_blk_in_use(pkc, pbd)) {
1058 			/* Can't record this packet */
1059 			return NULL;
1060 		} else {
1061 			/*
1062 			 * Ok, the block was released by user-space.
1063 			 * Now let's open that block.
1064 			 * opening a block also thaws the queue.
1065 			 * Thawing is a side effect.
1066 			 */
1067 			prb_open_block(pkc, pbd);
1068 		}
1069 	}
1070 
1071 	smp_mb();
1072 	curr = pkc->nxt_offset;
1073 	pkc->skb = skb;
1074 	end = (char *)pbd + pkc->kblk_size;
1075 
1076 	/* first try the current block */
1077 	if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1078 		prb_fill_curr_block(curr, pkc, pbd, len);
1079 		return (void *)curr;
1080 	}
1081 
1082 	/* Ok, close the current block */
1083 	prb_retire_current_block(pkc, po, 0);
1084 
1085 	/* Now, try to dispatch the next block */
1086 	curr = (char *)prb_dispatch_next_block(pkc, po);
1087 	if (curr) {
1088 		pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1089 		prb_fill_curr_block(curr, pkc, pbd, len);
1090 		return (void *)curr;
1091 	}
1092 
1093 	/*
1094 	 * No free blocks are available.user_space hasn't caught up yet.
1095 	 * Queue was just frozen and now this packet will get dropped.
1096 	 */
1097 	return NULL;
1098 }
1099 
1100 static void *packet_current_rx_frame(struct packet_sock *po,
1101 					    struct sk_buff *skb,
1102 					    int status, unsigned int len)
1103 {
1104 	char *curr = NULL;
1105 	switch (po->tp_version) {
1106 	case TPACKET_V1:
1107 	case TPACKET_V2:
1108 		curr = packet_lookup_frame(po, &po->rx_ring,
1109 					po->rx_ring.head, status);
1110 		return curr;
1111 	case TPACKET_V3:
1112 		return __packet_lookup_frame_in_block(po, skb, status, len);
1113 	default:
1114 		WARN(1, "TPACKET version not supported\n");
1115 		BUG();
1116 		return NULL;
1117 	}
1118 }
1119 
1120 static void *prb_lookup_block(struct packet_sock *po,
1121 				     struct packet_ring_buffer *rb,
1122 				     unsigned int idx,
1123 				     int status)
1124 {
1125 	struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1126 	struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1127 
1128 	if (status != BLOCK_STATUS(pbd))
1129 		return NULL;
1130 	return pbd;
1131 }
1132 
1133 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1134 {
1135 	unsigned int prev;
1136 	if (rb->prb_bdqc.kactive_blk_num)
1137 		prev = rb->prb_bdqc.kactive_blk_num-1;
1138 	else
1139 		prev = rb->prb_bdqc.knum_blocks-1;
1140 	return prev;
1141 }
1142 
1143 /* Assumes caller has held the rx_queue.lock */
1144 static void *__prb_previous_block(struct packet_sock *po,
1145 					 struct packet_ring_buffer *rb,
1146 					 int status)
1147 {
1148 	unsigned int previous = prb_previous_blk_num(rb);
1149 	return prb_lookup_block(po, rb, previous, status);
1150 }
1151 
1152 static void *packet_previous_rx_frame(struct packet_sock *po,
1153 					     struct packet_ring_buffer *rb,
1154 					     int status)
1155 {
1156 	if (po->tp_version <= TPACKET_V2)
1157 		return packet_previous_frame(po, rb, status);
1158 
1159 	return __prb_previous_block(po, rb, status);
1160 }
1161 
1162 static void packet_increment_rx_head(struct packet_sock *po,
1163 					    struct packet_ring_buffer *rb)
1164 {
1165 	switch (po->tp_version) {
1166 	case TPACKET_V1:
1167 	case TPACKET_V2:
1168 		return packet_increment_head(rb);
1169 	case TPACKET_V3:
1170 	default:
1171 		WARN(1, "TPACKET version not supported.\n");
1172 		BUG();
1173 		return;
1174 	}
1175 }
1176 
1177 static void *packet_previous_frame(struct packet_sock *po,
1178 		struct packet_ring_buffer *rb,
1179 		int status)
1180 {
1181 	unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1182 	return packet_lookup_frame(po, rb, previous, status);
1183 }
1184 
1185 static void packet_increment_head(struct packet_ring_buffer *buff)
1186 {
1187 	buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1188 }
1189 
1190 static void packet_inc_pending(struct packet_ring_buffer *rb)
1191 {
1192 	this_cpu_inc(*rb->pending_refcnt);
1193 }
1194 
1195 static void packet_dec_pending(struct packet_ring_buffer *rb)
1196 {
1197 	this_cpu_dec(*rb->pending_refcnt);
1198 }
1199 
1200 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1201 {
1202 	unsigned int refcnt = 0;
1203 	int cpu;
1204 
1205 	/* We don't use pending refcount in rx_ring. */
1206 	if (rb->pending_refcnt == NULL)
1207 		return 0;
1208 
1209 	for_each_possible_cpu(cpu)
1210 		refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1211 
1212 	return refcnt;
1213 }
1214 
1215 static int packet_alloc_pending(struct packet_sock *po)
1216 {
1217 	po->rx_ring.pending_refcnt = NULL;
1218 
1219 	po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1220 	if (unlikely(po->tx_ring.pending_refcnt == NULL))
1221 		return -ENOBUFS;
1222 
1223 	return 0;
1224 }
1225 
1226 static void packet_free_pending(struct packet_sock *po)
1227 {
1228 	free_percpu(po->tx_ring.pending_refcnt);
1229 }
1230 
1231 static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1232 {
1233 	struct sock *sk = &po->sk;
1234 	bool has_room;
1235 
1236 	if (po->prot_hook.func != tpacket_rcv)
1237 		return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1238 			<= sk->sk_rcvbuf;
1239 
1240 	spin_lock(&sk->sk_receive_queue.lock);
1241 	if (po->tp_version == TPACKET_V3)
1242 		has_room = prb_lookup_block(po, &po->rx_ring,
1243 					    po->rx_ring.prb_bdqc.kactive_blk_num,
1244 					    TP_STATUS_KERNEL);
1245 	else
1246 		has_room = packet_lookup_frame(po, &po->rx_ring,
1247 					       po->rx_ring.head,
1248 					       TP_STATUS_KERNEL);
1249 	spin_unlock(&sk->sk_receive_queue.lock);
1250 
1251 	return has_room;
1252 }
1253 
1254 static void packet_sock_destruct(struct sock *sk)
1255 {
1256 	skb_queue_purge(&sk->sk_error_queue);
1257 
1258 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1259 	WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1260 
1261 	if (!sock_flag(sk, SOCK_DEAD)) {
1262 		pr_err("Attempt to release alive packet socket: %p\n", sk);
1263 		return;
1264 	}
1265 
1266 	sk_refcnt_debug_dec(sk);
1267 }
1268 
1269 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1270 {
1271 	int x = atomic_read(&f->rr_cur) + 1;
1272 
1273 	if (x >= num)
1274 		x = 0;
1275 
1276 	return x;
1277 }
1278 
1279 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1280 				      struct sk_buff *skb,
1281 				      unsigned int num)
1282 {
1283 	return reciprocal_scale(skb_get_hash(skb), num);
1284 }
1285 
1286 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1287 				    struct sk_buff *skb,
1288 				    unsigned int num)
1289 {
1290 	int cur, old;
1291 
1292 	cur = atomic_read(&f->rr_cur);
1293 	while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1294 				     fanout_rr_next(f, num))) != cur)
1295 		cur = old;
1296 	return cur;
1297 }
1298 
1299 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1300 				     struct sk_buff *skb,
1301 				     unsigned int num)
1302 {
1303 	return smp_processor_id() % num;
1304 }
1305 
1306 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1307 				     struct sk_buff *skb,
1308 				     unsigned int num)
1309 {
1310 	return prandom_u32_max(num);
1311 }
1312 
1313 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1314 					  struct sk_buff *skb,
1315 					  unsigned int idx, unsigned int skip,
1316 					  unsigned int num)
1317 {
1318 	unsigned int i, j;
1319 
1320 	i = j = min_t(int, f->next[idx], num - 1);
1321 	do {
1322 		if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1323 			if (i != j)
1324 				f->next[idx] = i;
1325 			return i;
1326 		}
1327 		if (++i == num)
1328 			i = 0;
1329 	} while (i != j);
1330 
1331 	return idx;
1332 }
1333 
1334 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1335 				    struct sk_buff *skb,
1336 				    unsigned int num)
1337 {
1338 	return skb_get_queue_mapping(skb) % num;
1339 }
1340 
1341 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1342 {
1343 	return f->flags & (flag >> 8);
1344 }
1345 
1346 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1347 			     struct packet_type *pt, struct net_device *orig_dev)
1348 {
1349 	struct packet_fanout *f = pt->af_packet_priv;
1350 	unsigned int num = f->num_members;
1351 	struct packet_sock *po;
1352 	unsigned int idx;
1353 
1354 	if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1355 	    !num) {
1356 		kfree_skb(skb);
1357 		return 0;
1358 	}
1359 
1360 	if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1361 		skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1362 		if (!skb)
1363 			return 0;
1364 	}
1365 	switch (f->type) {
1366 	case PACKET_FANOUT_HASH:
1367 	default:
1368 		idx = fanout_demux_hash(f, skb, num);
1369 		break;
1370 	case PACKET_FANOUT_LB:
1371 		idx = fanout_demux_lb(f, skb, num);
1372 		break;
1373 	case PACKET_FANOUT_CPU:
1374 		idx = fanout_demux_cpu(f, skb, num);
1375 		break;
1376 	case PACKET_FANOUT_RND:
1377 		idx = fanout_demux_rnd(f, skb, num);
1378 		break;
1379 	case PACKET_FANOUT_QM:
1380 		idx = fanout_demux_qm(f, skb, num);
1381 		break;
1382 	case PACKET_FANOUT_ROLLOVER:
1383 		idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1384 		break;
1385 	}
1386 
1387 	po = pkt_sk(f->arr[idx]);
1388 	if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1389 	    unlikely(!packet_rcv_has_room(po, skb))) {
1390 		idx = fanout_demux_rollover(f, skb, idx, idx, num);
1391 		po = pkt_sk(f->arr[idx]);
1392 	}
1393 
1394 	return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1395 }
1396 
1397 DEFINE_MUTEX(fanout_mutex);
1398 EXPORT_SYMBOL_GPL(fanout_mutex);
1399 static LIST_HEAD(fanout_list);
1400 
1401 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1402 {
1403 	struct packet_fanout *f = po->fanout;
1404 
1405 	spin_lock(&f->lock);
1406 	f->arr[f->num_members] = sk;
1407 	smp_wmb();
1408 	f->num_members++;
1409 	spin_unlock(&f->lock);
1410 }
1411 
1412 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1413 {
1414 	struct packet_fanout *f = po->fanout;
1415 	int i;
1416 
1417 	spin_lock(&f->lock);
1418 	for (i = 0; i < f->num_members; i++) {
1419 		if (f->arr[i] == sk)
1420 			break;
1421 	}
1422 	BUG_ON(i >= f->num_members);
1423 	f->arr[i] = f->arr[f->num_members - 1];
1424 	f->num_members--;
1425 	spin_unlock(&f->lock);
1426 }
1427 
1428 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1429 {
1430 	if (ptype->af_packet_priv == (void *)((struct packet_sock *)sk)->fanout)
1431 		return true;
1432 
1433 	return false;
1434 }
1435 
1436 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1437 {
1438 	struct packet_sock *po = pkt_sk(sk);
1439 	struct packet_fanout *f, *match;
1440 	u8 type = type_flags & 0xff;
1441 	u8 flags = type_flags >> 8;
1442 	int err;
1443 
1444 	switch (type) {
1445 	case PACKET_FANOUT_ROLLOVER:
1446 		if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1447 			return -EINVAL;
1448 	case PACKET_FANOUT_HASH:
1449 	case PACKET_FANOUT_LB:
1450 	case PACKET_FANOUT_CPU:
1451 	case PACKET_FANOUT_RND:
1452 	case PACKET_FANOUT_QM:
1453 		break;
1454 	default:
1455 		return -EINVAL;
1456 	}
1457 
1458 	if (!po->running)
1459 		return -EINVAL;
1460 
1461 	if (po->fanout)
1462 		return -EALREADY;
1463 
1464 	mutex_lock(&fanout_mutex);
1465 	match = NULL;
1466 	list_for_each_entry(f, &fanout_list, list) {
1467 		if (f->id == id &&
1468 		    read_pnet(&f->net) == sock_net(sk)) {
1469 			match = f;
1470 			break;
1471 		}
1472 	}
1473 	err = -EINVAL;
1474 	if (match && match->flags != flags)
1475 		goto out;
1476 	if (!match) {
1477 		err = -ENOMEM;
1478 		match = kzalloc(sizeof(*match), GFP_KERNEL);
1479 		if (!match)
1480 			goto out;
1481 		write_pnet(&match->net, sock_net(sk));
1482 		match->id = id;
1483 		match->type = type;
1484 		match->flags = flags;
1485 		atomic_set(&match->rr_cur, 0);
1486 		INIT_LIST_HEAD(&match->list);
1487 		spin_lock_init(&match->lock);
1488 		atomic_set(&match->sk_ref, 0);
1489 		match->prot_hook.type = po->prot_hook.type;
1490 		match->prot_hook.dev = po->prot_hook.dev;
1491 		match->prot_hook.func = packet_rcv_fanout;
1492 		match->prot_hook.af_packet_priv = match;
1493 		match->prot_hook.id_match = match_fanout_group;
1494 		dev_add_pack(&match->prot_hook);
1495 		list_add(&match->list, &fanout_list);
1496 	}
1497 	err = -EINVAL;
1498 	if (match->type == type &&
1499 	    match->prot_hook.type == po->prot_hook.type &&
1500 	    match->prot_hook.dev == po->prot_hook.dev) {
1501 		err = -ENOSPC;
1502 		if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1503 			__dev_remove_pack(&po->prot_hook);
1504 			po->fanout = match;
1505 			atomic_inc(&match->sk_ref);
1506 			__fanout_link(sk, po);
1507 			err = 0;
1508 		}
1509 	}
1510 out:
1511 	mutex_unlock(&fanout_mutex);
1512 	return err;
1513 }
1514 
1515 static void fanout_release(struct sock *sk)
1516 {
1517 	struct packet_sock *po = pkt_sk(sk);
1518 	struct packet_fanout *f;
1519 
1520 	f = po->fanout;
1521 	if (!f)
1522 		return;
1523 
1524 	mutex_lock(&fanout_mutex);
1525 	po->fanout = NULL;
1526 
1527 	if (atomic_dec_and_test(&f->sk_ref)) {
1528 		list_del(&f->list);
1529 		dev_remove_pack(&f->prot_hook);
1530 		kfree(f);
1531 	}
1532 	mutex_unlock(&fanout_mutex);
1533 }
1534 
1535 static const struct proto_ops packet_ops;
1536 
1537 static const struct proto_ops packet_ops_spkt;
1538 
1539 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1540 			   struct packet_type *pt, struct net_device *orig_dev)
1541 {
1542 	struct sock *sk;
1543 	struct sockaddr_pkt *spkt;
1544 
1545 	/*
1546 	 *	When we registered the protocol we saved the socket in the data
1547 	 *	field for just this event.
1548 	 */
1549 
1550 	sk = pt->af_packet_priv;
1551 
1552 	/*
1553 	 *	Yank back the headers [hope the device set this
1554 	 *	right or kerboom...]
1555 	 *
1556 	 *	Incoming packets have ll header pulled,
1557 	 *	push it back.
1558 	 *
1559 	 *	For outgoing ones skb->data == skb_mac_header(skb)
1560 	 *	so that this procedure is noop.
1561 	 */
1562 
1563 	if (skb->pkt_type == PACKET_LOOPBACK)
1564 		goto out;
1565 
1566 	if (!net_eq(dev_net(dev), sock_net(sk)))
1567 		goto out;
1568 
1569 	skb = skb_share_check(skb, GFP_ATOMIC);
1570 	if (skb == NULL)
1571 		goto oom;
1572 
1573 	/* drop any routing info */
1574 	skb_dst_drop(skb);
1575 
1576 	/* drop conntrack reference */
1577 	nf_reset(skb);
1578 
1579 	spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1580 
1581 	skb_push(skb, skb->data - skb_mac_header(skb));
1582 
1583 	/*
1584 	 *	The SOCK_PACKET socket receives _all_ frames.
1585 	 */
1586 
1587 	spkt->spkt_family = dev->type;
1588 	strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1589 	spkt->spkt_protocol = skb->protocol;
1590 
1591 	/*
1592 	 *	Charge the memory to the socket. This is done specifically
1593 	 *	to prevent sockets using all the memory up.
1594 	 */
1595 
1596 	if (sock_queue_rcv_skb(sk, skb) == 0)
1597 		return 0;
1598 
1599 out:
1600 	kfree_skb(skb);
1601 oom:
1602 	return 0;
1603 }
1604 
1605 
1606 /*
1607  *	Output a raw packet to a device layer. This bypasses all the other
1608  *	protocol layers and you must therefore supply it with a complete frame
1609  */
1610 
1611 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1612 			       struct msghdr *msg, size_t len)
1613 {
1614 	struct sock *sk = sock->sk;
1615 	DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1616 	struct sk_buff *skb = NULL;
1617 	struct net_device *dev;
1618 	__be16 proto = 0;
1619 	int err;
1620 	int extra_len = 0;
1621 
1622 	/*
1623 	 *	Get and verify the address.
1624 	 */
1625 
1626 	if (saddr) {
1627 		if (msg->msg_namelen < sizeof(struct sockaddr))
1628 			return -EINVAL;
1629 		if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1630 			proto = saddr->spkt_protocol;
1631 	} else
1632 		return -ENOTCONN;	/* SOCK_PACKET must be sent giving an address */
1633 
1634 	/*
1635 	 *	Find the device first to size check it
1636 	 */
1637 
1638 	saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1639 retry:
1640 	rcu_read_lock();
1641 	dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1642 	err = -ENODEV;
1643 	if (dev == NULL)
1644 		goto out_unlock;
1645 
1646 	err = -ENETDOWN;
1647 	if (!(dev->flags & IFF_UP))
1648 		goto out_unlock;
1649 
1650 	/*
1651 	 * You may not queue a frame bigger than the mtu. This is the lowest level
1652 	 * raw protocol and you must do your own fragmentation at this level.
1653 	 */
1654 
1655 	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1656 		if (!netif_supports_nofcs(dev)) {
1657 			err = -EPROTONOSUPPORT;
1658 			goto out_unlock;
1659 		}
1660 		extra_len = 4; /* We're doing our own CRC */
1661 	}
1662 
1663 	err = -EMSGSIZE;
1664 	if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1665 		goto out_unlock;
1666 
1667 	if (!skb) {
1668 		size_t reserved = LL_RESERVED_SPACE(dev);
1669 		int tlen = dev->needed_tailroom;
1670 		unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1671 
1672 		rcu_read_unlock();
1673 		skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1674 		if (skb == NULL)
1675 			return -ENOBUFS;
1676 		/* FIXME: Save some space for broken drivers that write a hard
1677 		 * header at transmission time by themselves. PPP is the notable
1678 		 * one here. This should really be fixed at the driver level.
1679 		 */
1680 		skb_reserve(skb, reserved);
1681 		skb_reset_network_header(skb);
1682 
1683 		/* Try to align data part correctly */
1684 		if (hhlen) {
1685 			skb->data -= hhlen;
1686 			skb->tail -= hhlen;
1687 			if (len < hhlen)
1688 				skb_reset_network_header(skb);
1689 		}
1690 		err = memcpy_from_msg(skb_put(skb, len), msg, len);
1691 		if (err)
1692 			goto out_free;
1693 		goto retry;
1694 	}
1695 
1696 	if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1697 		/* Earlier code assumed this would be a VLAN pkt,
1698 		 * double-check this now that we have the actual
1699 		 * packet in hand.
1700 		 */
1701 		struct ethhdr *ehdr;
1702 		skb_reset_mac_header(skb);
1703 		ehdr = eth_hdr(skb);
1704 		if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1705 			err = -EMSGSIZE;
1706 			goto out_unlock;
1707 		}
1708 	}
1709 
1710 	skb->protocol = proto;
1711 	skb->dev = dev;
1712 	skb->priority = sk->sk_priority;
1713 	skb->mark = sk->sk_mark;
1714 
1715 	sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1716 
1717 	if (unlikely(extra_len == 4))
1718 		skb->no_fcs = 1;
1719 
1720 	skb_probe_transport_header(skb, 0);
1721 
1722 	dev_queue_xmit(skb);
1723 	rcu_read_unlock();
1724 	return len;
1725 
1726 out_unlock:
1727 	rcu_read_unlock();
1728 out_free:
1729 	kfree_skb(skb);
1730 	return err;
1731 }
1732 
1733 static unsigned int run_filter(const struct sk_buff *skb,
1734 				      const struct sock *sk,
1735 				      unsigned int res)
1736 {
1737 	struct sk_filter *filter;
1738 
1739 	rcu_read_lock();
1740 	filter = rcu_dereference(sk->sk_filter);
1741 	if (filter != NULL)
1742 		res = SK_RUN_FILTER(filter, skb);
1743 	rcu_read_unlock();
1744 
1745 	return res;
1746 }
1747 
1748 /*
1749  * This function makes lazy skb cloning in hope that most of packets
1750  * are discarded by BPF.
1751  *
1752  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1753  * and skb->cb are mangled. It works because (and until) packets
1754  * falling here are owned by current CPU. Output packets are cloned
1755  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1756  * sequencially, so that if we return skb to original state on exit,
1757  * we will not harm anyone.
1758  */
1759 
1760 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1761 		      struct packet_type *pt, struct net_device *orig_dev)
1762 {
1763 	struct sock *sk;
1764 	struct sockaddr_ll *sll;
1765 	struct packet_sock *po;
1766 	u8 *skb_head = skb->data;
1767 	int skb_len = skb->len;
1768 	unsigned int snaplen, res;
1769 
1770 	if (skb->pkt_type == PACKET_LOOPBACK)
1771 		goto drop;
1772 
1773 	sk = pt->af_packet_priv;
1774 	po = pkt_sk(sk);
1775 
1776 	if (!net_eq(dev_net(dev), sock_net(sk)))
1777 		goto drop;
1778 
1779 	skb->dev = dev;
1780 
1781 	if (dev->header_ops) {
1782 		/* The device has an explicit notion of ll header,
1783 		 * exported to higher levels.
1784 		 *
1785 		 * Otherwise, the device hides details of its frame
1786 		 * structure, so that corresponding packet head is
1787 		 * never delivered to user.
1788 		 */
1789 		if (sk->sk_type != SOCK_DGRAM)
1790 			skb_push(skb, skb->data - skb_mac_header(skb));
1791 		else if (skb->pkt_type == PACKET_OUTGOING) {
1792 			/* Special case: outgoing packets have ll header at head */
1793 			skb_pull(skb, skb_network_offset(skb));
1794 		}
1795 	}
1796 
1797 	snaplen = skb->len;
1798 
1799 	res = run_filter(skb, sk, snaplen);
1800 	if (!res)
1801 		goto drop_n_restore;
1802 	if (snaplen > res)
1803 		snaplen = res;
1804 
1805 	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1806 		goto drop_n_acct;
1807 
1808 	if (skb_shared(skb)) {
1809 		struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1810 		if (nskb == NULL)
1811 			goto drop_n_acct;
1812 
1813 		if (skb_head != skb->data) {
1814 			skb->data = skb_head;
1815 			skb->len = skb_len;
1816 		}
1817 		consume_skb(skb);
1818 		skb = nskb;
1819 	}
1820 
1821 	BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1822 		     sizeof(skb->cb));
1823 
1824 	sll = &PACKET_SKB_CB(skb)->sa.ll;
1825 	sll->sll_family = AF_PACKET;
1826 	sll->sll_hatype = dev->type;
1827 	sll->sll_protocol = skb->protocol;
1828 	sll->sll_pkttype = skb->pkt_type;
1829 	if (unlikely(po->origdev))
1830 		sll->sll_ifindex = orig_dev->ifindex;
1831 	else
1832 		sll->sll_ifindex = dev->ifindex;
1833 
1834 	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1835 
1836 	PACKET_SKB_CB(skb)->origlen = skb->len;
1837 
1838 	if (pskb_trim(skb, snaplen))
1839 		goto drop_n_acct;
1840 
1841 	skb_set_owner_r(skb, sk);
1842 	skb->dev = NULL;
1843 	skb_dst_drop(skb);
1844 
1845 	/* drop conntrack reference */
1846 	nf_reset(skb);
1847 
1848 	spin_lock(&sk->sk_receive_queue.lock);
1849 	po->stats.stats1.tp_packets++;
1850 	skb->dropcount = atomic_read(&sk->sk_drops);
1851 	__skb_queue_tail(&sk->sk_receive_queue, skb);
1852 	spin_unlock(&sk->sk_receive_queue.lock);
1853 	sk->sk_data_ready(sk);
1854 	return 0;
1855 
1856 drop_n_acct:
1857 	spin_lock(&sk->sk_receive_queue.lock);
1858 	po->stats.stats1.tp_drops++;
1859 	atomic_inc(&sk->sk_drops);
1860 	spin_unlock(&sk->sk_receive_queue.lock);
1861 
1862 drop_n_restore:
1863 	if (skb_head != skb->data && skb_shared(skb)) {
1864 		skb->data = skb_head;
1865 		skb->len = skb_len;
1866 	}
1867 drop:
1868 	consume_skb(skb);
1869 	return 0;
1870 }
1871 
1872 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1873 		       struct packet_type *pt, struct net_device *orig_dev)
1874 {
1875 	struct sock *sk;
1876 	struct packet_sock *po;
1877 	struct sockaddr_ll *sll;
1878 	union tpacket_uhdr h;
1879 	u8 *skb_head = skb->data;
1880 	int skb_len = skb->len;
1881 	unsigned int snaplen, res;
1882 	unsigned long status = TP_STATUS_USER;
1883 	unsigned short macoff, netoff, hdrlen;
1884 	struct sk_buff *copy_skb = NULL;
1885 	struct timespec ts;
1886 	__u32 ts_status;
1887 
1888 	/* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
1889 	 * We may add members to them until current aligned size without forcing
1890 	 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
1891 	 */
1892 	BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
1893 	BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
1894 
1895 	if (skb->pkt_type == PACKET_LOOPBACK)
1896 		goto drop;
1897 
1898 	sk = pt->af_packet_priv;
1899 	po = pkt_sk(sk);
1900 
1901 	if (!net_eq(dev_net(dev), sock_net(sk)))
1902 		goto drop;
1903 
1904 	if (dev->header_ops) {
1905 		if (sk->sk_type != SOCK_DGRAM)
1906 			skb_push(skb, skb->data - skb_mac_header(skb));
1907 		else if (skb->pkt_type == PACKET_OUTGOING) {
1908 			/* Special case: outgoing packets have ll header at head */
1909 			skb_pull(skb, skb_network_offset(skb));
1910 		}
1911 	}
1912 
1913 	if (skb->ip_summed == CHECKSUM_PARTIAL)
1914 		status |= TP_STATUS_CSUMNOTREADY;
1915 
1916 	snaplen = skb->len;
1917 
1918 	res = run_filter(skb, sk, snaplen);
1919 	if (!res)
1920 		goto drop_n_restore;
1921 	if (snaplen > res)
1922 		snaplen = res;
1923 
1924 	if (sk->sk_type == SOCK_DGRAM) {
1925 		macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1926 				  po->tp_reserve;
1927 	} else {
1928 		unsigned int maclen = skb_network_offset(skb);
1929 		netoff = TPACKET_ALIGN(po->tp_hdrlen +
1930 				       (maclen < 16 ? 16 : maclen)) +
1931 			po->tp_reserve;
1932 		macoff = netoff - maclen;
1933 	}
1934 	if (po->tp_version <= TPACKET_V2) {
1935 		if (macoff + snaplen > po->rx_ring.frame_size) {
1936 			if (po->copy_thresh &&
1937 			    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1938 				if (skb_shared(skb)) {
1939 					copy_skb = skb_clone(skb, GFP_ATOMIC);
1940 				} else {
1941 					copy_skb = skb_get(skb);
1942 					skb_head = skb->data;
1943 				}
1944 				if (copy_skb)
1945 					skb_set_owner_r(copy_skb, sk);
1946 			}
1947 			snaplen = po->rx_ring.frame_size - macoff;
1948 			if ((int)snaplen < 0)
1949 				snaplen = 0;
1950 		}
1951 	} else if (unlikely(macoff + snaplen >
1952 			    GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
1953 		u32 nval;
1954 
1955 		nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
1956 		pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
1957 			    snaplen, nval, macoff);
1958 		snaplen = nval;
1959 		if (unlikely((int)snaplen < 0)) {
1960 			snaplen = 0;
1961 			macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
1962 		}
1963 	}
1964 	spin_lock(&sk->sk_receive_queue.lock);
1965 	h.raw = packet_current_rx_frame(po, skb,
1966 					TP_STATUS_KERNEL, (macoff+snaplen));
1967 	if (!h.raw)
1968 		goto ring_is_full;
1969 	if (po->tp_version <= TPACKET_V2) {
1970 		packet_increment_rx_head(po, &po->rx_ring);
1971 	/*
1972 	 * LOSING will be reported till you read the stats,
1973 	 * because it's COR - Clear On Read.
1974 	 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1975 	 * at packet level.
1976 	 */
1977 		if (po->stats.stats1.tp_drops)
1978 			status |= TP_STATUS_LOSING;
1979 	}
1980 	po->stats.stats1.tp_packets++;
1981 	if (copy_skb) {
1982 		status |= TP_STATUS_COPY;
1983 		__skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1984 	}
1985 	spin_unlock(&sk->sk_receive_queue.lock);
1986 
1987 	skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1988 
1989 	if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
1990 		getnstimeofday(&ts);
1991 
1992 	status |= ts_status;
1993 
1994 	switch (po->tp_version) {
1995 	case TPACKET_V1:
1996 		h.h1->tp_len = skb->len;
1997 		h.h1->tp_snaplen = snaplen;
1998 		h.h1->tp_mac = macoff;
1999 		h.h1->tp_net = netoff;
2000 		h.h1->tp_sec = ts.tv_sec;
2001 		h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2002 		hdrlen = sizeof(*h.h1);
2003 		break;
2004 	case TPACKET_V2:
2005 		h.h2->tp_len = skb->len;
2006 		h.h2->tp_snaplen = snaplen;
2007 		h.h2->tp_mac = macoff;
2008 		h.h2->tp_net = netoff;
2009 		h.h2->tp_sec = ts.tv_sec;
2010 		h.h2->tp_nsec = ts.tv_nsec;
2011 		if (skb_vlan_tag_present(skb)) {
2012 			h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2013 			h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2014 			status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2015 		} else {
2016 			h.h2->tp_vlan_tci = 0;
2017 			h.h2->tp_vlan_tpid = 0;
2018 		}
2019 		memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2020 		hdrlen = sizeof(*h.h2);
2021 		break;
2022 	case TPACKET_V3:
2023 		/* tp_nxt_offset,vlan are already populated above.
2024 		 * So DONT clear those fields here
2025 		 */
2026 		h.h3->tp_status |= status;
2027 		h.h3->tp_len = skb->len;
2028 		h.h3->tp_snaplen = snaplen;
2029 		h.h3->tp_mac = macoff;
2030 		h.h3->tp_net = netoff;
2031 		h.h3->tp_sec  = ts.tv_sec;
2032 		h.h3->tp_nsec = ts.tv_nsec;
2033 		memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2034 		hdrlen = sizeof(*h.h3);
2035 		break;
2036 	default:
2037 		BUG();
2038 	}
2039 
2040 	sll = h.raw + TPACKET_ALIGN(hdrlen);
2041 	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2042 	sll->sll_family = AF_PACKET;
2043 	sll->sll_hatype = dev->type;
2044 	sll->sll_protocol = skb->protocol;
2045 	sll->sll_pkttype = skb->pkt_type;
2046 	if (unlikely(po->origdev))
2047 		sll->sll_ifindex = orig_dev->ifindex;
2048 	else
2049 		sll->sll_ifindex = dev->ifindex;
2050 
2051 	smp_mb();
2052 
2053 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2054 	if (po->tp_version <= TPACKET_V2) {
2055 		u8 *start, *end;
2056 
2057 		end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2058 					macoff + snaplen);
2059 
2060 		for (start = h.raw; start < end; start += PAGE_SIZE)
2061 			flush_dcache_page(pgv_to_page(start));
2062 	}
2063 	smp_wmb();
2064 #endif
2065 
2066 	if (po->tp_version <= TPACKET_V2) {
2067 		__packet_set_status(po, h.raw, status);
2068 		sk->sk_data_ready(sk);
2069 	} else {
2070 		prb_clear_blk_fill_status(&po->rx_ring);
2071 	}
2072 
2073 drop_n_restore:
2074 	if (skb_head != skb->data && skb_shared(skb)) {
2075 		skb->data = skb_head;
2076 		skb->len = skb_len;
2077 	}
2078 drop:
2079 	kfree_skb(skb);
2080 	return 0;
2081 
2082 ring_is_full:
2083 	po->stats.stats1.tp_drops++;
2084 	spin_unlock(&sk->sk_receive_queue.lock);
2085 
2086 	sk->sk_data_ready(sk);
2087 	kfree_skb(copy_skb);
2088 	goto drop_n_restore;
2089 }
2090 
2091 static void tpacket_destruct_skb(struct sk_buff *skb)
2092 {
2093 	struct packet_sock *po = pkt_sk(skb->sk);
2094 
2095 	if (likely(po->tx_ring.pg_vec)) {
2096 		void *ph;
2097 		__u32 ts;
2098 
2099 		ph = skb_shinfo(skb)->destructor_arg;
2100 		packet_dec_pending(&po->tx_ring);
2101 
2102 		ts = __packet_set_timestamp(po, ph, skb);
2103 		__packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2104 	}
2105 
2106 	sock_wfree(skb);
2107 }
2108 
2109 static bool ll_header_truncated(const struct net_device *dev, int len)
2110 {
2111 	/* net device doesn't like empty head */
2112 	if (unlikely(len <= dev->hard_header_len)) {
2113 		net_warn_ratelimited("%s: packet size is too short (%d <= %d)\n",
2114 				     current->comm, len, dev->hard_header_len);
2115 		return true;
2116 	}
2117 
2118 	return false;
2119 }
2120 
2121 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2122 		void *frame, struct net_device *dev, int size_max,
2123 		__be16 proto, unsigned char *addr, int hlen)
2124 {
2125 	union tpacket_uhdr ph;
2126 	int to_write, offset, len, tp_len, nr_frags, len_max;
2127 	struct socket *sock = po->sk.sk_socket;
2128 	struct page *page;
2129 	void *data;
2130 	int err;
2131 
2132 	ph.raw = frame;
2133 
2134 	skb->protocol = proto;
2135 	skb->dev = dev;
2136 	skb->priority = po->sk.sk_priority;
2137 	skb->mark = po->sk.sk_mark;
2138 	sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2139 	skb_shinfo(skb)->destructor_arg = ph.raw;
2140 
2141 	switch (po->tp_version) {
2142 	case TPACKET_V2:
2143 		tp_len = ph.h2->tp_len;
2144 		break;
2145 	default:
2146 		tp_len = ph.h1->tp_len;
2147 		break;
2148 	}
2149 	if (unlikely(tp_len > size_max)) {
2150 		pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2151 		return -EMSGSIZE;
2152 	}
2153 
2154 	skb_reserve(skb, hlen);
2155 	skb_reset_network_header(skb);
2156 
2157 	if (!packet_use_direct_xmit(po))
2158 		skb_probe_transport_header(skb, 0);
2159 	if (unlikely(po->tp_tx_has_off)) {
2160 		int off_min, off_max, off;
2161 		off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2162 		off_max = po->tx_ring.frame_size - tp_len;
2163 		if (sock->type == SOCK_DGRAM) {
2164 			switch (po->tp_version) {
2165 			case TPACKET_V2:
2166 				off = ph.h2->tp_net;
2167 				break;
2168 			default:
2169 				off = ph.h1->tp_net;
2170 				break;
2171 			}
2172 		} else {
2173 			switch (po->tp_version) {
2174 			case TPACKET_V2:
2175 				off = ph.h2->tp_mac;
2176 				break;
2177 			default:
2178 				off = ph.h1->tp_mac;
2179 				break;
2180 			}
2181 		}
2182 		if (unlikely((off < off_min) || (off_max < off)))
2183 			return -EINVAL;
2184 		data = ph.raw + off;
2185 	} else {
2186 		data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2187 	}
2188 	to_write = tp_len;
2189 
2190 	if (sock->type == SOCK_DGRAM) {
2191 		err = dev_hard_header(skb, dev, ntohs(proto), addr,
2192 				NULL, tp_len);
2193 		if (unlikely(err < 0))
2194 			return -EINVAL;
2195 	} else if (dev->hard_header_len) {
2196 		if (ll_header_truncated(dev, tp_len))
2197 			return -EINVAL;
2198 
2199 		skb_push(skb, dev->hard_header_len);
2200 		err = skb_store_bits(skb, 0, data,
2201 				dev->hard_header_len);
2202 		if (unlikely(err))
2203 			return err;
2204 
2205 		data += dev->hard_header_len;
2206 		to_write -= dev->hard_header_len;
2207 	}
2208 
2209 	offset = offset_in_page(data);
2210 	len_max = PAGE_SIZE - offset;
2211 	len = ((to_write > len_max) ? len_max : to_write);
2212 
2213 	skb->data_len = to_write;
2214 	skb->len += to_write;
2215 	skb->truesize += to_write;
2216 	atomic_add(to_write, &po->sk.sk_wmem_alloc);
2217 
2218 	while (likely(to_write)) {
2219 		nr_frags = skb_shinfo(skb)->nr_frags;
2220 
2221 		if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2222 			pr_err("Packet exceed the number of skb frags(%lu)\n",
2223 			       MAX_SKB_FRAGS);
2224 			return -EFAULT;
2225 		}
2226 
2227 		page = pgv_to_page(data);
2228 		data += len;
2229 		flush_dcache_page(page);
2230 		get_page(page);
2231 		skb_fill_page_desc(skb, nr_frags, page, offset, len);
2232 		to_write -= len;
2233 		offset = 0;
2234 		len_max = PAGE_SIZE;
2235 		len = ((to_write > len_max) ? len_max : to_write);
2236 	}
2237 
2238 	return tp_len;
2239 }
2240 
2241 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2242 {
2243 	struct sk_buff *skb;
2244 	struct net_device *dev;
2245 	__be16 proto;
2246 	int err, reserve = 0;
2247 	void *ph;
2248 	DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2249 	bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2250 	int tp_len, size_max;
2251 	unsigned char *addr;
2252 	int len_sum = 0;
2253 	int status = TP_STATUS_AVAILABLE;
2254 	int hlen, tlen;
2255 
2256 	mutex_lock(&po->pg_vec_lock);
2257 
2258 	if (likely(saddr == NULL)) {
2259 		dev	= packet_cached_dev_get(po);
2260 		proto	= po->num;
2261 		addr	= NULL;
2262 	} else {
2263 		err = -EINVAL;
2264 		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2265 			goto out;
2266 		if (msg->msg_namelen < (saddr->sll_halen
2267 					+ offsetof(struct sockaddr_ll,
2268 						sll_addr)))
2269 			goto out;
2270 		proto	= saddr->sll_protocol;
2271 		addr	= saddr->sll_addr;
2272 		dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2273 	}
2274 
2275 	err = -ENXIO;
2276 	if (unlikely(dev == NULL))
2277 		goto out;
2278 	err = -ENETDOWN;
2279 	if (unlikely(!(dev->flags & IFF_UP)))
2280 		goto out_put;
2281 
2282 	reserve = dev->hard_header_len + VLAN_HLEN;
2283 	size_max = po->tx_ring.frame_size
2284 		- (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2285 
2286 	if (size_max > dev->mtu + reserve)
2287 		size_max = dev->mtu + reserve;
2288 
2289 	do {
2290 		ph = packet_current_frame(po, &po->tx_ring,
2291 					  TP_STATUS_SEND_REQUEST);
2292 		if (unlikely(ph == NULL)) {
2293 			if (need_wait && need_resched())
2294 				schedule();
2295 			continue;
2296 		}
2297 
2298 		status = TP_STATUS_SEND_REQUEST;
2299 		hlen = LL_RESERVED_SPACE(dev);
2300 		tlen = dev->needed_tailroom;
2301 		skb = sock_alloc_send_skb(&po->sk,
2302 				hlen + tlen + sizeof(struct sockaddr_ll),
2303 				0, &err);
2304 
2305 		if (unlikely(skb == NULL))
2306 			goto out_status;
2307 
2308 		tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2309 					  addr, hlen);
2310 		if (tp_len > dev->mtu + dev->hard_header_len) {
2311 			struct ethhdr *ehdr;
2312 			/* Earlier code assumed this would be a VLAN pkt,
2313 			 * double-check this now that we have the actual
2314 			 * packet in hand.
2315 			 */
2316 
2317 			skb_reset_mac_header(skb);
2318 			ehdr = eth_hdr(skb);
2319 			if (ehdr->h_proto != htons(ETH_P_8021Q))
2320 				tp_len = -EMSGSIZE;
2321 		}
2322 		if (unlikely(tp_len < 0)) {
2323 			if (po->tp_loss) {
2324 				__packet_set_status(po, ph,
2325 						TP_STATUS_AVAILABLE);
2326 				packet_increment_head(&po->tx_ring);
2327 				kfree_skb(skb);
2328 				continue;
2329 			} else {
2330 				status = TP_STATUS_WRONG_FORMAT;
2331 				err = tp_len;
2332 				goto out_status;
2333 			}
2334 		}
2335 
2336 		packet_pick_tx_queue(dev, skb);
2337 
2338 		skb->destructor = tpacket_destruct_skb;
2339 		__packet_set_status(po, ph, TP_STATUS_SENDING);
2340 		packet_inc_pending(&po->tx_ring);
2341 
2342 		status = TP_STATUS_SEND_REQUEST;
2343 		err = po->xmit(skb);
2344 		if (unlikely(err > 0)) {
2345 			err = net_xmit_errno(err);
2346 			if (err && __packet_get_status(po, ph) ==
2347 				   TP_STATUS_AVAILABLE) {
2348 				/* skb was destructed already */
2349 				skb = NULL;
2350 				goto out_status;
2351 			}
2352 			/*
2353 			 * skb was dropped but not destructed yet;
2354 			 * let's treat it like congestion or err < 0
2355 			 */
2356 			err = 0;
2357 		}
2358 		packet_increment_head(&po->tx_ring);
2359 		len_sum += tp_len;
2360 	} while (likely((ph != NULL) ||
2361 		/* Note: packet_read_pending() might be slow if we have
2362 		 * to call it as it's per_cpu variable, but in fast-path
2363 		 * we already short-circuit the loop with the first
2364 		 * condition, and luckily don't have to go that path
2365 		 * anyway.
2366 		 */
2367 		 (need_wait && packet_read_pending(&po->tx_ring))));
2368 
2369 	err = len_sum;
2370 	goto out_put;
2371 
2372 out_status:
2373 	__packet_set_status(po, ph, status);
2374 	kfree_skb(skb);
2375 out_put:
2376 	dev_put(dev);
2377 out:
2378 	mutex_unlock(&po->pg_vec_lock);
2379 	return err;
2380 }
2381 
2382 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2383 				        size_t reserve, size_t len,
2384 				        size_t linear, int noblock,
2385 				        int *err)
2386 {
2387 	struct sk_buff *skb;
2388 
2389 	/* Under a page?  Don't bother with paged skb. */
2390 	if (prepad + len < PAGE_SIZE || !linear)
2391 		linear = len;
2392 
2393 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2394 				   err, 0);
2395 	if (!skb)
2396 		return NULL;
2397 
2398 	skb_reserve(skb, reserve);
2399 	skb_put(skb, linear);
2400 	skb->data_len = len - linear;
2401 	skb->len += len - linear;
2402 
2403 	return skb;
2404 }
2405 
2406 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2407 {
2408 	struct sock *sk = sock->sk;
2409 	DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2410 	struct sk_buff *skb;
2411 	struct net_device *dev;
2412 	__be16 proto;
2413 	unsigned char *addr;
2414 	int err, reserve = 0;
2415 	struct virtio_net_hdr vnet_hdr = { 0 };
2416 	int offset = 0;
2417 	int vnet_hdr_len;
2418 	struct packet_sock *po = pkt_sk(sk);
2419 	unsigned short gso_type = 0;
2420 	int hlen, tlen;
2421 	int extra_len = 0;
2422 	ssize_t n;
2423 
2424 	/*
2425 	 *	Get and verify the address.
2426 	 */
2427 
2428 	if (likely(saddr == NULL)) {
2429 		dev	= packet_cached_dev_get(po);
2430 		proto	= po->num;
2431 		addr	= NULL;
2432 	} else {
2433 		err = -EINVAL;
2434 		if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2435 			goto out;
2436 		if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2437 			goto out;
2438 		proto	= saddr->sll_protocol;
2439 		addr	= saddr->sll_addr;
2440 		dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2441 	}
2442 
2443 	err = -ENXIO;
2444 	if (unlikely(dev == NULL))
2445 		goto out_unlock;
2446 	err = -ENETDOWN;
2447 	if (unlikely(!(dev->flags & IFF_UP)))
2448 		goto out_unlock;
2449 
2450 	if (sock->type == SOCK_RAW)
2451 		reserve = dev->hard_header_len;
2452 	if (po->has_vnet_hdr) {
2453 		vnet_hdr_len = sizeof(vnet_hdr);
2454 
2455 		err = -EINVAL;
2456 		if (len < vnet_hdr_len)
2457 			goto out_unlock;
2458 
2459 		len -= vnet_hdr_len;
2460 
2461 		err = -EFAULT;
2462 		n = copy_from_iter(&vnet_hdr, vnet_hdr_len, &msg->msg_iter);
2463 		if (n != vnet_hdr_len)
2464 			goto out_unlock;
2465 
2466 		if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2467 		    (__virtio16_to_cpu(false, vnet_hdr.csum_start) +
2468 		     __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2 >
2469 		      __virtio16_to_cpu(false, vnet_hdr.hdr_len)))
2470 			vnet_hdr.hdr_len = __cpu_to_virtio16(false,
2471 				 __virtio16_to_cpu(false, vnet_hdr.csum_start) +
2472 				__virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2);
2473 
2474 		err = -EINVAL;
2475 		if (__virtio16_to_cpu(false, vnet_hdr.hdr_len) > len)
2476 			goto out_unlock;
2477 
2478 		if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2479 			switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2480 			case VIRTIO_NET_HDR_GSO_TCPV4:
2481 				gso_type = SKB_GSO_TCPV4;
2482 				break;
2483 			case VIRTIO_NET_HDR_GSO_TCPV6:
2484 				gso_type = SKB_GSO_TCPV6;
2485 				break;
2486 			case VIRTIO_NET_HDR_GSO_UDP:
2487 				gso_type = SKB_GSO_UDP;
2488 				break;
2489 			default:
2490 				goto out_unlock;
2491 			}
2492 
2493 			if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2494 				gso_type |= SKB_GSO_TCP_ECN;
2495 
2496 			if (vnet_hdr.gso_size == 0)
2497 				goto out_unlock;
2498 
2499 		}
2500 	}
2501 
2502 	if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2503 		if (!netif_supports_nofcs(dev)) {
2504 			err = -EPROTONOSUPPORT;
2505 			goto out_unlock;
2506 		}
2507 		extra_len = 4; /* We're doing our own CRC */
2508 	}
2509 
2510 	err = -EMSGSIZE;
2511 	if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2512 		goto out_unlock;
2513 
2514 	err = -ENOBUFS;
2515 	hlen = LL_RESERVED_SPACE(dev);
2516 	tlen = dev->needed_tailroom;
2517 	skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
2518 			       __virtio16_to_cpu(false, vnet_hdr.hdr_len),
2519 			       msg->msg_flags & MSG_DONTWAIT, &err);
2520 	if (skb == NULL)
2521 		goto out_unlock;
2522 
2523 	skb_set_network_header(skb, reserve);
2524 
2525 	err = -EINVAL;
2526 	if (sock->type == SOCK_DGRAM) {
2527 		offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2528 		if (unlikely(offset < 0))
2529 			goto out_free;
2530 	} else {
2531 		if (ll_header_truncated(dev, len))
2532 			goto out_free;
2533 	}
2534 
2535 	/* Returns -EFAULT on error */
2536 	err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2537 	if (err)
2538 		goto out_free;
2539 
2540 	sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2541 
2542 	if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2543 		/* Earlier code assumed this would be a VLAN pkt,
2544 		 * double-check this now that we have the actual
2545 		 * packet in hand.
2546 		 */
2547 		struct ethhdr *ehdr;
2548 		skb_reset_mac_header(skb);
2549 		ehdr = eth_hdr(skb);
2550 		if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2551 			err = -EMSGSIZE;
2552 			goto out_free;
2553 		}
2554 	}
2555 
2556 	skb->protocol = proto;
2557 	skb->dev = dev;
2558 	skb->priority = sk->sk_priority;
2559 	skb->mark = sk->sk_mark;
2560 
2561 	packet_pick_tx_queue(dev, skb);
2562 
2563 	if (po->has_vnet_hdr) {
2564 		if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2565 			u16 s = __virtio16_to_cpu(false, vnet_hdr.csum_start);
2566 			u16 o = __virtio16_to_cpu(false, vnet_hdr.csum_offset);
2567 			if (!skb_partial_csum_set(skb, s, o)) {
2568 				err = -EINVAL;
2569 				goto out_free;
2570 			}
2571 		}
2572 
2573 		skb_shinfo(skb)->gso_size =
2574 			__virtio16_to_cpu(false, vnet_hdr.gso_size);
2575 		skb_shinfo(skb)->gso_type = gso_type;
2576 
2577 		/* Header must be checked, and gso_segs computed. */
2578 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2579 		skb_shinfo(skb)->gso_segs = 0;
2580 
2581 		len += vnet_hdr_len;
2582 	}
2583 
2584 	if (!packet_use_direct_xmit(po))
2585 		skb_probe_transport_header(skb, reserve);
2586 	if (unlikely(extra_len == 4))
2587 		skb->no_fcs = 1;
2588 
2589 	err = po->xmit(skb);
2590 	if (err > 0 && (err = net_xmit_errno(err)) != 0)
2591 		goto out_unlock;
2592 
2593 	dev_put(dev);
2594 
2595 	return len;
2596 
2597 out_free:
2598 	kfree_skb(skb);
2599 out_unlock:
2600 	if (dev)
2601 		dev_put(dev);
2602 out:
2603 	return err;
2604 }
2605 
2606 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2607 		struct msghdr *msg, size_t len)
2608 {
2609 	struct sock *sk = sock->sk;
2610 	struct packet_sock *po = pkt_sk(sk);
2611 
2612 	if (po->tx_ring.pg_vec)
2613 		return tpacket_snd(po, msg);
2614 	else
2615 		return packet_snd(sock, msg, len);
2616 }
2617 
2618 /*
2619  *	Close a PACKET socket. This is fairly simple. We immediately go
2620  *	to 'closed' state and remove our protocol entry in the device list.
2621  */
2622 
2623 static int packet_release(struct socket *sock)
2624 {
2625 	struct sock *sk = sock->sk;
2626 	struct packet_sock *po;
2627 	struct net *net;
2628 	union tpacket_req_u req_u;
2629 
2630 	if (!sk)
2631 		return 0;
2632 
2633 	net = sock_net(sk);
2634 	po = pkt_sk(sk);
2635 
2636 	mutex_lock(&net->packet.sklist_lock);
2637 	sk_del_node_init_rcu(sk);
2638 	mutex_unlock(&net->packet.sklist_lock);
2639 
2640 	preempt_disable();
2641 	sock_prot_inuse_add(net, sk->sk_prot, -1);
2642 	preempt_enable();
2643 
2644 	spin_lock(&po->bind_lock);
2645 	unregister_prot_hook(sk, false);
2646 	packet_cached_dev_reset(po);
2647 
2648 	if (po->prot_hook.dev) {
2649 		dev_put(po->prot_hook.dev);
2650 		po->prot_hook.dev = NULL;
2651 	}
2652 	spin_unlock(&po->bind_lock);
2653 
2654 	packet_flush_mclist(sk);
2655 
2656 	if (po->rx_ring.pg_vec) {
2657 		memset(&req_u, 0, sizeof(req_u));
2658 		packet_set_ring(sk, &req_u, 1, 0);
2659 	}
2660 
2661 	if (po->tx_ring.pg_vec) {
2662 		memset(&req_u, 0, sizeof(req_u));
2663 		packet_set_ring(sk, &req_u, 1, 1);
2664 	}
2665 
2666 	fanout_release(sk);
2667 
2668 	synchronize_net();
2669 	/*
2670 	 *	Now the socket is dead. No more input will appear.
2671 	 */
2672 	sock_orphan(sk);
2673 	sock->sk = NULL;
2674 
2675 	/* Purge queues */
2676 
2677 	skb_queue_purge(&sk->sk_receive_queue);
2678 	packet_free_pending(po);
2679 	sk_refcnt_debug_release(sk);
2680 
2681 	sock_put(sk);
2682 	return 0;
2683 }
2684 
2685 /*
2686  *	Attach a packet hook.
2687  */
2688 
2689 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 proto)
2690 {
2691 	struct packet_sock *po = pkt_sk(sk);
2692 	const struct net_device *dev_curr;
2693 	__be16 proto_curr;
2694 	bool need_rehook;
2695 
2696 	if (po->fanout) {
2697 		if (dev)
2698 			dev_put(dev);
2699 
2700 		return -EINVAL;
2701 	}
2702 
2703 	lock_sock(sk);
2704 	spin_lock(&po->bind_lock);
2705 
2706 	proto_curr = po->prot_hook.type;
2707 	dev_curr = po->prot_hook.dev;
2708 
2709 	need_rehook = proto_curr != proto || dev_curr != dev;
2710 
2711 	if (need_rehook) {
2712 		unregister_prot_hook(sk, true);
2713 
2714 		po->num = proto;
2715 		po->prot_hook.type = proto;
2716 
2717 		if (po->prot_hook.dev)
2718 			dev_put(po->prot_hook.dev);
2719 
2720 		po->prot_hook.dev = dev;
2721 
2722 		po->ifindex = dev ? dev->ifindex : 0;
2723 		packet_cached_dev_assign(po, dev);
2724 	}
2725 
2726 	if (proto == 0 || !need_rehook)
2727 		goto out_unlock;
2728 
2729 	if (!dev || (dev->flags & IFF_UP)) {
2730 		register_prot_hook(sk);
2731 	} else {
2732 		sk->sk_err = ENETDOWN;
2733 		if (!sock_flag(sk, SOCK_DEAD))
2734 			sk->sk_error_report(sk);
2735 	}
2736 
2737 out_unlock:
2738 	spin_unlock(&po->bind_lock);
2739 	release_sock(sk);
2740 	return 0;
2741 }
2742 
2743 /*
2744  *	Bind a packet socket to a device
2745  */
2746 
2747 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2748 			    int addr_len)
2749 {
2750 	struct sock *sk = sock->sk;
2751 	char name[15];
2752 	struct net_device *dev;
2753 	int err = -ENODEV;
2754 
2755 	/*
2756 	 *	Check legality
2757 	 */
2758 
2759 	if (addr_len != sizeof(struct sockaddr))
2760 		return -EINVAL;
2761 	strlcpy(name, uaddr->sa_data, sizeof(name));
2762 
2763 	dev = dev_get_by_name(sock_net(sk), name);
2764 	if (dev)
2765 		err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2766 	return err;
2767 }
2768 
2769 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2770 {
2771 	struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2772 	struct sock *sk = sock->sk;
2773 	struct net_device *dev = NULL;
2774 	int err;
2775 
2776 
2777 	/*
2778 	 *	Check legality
2779 	 */
2780 
2781 	if (addr_len < sizeof(struct sockaddr_ll))
2782 		return -EINVAL;
2783 	if (sll->sll_family != AF_PACKET)
2784 		return -EINVAL;
2785 
2786 	if (sll->sll_ifindex) {
2787 		err = -ENODEV;
2788 		dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2789 		if (dev == NULL)
2790 			goto out;
2791 	}
2792 	err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2793 
2794 out:
2795 	return err;
2796 }
2797 
2798 static struct proto packet_proto = {
2799 	.name	  = "PACKET",
2800 	.owner	  = THIS_MODULE,
2801 	.obj_size = sizeof(struct packet_sock),
2802 };
2803 
2804 /*
2805  *	Create a packet of type SOCK_PACKET.
2806  */
2807 
2808 static int packet_create(struct net *net, struct socket *sock, int protocol,
2809 			 int kern)
2810 {
2811 	struct sock *sk;
2812 	struct packet_sock *po;
2813 	__be16 proto = (__force __be16)protocol; /* weird, but documented */
2814 	int err;
2815 
2816 	if (!ns_capable(net->user_ns, CAP_NET_RAW))
2817 		return -EPERM;
2818 	if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2819 	    sock->type != SOCK_PACKET)
2820 		return -ESOCKTNOSUPPORT;
2821 
2822 	sock->state = SS_UNCONNECTED;
2823 
2824 	err = -ENOBUFS;
2825 	sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2826 	if (sk == NULL)
2827 		goto out;
2828 
2829 	sock->ops = &packet_ops;
2830 	if (sock->type == SOCK_PACKET)
2831 		sock->ops = &packet_ops_spkt;
2832 
2833 	sock_init_data(sock, sk);
2834 
2835 	po = pkt_sk(sk);
2836 	sk->sk_family = PF_PACKET;
2837 	po->num = proto;
2838 	po->xmit = dev_queue_xmit;
2839 
2840 	err = packet_alloc_pending(po);
2841 	if (err)
2842 		goto out2;
2843 
2844 	packet_cached_dev_reset(po);
2845 
2846 	sk->sk_destruct = packet_sock_destruct;
2847 	sk_refcnt_debug_inc(sk);
2848 
2849 	/*
2850 	 *	Attach a protocol block
2851 	 */
2852 
2853 	spin_lock_init(&po->bind_lock);
2854 	mutex_init(&po->pg_vec_lock);
2855 	po->prot_hook.func = packet_rcv;
2856 
2857 	if (sock->type == SOCK_PACKET)
2858 		po->prot_hook.func = packet_rcv_spkt;
2859 
2860 	po->prot_hook.af_packet_priv = sk;
2861 
2862 	if (proto) {
2863 		po->prot_hook.type = proto;
2864 		register_prot_hook(sk);
2865 	}
2866 
2867 	mutex_lock(&net->packet.sklist_lock);
2868 	sk_add_node_rcu(sk, &net->packet.sklist);
2869 	mutex_unlock(&net->packet.sklist_lock);
2870 
2871 	preempt_disable();
2872 	sock_prot_inuse_add(net, &packet_proto, 1);
2873 	preempt_enable();
2874 
2875 	return 0;
2876 out2:
2877 	sk_free(sk);
2878 out:
2879 	return err;
2880 }
2881 
2882 /*
2883  *	Pull a packet from our receive queue and hand it to the user.
2884  *	If necessary we block.
2885  */
2886 
2887 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2888 			  struct msghdr *msg, size_t len, int flags)
2889 {
2890 	struct sock *sk = sock->sk;
2891 	struct sk_buff *skb;
2892 	int copied, err;
2893 	int vnet_hdr_len = 0;
2894 
2895 	err = -EINVAL;
2896 	if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2897 		goto out;
2898 
2899 #if 0
2900 	/* What error should we return now? EUNATTACH? */
2901 	if (pkt_sk(sk)->ifindex < 0)
2902 		return -ENODEV;
2903 #endif
2904 
2905 	if (flags & MSG_ERRQUEUE) {
2906 		err = sock_recv_errqueue(sk, msg, len,
2907 					 SOL_PACKET, PACKET_TX_TIMESTAMP);
2908 		goto out;
2909 	}
2910 
2911 	/*
2912 	 *	Call the generic datagram receiver. This handles all sorts
2913 	 *	of horrible races and re-entrancy so we can forget about it
2914 	 *	in the protocol layers.
2915 	 *
2916 	 *	Now it will return ENETDOWN, if device have just gone down,
2917 	 *	but then it will block.
2918 	 */
2919 
2920 	skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2921 
2922 	/*
2923 	 *	An error occurred so return it. Because skb_recv_datagram()
2924 	 *	handles the blocking we don't see and worry about blocking
2925 	 *	retries.
2926 	 */
2927 
2928 	if (skb == NULL)
2929 		goto out;
2930 
2931 	if (pkt_sk(sk)->has_vnet_hdr) {
2932 		struct virtio_net_hdr vnet_hdr = { 0 };
2933 
2934 		err = -EINVAL;
2935 		vnet_hdr_len = sizeof(vnet_hdr);
2936 		if (len < vnet_hdr_len)
2937 			goto out_free;
2938 
2939 		len -= vnet_hdr_len;
2940 
2941 		if (skb_is_gso(skb)) {
2942 			struct skb_shared_info *sinfo = skb_shinfo(skb);
2943 
2944 			/* This is a hint as to how much should be linear. */
2945 			vnet_hdr.hdr_len =
2946 				__cpu_to_virtio16(false, skb_headlen(skb));
2947 			vnet_hdr.gso_size =
2948 				__cpu_to_virtio16(false, sinfo->gso_size);
2949 			if (sinfo->gso_type & SKB_GSO_TCPV4)
2950 				vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2951 			else if (sinfo->gso_type & SKB_GSO_TCPV6)
2952 				vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2953 			else if (sinfo->gso_type & SKB_GSO_UDP)
2954 				vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2955 			else if (sinfo->gso_type & SKB_GSO_FCOE)
2956 				goto out_free;
2957 			else
2958 				BUG();
2959 			if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2960 				vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2961 		} else
2962 			vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2963 
2964 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
2965 			vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2966 			vnet_hdr.csum_start = __cpu_to_virtio16(false,
2967 					  skb_checksum_start_offset(skb));
2968 			vnet_hdr.csum_offset = __cpu_to_virtio16(false,
2969 							 skb->csum_offset);
2970 		} else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2971 			vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2972 		} /* else everything is zero */
2973 
2974 		err = memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_len);
2975 		if (err < 0)
2976 			goto out_free;
2977 	}
2978 
2979 	/* You lose any data beyond the buffer you gave. If it worries
2980 	 * a user program they can ask the device for its MTU
2981 	 * anyway.
2982 	 */
2983 	copied = skb->len;
2984 	if (copied > len) {
2985 		copied = len;
2986 		msg->msg_flags |= MSG_TRUNC;
2987 	}
2988 
2989 	err = skb_copy_datagram_msg(skb, 0, msg, copied);
2990 	if (err)
2991 		goto out_free;
2992 
2993 	sock_recv_ts_and_drops(msg, sk, skb);
2994 
2995 	if (msg->msg_name) {
2996 		/* If the address length field is there to be filled
2997 		 * in, we fill it in now.
2998 		 */
2999 		if (sock->type == SOCK_PACKET) {
3000 			__sockaddr_check_size(sizeof(struct sockaddr_pkt));
3001 			msg->msg_namelen = sizeof(struct sockaddr_pkt);
3002 		} else {
3003 			struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3004 			msg->msg_namelen = sll->sll_halen +
3005 				offsetof(struct sockaddr_ll, sll_addr);
3006 		}
3007 		memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3008 		       msg->msg_namelen);
3009 	}
3010 
3011 	if (pkt_sk(sk)->auxdata) {
3012 		struct tpacket_auxdata aux;
3013 
3014 		aux.tp_status = TP_STATUS_USER;
3015 		if (skb->ip_summed == CHECKSUM_PARTIAL)
3016 			aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3017 		aux.tp_len = PACKET_SKB_CB(skb)->origlen;
3018 		aux.tp_snaplen = skb->len;
3019 		aux.tp_mac = 0;
3020 		aux.tp_net = skb_network_offset(skb);
3021 		if (skb_vlan_tag_present(skb)) {
3022 			aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3023 			aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3024 			aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3025 		} else {
3026 			aux.tp_vlan_tci = 0;
3027 			aux.tp_vlan_tpid = 0;
3028 		}
3029 		put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3030 	}
3031 
3032 	/*
3033 	 *	Free or return the buffer as appropriate. Again this
3034 	 *	hides all the races and re-entrancy issues from us.
3035 	 */
3036 	err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3037 
3038 out_free:
3039 	skb_free_datagram(sk, skb);
3040 out:
3041 	return err;
3042 }
3043 
3044 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3045 			       int *uaddr_len, int peer)
3046 {
3047 	struct net_device *dev;
3048 	struct sock *sk	= sock->sk;
3049 
3050 	if (peer)
3051 		return -EOPNOTSUPP;
3052 
3053 	uaddr->sa_family = AF_PACKET;
3054 	memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3055 	rcu_read_lock();
3056 	dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3057 	if (dev)
3058 		strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3059 	rcu_read_unlock();
3060 	*uaddr_len = sizeof(*uaddr);
3061 
3062 	return 0;
3063 }
3064 
3065 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3066 			  int *uaddr_len, int peer)
3067 {
3068 	struct net_device *dev;
3069 	struct sock *sk = sock->sk;
3070 	struct packet_sock *po = pkt_sk(sk);
3071 	DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3072 
3073 	if (peer)
3074 		return -EOPNOTSUPP;
3075 
3076 	sll->sll_family = AF_PACKET;
3077 	sll->sll_ifindex = po->ifindex;
3078 	sll->sll_protocol = po->num;
3079 	sll->sll_pkttype = 0;
3080 	rcu_read_lock();
3081 	dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3082 	if (dev) {
3083 		sll->sll_hatype = dev->type;
3084 		sll->sll_halen = dev->addr_len;
3085 		memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3086 	} else {
3087 		sll->sll_hatype = 0;	/* Bad: we have no ARPHRD_UNSPEC */
3088 		sll->sll_halen = 0;
3089 	}
3090 	rcu_read_unlock();
3091 	*uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3092 
3093 	return 0;
3094 }
3095 
3096 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3097 			 int what)
3098 {
3099 	switch (i->type) {
3100 	case PACKET_MR_MULTICAST:
3101 		if (i->alen != dev->addr_len)
3102 			return -EINVAL;
3103 		if (what > 0)
3104 			return dev_mc_add(dev, i->addr);
3105 		else
3106 			return dev_mc_del(dev, i->addr);
3107 		break;
3108 	case PACKET_MR_PROMISC:
3109 		return dev_set_promiscuity(dev, what);
3110 	case PACKET_MR_ALLMULTI:
3111 		return dev_set_allmulti(dev, what);
3112 	case PACKET_MR_UNICAST:
3113 		if (i->alen != dev->addr_len)
3114 			return -EINVAL;
3115 		if (what > 0)
3116 			return dev_uc_add(dev, i->addr);
3117 		else
3118 			return dev_uc_del(dev, i->addr);
3119 		break;
3120 	default:
3121 		break;
3122 	}
3123 	return 0;
3124 }
3125 
3126 static void packet_dev_mclist_delete(struct net_device *dev,
3127 				     struct packet_mclist **mlp)
3128 {
3129 	struct packet_mclist *ml;
3130 
3131 	while ((ml = *mlp) != NULL) {
3132 		if (ml->ifindex == dev->ifindex) {
3133 			packet_dev_mc(dev, ml, -1);
3134 			*mlp = ml->next;
3135 			kfree(ml);
3136 		} else
3137 			mlp = &ml->next;
3138 	}
3139 }
3140 
3141 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3142 {
3143 	struct packet_sock *po = pkt_sk(sk);
3144 	struct packet_mclist *ml, *i;
3145 	struct net_device *dev;
3146 	int err;
3147 
3148 	rtnl_lock();
3149 
3150 	err = -ENODEV;
3151 	dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3152 	if (!dev)
3153 		goto done;
3154 
3155 	err = -EINVAL;
3156 	if (mreq->mr_alen > dev->addr_len)
3157 		goto done;
3158 
3159 	err = -ENOBUFS;
3160 	i = kmalloc(sizeof(*i), GFP_KERNEL);
3161 	if (i == NULL)
3162 		goto done;
3163 
3164 	err = 0;
3165 	for (ml = po->mclist; ml; ml = ml->next) {
3166 		if (ml->ifindex == mreq->mr_ifindex &&
3167 		    ml->type == mreq->mr_type &&
3168 		    ml->alen == mreq->mr_alen &&
3169 		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3170 			ml->count++;
3171 			/* Free the new element ... */
3172 			kfree(i);
3173 			goto done;
3174 		}
3175 	}
3176 
3177 	i->type = mreq->mr_type;
3178 	i->ifindex = mreq->mr_ifindex;
3179 	i->alen = mreq->mr_alen;
3180 	memcpy(i->addr, mreq->mr_address, i->alen);
3181 	i->count = 1;
3182 	i->next = po->mclist;
3183 	po->mclist = i;
3184 	err = packet_dev_mc(dev, i, 1);
3185 	if (err) {
3186 		po->mclist = i->next;
3187 		kfree(i);
3188 	}
3189 
3190 done:
3191 	rtnl_unlock();
3192 	return err;
3193 }
3194 
3195 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3196 {
3197 	struct packet_mclist *ml, **mlp;
3198 
3199 	rtnl_lock();
3200 
3201 	for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3202 		if (ml->ifindex == mreq->mr_ifindex &&
3203 		    ml->type == mreq->mr_type &&
3204 		    ml->alen == mreq->mr_alen &&
3205 		    memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3206 			if (--ml->count == 0) {
3207 				struct net_device *dev;
3208 				*mlp = ml->next;
3209 				dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3210 				if (dev)
3211 					packet_dev_mc(dev, ml, -1);
3212 				kfree(ml);
3213 			}
3214 			break;
3215 		}
3216 	}
3217 	rtnl_unlock();
3218 	return 0;
3219 }
3220 
3221 static void packet_flush_mclist(struct sock *sk)
3222 {
3223 	struct packet_sock *po = pkt_sk(sk);
3224 	struct packet_mclist *ml;
3225 
3226 	if (!po->mclist)
3227 		return;
3228 
3229 	rtnl_lock();
3230 	while ((ml = po->mclist) != NULL) {
3231 		struct net_device *dev;
3232 
3233 		po->mclist = ml->next;
3234 		dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3235 		if (dev != NULL)
3236 			packet_dev_mc(dev, ml, -1);
3237 		kfree(ml);
3238 	}
3239 	rtnl_unlock();
3240 }
3241 
3242 static int
3243 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3244 {
3245 	struct sock *sk = sock->sk;
3246 	struct packet_sock *po = pkt_sk(sk);
3247 	int ret;
3248 
3249 	if (level != SOL_PACKET)
3250 		return -ENOPROTOOPT;
3251 
3252 	switch (optname) {
3253 	case PACKET_ADD_MEMBERSHIP:
3254 	case PACKET_DROP_MEMBERSHIP:
3255 	{
3256 		struct packet_mreq_max mreq;
3257 		int len = optlen;
3258 		memset(&mreq, 0, sizeof(mreq));
3259 		if (len < sizeof(struct packet_mreq))
3260 			return -EINVAL;
3261 		if (len > sizeof(mreq))
3262 			len = sizeof(mreq);
3263 		if (copy_from_user(&mreq, optval, len))
3264 			return -EFAULT;
3265 		if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3266 			return -EINVAL;
3267 		if (optname == PACKET_ADD_MEMBERSHIP)
3268 			ret = packet_mc_add(sk, &mreq);
3269 		else
3270 			ret = packet_mc_drop(sk, &mreq);
3271 		return ret;
3272 	}
3273 
3274 	case PACKET_RX_RING:
3275 	case PACKET_TX_RING:
3276 	{
3277 		union tpacket_req_u req_u;
3278 		int len;
3279 
3280 		switch (po->tp_version) {
3281 		case TPACKET_V1:
3282 		case TPACKET_V2:
3283 			len = sizeof(req_u.req);
3284 			break;
3285 		case TPACKET_V3:
3286 		default:
3287 			len = sizeof(req_u.req3);
3288 			break;
3289 		}
3290 		if (optlen < len)
3291 			return -EINVAL;
3292 		if (pkt_sk(sk)->has_vnet_hdr)
3293 			return -EINVAL;
3294 		if (copy_from_user(&req_u.req, optval, len))
3295 			return -EFAULT;
3296 		return packet_set_ring(sk, &req_u, 0,
3297 			optname == PACKET_TX_RING);
3298 	}
3299 	case PACKET_COPY_THRESH:
3300 	{
3301 		int val;
3302 
3303 		if (optlen != sizeof(val))
3304 			return -EINVAL;
3305 		if (copy_from_user(&val, optval, sizeof(val)))
3306 			return -EFAULT;
3307 
3308 		pkt_sk(sk)->copy_thresh = val;
3309 		return 0;
3310 	}
3311 	case PACKET_VERSION:
3312 	{
3313 		int val;
3314 
3315 		if (optlen != sizeof(val))
3316 			return -EINVAL;
3317 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3318 			return -EBUSY;
3319 		if (copy_from_user(&val, optval, sizeof(val)))
3320 			return -EFAULT;
3321 		switch (val) {
3322 		case TPACKET_V1:
3323 		case TPACKET_V2:
3324 		case TPACKET_V3:
3325 			po->tp_version = val;
3326 			return 0;
3327 		default:
3328 			return -EINVAL;
3329 		}
3330 	}
3331 	case PACKET_RESERVE:
3332 	{
3333 		unsigned int val;
3334 
3335 		if (optlen != sizeof(val))
3336 			return -EINVAL;
3337 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3338 			return -EBUSY;
3339 		if (copy_from_user(&val, optval, sizeof(val)))
3340 			return -EFAULT;
3341 		po->tp_reserve = val;
3342 		return 0;
3343 	}
3344 	case PACKET_LOSS:
3345 	{
3346 		unsigned int val;
3347 
3348 		if (optlen != sizeof(val))
3349 			return -EINVAL;
3350 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3351 			return -EBUSY;
3352 		if (copy_from_user(&val, optval, sizeof(val)))
3353 			return -EFAULT;
3354 		po->tp_loss = !!val;
3355 		return 0;
3356 	}
3357 	case PACKET_AUXDATA:
3358 	{
3359 		int val;
3360 
3361 		if (optlen < sizeof(val))
3362 			return -EINVAL;
3363 		if (copy_from_user(&val, optval, sizeof(val)))
3364 			return -EFAULT;
3365 
3366 		po->auxdata = !!val;
3367 		return 0;
3368 	}
3369 	case PACKET_ORIGDEV:
3370 	{
3371 		int val;
3372 
3373 		if (optlen < sizeof(val))
3374 			return -EINVAL;
3375 		if (copy_from_user(&val, optval, sizeof(val)))
3376 			return -EFAULT;
3377 
3378 		po->origdev = !!val;
3379 		return 0;
3380 	}
3381 	case PACKET_VNET_HDR:
3382 	{
3383 		int val;
3384 
3385 		if (sock->type != SOCK_RAW)
3386 			return -EINVAL;
3387 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3388 			return -EBUSY;
3389 		if (optlen < sizeof(val))
3390 			return -EINVAL;
3391 		if (copy_from_user(&val, optval, sizeof(val)))
3392 			return -EFAULT;
3393 
3394 		po->has_vnet_hdr = !!val;
3395 		return 0;
3396 	}
3397 	case PACKET_TIMESTAMP:
3398 	{
3399 		int val;
3400 
3401 		if (optlen != sizeof(val))
3402 			return -EINVAL;
3403 		if (copy_from_user(&val, optval, sizeof(val)))
3404 			return -EFAULT;
3405 
3406 		po->tp_tstamp = val;
3407 		return 0;
3408 	}
3409 	case PACKET_FANOUT:
3410 	{
3411 		int val;
3412 
3413 		if (optlen != sizeof(val))
3414 			return -EINVAL;
3415 		if (copy_from_user(&val, optval, sizeof(val)))
3416 			return -EFAULT;
3417 
3418 		return fanout_add(sk, val & 0xffff, val >> 16);
3419 	}
3420 	case PACKET_TX_HAS_OFF:
3421 	{
3422 		unsigned int val;
3423 
3424 		if (optlen != sizeof(val))
3425 			return -EINVAL;
3426 		if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3427 			return -EBUSY;
3428 		if (copy_from_user(&val, optval, sizeof(val)))
3429 			return -EFAULT;
3430 		po->tp_tx_has_off = !!val;
3431 		return 0;
3432 	}
3433 	case PACKET_QDISC_BYPASS:
3434 	{
3435 		int val;
3436 
3437 		if (optlen != sizeof(val))
3438 			return -EINVAL;
3439 		if (copy_from_user(&val, optval, sizeof(val)))
3440 			return -EFAULT;
3441 
3442 		po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3443 		return 0;
3444 	}
3445 	default:
3446 		return -ENOPROTOOPT;
3447 	}
3448 }
3449 
3450 static int packet_getsockopt(struct socket *sock, int level, int optname,
3451 			     char __user *optval, int __user *optlen)
3452 {
3453 	int len;
3454 	int val, lv = sizeof(val);
3455 	struct sock *sk = sock->sk;
3456 	struct packet_sock *po = pkt_sk(sk);
3457 	void *data = &val;
3458 	union tpacket_stats_u st;
3459 
3460 	if (level != SOL_PACKET)
3461 		return -ENOPROTOOPT;
3462 
3463 	if (get_user(len, optlen))
3464 		return -EFAULT;
3465 
3466 	if (len < 0)
3467 		return -EINVAL;
3468 
3469 	switch (optname) {
3470 	case PACKET_STATISTICS:
3471 		spin_lock_bh(&sk->sk_receive_queue.lock);
3472 		memcpy(&st, &po->stats, sizeof(st));
3473 		memset(&po->stats, 0, sizeof(po->stats));
3474 		spin_unlock_bh(&sk->sk_receive_queue.lock);
3475 
3476 		if (po->tp_version == TPACKET_V3) {
3477 			lv = sizeof(struct tpacket_stats_v3);
3478 			st.stats3.tp_packets += st.stats3.tp_drops;
3479 			data = &st.stats3;
3480 		} else {
3481 			lv = sizeof(struct tpacket_stats);
3482 			st.stats1.tp_packets += st.stats1.tp_drops;
3483 			data = &st.stats1;
3484 		}
3485 
3486 		break;
3487 	case PACKET_AUXDATA:
3488 		val = po->auxdata;
3489 		break;
3490 	case PACKET_ORIGDEV:
3491 		val = po->origdev;
3492 		break;
3493 	case PACKET_VNET_HDR:
3494 		val = po->has_vnet_hdr;
3495 		break;
3496 	case PACKET_VERSION:
3497 		val = po->tp_version;
3498 		break;
3499 	case PACKET_HDRLEN:
3500 		if (len > sizeof(int))
3501 			len = sizeof(int);
3502 		if (copy_from_user(&val, optval, len))
3503 			return -EFAULT;
3504 		switch (val) {
3505 		case TPACKET_V1:
3506 			val = sizeof(struct tpacket_hdr);
3507 			break;
3508 		case TPACKET_V2:
3509 			val = sizeof(struct tpacket2_hdr);
3510 			break;
3511 		case TPACKET_V3:
3512 			val = sizeof(struct tpacket3_hdr);
3513 			break;
3514 		default:
3515 			return -EINVAL;
3516 		}
3517 		break;
3518 	case PACKET_RESERVE:
3519 		val = po->tp_reserve;
3520 		break;
3521 	case PACKET_LOSS:
3522 		val = po->tp_loss;
3523 		break;
3524 	case PACKET_TIMESTAMP:
3525 		val = po->tp_tstamp;
3526 		break;
3527 	case PACKET_FANOUT:
3528 		val = (po->fanout ?
3529 		       ((u32)po->fanout->id |
3530 			((u32)po->fanout->type << 16) |
3531 			((u32)po->fanout->flags << 24)) :
3532 		       0);
3533 		break;
3534 	case PACKET_TX_HAS_OFF:
3535 		val = po->tp_tx_has_off;
3536 		break;
3537 	case PACKET_QDISC_BYPASS:
3538 		val = packet_use_direct_xmit(po);
3539 		break;
3540 	default:
3541 		return -ENOPROTOOPT;
3542 	}
3543 
3544 	if (len > lv)
3545 		len = lv;
3546 	if (put_user(len, optlen))
3547 		return -EFAULT;
3548 	if (copy_to_user(optval, data, len))
3549 		return -EFAULT;
3550 	return 0;
3551 }
3552 
3553 
3554 static int packet_notifier(struct notifier_block *this,
3555 			   unsigned long msg, void *ptr)
3556 {
3557 	struct sock *sk;
3558 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3559 	struct net *net = dev_net(dev);
3560 
3561 	rcu_read_lock();
3562 	sk_for_each_rcu(sk, &net->packet.sklist) {
3563 		struct packet_sock *po = pkt_sk(sk);
3564 
3565 		switch (msg) {
3566 		case NETDEV_UNREGISTER:
3567 			if (po->mclist)
3568 				packet_dev_mclist_delete(dev, &po->mclist);
3569 			/* fallthrough */
3570 
3571 		case NETDEV_DOWN:
3572 			if (dev->ifindex == po->ifindex) {
3573 				spin_lock(&po->bind_lock);
3574 				if (po->running) {
3575 					__unregister_prot_hook(sk, false);
3576 					sk->sk_err = ENETDOWN;
3577 					if (!sock_flag(sk, SOCK_DEAD))
3578 						sk->sk_error_report(sk);
3579 				}
3580 				if (msg == NETDEV_UNREGISTER) {
3581 					packet_cached_dev_reset(po);
3582 					po->ifindex = -1;
3583 					if (po->prot_hook.dev)
3584 						dev_put(po->prot_hook.dev);
3585 					po->prot_hook.dev = NULL;
3586 				}
3587 				spin_unlock(&po->bind_lock);
3588 			}
3589 			break;
3590 		case NETDEV_UP:
3591 			if (dev->ifindex == po->ifindex) {
3592 				spin_lock(&po->bind_lock);
3593 				if (po->num)
3594 					register_prot_hook(sk);
3595 				spin_unlock(&po->bind_lock);
3596 			}
3597 			break;
3598 		}
3599 	}
3600 	rcu_read_unlock();
3601 	return NOTIFY_DONE;
3602 }
3603 
3604 
3605 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3606 			unsigned long arg)
3607 {
3608 	struct sock *sk = sock->sk;
3609 
3610 	switch (cmd) {
3611 	case SIOCOUTQ:
3612 	{
3613 		int amount = sk_wmem_alloc_get(sk);
3614 
3615 		return put_user(amount, (int __user *)arg);
3616 	}
3617 	case SIOCINQ:
3618 	{
3619 		struct sk_buff *skb;
3620 		int amount = 0;
3621 
3622 		spin_lock_bh(&sk->sk_receive_queue.lock);
3623 		skb = skb_peek(&sk->sk_receive_queue);
3624 		if (skb)
3625 			amount = skb->len;
3626 		spin_unlock_bh(&sk->sk_receive_queue.lock);
3627 		return put_user(amount, (int __user *)arg);
3628 	}
3629 	case SIOCGSTAMP:
3630 		return sock_get_timestamp(sk, (struct timeval __user *)arg);
3631 	case SIOCGSTAMPNS:
3632 		return sock_get_timestampns(sk, (struct timespec __user *)arg);
3633 
3634 #ifdef CONFIG_INET
3635 	case SIOCADDRT:
3636 	case SIOCDELRT:
3637 	case SIOCDARP:
3638 	case SIOCGARP:
3639 	case SIOCSARP:
3640 	case SIOCGIFADDR:
3641 	case SIOCSIFADDR:
3642 	case SIOCGIFBRDADDR:
3643 	case SIOCSIFBRDADDR:
3644 	case SIOCGIFNETMASK:
3645 	case SIOCSIFNETMASK:
3646 	case SIOCGIFDSTADDR:
3647 	case SIOCSIFDSTADDR:
3648 	case SIOCSIFFLAGS:
3649 		return inet_dgram_ops.ioctl(sock, cmd, arg);
3650 #endif
3651 
3652 	default:
3653 		return -ENOIOCTLCMD;
3654 	}
3655 	return 0;
3656 }
3657 
3658 static unsigned int packet_poll(struct file *file, struct socket *sock,
3659 				poll_table *wait)
3660 {
3661 	struct sock *sk = sock->sk;
3662 	struct packet_sock *po = pkt_sk(sk);
3663 	unsigned int mask = datagram_poll(file, sock, wait);
3664 
3665 	spin_lock_bh(&sk->sk_receive_queue.lock);
3666 	if (po->rx_ring.pg_vec) {
3667 		if (!packet_previous_rx_frame(po, &po->rx_ring,
3668 			TP_STATUS_KERNEL))
3669 			mask |= POLLIN | POLLRDNORM;
3670 	}
3671 	spin_unlock_bh(&sk->sk_receive_queue.lock);
3672 	spin_lock_bh(&sk->sk_write_queue.lock);
3673 	if (po->tx_ring.pg_vec) {
3674 		if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3675 			mask |= POLLOUT | POLLWRNORM;
3676 	}
3677 	spin_unlock_bh(&sk->sk_write_queue.lock);
3678 	return mask;
3679 }
3680 
3681 
3682 /* Dirty? Well, I still did not learn better way to account
3683  * for user mmaps.
3684  */
3685 
3686 static void packet_mm_open(struct vm_area_struct *vma)
3687 {
3688 	struct file *file = vma->vm_file;
3689 	struct socket *sock = file->private_data;
3690 	struct sock *sk = sock->sk;
3691 
3692 	if (sk)
3693 		atomic_inc(&pkt_sk(sk)->mapped);
3694 }
3695 
3696 static void packet_mm_close(struct vm_area_struct *vma)
3697 {
3698 	struct file *file = vma->vm_file;
3699 	struct socket *sock = file->private_data;
3700 	struct sock *sk = sock->sk;
3701 
3702 	if (sk)
3703 		atomic_dec(&pkt_sk(sk)->mapped);
3704 }
3705 
3706 static const struct vm_operations_struct packet_mmap_ops = {
3707 	.open	=	packet_mm_open,
3708 	.close	=	packet_mm_close,
3709 };
3710 
3711 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3712 			unsigned int len)
3713 {
3714 	int i;
3715 
3716 	for (i = 0; i < len; i++) {
3717 		if (likely(pg_vec[i].buffer)) {
3718 			if (is_vmalloc_addr(pg_vec[i].buffer))
3719 				vfree(pg_vec[i].buffer);
3720 			else
3721 				free_pages((unsigned long)pg_vec[i].buffer,
3722 					   order);
3723 			pg_vec[i].buffer = NULL;
3724 		}
3725 	}
3726 	kfree(pg_vec);
3727 }
3728 
3729 static char *alloc_one_pg_vec_page(unsigned long order)
3730 {
3731 	char *buffer;
3732 	gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3733 			  __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3734 
3735 	buffer = (char *) __get_free_pages(gfp_flags, order);
3736 	if (buffer)
3737 		return buffer;
3738 
3739 	/* __get_free_pages failed, fall back to vmalloc */
3740 	buffer = vzalloc((1 << order) * PAGE_SIZE);
3741 	if (buffer)
3742 		return buffer;
3743 
3744 	/* vmalloc failed, lets dig into swap here */
3745 	gfp_flags &= ~__GFP_NORETRY;
3746 	buffer = (char *) __get_free_pages(gfp_flags, order);
3747 	if (buffer)
3748 		return buffer;
3749 
3750 	/* complete and utter failure */
3751 	return NULL;
3752 }
3753 
3754 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3755 {
3756 	unsigned int block_nr = req->tp_block_nr;
3757 	struct pgv *pg_vec;
3758 	int i;
3759 
3760 	pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3761 	if (unlikely(!pg_vec))
3762 		goto out;
3763 
3764 	for (i = 0; i < block_nr; i++) {
3765 		pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3766 		if (unlikely(!pg_vec[i].buffer))
3767 			goto out_free_pgvec;
3768 	}
3769 
3770 out:
3771 	return pg_vec;
3772 
3773 out_free_pgvec:
3774 	free_pg_vec(pg_vec, order, block_nr);
3775 	pg_vec = NULL;
3776 	goto out;
3777 }
3778 
3779 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3780 		int closing, int tx_ring)
3781 {
3782 	struct pgv *pg_vec = NULL;
3783 	struct packet_sock *po = pkt_sk(sk);
3784 	int was_running, order = 0;
3785 	struct packet_ring_buffer *rb;
3786 	struct sk_buff_head *rb_queue;
3787 	__be16 num;
3788 	int err = -EINVAL;
3789 	/* Added to avoid minimal code churn */
3790 	struct tpacket_req *req = &req_u->req;
3791 
3792 	/* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3793 	if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3794 		WARN(1, "Tx-ring is not supported.\n");
3795 		goto out;
3796 	}
3797 
3798 	rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3799 	rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3800 
3801 	err = -EBUSY;
3802 	if (!closing) {
3803 		if (atomic_read(&po->mapped))
3804 			goto out;
3805 		if (packet_read_pending(rb))
3806 			goto out;
3807 	}
3808 
3809 	if (req->tp_block_nr) {
3810 		/* Sanity tests and some calculations */
3811 		err = -EBUSY;
3812 		if (unlikely(rb->pg_vec))
3813 			goto out;
3814 
3815 		switch (po->tp_version) {
3816 		case TPACKET_V1:
3817 			po->tp_hdrlen = TPACKET_HDRLEN;
3818 			break;
3819 		case TPACKET_V2:
3820 			po->tp_hdrlen = TPACKET2_HDRLEN;
3821 			break;
3822 		case TPACKET_V3:
3823 			po->tp_hdrlen = TPACKET3_HDRLEN;
3824 			break;
3825 		}
3826 
3827 		err = -EINVAL;
3828 		if (unlikely((int)req->tp_block_size <= 0))
3829 			goto out;
3830 		if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3831 			goto out;
3832 		if (po->tp_version >= TPACKET_V3 &&
3833 		    (int)(req->tp_block_size -
3834 			  BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
3835 			goto out;
3836 		if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3837 					po->tp_reserve))
3838 			goto out;
3839 		if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3840 			goto out;
3841 
3842 		rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3843 		if (unlikely(rb->frames_per_block <= 0))
3844 			goto out;
3845 		if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3846 					req->tp_frame_nr))
3847 			goto out;
3848 
3849 		err = -ENOMEM;
3850 		order = get_order(req->tp_block_size);
3851 		pg_vec = alloc_pg_vec(req, order);
3852 		if (unlikely(!pg_vec))
3853 			goto out;
3854 		switch (po->tp_version) {
3855 		case TPACKET_V3:
3856 		/* Transmit path is not supported. We checked
3857 		 * it above but just being paranoid
3858 		 */
3859 			if (!tx_ring)
3860 				init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3861 			break;
3862 		default:
3863 			break;
3864 		}
3865 	}
3866 	/* Done */
3867 	else {
3868 		err = -EINVAL;
3869 		if (unlikely(req->tp_frame_nr))
3870 			goto out;
3871 	}
3872 
3873 	lock_sock(sk);
3874 
3875 	/* Detach socket from network */
3876 	spin_lock(&po->bind_lock);
3877 	was_running = po->running;
3878 	num = po->num;
3879 	if (was_running) {
3880 		po->num = 0;
3881 		__unregister_prot_hook(sk, false);
3882 	}
3883 	spin_unlock(&po->bind_lock);
3884 
3885 	synchronize_net();
3886 
3887 	err = -EBUSY;
3888 	mutex_lock(&po->pg_vec_lock);
3889 	if (closing || atomic_read(&po->mapped) == 0) {
3890 		err = 0;
3891 		spin_lock_bh(&rb_queue->lock);
3892 		swap(rb->pg_vec, pg_vec);
3893 		rb->frame_max = (req->tp_frame_nr - 1);
3894 		rb->head = 0;
3895 		rb->frame_size = req->tp_frame_size;
3896 		spin_unlock_bh(&rb_queue->lock);
3897 
3898 		swap(rb->pg_vec_order, order);
3899 		swap(rb->pg_vec_len, req->tp_block_nr);
3900 
3901 		rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3902 		po->prot_hook.func = (po->rx_ring.pg_vec) ?
3903 						tpacket_rcv : packet_rcv;
3904 		skb_queue_purge(rb_queue);
3905 		if (atomic_read(&po->mapped))
3906 			pr_err("packet_mmap: vma is busy: %d\n",
3907 			       atomic_read(&po->mapped));
3908 	}
3909 	mutex_unlock(&po->pg_vec_lock);
3910 
3911 	spin_lock(&po->bind_lock);
3912 	if (was_running) {
3913 		po->num = num;
3914 		register_prot_hook(sk);
3915 	}
3916 	spin_unlock(&po->bind_lock);
3917 	if (closing && (po->tp_version > TPACKET_V2)) {
3918 		/* Because we don't support block-based V3 on tx-ring */
3919 		if (!tx_ring)
3920 			prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3921 	}
3922 	release_sock(sk);
3923 
3924 	if (pg_vec)
3925 		free_pg_vec(pg_vec, order, req->tp_block_nr);
3926 out:
3927 	return err;
3928 }
3929 
3930 static int packet_mmap(struct file *file, struct socket *sock,
3931 		struct vm_area_struct *vma)
3932 {
3933 	struct sock *sk = sock->sk;
3934 	struct packet_sock *po = pkt_sk(sk);
3935 	unsigned long size, expected_size;
3936 	struct packet_ring_buffer *rb;
3937 	unsigned long start;
3938 	int err = -EINVAL;
3939 	int i;
3940 
3941 	if (vma->vm_pgoff)
3942 		return -EINVAL;
3943 
3944 	mutex_lock(&po->pg_vec_lock);
3945 
3946 	expected_size = 0;
3947 	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3948 		if (rb->pg_vec) {
3949 			expected_size += rb->pg_vec_len
3950 						* rb->pg_vec_pages
3951 						* PAGE_SIZE;
3952 		}
3953 	}
3954 
3955 	if (expected_size == 0)
3956 		goto out;
3957 
3958 	size = vma->vm_end - vma->vm_start;
3959 	if (size != expected_size)
3960 		goto out;
3961 
3962 	start = vma->vm_start;
3963 	for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3964 		if (rb->pg_vec == NULL)
3965 			continue;
3966 
3967 		for (i = 0; i < rb->pg_vec_len; i++) {
3968 			struct page *page;
3969 			void *kaddr = rb->pg_vec[i].buffer;
3970 			int pg_num;
3971 
3972 			for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3973 				page = pgv_to_page(kaddr);
3974 				err = vm_insert_page(vma, start, page);
3975 				if (unlikely(err))
3976 					goto out;
3977 				start += PAGE_SIZE;
3978 				kaddr += PAGE_SIZE;
3979 			}
3980 		}
3981 	}
3982 
3983 	atomic_inc(&po->mapped);
3984 	vma->vm_ops = &packet_mmap_ops;
3985 	err = 0;
3986 
3987 out:
3988 	mutex_unlock(&po->pg_vec_lock);
3989 	return err;
3990 }
3991 
3992 static const struct proto_ops packet_ops_spkt = {
3993 	.family =	PF_PACKET,
3994 	.owner =	THIS_MODULE,
3995 	.release =	packet_release,
3996 	.bind =		packet_bind_spkt,
3997 	.connect =	sock_no_connect,
3998 	.socketpair =	sock_no_socketpair,
3999 	.accept =	sock_no_accept,
4000 	.getname =	packet_getname_spkt,
4001 	.poll =		datagram_poll,
4002 	.ioctl =	packet_ioctl,
4003 	.listen =	sock_no_listen,
4004 	.shutdown =	sock_no_shutdown,
4005 	.setsockopt =	sock_no_setsockopt,
4006 	.getsockopt =	sock_no_getsockopt,
4007 	.sendmsg =	packet_sendmsg_spkt,
4008 	.recvmsg =	packet_recvmsg,
4009 	.mmap =		sock_no_mmap,
4010 	.sendpage =	sock_no_sendpage,
4011 };
4012 
4013 static const struct proto_ops packet_ops = {
4014 	.family =	PF_PACKET,
4015 	.owner =	THIS_MODULE,
4016 	.release =	packet_release,
4017 	.bind =		packet_bind,
4018 	.connect =	sock_no_connect,
4019 	.socketpair =	sock_no_socketpair,
4020 	.accept =	sock_no_accept,
4021 	.getname =	packet_getname,
4022 	.poll =		packet_poll,
4023 	.ioctl =	packet_ioctl,
4024 	.listen =	sock_no_listen,
4025 	.shutdown =	sock_no_shutdown,
4026 	.setsockopt =	packet_setsockopt,
4027 	.getsockopt =	packet_getsockopt,
4028 	.sendmsg =	packet_sendmsg,
4029 	.recvmsg =	packet_recvmsg,
4030 	.mmap =		packet_mmap,
4031 	.sendpage =	sock_no_sendpage,
4032 };
4033 
4034 static const struct net_proto_family packet_family_ops = {
4035 	.family =	PF_PACKET,
4036 	.create =	packet_create,
4037 	.owner	=	THIS_MODULE,
4038 };
4039 
4040 static struct notifier_block packet_netdev_notifier = {
4041 	.notifier_call =	packet_notifier,
4042 };
4043 
4044 #ifdef CONFIG_PROC_FS
4045 
4046 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4047 	__acquires(RCU)
4048 {
4049 	struct net *net = seq_file_net(seq);
4050 
4051 	rcu_read_lock();
4052 	return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4053 }
4054 
4055 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4056 {
4057 	struct net *net = seq_file_net(seq);
4058 	return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4059 }
4060 
4061 static void packet_seq_stop(struct seq_file *seq, void *v)
4062 	__releases(RCU)
4063 {
4064 	rcu_read_unlock();
4065 }
4066 
4067 static int packet_seq_show(struct seq_file *seq, void *v)
4068 {
4069 	if (v == SEQ_START_TOKEN)
4070 		seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
4071 	else {
4072 		struct sock *s = sk_entry(v);
4073 		const struct packet_sock *po = pkt_sk(s);
4074 
4075 		seq_printf(seq,
4076 			   "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4077 			   s,
4078 			   atomic_read(&s->sk_refcnt),
4079 			   s->sk_type,
4080 			   ntohs(po->num),
4081 			   po->ifindex,
4082 			   po->running,
4083 			   atomic_read(&s->sk_rmem_alloc),
4084 			   from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4085 			   sock_i_ino(s));
4086 	}
4087 
4088 	return 0;
4089 }
4090 
4091 static const struct seq_operations packet_seq_ops = {
4092 	.start	= packet_seq_start,
4093 	.next	= packet_seq_next,
4094 	.stop	= packet_seq_stop,
4095 	.show	= packet_seq_show,
4096 };
4097 
4098 static int packet_seq_open(struct inode *inode, struct file *file)
4099 {
4100 	return seq_open_net(inode, file, &packet_seq_ops,
4101 			    sizeof(struct seq_net_private));
4102 }
4103 
4104 static const struct file_operations packet_seq_fops = {
4105 	.owner		= THIS_MODULE,
4106 	.open		= packet_seq_open,
4107 	.read		= seq_read,
4108 	.llseek		= seq_lseek,
4109 	.release	= seq_release_net,
4110 };
4111 
4112 #endif
4113 
4114 static int __net_init packet_net_init(struct net *net)
4115 {
4116 	mutex_init(&net->packet.sklist_lock);
4117 	INIT_HLIST_HEAD(&net->packet.sklist);
4118 
4119 	if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4120 		return -ENOMEM;
4121 
4122 	return 0;
4123 }
4124 
4125 static void __net_exit packet_net_exit(struct net *net)
4126 {
4127 	remove_proc_entry("packet", net->proc_net);
4128 }
4129 
4130 static struct pernet_operations packet_net_ops = {
4131 	.init = packet_net_init,
4132 	.exit = packet_net_exit,
4133 };
4134 
4135 
4136 static void __exit packet_exit(void)
4137 {
4138 	unregister_netdevice_notifier(&packet_netdev_notifier);
4139 	unregister_pernet_subsys(&packet_net_ops);
4140 	sock_unregister(PF_PACKET);
4141 	proto_unregister(&packet_proto);
4142 }
4143 
4144 static int __init packet_init(void)
4145 {
4146 	int rc = proto_register(&packet_proto, 0);
4147 
4148 	if (rc != 0)
4149 		goto out;
4150 
4151 	sock_register(&packet_family_ops);
4152 	register_pernet_subsys(&packet_net_ops);
4153 	register_netdevice_notifier(&packet_netdev_notifier);
4154 out:
4155 	return rc;
4156 }
4157 
4158 module_init(packet_init);
4159 module_exit(packet_exit);
4160 MODULE_LICENSE("GPL");
4161 MODULE_ALIAS_NETPROTO(PF_PACKET);
4162